Microwave and RF Vacuum Electronic Power Sources

Do you design and build vacuum electron devices, or work with the systems that
use them? Quickly develop a solid understanding of how these devices work with
this authoritative guide, written by an author with over 50 years of experience in
the field. Rigorous in its approach, it focuses on the theory and design of com-
mercially significant types of gridded, linear-​beam, crossed-​field, and fast-​wave
tubes. Essential components such as waveguides, resonators, slow-​wave structures,
electron guns, beams, magnets, and collectors are also covered, as well as the inte-
gration and reliable operation of devices in microwave and RF systems. Complex
mathematical analysis is kept to a minimum, and Mathcad worksheets supporting
the book online aid understanding of key concepts and connect the theory with
practice. Including coverage of primary sources and current research trends, this is
essential reading for researchers, practitioners, and graduate students working on
vacuum electron devices.

Richard G.  Carter is a Professor Emeritus in the Department of Engineering at

Lancaster University, and a Fellow of the IET.
ii

‘Vacuum electronic devices are marvels of technology that achieve amazing performance
advantages by a sophisticated combination of a vast diversity of physics, materials, compo-
nents, and engineering principles. This book finds a worthy value-​add role in the pantheon of
great vacuum electronics (VE) literature by masterfully marrying a comprehensive treatment
with carefully crafted choices of appropriate—​but not excessive—​depth.  Thus it informs,
provides rigor, but is extremely accessible to many levels of expertise. The background treat-
ment of electromagnetics is especially noteworthy—​with its mapping of structure proper-
ties to equivalent circuit models—​since the start for understanding VE is the science and
design principles of electromagnetic fields in waveguides and resonators. The coverage of
VE devices and critical materials is current and thus state-​of-​the-​art, which is important
since research discovery advances keep pushing the limits of the performance capabilities of
modern VE. I wholeheartedly recommend this book to VE design engineers, and especially
anyone who is new to this exciting field, as well as experts who are seeking new ways to teach
its fundamental principles.’

John H. Booske, University of Wisconsin-​Madison

‘In spite of significant progress with solid state power amplifiers during the last decades,
large zones in parameter space remain inaccessible by this technology and will continue to be
the realm of vacuum electronics. Vacuum electronics however seems to have the undeserved
reputation of being less “modern” and thus often is given less attention in University classes.
The book Microwave and RF Vacuum Electronic Power Sources responds to the urgent need
to keep and build expertise in this field. Carter accomplishes the difficult task to present the
matter well-​structured didactically and in sufficient depth to make this an excellent textbook,
both for self-​education and to accompany lectures. At the same time and thanks to its com-
pleteness, accomplished experts would certainly want it in their bookshelves as a modern
reference.’

Erk Jensen, Head of RF group, CERN

‘This book is a tour de force on the subject of high-​power vacuum electronic devices and a
“must have” for anyone in the field. Professor Carter is a leading authority on the subject
and has spent his career in equal parts developing the technology and teaching it in clear
and intuitive style. The book contains a comprehensive review of the principles and conven-
tions of all types of vacuum tubes with detailed explanations and examples of the most
common varieties. Part history lesson, part “how to” guide this book will be as useful to
newcomers entering the field as it will be to the most experienced practitioners. The Mathcad
Worksheets available online will be invaluable for those new to the subject and a starting off
point for further development.’

Robert Rimmer, JLab
‘Combining pedagogy, science and practice, Professor Carter brings up step by step deeper
in the detailed understanding and design methods of all types of RF & Microwave Electron
Devices.
A precious book for students, designers and system users who will acquire practical theory
based on most advanced analytical formulas and supported by a huge list of references, as
well as Mathcad worksheets.  Enjoy this journey in electromagnetics and vacuum.’

Philippe Thouvenin, Thales –​Microwave & Imaging

Sub-​Systems, Technical Director –​Microwave Sub-​Systems
‘Vacuum Electronics is a dynamic and exciting field and Vacuum Electron Devices in par-
ticular remain vital building blocks in numerous critical electromagnetic systems. The new
iii

book by Professor Richard Carter provides a solid mathematical and technological founda-
tion for students in the field and a large knowledge base for seasoned professionals. The
book includes relativistic formulations of many basic working equations, online Mathcad
worksheets for instructional and practical use, and device performance characteristics help-
ful to systems engineers and other users.’

Richard True, L3 Technologies (Retired)

‘In high-​power and high-​frequency systems for telecommunication, radar, remote sensing,
broadcasting, processing of materials and scientific applications, vacuum electronic devices
(tubes) are the only radiofrequency (RF), microwave, and millimeter-​wave sources. This
excellent and comprehensive monograph focuses on the theory, design, state-​of-​the-​art and
current research trends of commercially important types of gridded tubes (triodes, tetrodes,
inductive output tubes), linear electron beam tubes (single-​, multi-​and sheet beam klystrons,
travelling wave tubes), crossed field tubes (magnetrons, crossed-​field amplifiers), and fast-​
wave microwave devices (gyro-​oscillators and amplifiers, free electron masers). It is devoted
to researchers, practitioners and graduate students working in the fields of R&D and appli-
cation of vacuum electron devices. After the introduction of essential tube components such
as electron guns, electron beam guiding systems, waveguide and slow-​wave structures, reso-
nators, output couplers, dielectric windows, and electron collectors, beam-​wave interaction
and the different types of microwave and RF vacuum electronic power sources are discussed
in detail. Well-​selected case studies for the various types of tubes and Mathcad worksheets
supporting the book online (can be downloaded from the publisher’s website) aid under-
standing of key concepts and link theory with practice. Concluding chapters on emission
and breakdown phenomena, different types of magnets and system integration may help to
achieve reliable operation of microwave and RF systems.’

Manfred Thumm, Karlsruhe Institute of Technology (KIT)

‘This book is the author’s life-​long contribution to the microwave tubes’ community; it is
a comprehensive and systematic book including passive components and all kinds of RF
vacuum devices with the analysis of cathodes, electron optics, reliability, and the system
integration. This book gives both clear physical imaging, deep understanding, the author’s
new insight of the high frequency structure and beam-​wave interaction as well as concise
physical models and worksheet for engineering design with both circuit and field method
using modern computer facility.’

Jinjun Feng, Tsinghua University

‘Richard Carter’s fifty years of internationally leading expertise is distilled in this essential
reading for everyone working on high power microwave and RF sources. Researchers, gradu-
ate students and practitioners will find this monograph indispensable.
Not only does it provide an in-​depth understanding of the many subtleties of the subject
that will satisfy the experienced practitioner, its careful construction and clarity of expres-
sion offers the student entering this field an excellent introduction that they will continue to
value and retain on their shelves as their own expertise develops.
To achieve a worthwhile understanding of this field some use of mathematics is essential
and this text very thoughtfully employs the necessary mathematics to support the theoretical
discussions in ways that are accessible to science and engineering graduates.
The coverage includes the most significant families of vacuum electronic microwave and
RF power oscillators and amplifiers that both those working on research and development
of the sources, or needing to use these sources in applications are most likely to encounter.
This comprehensive list includes slow-​wave sources, gridded tubes, triodes, tetrodes, induc-
tive output tubes (IOTs), klystrons, travelling-​wave tubes (TWTs), magnetrons, crossed-​field
iv

amplifiers as well as fast-​wave sources such as electron cyclotron masers, gyrotrons, cyclo-
tron auto-​resonance masers, gyro-​amplifers, peniotrons, ubitrons and free electron lasers.
The treatments of the slow-​wave devices tend to be more extensive than some of the fast-​
wave devices. For example although electron cyclotron masers and gyrotrons are covered in
some detail the author intentionally leaves a detailed treatment of free electron lasers to the
referencing of some existing publications. This balance of the coverage is well-​justified, as
it reflects the fact that the majority of the commercial applications of microwave and RF
vacuum electronic power sources presently use, and are likely to continue to use, slow-​wave
devices. Consequently the students, researchers and practitioners in this field tend to be well-​
motivated to study and understand the range of sources that are covered in this book.
This is a “stand-​alone” book since it does not assume that the reader is familiar with all of
the specialized components from which vacuum electronic microwave and RF power sources
are constructed. The main components such as waveguides, resonators, slow-​wave structures,
electron guns, beams, magnets and collectors are expertly covered in dedicated sections that
add extra value. The coverage of multipactor discharges and system integration will be found
to be especially useful to those interested in applying these sources.
This text is highly recommended for graduate students, practising scientists and engineers
and for libraries.’

Alan D. R. Phelps, FRSE (Prof), University of Strathclyde

‘This book by Richard Carter is a comprehensive, complete, and excellent exploration of
the microwave and RF vacuum electronic power source field. The content of the book has
been refined and enhanced by invited lectures given in many relevant international institutes.
The subject matter is not only important in its own right but also provides a splendid insight
to comprehending the wider literature in the field. The mathematical analyses are kept to a
minimum which encourages the reader to master the principles and establish the basis for
advanced developments. The reader’s understanding is enhanced by the accessibility to the
worksheets used for the calculations throughout the book. Unusually if not uniquely, the
author uses Mathcad as a powerful and flexible tool for modelling problems in vacuum elec-
tronics. These models are made available to the readers.
In summary, this work is an indispensable reference handbook in the field of microwaves
and RF power, both for beginners and experienced professionals alike.’

Stephen Myers, Executive Chair ADAM SA, Former CERN Director of

Accelerators and Technology
‘Destined to be a classic reference in the field of RF vacuum electronic device analysis and
design. From density modulated power grid and induction output tubes, to slow wave linear
and crossed field devices and fast wave high power gyrotrons, and their key components,
every subject is treated with technical rigor interspersed with sage practical advice obtained
over the course of Prof. Carter’s distinguished career in the field. Notable features include
the strong use of transmission line analysis for device modelling providing insight not always
available with standard field analysis or computer simulation, a detailed list of references at
the end of every chapter for advanced reading, and online worksheets for invaluable hands-​
on experience with the physics behind the equations bringing home the necessary design
trades that always need to be made –​an excellent learning tool. Enthusiastically recommend
for both the intern/​new hire as well as the seasoned RF vacuum electronics engineer and
researcher.
Professor Carter says in the Introduction that the book can be considered to be fifty years
in the making. I would add that it was well worth the wait –​a monumental achievement.’

Carter M. Armstrong, L3 Technologies

v

The Cambridge RF and Microwave Engineering Series

Series Editor

Steve C. Cripps, Distinguished Research Professor, Cardiff University

Editorial Advisory Board

James F. Buckwalter, UCSB
Jenshan Lin, University of Florida
John Wood, Maxim Integrated Products

Peter Aaen, Jaime Plá and John Wood, Modeling and Characterization of RF and Microwave
Power FETs
Dominique Schreurs, Máirtín O’Droma, Anthony A. Goacher and Michael Gadringer
(Eds), RF Amplifier Behavioral Modeling
Fan Yang and Yahya Rahmat-​Samii, Electromagnetic Band Gap Structures in Antenna
Engineering
Enrico Rubiola, Phase Noise and Frequency Stability in Oscillators
Earl McCune, Practical Digital Wireless Signals
Stepan Lucyszyn, Advanced RF MEMS
Patrick Roblin, Nonlinear RF Circuits and the Large-​Signal Network Analyzer
Matthias Rudolph, Christian Fager and David E. Root (Eds), Nonlinear Transistor Model
Parameter Extraction Techniques
John L. B. Walker (Ed.), Handbook of RF and Microwave Solid-​State Power Amplifiers
Anh-​Vu H. Pham, Morgan J. Chen and Kunia Aihara, LCP for Microwave Packages and
Modules
Sorin Voinigescu, High-​Frequency Integrated Circuits
Richard Collier, Transmission Lines
Valeria Teppati, Andrea Ferrero and Mohamed Sayed (Eds), Modern RF and Microwave
Measurement Techniques
Nuno Borges Carvalho and Dominique Schreurs, Microwave and Wireless Measurement
Techniques
David E. Root, Jason Horn, Jan Verspecht and Mihai Marcu, X-​Parameters
Earl McCune, Dynamic Power Supply Transmitters
Hossein Hashemi and Sanjay Raman (Eds), Silicon mm-​ Wave Power Amplifiers and
Transmitters
T. Mitch Wallis and Pavel Kabos, Measurement Techniques for Radio Frequency
Nanoelectronics
Giovanni Ghione and Marco Pirola, Microwave Electronics
Isar Mostafanezhad, Olga Boric-Lubecke and Jenshan Lin (Eds), Medical and Biological
Microwave Sensors
Richard G. Carter, Microwave and RF Vacuum Electronic Power Sources

Forthcoming
José Carlos Pedro, David Root, Jianjun Xu and Luis Cotimos Nunes, Nonlinear Circuit
Simulation and Modeling
Michael Schröter and Martin Claus, Carbon Nanotube Electronics for Analog Radio-​
Frequency Applications
vi
vi

Microwave and RF Vacuum

Electronic Power Sources
R IC H ARD G .   CARTER
Lancaster University
vi

University Printing House, Cambridge CB2 8BS, United Kingdom

One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–​321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi –​110025, India
79 Anson Road, #06-​04/​06, Singapore 079906

Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9780521198622
DOI: 10.1017/​9780511979231
© Cambridge University Press 2018
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2018
Printed in the United Kingdom by Clays, St Ives plc
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Names: Carter, R. G. (Richard Geoffrey), author.
Title: Microwave and RF vacuum electronic power sources / Richard G. Carter.
Description: Cambridge, United Kingdom ; New York, NY :
Cambridge University Press, 2018. |
Series: The Cambridge RF and microwave engineering series |
Includes bibliographical references and index.
Identifiers: LCCN 2017049391 | ISBN 9780521198622 (hardback)
Subjects: LCSH: Vacuum-tube circuits. | Microwave circuits. | Radio circuits. |
Microwave devices–Power supply.
Classification: LCC TK7871.72.C37 2018 | DDC 621.381/32–dc23
LC record available at https://lccn.loc.gov/2017049391
ISBN 978-​0-​521-​19862-​2 Hardback
Additional resources for this publication at www.cambridge.org/9780521198622.
Cambridge University Press has no responsibility for the persistence or accuracy
of URLs for external or third-​party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
ix

To my dear wife Awena
without whose loving support
this book could never have been written
x
xi

Contents

Preface page xxiii

Principal Roman Symbols xxvii
Principal Greek Symbols xxxi
List of Abbreviations xxxiii

1 Overview 1
1.1 Introduction 1
1.2 Vacuum Electronic and Solid-​State Technologies 2
1.3 Principles of Operation 4
1.3.1 Geometry 5
1.3.2 Electron Dynamics 5
1.3.3 Modulation of the Electron Current 7
1.3.4 Amplification, Gain, and Linearity 7
1.3.5 Power Output and Efficiency 9
1.3.6 Bandwidth 11
1.3.7 The Electromagnetic Structure 11
1.3.8 Coupled-​Mode Theory 12
1.3.9 Classification of Vacuum Tubes 17
1.4 Applications of Vacuum Tubes 17
1.5 The Statement of Requirements 19
1.6 Signals and Noise 20
1.6.1 Noise 22
1.6.2 Analogue Modulation 23
1.6.3 Digital Modulation 26
1.6.4 Multiplexing 28
1.7 Engineering Design 32
1.7.1 Dimensionless Parameters and Scaling 33
1.7.2 Modelling 36
xi

xii Contents

2 Waveguides 41
2.1 Introduction 41
2.2 Waveguide Theory 41
2.2.1 The Transverse Electric and Magnetic Mode 45
2.2.2 Transverse Electric Modes 46
2.2.3 Transverse Magnetic Modes 49
2.3 Practical Waveguides 52
2.3.1 Coaxial Lines 52
2.3.2 Rectangular Waveguides 58
2.3.3 Ridged Waveguides 60
2.3.4 Circular Waveguides 64
2.3.5 Summary of Waveguide Impedances 65
2.4 Waveguide Discontinuities 65
2.4.1 Height Step in a Rectangular Waveguide 70
2.4.2 Capacitive Iris in a Rectangular Waveguide 72
2.4.3 Inductive Iris in a Rectangular Waveguide 74
2.5 Matching Techniques 75
2.5.1 Stub Matching 75
2.5.2 Broad-​band Matching 76
2.5.3 Stepped Impedance Transformers 79
2.6 Coupling without Change of Mode 79
2.7 Coupling with Change of Mode 81
2.8 Windows 83
2.8.1 Windows in Coaxial Lines 86
2.8.2 Windows in Rectangular Waveguide 87

3 Resonators 93

3.1 Introduction 93
3.2 Resonant Circuits 94
3.2.1 The Properties of Resonant Circuits 95
3.2.2 External Loading of Resonant Circuits 98
3.2.3 Excitation of Resonant Circuits 99
3.2.4 Coupled Resonators 102
3.3 Pill-​Box Cavity Resonators 104
3.3.1 Effects of Surface Roughness 107
3.3.2 Higher-Order Modes 109
3.4 Rectangular Cavity Resonators 110
3.5 Re-​entrant Cavities 111
3.5.1 Method of Moments Model of Re-​entrant
Cavities 112
3.5.2 Fujisawa’s Model of Re-​entrant Cavities 115
3.5.3 The Interaction Field 118
3.5.4 Practical Re-​entrant Cavities 120
xi

Contents xiii

3.6 External Coupling to Cavities 120

3.6.1 Loop Coupling 122
3.6.2 Iris Coupling 124
3.7 Measurement of Cavity Parameters 129

4 Slow-​Wave Structures 134

4.1 Introduction 134
4.1.1 Uniform Slow-​Wave Structures 134
4.1.2 Periodic Slow-​Wave Structures 139
4.1.3 Space Harmonics 141
4.2 Planar Slow-​Wave Structures 145
4.2.1 Ladder Line 148
4.2.2 Meander Line 149
4.2.3 Inter-digital Line 152
4.3 Helix Slow-​Wave Structures 152
4.3.1 The Sheath Helix Model 154
4.3.2 Dispersion Shaping 158
4.3.3 Tape Helix Slow-​Wave Structures 159
4.3.4 Equivalent Circuit of Helix Slow-​Wave Structures 162
4.3.5 Couplers and Attenuators 165
4.4 Ring-​Bar and Ring-​Loop Structures 166
4.5 Waveguide Slow-​Wave Structures 169
4.5.1 The Folded Waveguide Structure 169
4.5.2 Helical Waveguides 173
4.6 Coupled-​Cavity Slow-​Wave Structures 173
4.6.1 Space Harmonic Structures 174
4.6.2 The Cloverleaf Structure 176
4.6.3 The Centipede Structure 179
4.6.4 Termination of Coupled-​Cavity Slow-​Wave Structures 180
4.7 Measurement of the Properties of Slow-​Wave Structures 181
4.7.1 Measurements on Coupled-​Cavity Slow-​Wave Structures 181
4.7.2 Measurements on Helix Slow-​Wave Structures 183

5 Thermionic Diodes 190

5.1 Introduction 190
5.1.1 Dimensional Analysis of Thermionic Diodes 191
5.1.2 Current Limitation 193
5.2 The Planar Space-​Charge Limited Diode 194
5.3 The Planar Diode Including the Effects of Thermal Velocities 196
5.3.1 Electron Flow between the Potential Minimum and the Anode 197
5.3.2 Electron Flow between the Cathode and the Potential
Minimum 199
5.3.3 Numerical Evaluation 201
vxi

xiv Contents

5.4 The Planar Diode Including the Effects of Relativity 204

5.5 The Cylindrical Space-​Charge Limited Diode 206
5.6 The Spherical Space-​Charge Limited Diode 209
5.7 Transit-​Time Effects in a Planar Diode 211
5.8 Injection of Electrons into a Planar Diode 214
5.9 Diodes with Two-​Dimensional Flow of Current 219

6 Triodes and Tetrodes 224

6.1 Introduction 224
6.2 Electrostatic Models of Triodes 225
6.3 Penetration Factor in a Planar Triode 228
6.3.1 A Triode with Uniform Electric Field on the Cathode 230
6.3.2 A Triode with Non-​Uniform Electric Field
on the Cathode 232
6.3.3 Calculation of Penetration Factors Using
Numerical Methods 234
6.4 Static Characteristics of Triodes 234
6.4.1 Grid Current 238
6.4.2 Triodes with Island Formation 240
6.5 Electrostatic Models of Tetrodes 242
6.6 Penetration Factors in Tetrodes 243
6.7 Static Characteristics of Tetrodes 245
6.7.1 Grid Currents in Tetrodes 246
6.7.2 Effect of Space-​Charge between the Screen Grid
and the Anode 246

7 Linear Electron Beams 250

7.1 Introduction 250
7.2 Cylindrical Electron Beams 250
7.3 Electron Optics without Space-​Charge 252
7.3.1 The Paraxial Ray Equation of Electrostatic
Electron Optics 253
7.3.2 Thin Electrostatic Lenses 254
7.3.3 Busch’s Theorem 256
7.3.4 Magnetostatic Electron Optics without
Space Charge 257
7.3.5 Thin Magnetic Lenses 258
7.4 Electron Optics with Space-​Charge 259
7.4.1 Solenoid Focusing 260
7.4.2 Scalloping 262
7.4.3 Beam Stiffness 263
7.5 Beam Spreading 266
7.5.1 The Universal Beam-​Spreading Curve 266
7.5.2 Spreading of Rotating Beams 268
xv

Contents xv

7.6 Periodic Focusing 269

7.6.1 Periodic Permanent Magnet (PPM) Focusing 270
7.6.2 Practical PPM Focusing Systems 276
7.6.3 Periodic Electrostatic Focusing 277
7.7 Other Forms of Linear Electron Beam 280
7.7.1 Sheet Electron Beams 281
7.7.2 Annular Electron Beams 281
7.8 Imperfections in Electron Beams 281
7.8.1 Thermal Velocities 282
7.8.2 Trapped Ions 282

8 Electron Flow in Crossed Fields 287

8.1 Introduction 287
8.2 Crossed-​Field Electron Flow in Planar Geometry 288
8.2.1 Electron Motion without Space-​Charge 289
8.2.2 Injected Beam with Space-​Charge 292
8.3 The Planar Magnetron Diode 294
8.3.1 The Diode Is Conducting 294
8.3.2 The Diode Is Cut-Off 298
8.4 Crossed-​Field Electron Flow in Cylindrical Geometry 301
8.4.1 Electron Motion without Space-​Charge 301
8.4.2 Injected Beam with Space-​Charge 303
8.5 The Cylindrical Magnetron Diode 305
8.6 Experimental Behaviour of Magnetron Diodes 308
8.7 The Magnetron Problem 314

9 Electron Guns 317

9.1 Introduction 317
9.2 The Pierce Electron Gun 318
9.2.1 Electrostatic Theory of the Pierce Electron Gun 318
9.2.2 The Focus Electrode and Anode Nose 322
9.2.3 Improved Model of the Anode Lens 324
9.2.4 The Effects of Thermal Velocities 328
9.2.5 Electrostatic Design of a Pierce Electron Gun 331
9.3 Magnetic Field Design for a Pierce Electron Gun 333
9.3.1 Solenoid Focusing 333
9.3.2 PPM Focusing 338
9.4 Other Pierce Guns 340
9.4.1 Guns for Sheet Beams 340
9.4.2 Guns for Hollow Beams 340
9.5 Beam Control Electrodes 341
9.5.1 Modulating Anode and Control Focus Electrode 341
9.5.2 Intercepting Control Grid 342
9.5.3 Non-​Intercepting Control Grid 343
xvi

xvi Contents

9.6 Crossed-​Field Electron Guns 344

9.6.1 Kino Gun 345
9.6.2 Magnetron Injection Gun 346

10 Electron Collectors and Cooling 352

10.1 Introduction 352

10.2 Linear Beam Tube Collectors 353
10.3 Collector Depression 358
10.3.1 Multi-​Element Depressed Collectors 360
10.3.2 Non-​Ideal Multi-​Element Depressed Collectors 363
10.4 Design of Multi-​Element Depressed Collectors 366
10.4.1 Suppression of Secondary Electrons 369
10.4.2 Reconditioning the Spent Electron Beam 370
10.5 Cooling 371
10.5.1 Conduction Cooling 371
10.5.2 Air Cooling 371
10.5.3 Liquid Cooling 372
10.5.4 Vapour Phase Cooling 372

11 Beam-​Wave Interaction 375

11.1 Introduction 375

11.2 Ballistic Theory of Interaction with a Gap 376
11.2.1 Beam Modulation by a Gridded Gap 376
11.2.2 Ballistic Electron Bunching 378
11.2.3 Beam Loading of a Gridded Gap 380
11.2.4 Beam Modulation by a Gridless Gap 382
11.3 Space Charge Waves on Linear Electron Beams 385
11.3.1 Effect of Radial Boundaries 388
11.3.2 Induced Current 390
11.3.3 Transmission Line Representation of Space-​Charge
Waves 394
11.3.4 Space-​Charge Waves on Non-​Ideal Electron Beams 396
11.3.5 Higher-Order Modes 396
11.3.6 Cyclotron Waves 397
11.4 Space-​Charge Wave Theory of the Interaction between
a Beam and a Gap 397
11.4.1 Current Induced in a Gap by Space-​Charge Waves 399
11.4.2 Beam Loading of a Gridless Gap 400
11.4.3 Beam Interaction with a Passive Gridless Gap 402
11.5 Continuous Interaction with a Slow-​Wave Structure 404
11.6 Discrete Interaction with a Slow-​Wave Structure 411
11.7 Backward-​Wave Interactions 415
xvi

Contents xvii

11.8 Large-​Signal Modelling of Beam-​Wave Interactions 417

11.8.1 Large-​Signal Model of the Beam–​Gap Interaction 419
11.8.2 Modulation of an Electron Beam by a Gap 422
11.8.3 Current Induced in a Passive Gap by a
Modulated Beam 423
11.8.4 Power Transfer in an Output Gap 426

12 Gridded Tubes 433

12.1 Introduction 433

12.1.1 Gridded Tube Amplifiers 433
12.1.2 Classes of Amplification 434
12.2 Triodes 439
12.2.1 Case Study: The ML-​5681 Triode 441
12.3 Tetrodes 443
12.3.1 Case Study: The RS 2058 Tetrode 444
12.4 Design of Triodes and Tetrodes 448
12.5 Design of Triode and Tetrode Amplifiers 451
12.5.1 Practical Details 454
12.6 Inductive Output Tubes (IOTs) 455
12.6.1 Bunch Formation 456
12.6.2 Space-​Charge Debunching 457
12.6.3 Power Transfer in the Output Gap 458
12.6.4 IOT Collectors 459
12.6.5 Case Study: The 116LS IOT 460

13 Klystrons 466
13.1 Introduction 466
13.2 Small-​Signal Klystron Theory 468
13.2.1 Input Cavity 469
13.2.2 Idler Cavities 470
13.2.3 Output Cavity 471
13.2.4 Simplified Small-​Signal Model 474
13.2.5 Overall Performance 475
13.3 Large-​Signal Behaviour of Klystrons 475
13.3.1 Klystron Sections 477
13.3.2 Initial Bunching Section 478
13.3.3 Final Bunching Section 479
13.3.4 Output Section 481
13.3.5 Output Coupling 483
13.3.6 Theoretical Limits to Efficiency 485
13.3.7 Electron Collection 488
13.3.8 Terminal Characteristics 490
xvii

xviii Contents

13.4 Klystron Design 491

13.4.1 Broad-​band Klystrons 493
13.4.2 High-​Efficiency Klystrons 494
13.4.3 Case Study: The SLAC 5045 Klystron 497
13.5 Other Klystrons 499
13.5.1 Multiple-​Beam Klystrons 500
13.5.2 Sheet Beam Klystrons 502

14 Travelling-​Wave Tubes 507

14.1 Introduction 507
14.1.1 Helix and Helix-​Derived TWTs 508
14.1.2 Coupled-​Cavity TWTs 509
14.1.3 Energy Conversion in a TWT 510
14.2 Small-​Signal Theory 513
14.2.1 Small-​Signal Theory of Helix TWTs 514
14.2.2 Small-​Signal Theory of Coupled-​Cavity TWTs 520
14.3 Large Signal Effects 521
14.3.1 Dimensionless Parameters 526
14.3.2 Dependence of Efficiency on Normalised
Parameters 528
14.3.3 Dependence of Efficiency on the Operating Point 531
14.3.4 Effect of a Sever 534
14.3.5 Harmonics 537
14.3.6 Transfer Characteristics 538
14.3.7 Tapers 542
14.3.8 Stability 543
14.4 TWT Design 544
14.4.1 Case Study: An Octave Bandwidth Helix TWT 545
14.4.2 Millimetre-​Wave Helix TWTs 549
14.4.3 High Efficiency Helix TWTs 550
14.4.4 Ultra-​Broad-​band TWTs 553
14.4.5 Coupled-​Cavity TWTs 555
14.4.6 Hybrid Tubes 556

15 Magnetrons 565
15.1 Introduction 565
15.2 Basic Principles 566
15.2.1 Hub Model 569
15.2.2 Interaction Field 569
15.2.3 Threshold Condition for Oscillations 573
15.2.4 Electronic Efficiency 580
15.3 Magnetron Anodes 580
15.3.1 Strapped Anodes 587
15.3.2 Rising Sun Anodes 590
xi

Contents xix

15.3.3 Coaxial Anodes 593

15.3.4 Long Anodes 593
15.4 Magnetron Properties 596
15.4.1 The Performance Chart 596
15.4.2 Frequency Pushing 597
15.4.3 Frequency Pulling 598
15.4.4 Spectrum 599
15.4.5 Mode Selection, Priming, and Locking 601
15.5 Particle in Cell Magnetron Models 603
15.6 Simple Magnetron Models 604
15.6.1 The Space-​Charge Hub 606
15.6.2 The Rigid Spoke Model 608
15.6.3 Guiding Centre Orbits 609
15.6.4 Electron Trajectory Model 612
15.6.5 Calculation of the Output Power 614
15.6.6 The Rieke Diagram 616
15.6.7 Frequency Pushing 616
15.7 Magnetron Design 619
15.7.1 Dimensionless Parameters 619
15.7.2 Design Parameters 620
15.7.3 Design Case Study 620
15.7.4 Other Considerations 624

16 Crossed-​Field Amplifiers 629

16.1 Introduction 629
16.1.1 Emitting Cathode CFAs 631
16.1.2 Injected Beam CFAs 633
16.2 CFA Construction 633
16.2.1 Slow-​Wave Structures 634
16.2.2 Cathodes 634
16.2.3 Cathode-​Driven CFAs 635
16.3 Basic Principles 636
16.4 CFA Characteristics 639
16.4.1 Performance Chart 639
16.4.2 Modulation 641
16.4.3 Transfer Characteristics 642
16.4.4 Signal Growth and Anode Dissipation 645
16.5 Theoretical Models of CFAs 645
16.5.1 PIC Codes 645
16.5.2 Soliton Theory 646
16.5.3 Guiding Centre Theory 646
16.5.4 Non-​Linear Fluid Mechanics 647
16.5.5 Rigid Spoke Model 647
16.6 CFA Design 654
x

xx Contents

17 Fast-​Wave Devices 659

17.1 Introduction 659
17.2 Electron Cyclotron Masers 660
17.2.1 Small-​Signal Theory of ECM Interactions 664
17.3 Gyrotron Oscillators 668
17.3.1 Large-​Signal Interaction Model 671
17.3.2 Case Study: A 140 GHz, 1 MW CW, Gyrotron 677
17.3.3 Design of Gyrotron Oscillators 679
17.3.4 Cyclotron Auto-​Resonance Masers 683
17.3.5 Tuneable Gyrotrons 683
17.4 Gyro-​Amplifiers 684
17.5 Peniotrons 685
17.6 Ubitrons (Free Electron Lasers) 686

18 Emission and Breakdown Phenomena 694

18.1 Introduction 694
18.2 Emission of Electrons from Metal Surfaces 694
18.2.1 Thermionic Emission 696
18.2.2 Field-​Enhanced Emission (the Schottky Effect) 697
18.2.3 Field Emission 698
18.2.4 Photo-​Electric Emission 699
18.3 Secondary Electron Emission 700
18.3.1 Modelling Secondary Electron Emission 704
18.4 X-​ray Emission 706
18.5 Thermionic Cathodes 708
18.5.1 Metal Emitters 709
18.5.2 Oxide Cathodes 710
18.5.3 Dispenser Cathodes 710
18.6 Field Emission Cathodes 712
18.7 Voltage Breakdown 713
18.7.1 Voltage Breakdown in Vacuum 713
18.7.2 Voltage Breakdown in Gases 715
18.7.3 Voltage Breakdown on Insulators 720
18.8 Multipactor Discharges 721
18.8.1 Theory of Multipactor Discharges between Parallel Plates 723
18.8.2 Multipactor Discharges between Coaxial Cylinders 727
18.8.3 Multipactor Discharges in Crossed Fields 727
18.8.4 Modelling Multipactor Discharges 727

19 Magnets 735
19.1 Introduction 735
19.2 Review of Theory 736
19.2.1 Ferromagnetism 737
19.2.2 Conduction of Magnetic Flux by Soft Magnetic Materials 738
xxi

Contents xxi

19.3 Magnetic Circuits 739

19.3.1 Circuits Including Permanent Magnets 742
19.4 Magnetic Materials 744
19.4.1 Soft Magnetic Materials 744
19.4.2 Permanent Magnet Materials 745
19.5 Coil Dominated Magnets 747
19.5.1 Arrays of Coils 750
19.5.2 Solenoids 751
19.6 Iron Dominated Magnets 753
19.7 Permanent Magnet Design 754
19.7.1 Permanent Magnets for Magnetrons and CFAs 754
19.7.2 Permanent Magnets for Linear Beam Tubes 756
19.7.3 Periodic Permanent Magnet (PPM) Systems 758

20 System Integration 765

20.1 Introduction 765
20.2 DC Power Supplies 765
20.2.1 High-​Voltage Switches 766
20.2.2 Load Impedance 767
20.2.3 Electric Power Converters 768
20.3 Pulse Modulators 771
20.3.1 Active-​Switch Modulator with a Resistive Load 772
20.3.2 Active-​Switch Modulator with a Biased Diode Load 774
20.3.3 Line-​Type Modulators 777
20.4 RF Systems 779
20.5 Cooling System 780
20.6 Control System 781
20.6.1 Interlocks 781
20.6.2 Tube Protection 782
20.7 Care of Tubes 782
20.8 Safety 783
20.9 Reliability 784
20.10 Conclusion 785

Appendix Mathcad Worksheets (available online

at www.cambridge.org/9780521198622) 789
Index 791
xxi
xxii

Preface

It could be said that it has taken me fifty years to write this book. I entered the field
of vacuum electronics as a PhD student in 1965 at the University College of North
Wales, Bangor. That date coincided with the end of what I think of as the heroic
period of the field. The years from 1939 to 1965 saw contributions to the sub-
ject from people whose names are well-​known to those who followed them. Their
achievements in the exploration of a new technology were remarkable, especially
when one considers that nearly everything had to be done by theory and experi-
ment. Computer modelling was in its infancy. The period since 1965 has seen an
explosion in the power of computers and what can be achieved with them. I took
my first steps in computer programming on the EDSAC II vacuum tube computer
at Cambridge University using a very primitive autocode. The point has now been
reached where computing power is not only sufficient to model devices in great
detail, but also to permit automatic optimisation of them. During my career com-
puter modelling of tubes became increasingly important, and that was the main
focus of my own work, and of those who worked with me.
From Bangor I  moved to what was then the English Electric Valve Company
(now Teledyne e2v) in Chelmsford. The culture of the company at that period has
been well-​described by Norman Pond in his book The Tube Guys. There was no
separation between development and production so that we were always conscious
that the tubes we were developing had to be made by the people alongside whom
we worked. I was fortunate to be seconded to a development project at the Services
Electronics Research Laboratory, where, under the able guidance of Robin King,
I undertook the detailed design of most parts of a C-​band coupled-​cavity TWT.
Most of the design calculations were carried out on an electro-​mechanical calcu-
lator on my desk. But I did take over some FORTRAN programs for modelling
TWTs, which had been written by Harold Curnow. Those programs had to be run
via an unreliable landline, using a chattering teletype, on a mainframe computer 50
miles away.
After three years I  moved to the newly formed Engineering Department at
Lancaster University. The Department was based on the idea that engineering is
a single discipline, which fitted the multidisciplinary nature of vacuum electronics
rather well. I was encouraged to continue work in the field when I was offered an
appointment as a consultant by EEV. That remained an important relationship
for nearly thirty years, keeping me in touch with the problems that mattered to my
vxi

xxiv Preface

colleagues in industry. The relationship led to the creation of an innovative under-

graduate course in Physical Electronic Engineering, sponsored by EEV. I developed
and taught a module on vacuum electronics for this course. The students undertook
industrial placements with the company and my role in setting up, and monitoring,
those activities gave me contacts in nearly every department there. The creation
of the course was ably supported by Dr John Whitmore (at that time Manager of
the Light Conversion Division). John also encouraged me to start to build my own
research group, which became a major focus of my life until I retired in 2009. In
1989 I was invited by the company to prepare a set of twenty videotaped lectures to
be used for training purposes. The project was the brainchild of Maurice Esterson
and I  was given invaluable practical help by F.J. (Peter) Weaver. Those lectures
could be said to be the true foundation of the present book.
Over time my research group flourished. I  was especially grateful to Maurice
Esterson and Brian Cooper of EEV who supported research projects, and made a
point of involving me in the technical problems of their staff. Over the period until
I retired nearly fifty research students, assistants, and fellows worked with me, and
I owe a debt of gratitude to them for the way they forced me to try to think clearly
about their work in order to guide them. There were collaborative projects with
CEERI in India and with BVERI, and a number of other institutions, in China. It
is a matter of pride to me that several of the people who were members of the group
have since occupied senior positions in laboratories overseas. The material which
has gone on to form the content of this book was honed through lectures I was
invited to give at around sixteen institutions, in nine countries, on three continents.
These included contributions to five CERN Accelerator Schools to which I  was
introduced by colleagues at the Daresbury Laboratory. I also contributed lectures,
with colleagues at the universities of Lancaster and Strathclyde, to a joint MSc in
High Power Radiofrequency Engineering.
The idea of trying to collect my thoughts on vacuum electronics into the form
of a book has been around for a long time. But I was unable to find time to devote
to the task until after I retired. Throughout my career my interest has been in the
modelling and conceptual design of tubes. In teaching I have always tried to keep
close to a physical understanding of how tubes operate. And I have been interested
in developing simple models as an aid to understanding in the conceptual phase
of design, following the example of the pioneers in the field. I  have been influ-
enced, especially, by the books by Pierce, Slater, Gittins, Beck, and Gewartowski
and Watson. The literature in the field is now too extensive for any one person to
master. The references cited in this book are the sources which I consulted while
writing it. My aim was to try to bring this material together to give a comprehensive
and consistent account of the subject. The topics I have chosen to include are those
I have found useful. I hope that readers will find that this material is not only useful
in its own right but also provides a route to understanding the wider literature in
the field.
Microwave engineering has often been a happy hunting ground for those who
love advanced mathematics. Although it is not possible to discuss the topics in this
xv

Preface xxv

book without some use of mathematics I have tried to make the treatment acces-
sible. I  am impressed by the thought that engineers generally try to find simple
ways of modelling complex problems. We use ‘j-​notation’ to allow us to apply DC
circuit theory to AC problems. And we use transmission line, and equivalent circuit
methods to model problems involving electromagnetic fields. This way of thinking
about vacuum electronics remains as useful today as it was when I entered the field
fifty years ago. It provides insight into the physics of devices which cannot readily
be gained in any other way. I believe it to be an essential foundation for training new
entrants to the field whose knowledge will, otherwise, come largely from computer
simulation using general purpose electromagnetics software and particle-​in-​cell
codes. After writing in FORTRAN for many years I became an enthusiast for the
possibilities of modelling with software such as Mathcad. Although this approach
is not suitable for all problems it does provide a remarkably powerful and flex-
ible way of modelling problems in vacuum electronics. Over time I have developed
Mathcad models for many aspects of tube design and I have used these extensively
whilst writing this book. The models which I used are listed in the Appendix, and
can be downloaded from the publisher’s website. I would like to encourage my read-
ers to use those models to explore aspects of tube design. The use of Mathcad
makes it easy for the user to modify the models, and to add diagnostic features, as
part of their own learning process.
I have been grateful to many people for help, advice, and inspiration at different
stages of my career. There are too many to name them all, but I particularly want to
mention: Rob Newton, my PhD supervisor, from whom I learnt the value of space-​
charge wave theory and coupled-​mode analysis; Dudley Perring from ESTEC who
introduced me to the world of space TWTs, becoming simultaneously a project
sponsor and a research student until his untimely death; and Rodney Vaughan,
from whom I learnt that good mathematical models of tubes were best kept close
to the underlying physics. I would also like to acknowledge the colleagues at EEV,
TMD technologies, Thales at Ulm and Vélizy, CERN, DESY, and the Daresbury
Laboratory who generously gave time to advise on research projects. There have
been many fruitful discussions with my colleagues at Lancaster, especially Graeme
Burt, Amos Dexter, Claudio Paoloni, and Robin Tucker.
During the writing of this book I  have been most grateful to colleagues who
reviewed chapters in draft and made many helpful comments and suggestions. In
alphabetical order: Carter Armstrong, Heinz Bohlen, Bruce Carlsten, Jim Dayton,
Amos Dexter, Tushar Ghosh, Dan Goebel, Michel Grézaud, Neil Marks, John
Ospechuk, Alan Phelps, Alan Sangster, Manfred Thumm, Mike Tracy, Todd
Treado, and Dick True. I hope they will forgive me if I have not always taken their
advice. The book, as it stands, is entirely my responsibility. I would like to thank
Lancaster University for appointing me as an Emeritus Professor when I retired.
That gave me continued access to the information and computer resources of the
University. I am also grateful to Julie Lancashire and her colleagues at Cambridge
University Press who have supported me consistently in a project that has taken
much longer than either they or I expected.
xvi

xxvi Preface

Finally, I owe an immense debt of gratitude to my wife Dr Awena Carter, whose

name means ‘inspiration’ in Welsh, for her unstinting love and support throughout
the whole of my career. She has accompanied me to conferences, and on visits to
institutions, and is known to many of my colleagues in the field. She has borne
patiently (most of the time) with the long hours I have spent working on the book,
and she brought her considerable expertise in writing to bear by reading the whole
text in draft and pointing out where its clarity could be improved.
xxivi

Principal Roman Symbols

A Magnetic vector potential

A Amplification (voltage gain)
Area
a Waveguide width
Tunnel radius
Grid spacing
B Magnetic flux density = µH
BB Brillouin field
B0 Characteristic field
B Susceptance
Bandwidth
Bb Beam loading susceptance
b Waveguide height
Beam radius
Pierce velocity parameter
C Capacitance
Pierce gain parameter
c Velocity of light = 0.2998 × 109 m s−1
D Electric displacement = εE
d Dielectric radius
Distance
E Electric field
EF Fermi energy
Em Space harmonic amplitude
e Elementary charge = 1.602 × 10−19 C
Base of natural logarithms = 2.718
f Frequency
G Conductance = 1 R
Power gain
GdB Gain in decibels = 10 lnG
Gb Beam loading conductance
G0 = I 0 V0 Beam conductance
g Gap length
xxivi

xxviii Principal Roman Symbols

H Magnetic field
h Height
Planck’s constant = 6.626 × 10−34 J s
I Electric current
Bessel function
Intensity
Ia Anode current
Ig Grid current
I0 DC current
In Harmonic current n = 1, 2, 3
In (z) Modified Bessel function of the first kind
I± Space-​charge wave RF currents
i Current
ig Induced current in a gap
J Current density
Bessel function
Jn (z) Bessel function of the first kind
JC Child-​Langmuir current density
JS Schottky current density
K Bessel function
Coupling factor
Perveance
Ratio of flux linkages = 1 − 1 m2
Kn (z) Modified Bessel function of the second kind
k Boltzmann constant = 1.380 × 1023 J K−1
Coupling factor
Free space propagation factor = ω c
L Length
Self inductance
Cold loss
l Length
M Mutual inductance
Gap coupling factor
M Magneto-​motive force
m Mass
m Normalised magnetic flux density = B BB
m0 Rest mass of the electron = 9.109 × 10 −31 kg
n Harmonic number
P Power
Pressure
Permeance (= 1/​reluctance)
Pin Input power
Pout Output power
xi

Principal Roman Symbols xxix

p Pitch
Plasma frequency reduction factor
Q Quality factor of a resonator
QE External Q
QL Loaded Q
QU Unloaded Q
QC Pierce space charge parameter
q Induced current reduction factor
R Resistance
Spherical radius
Normalised radius
R Reluctance
Rc Cavity shunt impedance
Rs Surface resistance
r Radius
Resistance
ra Anode radius
rc Cathode radius
s Shield radius
T Absolute temperature (K)
Kinetic energy
t Thickness
u Velocity
u0 DC velocity
V Voltage
Volume
Va Anode voltage
Accelerating voltage
Vb Breakdown voltage
Vc Carrier amplitude
Collector potential with respect to the tube body
Vf Final voltage
Vg Grid voltage
Gap voltage
VH Hull cut-​off voltage
Vi Initial voltage
Vin Input voltage
Vout Output voltage
VR Rest energy of the electron = m0c 2 e = 511 keV
Vs Source voltage
VT Threshold (Hartree) voltage
Volt equivalent of temperature = kT e
V0 DC voltage
Characteristic voltage
x

xxx Principal Roman Symbols

V1 AC voltage
Beam kinetic voltage
V± Space-​charge wave kinetic voltages
v Voltage
Velocity
vg = d ω d β Group velocity
vp = ω β Phase velocity
W Stored energy
w Width
X Reactance
Y Admittance = 1 Z
Ye Electronic admittance
Z Impedance = 1 Y
Normalised position
ZC Coupling impedance
Zc Characteristic impedance
Ze Electronic impedance
Zg Waveguide impedance
ZP Pierce impedance
ZS Surface impedance
ZT Total impedance
Zw = µ ε Wave impedance
Z0 = µ 0 ε 0 Wave impedance of free space
xxi

Principal Greek Symbols

α Loss parameter
Time constant
β Propagation constant
βC = ωC c Waveguide cut-​off
βc = ω c u0 Cyclotron propagation constant
βe = ω u0 Electronic propagation constant
β g = βC2 − ω 2 c 2 Waveguide propagation constant
β L = ω L u0 Larmor propagation constant
β m = β0 + 2 mπ p Space harmonic propagation constant
β mn = ω mn c Waveguide cut-​off for the ( m, n ) mode
β p = ω p u0 Plasma propagation constant
βq = ω q u0 Reduced plasma propagation constant
βs Synchronous propagation constant
βw = ε r ω c Propagation constant in a dielectric
β0 = ω v p Propagation constant in material
Slow-​wave structure propagation constant
β0 = 2π L Periodic electrostatic focusing
Periodic permanent magnet focusing
β± Space-​charge wave propagation constants
γ Radial propagation constant = β2 − ω 2 c 2
Relativistic factor = 1 1 − u 2 c 2
δ Skin depth
Secondary electron emission coefficient
ε Permittivity
εr = ε ε0 Relative permittivity
ε0 Primary electric constant = 8.854 × 10–​12 F m−​1
η Efficiency
ηA Power added efficiency
ηc Circuit efficiency
ηe Electronic efficiency
ηh Harmonic efficiency
ηO Overall efficiency
ηrf RF efficiency
xxixi

xxxii Principal Greek Symbols

θ Angle
λ Wavelength
λC Waveguide cut-​off wavelength
λ g = 2π β g Guide wavelength
μ Permeability
Gap coupling factor
Amplification factor
Attenuation coefficient of X-​rays
µc Radial gap coupling factor
µd Axial gap coupling factor
µr Relative permeability
µ0 Primary magnetic constant 4 π × 10 −7 H m −1
ρ Charge density
Resistivity
Voltage reflection coefficient
Density
σ Conductivity
τ Time constant
Transit time
Radial propagation constant
Φ Magnetic flux
ϕ Angle
Phase
Work function
ψ Angle
Helix pitch angle
ω Angular frequency = 2 π f
ωC Waveguide cut-​off frequency
eB Cyclotron frequency
ωc =
m0
ωH Hull cut-​off frequency
ω L = ωc 2 Larmor frequency
ω mn Cut-​off frequency of waveguide mode ( m, n )
eρ Electron plasma frequency
ωp =
ε 0 m0
ω q = pω p Reduced plasma frequency
ωs Synchronous frequency
Synchronous angular velocity
ω0 Resonant frequency
xxixi

Abbreviations

AC Alternating current
AM Amplitude modulation
BWO Backward wave oscillator
CFA Crossed-​field amplifier
CVD Chemical vapour deposition
CW Continuous wave
DC Direct current
ECM Electron cyclotron maser
ECM Electronic counter-​measures
FEL Free electron laser
GTO Gate turn-​off thyristor
IGBT Insulated gate bipolar transistor
IOT Inductive output tube
PM Phase modulation
PPM Periodic permanent magnet
RF Radio frequency
RMS Root mean square
TWT Travelling-wave tube
xvxi
1

1 Overview

1.1 Introduction

Sources of power in the radiofrequency, microwave, and millimetre wave regions

of the electromagnetic spectrum are essential components in a wide range of sys-
tems for telecommunications, broadcasting, remote sensing, and processing of
materials1. Current research is extending the frequency range into the sub-​millimetre
region. These sources employ either vacuum electronic, or solid state, technologies.
At the higher frequencies and power levels, vacuum electronic devices (tubes2) are
the only sources available (see Section 1.2).
The purpose of this book is to provide a comprehensive introduction to the the-
ory and conceptual design of the types of tubes which are of continuing impor­
tance. The design and operation of vacuum tubes requires knowledge and skills
drawn both from electrical and electronic engineering and from physics. The treat-
ment here is intended to be accessible to those whose training has been in either
discipline. The use of advanced mathematics has been avoided as far as possible
with considerable emphasis on the use of simple numerical methods. The book is
designed to be a reference text for designers and users of vacuum tubes, and a text-
book for people who are new to the field.
This is a mature field in which much has been published since its first beginnings
in 1904 [1]. The sources cited here are those which have been used as the basis for
the book. They are believed to comprise most of the most important sources in the
field and the reader is invited to consult them for further information. References
have been included to sources that provide additional information on many of the
topics, but no attempt has been made to provide a comprehensive bibliography of
the subject and longer lists of references are to be found elsewhere [2]. A further aim
of this book is to provide the reader with the background necessary to read with
understanding other papers in the field.
In any book it is necessary to make choices about what should be included and
what excluded. The subjects covered are those the author believes to be important
for the practical business of designing and using vacuum tubes. Because the focus

1
For convenience, the term RF is used throughout this book to refer to all frequencies in the range 30 kHz to
300 GHz.
2
We shall call these devices (vacuum) tubes because that terminology is familiar and concise.
2

2 Overview

is on theory and conceptual design very little has been said about the technology
of tube construction, which is well treated elsewhere [3, 4]. Similarly, little has been
said specifically about vacuum tubes for use at sub-​millimetre wavelengths since
they employ the same principles as those at lower frequencies, and most of the chal-
lenges are in the technology of their construction.
This chapter provides an overview of the subject of the book. The next section
compares vacuum tubes with solid-​state devices to show how the technologies are
complementary. Section 1.3 provides an overview of the physical principles on
which vacuum tubes are based and definitions of the key terms used to describe
their performance. A tube converts the DC power in the initial electron stream into
RF output power by interaction with electromagnetic structures. These structures
support standing (resonant), or travelling, electromagnetic waves. Coupled-​mode
theory is introduced as a valuable conceptual tool for understanding the interac-
tions between electron streams and travelling electromagnetic waves. The section
concludes with a classification of the principal types of vacuum tube based on
the preceding discussion. The principal applications of vacuum tubes are reviewed
in Section 1.4 together with some of the factors which govern the availability of
tubes of different types. That leads, in Section 1.5 to consideration of the com-
munication between the designers and users of tubes in the form of a Statement of
Requirements. This statement specifies both the electrical performance required,
and the factors which constrain the design. Many tubes are required to amplify
modulated carrier signals whose properties are not normally familiar to people
whose primary discipline is Physics. An introduction to analogue and digital modu-
lation, noise and multiplexing is provided in Section 1.6. Finally, Section 1.7 consid-
ers some of the principles of the engineering design of tubes including dimensional
analysis and scaling, and the use of computer modelling.
The remainder of the book comprises four sections:
i) Chapters 2–​4 deal with the properties of the passive electromagnetic compo-
nents employed in vacuum tubes.
ii) Chapters 5–​11 are concerned with aspects of electron dynamics in vacuum that
are employed in tubes in a variety of ways. Chapter 10 also includes a discus-
sion of methods of cooling.
iii) Chapters 12–​17 show how the fundamental principles introduced earlier in the
book are applied to specific types of tube, and their conceptual design.
iv) Chapters 18–​20 provide an introduction to some technological issues which are
common to most types of tube and their successful use in systems.

1.2 Vacuum Electronic and Solid-​State Technologies

The characteristic size of any active RF device is determined by the distance travelled
by the charge carriers in one RF cycle. Thus the size of a device decreases as the fre-
quency increases and as the velocity of the charge carriers decreases. The application
3

Vacuum Electronic and Solid-State Technologies 3

of any electronic technology is limited, ultimately, by temperature since the oper-

ation of a device generates heat. Hence, the maximum continuous, or average, RF
power which can be generated by a single device is determined by the power dissi-
pation within it. At low power levels semiconductor devices have the advantages of
small size, and low voltage operation. But these become disadvantages at high power
levels because the current passing through the device is high. Thus there are large
conduction losses, generating heat, within a small volume. They can be reduced, to
some extent, by operating the device as a switch rather than in its active mode.
Transistors can currently deliver around 100 W of continuous power or 1 kW of
pulsed power at frequencies in the region of 1 GHz [5–​7]. Further developments,
including transistors using diamond as a semiconducting material, may increase
the power to several kW [8, 9]. High power amplifiers can be made by operating
many transistors in parallel but the penalties of increased complexity set limits to
this. The power combining can take place in space, as in active phased-​array radar,
where average powers up to tens of kilowatts can be achieved [10]. Alternatively,
a power combining network may be used as in the 190 kW, 352 MHz, amplifier at
Synchrotron SOLEIL [11]. This amplifier combines the power from four 50 kW
towers each containing twenty 2.5 kW units. A unit comprises eight 315 W modules
each having a pair of transistors operated in push-​pull. This arrangement means
that the loss of power from the failure of an individual transistor is small, and the
degradation of the amplifier from this cause is gradual. However, the power output
is still almost an order of magnitude less than that achieved by vacuum tubes at the
same frequency, and the frequency is at the lower end of the range for which vac-
uum tubes have been developed. The use of low DC voltages reduces problems of
reliability caused by voltage breakdown. But the consequent need for high currents
leads to DC losses in the connecting bus-​bars. A review of high power semicon-
ductor RF power technology is given in [12].
Vacuum tubes, in contrast, operate at high voltages and low currents. The charge
carriers have a much higher velocity than in a semiconducting material, and they
are not subject to energy loss through collisions as they pass through it. Thus the
active volume can be large with only RF losses within it. In many cases, the greater
part of the heat generated is dissipated on electrodes which are separate from the
active region, and whose size can be increased to reduce the power density. There
is a common misconception that vacuum tubes are fragile, short-​lived, unreliable,
and inefficient. In fact, modern vacuum devices are mechanically robust and able
to survive short-​term electrical overloads without damage. They have demonstrated
outstanding reliability and lifetimes in the very demanding environment of space.
Vacuum tube amplifiers can have maximum conversion efficiencies of up to 70%,
and 90% has been achieved in oscillators. However, the most reliable performance
of any technology is achieved by allowing generous design margins and operating
well within their limits. Vacuum tubes can be operated in parallel, in the same way
as semiconductor devices, but the number of parallel devices is usually quite small
[13, 14]. An exception to this is in phased-​array radar where the output from a larger
number of microwave power modules (see Section 20.1) is combined in space [15].
4

4 Overview

Figure 1.1:  Comparison of the performance of single vacuum electronic and solid state RF
power devices
(copyright 1999, IEEE, reproduced, with permission, from [16]).

Figure 1.1 shows a comparison between the performances of individual vacuum

and semiconductor RF power devices [16]. There have been advances in both tech-
nologies since that figure was drawn, but the general picture remains valid. A more
recent, but less detailed, figure is to be found in [17]. For further discussion of the
relative merits of vacuum tube and solid-​state RF power amplifiers see [18, 19]

1.3 Principles of Operation

The tubes which are the subject of this book are all power amplifiers and oscillators.
For our present purpose it is sufficient to regard oscillators as amplifiers in which
the RF input is provided by internal feedback. All vacuum tube amplifiers can be
understood in terms of the block diagram shown in Figure 1.2. A uniform stream
of electrons is emitted into the vacuum from the electron source and modulated
by the RF input voltage3. Radiofrequency energy is extracted from the modulated
stream of electrons, and their remaining energy is dissipated as heat on a collecting
electrode. The arrows show the direction of motion of the electrons; the conven-
tional current is, of course, in the opposite direction. The functions of the blocks
may be combined in various ways in different devices but the overall process is
essentially the same. The basic RF performance of an amplifier is defined in terms
of its gain, output power, efficiency, and instantaneous or tuneable bandwidth.

3
The word ‘stream’ is used here as being more general than the term electron beam which is used for specific
types of tube.
5

Principles of Operation 5

Figure 1.2:  Block diagram of a vacuum tube amplifier.

1.3.1 Geometry
The majority of practical vacuum tubes have geometries which are cylindrically
symmetrical, or close approximations to it. The flow of the DC current is radial or
axial, and driven by a static electric field in the same direction (see Chapters 5, 8,
and 9).

1.3.2 Electron Dynamics

The voltages employed in vacuum tubes are high enough, in many cases, for the
velocities of the electrons to be at least mildly relativistic. The kinetic energy of a
relativistic electron is [20]:

T = mc 2 − m0 c 2 , (1.1)

where m0 c 2 is the rest energy of the electron and the relativistic mass is

m0
m= = γ m0 , (1.2)
1 − u2 c2
where u is the velocity of the electron and c the velocity of light. If an electron starts
from rest at the cathode then its velocity at a point where the potential is V, relative
to the cathode,4 is found by using the principle of conservation of energy:

 1 
eV = m0 c 2  − 1 . (1.3)
 1− u c
2 2

This equation can be rearranged as
1
 1 2
u = c 1 −  , (1.4)
( )
2
 
1 + V V 
 
 R 

4
In this book voltages are referred to the cathode unless otherwise stated. In practice the tube body is nor-
mally at earth potential and the cathode potential is negative with respect to it.
6

6 Overview

where VR = m0 c 2 e = 511 kV is the rest energy of the electron expressed in electron

volts. When V VR  1 the fraction can be expanded by the Binomial Theorem to
give the approximate expression:

2V
u=c . (1.5)
VR
Figure  1.3 shows a comparison between the velocities calculated using the exact
and approximate equations. The error in the approximate velocity is 1% at 7 kV so
that the exact formula should be used for voltages higher than this.
The force acting on an electron is equal to the rate of change of momentum
d du dm
F= ( mu ) = m + u . (1.6)
dt dt dt
If the force acts in the direction of the motion of the electrons then

m0
F= ⋅ a = γ 3 m0 a, (1.7)
(1 − u 2 c2 )
32

where a is the acceleration and γ 3 m0 is sometimes called the longitudinal mass.

When the force acts at right angles to the direction of motion then
m0
F= ⋅ a = γ m0 a, (1.8)
(1 − u 2 c2 )
12

where γ m0 is the transverse mass. If the longitudinal velocity is approximately con-

stant and much greater than the transverse velocity then it is possible to use the clas-
sical equations of motion with relativistic corrections to the mass of the electron.
The DC electron current in a tube may be collimated by a magnetic field which
is either parallel to the direction of the DC current or perpendicular to it (see
Chapters 7 and 8). Tubes in which the field is parallel to the current are described as
‘Type O’ and those in which it is perpendicular as ‘Type M’. The current may also
be collimated by a static electric field but this is rather rare.

Figure 1.3:  Comparison between exact and approximate electron velocities as a function of

the accelerating potential.
7

Principles of Operation 7

1.3.3 Modulation of the Electron Current

For the present we will assume that the input RF voltage is purely sinusoidal with
amplitude V1 and frequency ω . Practical input signals are discussed in Section 1.6.
It can be shown that the three possible methods of modulation of the current by
the RF input voltage are [21]:
• Emission density modulation in which the current emitted from the source is varied.
• Deflection and concentration modulation in which the electrons are deflected
sideways.
• Transit-​time modulation in which the electron velocity is varied.
The effect of any of these methods of modulation, or combinations of them, is to
produce a bunched stream of electrons whose current varies with time at the frequency
of the input signal. Because the process is non-​linear the time variation of the current
is not sinusoidal but can be represented by a Fourier series. The amplitudes and phases
of the harmonic components depend upon the amplitude of the input voltage, and on
the distance from the source. The modulation may be either in the same direction as
the DC current flow, or normal to it. Thus the current at some point z can be written
∞
I ( z, t ) = I 0 + ∑ I n (Vin , z ) exp ( jnωt ) , (1.9)
n =1

where I 0 is the current in the unmodulated stream, I n (Vin , z ) are complex ampli-
tudes, and modulation in the axial direction has been assumed. The DC and time-​
varying parts of the current may be in different directions in space. We note that the
real current cannot be negative. Its maximum value, relative to I 0 , is determined by
the process of modulation, or by the maximum current which can be drawn from
the source. The ratio I0/I0 cannot exceed 2.0 (see sections 11.8.4 and 13.3.4).

1.3.4 Amplification, Gain, and Linearity

The RF output voltage is obtained by passing the modulated stream through a region
at an effective position z2 where energy is removed from the electron bunches by an
RF electric field. This can be represented by an impedance Zn (Vin ) which depends
on frequency and on the magnitude of the input signal. Thus the output voltage is

 ∞ 
Vout (t ) = Re  ∑ I n (Vin , z2 ) Zn (Vin ) exp ( jnωt ) , (1.10)
 n = 1 
where, for simplicity we assume that the characteristic impedances of the input and
output waveguides are the same. The impedance Zn is effectively zero above some
value of the harmonic number n determined by the nature of the output section.
Thus the number of harmonics at which there is appreciable output power is small
and, in some cases, limited to the fundamental. If the output power is to be radi-
ated it may be necessary to filter out the harmonics to comply with the regulations
8

8 Overview

determining the bandwidth available to the system, as specified by international

agreement and national regulations [22, 23]. The transfer characteristic of the ampli­
fier at the frequency ω is then

Vout ( ωt ) = A (Vin ) cos ( ωt + Φ (Vin )) , (1.11)

where A (Vin ) is the AM/​AM (amplitude modulation) characteristic which is usually

plotted on decibel scales and Φ (Vin ) is the AM/​PM (phase modulation) character-
istic of the amplifier.
The details of the transfer characteristics depend upon the type of amplifier but
many of them show the same features. Figure 1.4 shows, as an example, a typical
AM/​AM curve of a travelling-​wave tube (TWT). The output power is proportional
to the input power at low drive levels (slope 1 dB/​dB) but reaches saturation as the
input is increased. In the figure the input and output powers have been normalised
to their values at saturation. The gain of the amplifier in decibels is given by

Pout
GdB = 10 log , (1.12)
Pin

where Pin and Pout are, respectively, the input and output power. Since we have
assumed that the input and output waveguides have the same characteristic imped-
ances we can write

 A (Vin ) 
GdB = 20 log  . (1.13)
 Vin 

The difference between the linear (small-​signal) and the saturated gain is the gain
compression which may be used as a measure of non-​linearity. It is common for a
tube to be described in terms of its saturated output power, but the output power
available under normal operating conditions may be less than this. For example the
tubes used in particle accelerators are normally operated ‘backed-​off’ from satu­
ration to provide a control margin for the operation of the accelerator. In a TWT
there is normally some output at second, and higher, harmonic frequencies. This
also depends on the input drive level as shown in Figure 1.4. At low drive levels the
second harmonic output is proportional to the square of the input power, giving
a slope of 2 dB/​dB. The maximum second harmonic power does not necessarily
occur at the drive level that saturates the fundamental. The intersection between
the projections of the linear parts of the fundamental and second harmonic curves,
known as the second-order intercept point, is a measure of the second-order distor-
tion of the amplifier.
The corresponding AM/​PM curve for a TWT is shown in Figure 1.5. The phase
is plotted relative to the phase at low drive levels, and the input power is normal-
ised to the input power at saturation. The phase of the output signal relative to the
input is constant at low drive levels but changes as the drive level is increased. Both
the amplitude and phase of the output depend on frequency, and on the operating
9

Principles of Operation 9

Figure 1.4:  Typical curves of fundamental and second harmonic output power of an RF

amplifier plotted against the input power, normalised to saturation.

Figure 1.5:  Typical curve of the phase of the output voltage of an RF amplifier against input
power normalised to saturation.

conditions of the amplifier, including the voltages applied and the external RF
matches. Thus any ripple in the voltages applied may result in amplitude and phase
modulation of the output signal. In an oscillator, voltage ripple may also produce
frequency modulation.

1.3.5 Power Output and Efficiency

An important consideration for many applications of tubes is the efficiency with
which the DC input power is converted into useful RF power. The principle of
conservation of energy requires that, in the steady state, the total input and output
powers must balance, that is

PRF in + PDC in = PRF out + PHeat . (1.14)

The DC input power includes the power in the electron stream, the cathode heater,
and electromagnets. If the tube has a depressed collector (see Section 10.3) the
10

10 Overview

DC power is reduced by the power recovered by the collector. When a comparison

is made between alternative tube types, or between vacuum tube and solid-​state
amplifiers, then the DC input power should be specified at the input to the power
supply to allow for losses in it. Any power required for cooling fans or pumps must
also be included. The RF output power comprises power at the fundamental fre-
quency and its harmonics. Heat is generated by the impact of electrons on the col-
lector and the tube body and by RF losses in the tube body, connecting waveguides
and windows.
A number of different definitions of efficiency are in use and it is important to
distinguish between them. The overall efficiency is defined here as the ratio of the
fundamental RF output power ( P2 ) to the total input power to the tube

P2
ηO = . (1.15)
PDC + PRF in

If the gain of the amplifier is high, the RF input power is much smaller than the DC
input power so that approximately

P2
ηO ≈ . (1.16)
PDC
The heater and electromagnet powers are typically much smaller than the stream power
in continuous wave (CW) tubes but they can be comparable with the stream power in
pulsed tubes. The power added efficiency, sometimes used when the gain is small, is

P2 − PRF in
ηA = . (1.17)
PDC
This efficiency is effectively identical to that in (1.16) if the gain is 20 dB or more.
The efficiency of tubes with depressed collectors is discussed in Section 10.3. The RF
efficiency is defined here as the ratio of the useful RF output power to the power input
to the electron stream (less any recovered by the collector) plus the RF input power.

P2
ηrf = . (1.18)
Pstream − Precovered + PRF in

The electronic efficiency is the efficiency with which power is transferred from the
electron stream to the RF electric field of the output circuit.

PRF out + Ploss

ηe = , (1.19)
Pstream
where Ploss is the total RF power loss in the output circuit. If the output circuit is
resonant so that only power at the fundamental frequency is included

P2 + P2, loss
ηe = , (1.20)
Pstream
1

Principles of Operation 11

where P2,loss is the RF loss at the fundamental frequency. Then the circuit efficiency
can be defined by

P2
ηc = . (1.21)
P2 + P2, loss

If, in addition, the RF input power is small enough to be neglected, and the tube
does not have a depressed collector then the RF efficiency is

ηrf = ηe ηc . (1.22)

1.3.6 Bandwidth
The AM/AM, and AM/PM, curves depend upon the frequency of the input signal.
Then the transfer function of the amplifier may be written H ( ω,Vin ). Figure 1.6 shows
a typical graph of the RF output power of an amplifier as a function of frequency.
This graph may be plotted at constant input power, or showing the saturated output
power at each frequency. The bandwidth is defined in terms of the frequencies at which
the power falls below the maximum by a specified amount (e.g. 1 dB bandwidth). This
may be expressed in absolute terms or as a percentage of the centre frequency. The
figure also shows that the output power may vary within the band and this can affect
the system in which the tube is employed. These ripples are normally greatest when the
tube is not saturated and are reduced at saturation by the effects of gain compression.
The custom of using a decibel scale for this graph can be misleading. It is important
to remember that a change of −1 dB represents a reduction in power by 20% and a
change of −3 dB a reduction of 50%. Similar graphs of efficiency, small-​signal gain,
and saturated gain against frequency can also be plotted. The bandwidth specified for
a tube may be to permit rapid changes in frequency (as in frequency-​agile radar), or to
enable it to be used with multiple and modulated signals (see Section 1.6).
In addition to power at harmonics of the input frequency the output of a tube
may also include non-​ harmonic power under some operating conditions (for
example during the rise and fall of the pulse during pulsed operation). These out-​
of-​band emissions are undesirable, and limits are usually specified to avoid interfer-
ence with other systems. For example it has been known for out-​of-​band emissions
from marine radar, and from industrial microwave ovens, to interfere with micro-
wave communication links. The tube also contributes to the noise in the system as
discussed in Section 1.6.1.

1.3.7 The Electromagnetic Structure

The electromagnetic structures used to modulate the electron stream, and to extract
power from it, can be divided into three groups:
• Fast-​wave structures in which the phase velocity of travelling waves is greater
than the velocity of light (Chapter 2).
12

12 Overview

Figure 1.6:  Typical graph of the output power of an RF amplifier plotted against frequency.

• Resonant (standing wave) structures in which the RF electric field profile is time-​
varying but fixed in space (Chapter 3).
• Slow-​wave structures which employ the field of a travelling electromagnetic wave
whose phase velocity is less than the velocity of light (Chapter 4).
The bandwidth of a tube using a resonant structure is typically restricted to a few
percent by the loaded Q of the resonance (see Section 3.2.2). If the higher-order
modes of the structure do not coincide with harmonics of the input frequency then
the harmonic output of the tube is small.
The bandwidths of tubes using travelling-​wave structures can be broad because
the phase velocity and the impedance presented to the electron current by the struc-
ture vary slowly with frequency. The interaction is distributed so that the bunch-
ing section and the output section of the tube are effectively combined. For useful
interaction to take place the mean electron velocity must be approximately syn-
chronous with the phase velocity of the wave. There may be interaction at harmon-
ics of the input frequency leading to appreciable output power at the second and
higher harmonics as shown in Figure 1.4.

1.3.8 Coupled-​Mode Theory
Useful insights into the properties of travelling-​wave structures and their inter-
actions with electron streams under small-​signal (linear) conditions are given by
coupled-​mode theory [24–​28]. For this purpose the small-​signal modulation of the
electron stream is described by a pair of normal modes whose impedances have
opposite signs. The properties of space-​charge waves on electron beams are dis-
cussed in detail in Chapter 11.
The properties of a wave propagating as exp j ( ωt − βz ) are the solutions to the
dispersion equation

D ( ω, β ) = 0, (1.23)
13

Principles of Operation 13

where the frequency (w) and the propagation constant ( β) are real if there is no
attenuation.5 The solutions of the dispersion equation can be plotted on a disper-
sion ( ω, β ) diagram. In this diagram the phase velocity of the wave is

ω
vp = (1.24)
β
and the group velocity, which is taken to be the velocity of propagation of energy, is

dω
vg = . (1.25)
dβ
The phase velocity is positive in the first quadrant of the ( ω, β ) diagram and nega-
tive in the second quadrant. The group velocity can be either positive or negative
depending upon the properties of the mode. We shall see in Chapter 4 that waves
having positive phase velocities and negative group velocities occur in periodic
structures.
Where two waves propagate in the same region of space they may interact to
produce coupled modes if their dispersion curves intersect as shown in Figure 1.7.
The dispersion equation for the coupled modes can then be written in the form

D1 ( ω, β ) D2 ( ω, β ) = κ ( ω, β ) , (1.26)

where κ ( ω, β ) represents the coupling between the modes. The coupled modes are
the solutions to this equation. In order to investigate the properties of cou­pling
under different conditions we will redefine ω and β so that the two uncoupled
modes intersect at the origin. This transformation does not change the group veloc­
ities of the modes.
The properties of the different possible kinds of mode coupling can be illustrated
by assuming that the uncoupled dispersion curves are straight lines so that

D ( ω, β ) = (ω − βvg ) = 0. (1.27)
The dispersion equation for the coupled modes is then

(ω − βv ) (ω − βv ) = κ
g1 g2 0 , (1.28)

where κ 0 is regarded as a constant close to the origin. This equation can be expanded
to give

ω 2 − (vg1 + vg 2 ) ωβ + vg1vg 2β2 − κ 0 = 0, (1.29)

which is quadratic in both ω and β. Solving for ω  gives

The symbol β used by engineers is employed in this book in place of the symbol k normally used by
5

physicists.
14

14 Overview

Figure 1.7:  Dispersion diagram for two modes of propagation.

ω=
1
2 {
(vg1 + vg 2 ) β ± (v g1 }
− vg 2 ) β2 + 4κ 0 (1.30)
2

and for β gives

β=
1
2vg1vg 2 {(
vg1 + vg 2 ) ω ± (v g1 }
− vg 2 ) ω 2 + 4vg1vg 2 κ 0 . (1.31)
2

The constant κ 0 is positive if the signs of the impedances of the uncoupled waves are
the same, and negative if they are opposite. Likewise the signs of the group velocities
of the waves may be the same, or opposite. Hence all possible combinations are cov-
ered by four cases, which can be explored using Worksheet 1.1. It can be shown that
the properties of the coupled modes can be revealed by considering the solutions of
(1.30) when β is real and the solutions of (1.31) when ω is real as follows [26, 29]:

Case A: The group velocities and the impedances have the same sign (vg1vg 2 > 0, κ 0 > 0 ).
When β is real the solutions of (1.30) are always two real values of ω and, when
ω is real, the solutions of (1.31) are two real values of β. These solutions represent
travelling waves with the modified dispersion diagram shown in Figure  1.8. In
Figures 1.8 to 1.11 the uncoupled modes are shown by dotted lines, the real parts
of the solutions by solid lines and the imaginary parts by dashed lines.

Case B: The group velocities have opposite signs and the impedances have the same
sign (vg1vg 2 < 0, κ 0 > 0 ).
When β is real the solutions of (1.30) are always two real values of ω but, when ω
is real and close to the origin, the solutions of (1.31) are complex conjugate pairs of
values of β, as shown in Figure 1.9(b). Physically, the complex values of β represent
travelling waves whose amplitudes decay exponentially with distance. For example
see Figure 4.28(a) where the forward and backward waves in a waveguide are cou-
pled to one another by periodic discontinuities to produce frequency stop-​bands in
which there are decaying (evanescent) waves.
15

Principles of Operation 15

Figure 1.8:  Coupling between two forward waves with impedances having the same sign:
(a) ω for real β , and (b) β for real ω.

Figure 1.9:  Coupling between a forward wave and a backward wave with impedances having
the same sign: (a) ω for real β , and (b) β for real ω.

Case C: The group velocities have the same sign and the impedances have opposite
signs (vg1vg 2 > 0, κ 0 < 0 ).
The solutions for both ω and β, close to the origin, are complex conjugate pairs,
as shown in Figure  1.10. It can be shown that the complex solution for β corre­
sponds to a convective instability in which the waves grow and decay exponentially
in space (see Figure 1.10(b)) [26, 29]. An example of this type of coupling is the
growth of waves in a travelling-​wave tube as shown in Figure 11.19.

Case D: The group velocities and the impedances have opposite signs
(v
g1vg 2 < 0, κ 0 < 0 ).
16

16 Overview

Figure 1.10:  Coupling between two forward waves with impedances having opposite
signs: (a) ω for real β , and (b) β for real ω.

When β is real and close to the origin the solutions of (1.30) are a complex con-
jugate pair of values of ω and, when ω is real, the solutions of (1.31) are always
two real values of β, as shown in Figure 1.11. The solution with complex ω is a
non-​convective, or absolute, instability in which the two solutions grow and decay
exponentially in time. The condition for this instability to exist is that there is at
least one value ω = ω s in the upper half of the complex plane for which there are
coincident roots in β s [29]. For the case considered here the condition for the exis­
tence of coincident roots is found from (1.31)

vg1vg 2 κ 0
ωs = ± 2 j . (1.32)
vg1 + vg 2

The condition is satisfied in this case when κ 0 = 0 and ω = 0 so that an absolute

instability exists for all negative values of κ 0. The corresponding values of β can be
found from (1.31):

vg1 − vg 2
βs = ω s . (1.33)
2 vg1vg 2

Thus β s is complex when ω s is complex. In general the start-​oscillation condition

can be found by finding the least magnitude of the coupling for which coincident
roots of β exist when ω is real (see Figures  17.5 and 17.6) [30]. If the coupling
is increased beyond this point both the frequency and the propagation constant
are complex (see Section 11.7). Examples of this kind of instability are backward-​
wave oscillations in a travelling-​wave tube (Section 11.7) and gyrotron oscillators
(Section 17.3).
17

Applications of Vacuum Tubes 17

Figure 1.11:  Coupling between forward and backward waves with impedances having
opposite signs: (a) ω for real β , and (b) β for real ω.

The properties of coupled modes remain the same in more general cases where
the uncoupled dispersion diagrams are not straight lines, κ is a function of ω and β,
and more than two modes are involved in the coupling. Thus a qualitative descrip-
tion of the properties of the coupled system can be derived from an examination
of the dispersion diagram of the uncoupled modes and the form of the dispersion
equation for the coupled system [29].

1.3.9 Classification of Vacuum Tubes

The preceding sections have reviewed the different ways in which the functional
blocks in Figure 1.2 can be realised, and that can be used as a basis for classifying
tubes. The discussion in this book is restricted to the types of tube which are of
continuing practical importance. A summary of their principal features is given in
Table 1.1. A few variations on the main types of tube exist, for example: triodes and
tetrodes with DC and RF current flow in the axial direction, and hybrid tubes in
which the bunching section resembles a klystron and the output section a travelling-​
wave tube. It is possible to think of the inductive output tube as a hybrid tube with
a triode bunching section and a klystron output section. Many other possible types
of tube can be envisaged [31–​33].

1.4 Applications of Vacuum Tubes

High-​power vacuum tubes have found applications in many fields as summarised

in Table 1.2 [17]. A comprehensive summary of the historical development of tubes
for the microwave and millimetre-​wave region of the spectrum can be found in
[34, 35]. For the history of power gridded tubes and their applications see [36].
18

18 Overview

Table 1.1:  Classification of the principal types of vacuum tube

Type Class DC RF current Magnetic Electro-​magnetic Bunching

(Chapter) current field structure

Triode and Gridded Radial Radial None Resonant Radial

tetrode (12)
Inductive output Gridded Axial Axial Axial Resonant Axial
tube (12)
Klystron (13) Linear- Axial Axial Axial Resonant Axial
beam (O)
Travelling-​wave Linear- Axial Axial Axial Slow-​wave Axial
tube (14) beam (O)
Magnetron (15) Crossed- Radial Tangential Axial Resonant Tangential
field (M)
Crossed-​field Crossed- Radial Tangential Axial Slow-​wave Tangential
amplifier (16) field (M)
Gyrotron (17) Fast-​wave Axial Tangential Axial Fast-​wave Tangential

In general, the development of high power vacuum tubes has been driven by
the requirements of particular applications. The number of tubes of each type
needed has not been great enough to allow the use of high volume manufacturing
methods. The exception to this rule is the domestic microwave oven magnetron
where very large numbers have been made. The cost per tube is then dramatically
reduced when compared with magnetrons of other types. In some cases tubes,
which have been developed for one purpose, have found applications elsewhere.
For example high power tetrodes for radio transmitters and klystrons for tele-
vision and radar transmitters have been used in particle accelerators. However,
if the primary market declines, the manufacturers may not continue to produce
tubes for low-​volume secondary applications. This was a major motivation for
the development of high power solid state amplifiers for the SOLEIL synchro-
tron [11]. Thus the nature of the market for high power vacuum tubes dictates
the need for close communication between the engineers developing a tube and
those working on the system in which it will be used. This pattern, which is typ-
ical of other markets for low-​volume, high-​technology, products is quite different
from that for high-​volume products. The successful incorporation of complex
components, such as tubes, into a system requires the systems engineers to have
a detailed understanding of how they work. It is not sufficient to regard tubes as
‘black boxes’ whose performance can be described completely by characteristics
defined at their terminals. Tube engineers must, likewise, have a detailed under-
standing of the issues which are important for the systems in which their products
are used. Some of these are reviewed in this chapter and in Chapter 20. For fur-
ther information see [37].
19

The Statement of Requirements 19

Table 1.2:  The principal applications of vacuum tubes

Field of application Principal applications Principal tube types

Broadcasting Radio Triodes

Tetrodes
Television Inductive output tubes
Klystrons
Direct broadcasting from satellites Helix travelling-wave tubes
Telecommunications Point to point links Helix travelling-wave tubes
Satellite communications Helix travelling-wave tubes
Klystrons
Deep space communications Coupled-​cavity travelling-wave tubes
Civilian radar Marine radar Magnetrons
Airborne radar Magnetrons
Weather radar Magnetrons
Air traffic control Magnetrons
Coupled-​cavity travelling-wave tubes
Military Early warning radar Klystrons
Target identification radar Magnetrons
Klystrons
Travelling-wave tubes
Crossed-​field amplifiers
Electronic countermeasures Helix travelling-wave tubes
Industrial Industrial heating Triodes
Magnetrons
Domestic ovens Magnetrons
Scientific Scientific particle accelerators Tetrodes,
Inductive output tubes
Klystrons
Medical accelerators Klystrons
Magnetrons
Thermonuclear fusion reactors Tetrodes, Klystrons
Gyrotrons

1.5 The Statement of Requirements

A fundamental document for communication between the customers and the

contractors for any engineering product is the Statement of Requirements (also
known as a Requirements Specification or a Statement of Work). It is the source
from which two further documents are derived: the Manufacturing Specification
(or Manufacturing Data Package), which provides all the information required
to manufacture the product; and the Test Specification, which gives details of
20

20 Overview

the tests that the product must undergo to demonstrate its compliance with the
requirements. It is self-​evident that any omissions or ambiguities in these docu-
ments will potentially lead to failure to achieve the outcome desired. Thus it is
essential that the Statement of Requirements and the Test Specification are pro-
duced by an open dialogue between representatives of the customer and the con-
tractor, both of whom fully understand the issues involved. For this purpose it is
helpful to have a checklist of the issues to be discussed [37, 38]. These issues fall
into two main groups: the performance requirements; and the design constraints.
Typical top-​level requirements are shown in Tables 1.3 and 1.4. The following sec-
tions discuss some of the issues which are important in the specification of per-
formance requirements.

1.6 Signals and Noise

In order to transmit information for telecommunications (one to one) or broadcast-

ing (one to many) it is necessary for the properties of the sinusoidal carrier wave to
be modulated by varying them with time. The properties of the input signal which
may be varied are amplitude, phase, and frequency. Many different methods of
modulation are in use and only a brief summary is given here of the factors that are
important in the specification of vacuum tube amplifiers [39–​41]. The methods fall
into two classes: analogue modulation in which the properties of the carrier vary
continuously with time; and digital modulation in which they are switched between
a number of states.
The baseband input signal is typically defined as a time-​varying voltage. This
could represent a continuously varying, analogue, source such as a speech or music
waveform. Alternatively it could represent a digital source such as a computer data
file, or the digitised form of an analogue waveform. It is common, for conve­nience
of analysis, to assume that the input signal is random when observed over an
extended period of time, though this is frequently not the case. The frequency spec-
trum of the signal is obtained by taking the Fourier transform of the waveform.
Theoretically this spectrum extends to infinity but, in practice, it is assumed to be
restricted to a bandwidth that extends from DC up to some maximum frequency.
The upper limit may be somewhat arbitrary and be defined as the point at which
the power density in the spectrum falls below a certain limit. This still applies if the
bandwidth of the signal has been limited by a filter, since no practical filter has an
infinitely sharp cut-​off.
The basic digital signal is binary in which the voltage switches between two states
at a fixed clock rate. However, it is often useful to convert this into a form in which a
greater number of states is employed. This can be illustrated by considering Table 1.5,
which shows different possible representations of the symbols in a sixteen charac-
ter set. Let us suppose that we wish to transmit a message made up of the 16 characters
in the first column at a rate of 1 symbol/​sec. It is helpful to think of these characters
21

Signals and Noise 21

Table 1.3:  Typical performance requirements for vacuum tube amplifiers

Frequency
Bandwidth (instantaneous, tuneable)
Power output (peak, mean)
Gain
Linearity (amplitude, phase)
Noise and spurious emissions
Efficiency
RF matches

Table 1.4:  Typical design constraints for vacuum tube amplifiers

High-​voltage power supply
  (voltage, current, pulse shape)
Auxiliary power supplies
Size and weight
Cooling
Temperature range (storage, operation)
Shock and vibration
Atmosphere (pressure, humidity)
Reliability
Repairability
Lifetime cost

as representing 16 different voltage levels in a stepwise approximation to an ana-

logue waveform. A little thought shows that the highest frequency is obtained when
any pair of symbols is repeated alternately (e.g. ABABABA). The Fourier trans-
form of all other sequences has a lower maximum frequency.
If we now wish to transmit the same information in binary form we must send
four binary digits (bits) per second. The highest frequency is obtained when the
symbols 0 and 1 are repeated alternately (e.g. FFFF, or KKKK). Finally we can
represent the 16 symbols by a four-​symbol set by grouping the binary digits in pairs
as shown in the third column. Now the highest frequency is obtained when any pair
of symbols is repeated alternately (e.g. WXWX = BB). From this it can be seen that
the bandwidth required to transmit the same message depends upon the number
of symbols in the set used to represent it. Alternatively, we see that the maximum
data rate for a fixed bandwidth is increased by a factor of two if a four-​symbol set is
used, and by a factor of four with a 16-​character set. Now an ideal low-​pass chan-
nel of bandwidth B can transmit 2B pulses per second. Thus the maximum rate at
which data can be transmitted over this channel is given by Hartley’s law

R = 2 B log 2 m bits s −1, (1.34)

where m is the number of symbols used.

2

22 Overview

1.6.1 Noise
Any communications system adds white noise to the input signal so that the signal
received is corrupted to some extent [39]. The thermal noise power in watts caused
by the random motion of charge carriers in resistive materials that is delivered to a
matched load is
N p = kT B, (1.35)
where k is Boltzmann’s constant ( k = 1.380 × 10 −23 J K −1 ), T is the absolute tempera-
ture and B is the bandwidth of the system in Hz. The effectiveness of an analogue
communication system is measured by the signal to noise ratio (S p N p ) where S p is
the average signal power. This is commonly expressed in decibels as

 Sp 
SNR = 10 log10   . (1.36)
 Np 
The minimum acceptable SNR for reliable communication is generally taken to be
about 10 dB.
For digital communications the corruption of the signal may result in some of the
bits being received incorrectly. Shannon’s theorem states that a noisy channel will
theoretically support error-​free data transmission at a channel capacity given by

 Sp 
C = B log 2 1 +  . (1.37)
 Np 

Table 1.5:  Representation of digital signals

16 symbols 2 symbols (binary) 4 symbols

A 0000 00 00 (WW)
B 0001 00 01 (WX)
C 0010 00 10 (WY)
D 0011 00 11 (WZ)
E 0100 01 00 (XW)
F 0101 01 01 (XX)
G 0110 01 10 (XY)
H 0111 01 11 (XZ)
I 1000 10 00 (YW)
J 1001 10 01 (YX)
K 1010 10 10 (YY)
L 1011 10 11 (YZ)
M 1100 11 00 (ZW)
N 1101 11 01 (ZX)
O 1110 11 10 (ZY)
P 1111 11 11 (ZZ)
23

Signals and Noise 23

This can be expressed alternatively as

 E R
C = B log 2 1 + b ⋅  , (1.38)
 N0 B 

where Eb is the energy per bit, R is the mean bit rate, and N 0 is the spectral energy
density of the noise. Shannon’s law represents an ideal which cannot be achieved
in practice, but it remains useful for comparison with practical systems. In a digital
transmission system the energy per bit, which is related to the bit error rate, is
commonly used in place of the signal to noise ratio [40]. A  second performance
measure is the bandwidth efficiency ( R B ) in bits s−1 Hz−1.
Each part of a communications channel contributes some noise to the system.
Here we are concerned specifically with the contribution from the final power
ampli­fier. The noise generated by an amplifier can be described by its noise figure,
which is the ratio of the signal to noise ratio at the input to that at the output

SNRi
F= . (1.39)
SNRo

If an amplifier has a power gain Ap then the signal to noise ratio at the output is

Ap Si
SNRo = , (1.40)
Ap Ni + N a

where Si and Ni are the signal and noise powers at the input and N a is the noise
power added by the amplifier. Thus

Ap Ni + N a
F= . (1.41)
Ap Ni

The noise figure is standardised by fixing the input noise power as that given by
(1.35) when T = 290 K. The noise added by the amplifier can be described by an
equivalent noise source at its input such that

N a = kTe BAp , (1.42)

where Te is the effective noise temperature of the amplifier. The noise performance
can also be described by the total noise power at the output, or by the noise power
density (noise power per Hz).

1.6.2 Analogue Modulation

In order to transmit the baseband signal over a wireless channel it must be super-
imposed upon an RF carrier wave by a modulator. The properties of the carrier
wave which can be varied by the modulator are its amplitude, frequency, and phase,
or some combination of them. In general the modulated signal at the input to the
24

24 Overview

power amplifier can be described by a sinusoidal carrier whose amplitude and phase
are functions of time [42, 43]. Thus

V1 (t ) = r (t ) cos ( ω 0 t + ψ (t )) , (1.43)

where ω 0 is the carrier frequency. The rates of change with time of the amplitude
(r (t )) and phase ( ψ (t )) are normally much slower than that of the carrier wave so
that it is permissible to regard the instantaneous response of the amplifier as being
that for a single unmodulated carrier wave. The response to more complex signals
can therefore be deduced from the steady-​state, single-​carrier, transfer character-
istics. Figure  1.12 shows examples of simple analogue modulation schemes. For
purposes of illustration the modulation frequency has been taken to be one eighth
of the carrier frequency although the ratio is normally much smaller than that.
The simplest example to understand is the amplitude modulation shown in
Figure 1.12(a). The envelope of the carrier carries information such as an audio
waveform. In the time domain the modulated waveform is given by

f (t ) =  K + S (t ) cos ( ω 0 t ) , (1.44)

where K is the amplitude of the unmodulated carrier and S (t ) is the signal wave-
form. When the signal is sinusoidal with frequency ω m the Fourier transform of
f (t ) yields the spectrum in the frequency domain shown in Figure  1.12(b). This
representation shows the amplitudes, but not the phases, of the components of the
signal. The central line at the carrier frequency is flanked by two sidebands at fre-
quencies ( ω 0 ± ω m ). This is known as double sideband amplitude modulation. For a
more general modulating waveform the sidebands are the frequency-​shifted Fourier
transforms of the baseband waveform. Faithful transmission of the modulated
waveform requires the bandwidth of the amplifier to be at least 2B (the Nyquist
limit) where B is the baseband bandwidth. The instantaneous power in the modu-
lated signal varies with time, as can be seen from Figure 1.12(a). Faithful transmis-
sion of the signal requires the amplifier to be linear up to the highest instantaneous
power. Any non-​linearity of the amplifier produces a distortion of the signal in the
time domain which is reflected in changes in its spectrum (see Section 1.6.4).
Figures  1.12(c) and (d) show double sideband suppressed carrier modulation.
The modulated waveform is

f (t ) = S (t ) cos ( ω 0 t ) . (1.45)

The spectrum differs from that in Figure 1.12(b) in the absence of the carrier fre-
quency. The bandwidth requirements for the amplifier are the same as in the previ-
ous case and the dynamic range is greater. However, this modulation scheme is less
susceptible to interference from noise than the previous case because all the power
is in the sidebands.
Figures  1.12(e) and (f) show phase modulation. The amplitude envelope of
the signal is constant in the time domain so that the instantaneous power is con-
stant. This has the advantage that the working point of the amplifier is virtually
25

Signals and Noise 25

Figure 1.12:  Time and frequency domain representations of analogue modulated carriers: (a)

and (b) double sideband amplitude modulation; (c) and (d) double sideband suppressed carrier
amplitude modulation; (e) and (f) phase modulation.
26

26 Overview

constant, so that the output signal is only affected by the frequency dependence
of the gain and phase of the amplifier. The modulated signal, assuming unity
amplitude, is

f (t ) = cos ( ω 0 t + S (t )) , (1.46)

so that the dependence of the modulated signal on the modulation is non-​linear.

For sinusoidal modulation at frequency ω m (1.46) can be written

f (t ) = cos ( ω 0 t + β sin ω mt ) , (1.47)

where β is the modulation index. The spectrum for this waveform, shown in
Figure  1.12(f) for β = 1, has additional lines at frequencies ( ω 0 ± nω m ) where
n = 1, 2, 3,. Thus the bandwidth required for faithful amplification is greater than
for amplitude modulation schemes. A  rule of thumb for the transmission band-
width is Carson’s rule [40]

BT = 2 (β + 1) ω m . (1.48)

Frequency modulation can also be represented by (1.47) so that its properties are
similar to those of phase modulation. The properties of analogue modulation
schemes can be explored using Worksheet 1.2.

1.6.3 Digital Modulation

The simplest digital modulation schemes are amplitude shift keyed (ASK), phase
shift keyed (PSK), and frequency shift keyed (FSK) modulation [39]. Figure 1.13
shows time and frequency domain representations of these methods of modulation,
using the same carrier and modulation frequencies as before. It may be observed
that the bandwidths of these signals are greater than those for the equivalent ana-
logue modulation, with the exception of binary FSK modulation, which may be
considered a special case of frequency modulation. The properties of binary digital
modulation schemes can be explored using Worksheet 1.2.
We saw in Section 1.6.1 that the bit rate in a digital system having fixed bandwidth
can be increased by using a greater number of symbols. Two common implementa-
tions of this principle are illustrated in Figure 1.14 which shows the positions of
the symbols on phasor diagrams known as constellation plots. Figure 1.14(a) shows
binary phase-​shift keyed (BPSK) modulation with a phase shift of 180° correspond-
ing to Figure 1.13(c) and (d). Quadrature phase-​shift keyed (QPSK) modulation is
shown in Figure 1.14(b), and quadrature amplitude modulation with 16 symbols
(16-​QAM) is shown in Figure 1.14(c). The effect of noise in the transmission chan-
nel is to produce a scatter of points around the desired constellation points at the
receiver. If the scatter is too great then the receiver will fail to identify the symbols
correctly leading to errors in transmission. Table 1.6 shows the bandwidth efficien-
cies and Eb N 0 figures for these three channels at a bit error rate of 10 −6 . These
figures assume coherent modulation so that the clock signals at the transmitter and
27

Signals and Noise 27

Figure 1.13:  Time and frequency domain representations of simple digitally-​modulated carriers: (a) and
(b) binary ASK; (c) and (d) binary PSK; (e) and (f) binary FSK.
28

28 Overview

Table 1.6:  Comparison between examples of coherent digital modulation schemes [40]

Modulation type Bandwidth efficiency Eb N 0

(bits s−1 Hz−1)

BPSK 0.5 10.6 dB

QPSK 1.0 10.6 dB
16-​QAM 2.0 14.3 dB

Figure 1.14:  Examples of constellation diagrams for digitally-​modulated carriers: (a) binary

phase-​shift keyed (BPSK), (b) quadrature phase-​shift keyed (QPSK), and (c) quadrature
amplitude modulation with 16 symbols (16-​QAM).

the receiver are locked to one another. It can be seen that the bandwidth efficiency
increases with the number of symbols according to (1.34) as expected.
From the point of view of the design of the power amplifier it is important to note
that BPSK and QPSK are constant envelope schemes which allow the efficiency of
the power amplifier to be high, subject only to the need to maintain adequate linear-
ity. On the other hand the amplitude variation in 16-​QAM means that the amplifier
is only working at the highest power conversion efficiency at four of the constellation
points. In addition, the overall transmitter power must be high enough to achieve an
acceptable bit error rate for the innermost constellation points where the transmit-
ted power is least. If all constellation points are equally probable then the average
efficiency is about half of that at the outermost points. This makes 16-​QAM unsuit-
able for applications such as satellite downlinks where high efficiency is important.

1.6.4 Multiplexing
It is commonly the case that one telecommunications channel carries data from
multiple sources. The interleaving of the different data streams, known as multi-
plexing, can be achieved in many different ways [44]. Of these, the method which
has implications for the design of a vacuum tube amplifier is frequency domain
multiplexing (FDM) in which the input to the tube comprises a number of carriers
each of which is modulated by a separate data stream [42]. In a typical scheme the
available bandwidth is occupied by N carriers at frequency intervals 2∆ω. If the
centre frequency is ω 0 then the channel frequencies are

ω n = ω 0 + n∆ω for n = ±1, ± 3, ± 5 , (1.49)

29

Signals and Noise 29

where each of these sub-​carriers is frequency-​modulated. The non-​linearity of the

final power amplifier produces mixing between the carriers leading to co-​channel
interference. It is therefore necessary to have some way of calculating this interfer-
ence and specifying the maximum acceptable level.
The simplest approach is to consider the central pair of carriers at frequencies
ω = ω 0 ± ∆ω with equal amplitudes. The two carriers each contribute half of the
RF input power so that their amplitudes are related to that of a single carrier hav-
ing the same total power by Vc = V1 2. The input to the final power amplifier can
be written in the form given in (1.43)

Vin (t ) = 2Vc cos ( ∆ω t ) cos ( ω 0 t ) = r (t ) cos ( ω 0 t ) , (1.50)

where r (t ) is the time-​varying amplitude. Then, from (1.11), the output of the
amplifier is

( )
V2 (t ) = A r (t ) cos ω 0 t + Φ r (t ) . (1.51)

The non-​linearity of the amplifier produces signals at frequencies given by

ω m,n = m ( ω 0 + ∆ω ) ± n ( ω 0 − ∆ω ) , (1.52)

where m, n = 0,1, 2. The amplitudes of the signals at these frequencies can be
found by Fourier analysis [45]. Thus, setting m = 1 and n = 0 gives the carrier fre-
quency ω1,0 = ω 0 + ∆ω at which the amplitude is
π
V1,0 (V1 ) =
1
{ }
A r ( φ ) cos Φ r ( φ ) sin φ d φ, (1.53)
π ∫0 

where φ = ∆ω t. The analysis is conveniently carried out in the base band. Since the
AM/​AM and AM/​PM characteristics of the amplifier normally vary only slowly
with frequency it is possible to assume that they are constant, so that the amplitude
of the other carrier (V0,1 ) is the same. Hence the total RF output power at the fre-
quencies of the two carriers is given by replacing Vc in (1.50) by the equivalent sin-
gle carrier amplitude V1. This power can be plotted against the RF input power for
comparison with the single-​carrier AM/​AM characteristic as shown in Figure 1.15.
At small input powers the amplifier is linear and the power gain is identical to that
with one carrier. However, as the input power approaches the single-​carrier satu-
ration level the output power is less with two carriers than with one. The reason
for this is illustrated in Figure 1.16 which shows typical two-​tone input and out-
put waveforms for an amplifier, normalised to the saturated output voltage of the
amplifier. For clarity of explanation the carrier frequency shown is much lower rel-
ative to the frequency of the envelope than would normally be the case. The output
voltage is reduced when the instantaneous amplitude of the signal exceeds the sat-
uration level of the amplifier. Thus the saturated output power under multi-​carrier
operation is less than that for a single carrier at the same input power. Figure 1.15
also compares the effects of AM/​AM conversion alone (typical of klystrons) with
those in which AM/​PM conversion is also included (typical of TWTs) [45].
30

30 Overview

Figure 1.15:  Typical AM/​AM transfer characteristics for two-​carrier operation showing

third-order (IM3) and fifth-order (IM5) intermodulation products.

Figure 1.16:  Typical waveforms for two-​carrier operation of a non-​linear amplifier: (a) input,

and (b) output.
31

Signals and Noise 31

When we set m = 2 and n = 1 in (1.52) we obtain a third-order intermodulation

product at frequency ω 2,1 = ω 0 + 3∆ω whose amplitude is
π
1
V2,1 (V1 ) = A ( 2Vc cos φ ) cos Φ ( 2Vc cos φ ) sin 3φ d φ. (1.54)
π ∫0

In the same way, setting m = 3 and n = 2 gives a fifth-order intermodulation product

at ω3,2 = ω 0 + 5 ∆ω whose amplitude is
π
1
V3,2 (V1 ) = A ( 2Vc cos φ ) cos Φ ( 2Vc cos φ ) sin 5φ d φ. (1.55)
π ∫0

The amplitudes of the other intermodulation products at ω1,2 and ω 2,3 are the same.
Then the total third-order (IM3) and fifth-order (IM5) intermodulation powers
can be plotted as shown in Figure 1.15. The frequencies of these signals coincide
with those of possible adjacent carriers resulting in co-​channel interference. The
acceptable level of interference is commonly expressed as the ratio (C I3) of the
fundamental power to that of the third-order intermodulation product, expressed
in decibels. At low drive levels the slope of the graph of the third-​order product is
3 dB/​dB and that of the fifth-​order product is 5 dB/​dB. Thus the ratio C I3 can
be increased by operating the amplifier at a reduced drive level. This is commonly
expressed in terms of the output back-​off from saturation (in dB) necessary to
achieve an acceptable value of C I3. The linear parts of the curves for the inter-
modulation products can be extrapolated in the same manner as in Figure 1.4 to
meet the extrapolation of the fundamental curve at the third-​order and fifth-​order
intercept points. These may also be used to specify the linearity of the amplifier.
The functions A and Φ required for the evaluation of the integrals in (1.53) to
(1.55) may be specified numerically using interpolation on data points which have
been determined experimentally or by large-​signal modelling [45]. Alternatively a
suitable function may be fitted to the data points [42, 43, 46]. Polynomial expansions
do not give a good fit to the data unless a large number of terms is used. In par-
ticular these models fail for drive levels approaching saturation. They do, however,
reveal that the fundamental, and intermodulation, amplitudes depend only on the
odd terms of the series. These can be specified independently of the terms describing
the dependence of the even harmonics on the input RF power [46]. Simple functions
which are a good fit to experimental data for TWTs have been proposed by Saleh [43]:

αa r
A (r ) = (1.56)
1 + βa r 2
and

αφr2
Φ (r ) = , (1.57)
1 + βφ r 2

where α a , βa , α φ and βφ are empirical constants. If the amplifier has unity gain at
saturation, so that A (1) = 1, it can be shown that α a = 2 and βa = 1. These values are
32

32 Overview

close to the empirical figures so that it is possible to use them as a useful approxima-
tion. In the same way it may be tentatively suggested that the function required to
generate the even harmonics might be written

αhr2
B (r ) = . (1.58)
(1 + βh r 2 )2
The curves shown in Figures  1.4 and 1.15 were generated using this non-​linear
model (see Worksheet 1.3).
The ratio C I3 is commonly used for specifying the intermodulation distortion
in an amplifier, not least because it is straightforward to measure it experimentally.
However, it is at best, a proxy for the intermodulation effects because, when there
are more than two carriers, the overall performance depends upon the signal levels
of all the carriers. Since the phases of the phase-​modulated carriers vary in a quasi-​
random manner it may be assumed that all phase combinations are equally prob-
able. Where only two carriers are used it is sufficient to consider unmodulated sine
waves because all possible phases arise in one period of the difference frequency.
Then the peak power is twice the mean power. However, with three or more carriers
this is not the case, and it is necessary to use Monte Carlo methods to examine all
possible phases [47]. As the number of carriers increases, the probability that all the
signals will be in phase simultaneously decreases. In the limit, a large number of
uncorrelated phase-​modulated carriers can be represented by band-​limited white
noise [45, 48]. This leads to an alternative measure of intermodulation distortion in
which the input to the amplifier is band-​limited white noise from which the portion
in a small frequency range has been removed by a notch filter. The non-​linearity of
the amplifier causes some signals to be generated in this frequency range. The ratio
of the amplitude of these to that of the white noise at the output of the amplifier is
the Noise Power Ratio (in dB). This is a measure of the level of interference which
would be experienced by a signal at the centre of the notch [49, 50]. An approxima-
tion to this can be achieved experimentally, or in simulations, by using a comb of
carriers at equal frequency intervals such that one carrier has been omitted from
the centre of the set as shown in Figure 1.17(a) [47, 51]. Figure 17(b) shows a typi-
cal output spectrum and the definition of the noise power ratio. Intermodulation
products are also generated on the edges of the input spectrum so increasing its
bandwidth. If the power in these is great enough then it may cause interference with
other systems operating in adjacent frequency bands. For this reason the permitted
bandwidth may be expressed as a spectrum envelope [52].

1.7 Engineering Design

The design of any engineering product can be divided into three phases:

i) Conceptual design, in which the principal dimensions and operating param-

eters are established.
3

Engineering Design 33

Figure 1.17:  Amplification of multiple carriers: (a) input spectrum, and (b) output spectrum.

ii) Detailed design, involving refinement of the conceptual design by prototyping

and advanced modelling, leading to the preparation of the manufacturing data
package and the test specification.
iii) Design consolidation, incorporating further refinements suggested by product
testing and field experience.

The costs that are committed, and incurred, increase at each stage. It is therefore
desirable to produce a conceptual design which does not require further modifica-
tion at the detailed design stage; and a detailed design which does not have to be
modified as a result of unsatisfactory performance in the field. It should be noted
that performance and environmental testing is normally carried out under labora-
tory conditions, and care must be taken to ensure that tubes which pass these tests
do also perform well in the field. Differences between the testing and the real envi­
ronments, such as the impedances of power supplies and matches of RF compo-
nents, can affect how a tube performs.

1.7.1 Dimensionless Parameters and Scaling

Nearly all tubes are developed in organisations in which the design process can be
based upon accumulated knowledge and expertise. Many tubes are designed by
scaling from existing designs, or by other modifications to them, rather than start-
ing from a clean sheet of paper. It is, however, important that this process should be
underpinned by fundamental understanding.
34

34 Overview

The basis for scaling is dimensional analysis [53, 54]. Let us suppose that some
aspect of the performance of a tube is governed by an equation which can be writ-
ten in the general form

F (Q1 ,Q2 Qn , r ′, r ′′ ) = 0, (1.59)

where Q1 ,Q2 etc. are dimensional variables and r ′, r ′′ etc. are dimensionless ratios.
Very commonly these are the ratios of the leading dimensions of the tube to a single
dimension or to a dimensional variable such as the free-​space wavelength. If two
tubes are described by the same set of dimensionless ratios, so that all dimensions
are changed by the same multiplying factor, they are said to be geometrically similar
to one another. If geometrical similarity is assumed then the ratios are constants
and (1.59) can be written

F (Q1 ,Q2 Qn ) = 0. (1.60)

This equation is a complete description of the problem, provided that none of

the variables required has been overlooked. The choice of variable for inclusion
is based upon an understanding of the physics of the problem. When it involves
electromagnetics then the primary electric and magnetic constants ε 0 and µ 0 may
have to be included in the list. The ratio of the charge to the rest mass of an
electron ( e m0 ) is also required when the dynamics of electrons are part of the
problem. Equation (1.60) remains valid even if the algebraic form of the func-
tion is unknown. The dimensions of the variables Q can be expressed in terms
of the fundamental dimensions mass [M]‌, length [L], and time [T], together with
voltage [V] or charge for electrical problems. Thus there are, at most, four funda-
mental dimensions. Table 1.7 shows the dimensions of some quantities relevant
to vacuum tubes.
The variables Q may be combined together to form dimensionless groups denoted
by Π so that (1.60) can be written

ψ ( Π1 , Π 2 , Πi ) = 0. (1.61)

The number of groups required is related to the number of variables by

i = n − k , (1.62)

where k is the number of dimensions required. This is known as Buckingham’s

theorem. An immediate effect of using dimensionless variables (or dimensionless
groups) is that the number of variables required to define the problem has been
reduced. Now let us choose one of the groups to be a dependent variable whose
value is determined by the values of the others. Then (1.61) becomes

Π1 = Ψ ( Π 2 , Π3 , Πi ) . (1.63)
35

Engineering Design 35

Table 1.7:  Typical independent and dependent parameters for vacuum tubes

Parameter Dimensions

Constants Velocity of light in free space (c) [LT −1 ]

0.299792 × 109 m s−1
Primary electric constant ( ε 0 ) [MLT −2 V −2 ]
8.854187 × 10−12 F m−1
Primary magnetic constant (µ 0 ) [M−1L−3T 4 V 2 ]
1.256637 × 10−6 H m−1
Elementary charge (e) [ML2T −2 V −1 ]
1.602176 × 10−19 C
Rest mass of the electron (m0) [ M]
9.109382 × 10−31 kg
0.510998 MeV
Electron charge to mass ratio (e m0 ) [L2T −2 V −1 ]
1.758820 × 1011 C kg−1
Parameters Length [L]
Voltage [V]
Magnetic flux density [L−2TV ]
Frequency [T −1 ]
R.F. power [ML2T −3 ]
Current [ML2T −3V −1 ]
Electron velocity [LT −1 ]
Charge density [ML−1T −2 V −1 ]

If the values of the independent variables ( Π 2 , Π3 , Πi ) are fixed then so is the value
of the dependent variable Π1. Two tubes which are described by the same values of
the dimensionless variables are said to be dynamically similar to one another. Thus
the basis of scaling is the maintenance of both geometrical and dynamic similarity.
The process of forming dimensionless groups is aided by choosing normalisa-
tions which are based on the underlying physics of the problem. Some examples are
shown in Table 1.8.
A simple illustration is provided by the calculation of the velocity of an electron
(u0 ) from an applied voltage (Va ) and the dimensional constants. The dependent
dimensionless group is

u0
Π1 = (1.64)
c
and the independent dimensionless group is

eVa
Π2 = . (1.65)
m0 c 2
36

36 Overview

Table 1.8:  Examples of the formation of dimensionless groups

Variable Normalising quantity Dimensionless

group

Dimension (d) Wavelength ( λ ) d

Propagation constant (β ) or βd
λ
Velocity (u) Velocity of light (c) u u
or
Phase velocity (v p ) c vp

Voltage (V) Rest energy of the electron in eV (V0 = m0c 2 e ) eV

m0c 2
Electron plasma frequency (ω p ) Frequency (ω ) ωp ω
Electron cyclotron frequency (ω c ) and c
ω ω

In this case there are five parameters and three independent dimensions: [M], [ V ]
and [ LT −1 ] so that the problem is completely described by the two dimensionless
groups which have been formed. Note that in this case the dimensions [ L ] and [ T ]
do not occur separately so that the number of independent dimensions is reduced
by one. It follows that we can write

Π1 = Ψ ( Π 2 ) , (1.66)

where Ψ is an unknown function which can be determined empirically, even if its

form cannot be found by mathematical analysis. The value of the reduction of the
description of a problem to the form in (1.63) is greater than just that of scaling. If
the form of the function Ψ is known, even over a limited range of variation of the
variables then it can be used to explore the effects of design choices on Π1. In cases
where geometrical similarity is not maintained then one, or more, of the ratios r are
added as dimensionless variables. Examples of the application of this method to
particular problems in tube design are given later in this book.

1.7.2 Modelling
The trend in tube design is towards achieving near-​ideal RF performance coupled
with high reliability and low cost of ownership. For this reason, the design of a state-​
of-​the-​art vacuum tube is a complex business which involves considerable use of
computer simulations to achieve right-​first-​time design [55, 56]. The advanced com-
puter codes used for detailed design can take many hours to run, even on the most
powerful computers available. They are invaluable tools for accurate modelling, but
are not well-​suited to the conceptual design stage which is the focus of this book. For
that, the engineers must have a sound understanding of the theory of vacuum tubes,
together with simple computer models which can be used for rapid design iterations.
37

References 37

The availability of advanced computer modelling tools such as Particle in Cell (PIC)
codes can mislead the unwary into supposing that the best way of modelling a tube
is to create a complete model in such a code. This approach is wrong on two counts:
i) A PIC code finds a self-​consistent numerical solution of Maxwell’s equations
and the equations of motion in a series of time steps. However, the solution is
linear in the part of the problem space which is not occupied by electrons. It is
therefore more efficient to compute that part of the solution once and store it for
future use. The problem is then reduced to the solution of the non-​linear model
in the space containing electrons, subject to matching the stored solution on the
boundary. This is the method used in the large-​signal models of tubes which are
described later in this book.
ii) Even the best computer codes cannot be wholly accurate models of reality
because the complexity of the problem is too great, so that some parts of it may
not be known precisely. Assumptions and approximations are always required.
For this reason the mathematical modelling of tubes is as much an art as it
is a science [57]. The best results are obtained when the physics of the prob-
lem are well understood so that the assumptions and approximations are valid.
Any model is only useful when it has been validated by comparison with experi-
mental results. Even then it can only be used with caution outside the range of
parameters for which it has been validated.
A large number of simple models were created in Mathcad14® during the writing
of the book and used to generate many of the figures. These models can be used to
explore the behaviour of different aspects of tubes by changing the parameters. They
are available in electronic form (see the list of worksheets in the Appendix, availa-
ble online at www.cambridge.org/9780521198622). The comments included in them
should be sufficient for users to create their own models using other software if they
wish. The models are intended as educational tools and have not been validated for
use in tube design.

References

[1]‌ J. A. Fleming, ‘Improvements in instruments for detecting and measuring alternating

electric currents’, UK Patent GB190424850 (A) 1904.
[2]‌ R. J. Barker et  al., eds, Modern Microwave and Millimeter-​Wave Power Electronics.
Piscataway, NJ: IEEE Press; Hoboken, NJ: Wiley-Interscience, 2005.
[3]‌ W. H. Kohl, Handbook of Materials and Techniques for Vacuum Devices. New  York:
American Institute of Physics, 1995.
[4]‌ F. Rosebury, Handbook of Electron Tube and Vacuum Techniques. New York: American
Institute of Physics, 1993.
[5]‌ R. J. Trew, ‘SiC and GaN transistors –​is there one winner for microwave power applica-
tions?’, Proceedings of the IEEE, vol. 90, pp. 1032–​1047, 2002.
[6]‌ M. Meneghini et  al., Power GaN Devices:  Materials, Applications and Reliability.
Springer, 2016.
38

38 Overview

[7]‌ R. S. Pengelly et  al., ‘A review of GaN on SiC high electron-​mobility power transis-
tors and MMICs’, IEEE Transactions on Microwave Theory and Techniques, vol. 60,
pp. 1764–​1783, 2012.
[8]‌ M. Kasu et al., ‘Diamond-​based RF power transistors: fundamentals and applications’,
Diamond & Related Materials, vol. 16, pp. 1010–​1015, 2007.
[9]‌ V. Camarchia et al., ‘An overview on recent developments in RF and microwave power
H-​terminated diamond MESFET technology’, presented at the 2014 International
Workshop on Integrated Nonlinear Microwave and Millimetre-​ wave Circuits
(INMMiC), 2014.
[10] M. A. Richards et al., Principles of Modern Radar. Edison, NJ: SciTech, 2010.
[11] P. Marchand et  al., ‘High power 352 MHz solid state amplifiers developed at the
Synchrotron SOLEIL’, Physical Review Special Topics –​Accelerators and Beams, vol.
10, p. 112001, 2007.
[12] R. Caverly et  al., ‘Advancements at the lower end:  advances in HF, VHF, and UHF
systems and technology’, IEEE Microwave Magazine, vol. 16, pp. 28–​49, 2015.
[13] H. P. Kindermann et  al., ‘The RF power plant of the SPS’, IEEE Transactions on
Nuclear Science, Vol. NS-​30, No. 4, vol. NS-​30, pp. 3414–​3416, August 1983.
[14] M. Jensen, ‘Inductive output tube based 300 kW RF amplifier for the Diamond Light
Source’, in EPAC 2004, Lucerne, Switzerland, pp. 962–​964, 2004.
[15] J. H. Booske et al., ‘Traveling-​wave tubes’, in R. J. Barker et al., eds, Modern Microwave
and Millimetre-​Wave Power Electronics. Piscataway, NJ: IEEE, pp. 171–​245, 2005.
[16] V. Granatstein et  al., ‘Vacuum electronics at the dawn of the twenty-​first century’,
Proceedings of the IEEE, vol. 87, pp. 702–​716, 1999.
[17] G. Faillon et al., ‘Microwave tubes’, in J. A. Eichmeier and M. K. Thumm, eds, Vacuum
Electronics: Components and Devices. Berlin: Springer-​Verlag, pp. 1–​84, 2008.
[18] J. H. Booske and R. J. Barker, ‘Introduction and overview’, in R. J. Barker et al., eds,
Modern Microwave and Millimetre-​Wave Power Electronics. Piscataway, NJ: IEEE Press,
pp. 1–​33, 2005.
[19] N. Ayllon, ‘Microwave high power amplifier technologies for space-​borne applica-
tions’, presented at the 2015 IEEE 16th Annual Wireless and Microwave Technology
Conference (WAMICON), 2015.
[20] W. G.  V. Rosser, An Introduction to the Theory of Relativity. London:  Butterworths,
1964.
[21] D. Gabor, ‘Energy conversion in electronic devices’, Journal of the Institution of
Electrical Engineers –​Part III: Communication Engineering, including the Proceedings of
the Wireless Section of the Institution, vol. 91, pp. 128–​141, 1944.
[22] National Frequency Planning Group, ‘United Kingdom Frequency Allocation Table’,
Committee on UK Spectrum Strategy, 2013.
[23] National Telecommunications and Information Administration, ‘Frequency
Allocations’, in Manual of Regulations and Procedures for Federal Radio Frequency
Management, US Department of Commerce, 2014.
[24] J. R. Pierce, ‘Coupling of modes of propagation’, Journal of Applied Physics, vol. 25, pp.
179–​183, 1954.
[25] W. H. Louisell, Coupled Mode and Parametric Electronics. New York: Wiley, 1960.
[26] R. J. Briggs, Electron-​stream Interaction with Plasmas. Cambridge, MA: MIT Press, 1964.
[27] C. C. Johnson, Field and Wave Electrodynamics. New York: McGraw-​Hill, 1965.
[28] J. R. Pierce, ‘The wave picture of microwave tubes’, The Bell System Technical Journal,
vol. 33, pp. 1343–​1372, 1954.
39

References 39

[29] R. N. Sudan, ‘Classification of instabilities from their dispersion relations’, Physics of

Fluids, vol. 8, pp. 1899–​1904, 1965.
[30] K. R. Chu and A. T. Lin, ‘Gain and bandwidth of the gyro-​TWT and CARM ampli-
fiers’, IEEE Transactions on Plasma Science, vol. 16, pp. 90–​104, 1988.
[31] J. Feinstein and K. Felch, ‘Status review of research on millimeter-​wave tubes’, IEEE
Transactions on Electron Devices, vol. 34, pp. 461–​467, 1987.
[32] G. Bekefi, ‘Survey of physics research in microwave devices’, in International Electron
Devices Meeting, pp. 822–​825, 1984.
[33] A. F. Harvey, ‘Microwave tubes –​an introductory review with bibliography’, Proceedings
of the IEE –​Part C: Monographs, vol. 107, pp. 29–​59, 1960.
[34] N. C. Luhmann, Jr. et  al., ‘Historical highlights’, in R. J. Barker et  al., eds,
Modern Microwave and Millimeter-​Wave Power Electronics. Piscataway, NJ: IEEE Press,
pp. 35–​106, 2005.
[35] V. Granatstein et  al., ‘Vacuum electronics at the dawn of the twenty-​first century’,
Proceedings of the IEEE, vol. 87, pp. 702–​716, 2002.
[36] T. E. Yingst et al., ‘High-​power gridded tubes –​1972’, Proceedings of the IEEE, vol. 61,
pp. 357–​381, 1973.
[37] L. Sivan, Microwave Tube Transmitters. London: Chapman & Hall, 1994.
[38] NATO, AC243, Panel 3, RSG-​19, DRG Handbook:  Microwave and Millimetre Wave
Tubes and Power Supplies. NATO, 1998.
[39] J. Dunlop and D. G. Smith, Telecommunications Engineering. Wokingham, UK:  Van
Nostrand Reinhold (UK) Co. Ltd, 1984.
[40] R. Ziemer, ‘Modulation’, in R. A. Meyers, ed., Encyclopedia of Telecommunications.
San Diego, CA: Academic Press, pp. 201–​213, 1989.
[41] J. B. Anderson, Digital Transmission Engineering [electronic resource], 2nd ed.
Piscataway, NJ: IEEE Press; Hoboken, NJ: Wiley-​Interscience, 2005.
[42] A. Berman and C. E. Mahle, ‘Nonlinear phase shift in traveling-​wave tubes as applied
to multiple access communications satellites’, IEEE Transactions on Communication
Technology, vol. 18, pp. 37–​48, 1970.
[43] A. A.  M. Saleh, ‘Frequency-​ independent and frequency-​ dependent nonlinear
models of TWT amplifiers’, IEEE Transactions on Communications, vol. 29, pp.
1715–​1720, 1981.
[44] R. Ziemer, ‘Multiplexing’, in R. A. Meyers, ed., Encyclopedia of Telecommunications.
San Diego, CA: Academic Press, pp. 215–220, 1989.
[45] G. R. Stette, ‘Calculation of intermodulation from a single carrier amplitude character-
istic’, IEEE Transactions on Communications, vol. 22, pp. 319–​323, 1974.
[46] W. E. Kunz et al., ‘Traveling-​wave tube amplifier characteristics for communications’,
Microwave Journal, vol. 10, pp. 41–​46, 1967.
[47] R. G. Carter et  al., ‘Computer simulation of intermodulation distortion in traveling
wave tube amplifiers’, IEEE Transactions on Electron Devices, vol. 48, pp. 178–​180,
January 2001.
[48] A. Guida, ‘Accurate calculation of TWT intermodulation in the many-​carrier case’,
IEEE Transactions on Communications, vol. 35, pp. 685–​687, 1987.
[49] T. Reveyrand et al., ‘A novel experimental noise power ratio characterization method
for multicarrier microwave power amplifiers’, in 55th ARFTG Conference Digest-​Spring,
pp. 1–​5, 2000.
[50] A. Katz and R. Gray, ‘Noise power ratio measurement tutorial’, Linearizer Technology,
Inc., www.lintech.com, 2012.
40

40 Overview

[51] C. Loo, ‘Calculation of the suppression of signals and intermodulation noise when
multiple unequal carriers are amplified by a TWT’, Canadian Electrical Engineering
Journal, vol. 2, pp. 29–​32, 1977.
[52] F. Amoroso, ‘The bandwidth of digital data signals’, IEEE Communications Magazine,
vol. 18, pp. 13–​24, 1980.
[53] E. Buckingham, ‘On physically similar systems; illustrations of the use of dimensional
equations’, Physical Review, vol. 4, pp.  345–376, 1914.
[54] E. B. Wilson Jr., An Introduction to Scientific Research. New York: McGraw-​Hill, 1952.
[55] T. M. Antonsen, Jr. et al., ‘Advances in modeling and simulation of vacuum electronic
devices’, Proceedings of the IEEE, vol. 87, pp. 804–​839, 1999.
[56] R. G. Carter, ‘Computer modelling of microwave tubes  –​a review’, in 2nd IEEE
International Vacuum Electronics Conference 2001, Noordwijk, Netherlands, pp.
393–​396, 2001.
[57] P. B. Johns, ‘The art of modelling’, Electronics and Power, vol. 25, pp. 565–​569, 1979.
41

2 Waveguides

2.1 Introduction

Modern vacuum tubes are power amplifiers and oscillators which require the use
of waveguides or coaxial lines to convey RF power into and out of them. The pur-
pose of this chapter is to provide a summary of those topics which are important
for the design of vacuum tubes. It also provides a foundation for the discussions of
resonators in Chapter 3, and of slow-​wave structures in Chapter 4. Section 2.2 sum-
marises the theory of hollow metal waveguides, and two-​conductor transmission
lines, having uniform cross-​sections. This leads to a discussion of practical coaxial
lines, and rectangular, ridged and circular waveguides in Section 2.3. The properties
of simple discontinuities in rectangular waveguides are considered in Section 2.4
followed by a discussion of matching techniques in Section 2.5. Sections 2.6 and
2.7 examine methods of coupling between waveguides of different cross-​sections
without, and with, changes in the mode of propagation. The final section reviews
the different kinds of vacuum windows which are used in coaxial lines and rect-
angular waveguides. The theory and practice of waveguides and waveguide com-
ponents is covered by many books and the reader is referred to them for detailed
information [1–​5].

2.2 Waveguide Theory

The propagation of electromagnetic waves in a source-​free region, filled with a

uniform material of permittivity ε and permeability µ , is governed by the wave
equations
∂2 E
∇2 E − εµ = 0 (2.1)
∂t 2
and
∂2 H
∇2 H − εµ = 0, (2.2)
∂t 2
where E is the electric field and H is the magnetic field [1]. We shall assume that
the waves are guided in the z direction by conducting boundaries whose shape
42

42 Waveguides

does not vary with z. The vector operator can be decomposed into transverse and
longitudinal parts

∂
∇ = ∇T + zˆ , (2.3)
∂z
where ẑ is the unit vector in the z direction. We also assume the propagation of har-
monic waves with frequency ω in the z direction so that

E = E 0 exp j (ωt − β g z ) (2.4)

and

H = H 0 exp j (ωt − β g z ) , (2.5)

where the vector amplitudes depend only upon the spatial variables normal to the z
axis. Substituting for E and H in (2.1) and (2.2) yields

∇T2 E 0 + (βw2 − β2g ) E 0 = 0 (2.6)

and

∇T2 H 0 + (βw2 − β2g ) H 0 = 0, (2.7)

where

βw2 = εµω 2 (2.8)

is the propagation constant of plane electromagnetic waves in the material filling

the waveguide. We are only interested in the cases where the waveguide is empty, or
filled with a low-​loss dielectric material so that, effectively, µ = µ 0 . The phase
velocity of plane waves is

ω 1
vp0 = = . (2.9)
βw εµ

If the waveguide is empty then v p0 = c, the velocity of light in free space, and the
propagation constant is βw = ω c . The solutions of (2.6) and (2.7) are eigenfunc-
tions which satisfy the boundary conditions on the conducting surfaces. There is
an infinity of such solutions and each has a corresponding eigenvalue βC such that

βC2 = βw2 − β2g (2.10)

or

β2g = βw2 − βC2 . (2.11)

It is evident, from (2.11), that waves can only propagate when bg is real, that is,
when βw2 > βC2 . But, from (2.8), βw is proportional to the frequency of the waves
43

Waveguide Theory 43

and, therefore, waves can propagate only at frequencies above the cut-​off frequency
given by

ωC = βC v p 0 . (2.12)

Equation (2.11) can also be written

ω = ωC2 + β2g v 2p 0 . (2.13)

Every mode in a waveguide obeys (2.13) and has its own cut-​off frequency. The
plot of this curve, shown in normalised form in Figure 2.1, is the dispersion ( ω − β )
diagram of the waveguide mode. We shall see that dispersion diagrams are valuable
tools for understanding the properties of wave guiding systems in general.
The phase velocity of the wave in the waveguide is

ω
vp = (2.14)
βg

and this is represented by the slope of the line joining a point on the dispersion
curve to the origin. The dashed line shows the dispersion curve for plane waves in
unbounded material for which the phase velocity is v p0. It can be seen that for guided
waves v p > v p 0 at all frequencies. The group velocity, which is the velocity of propa-
gation of information along the waveguide, is the slope of the dispersion curve, i.e.

dω
vg = (2.15)
dβ g

Figure 2.1:  Normalised dispersion diagram for waves in a uniform waveguide.

4

44 Waveguides

and we see that vg < v p 0 and that vg → 0 at cut-​off. An important special case, which
is discussed below, occurs when βC = 0 and the dispersion curve is the straight
dashed line through the origin.
When ω < ωC the waveguide is cut off and (2.11) can be written

β g = ± j βw (ωC ω ) − 1,
2
(2.16)

where the negative sign is taken for waves travelling in the positive z direction. The mode
is then an evanescent mode and the amplitude of the wave then decays exponentially
with z. The attenuation of the cut-​off mode in decibels per free-​space wavelength is

  2π  
= −40 π log ( e ) (ωC ω ) − 1.
2
20 log  exp  − β g (2.17)
  βw  

Thus, for example, if ωC ω = 1.05 the attenuation in one free-​space wavelength is

17.5 dB. We shall explore the significance of this result when discussing higher-​
order mode effects in waveguides.
It is convenient to express the vector amplitudes of the waves as the sum of lon-
gitudinal and transverse vectors [3] so that
E 0 = ET + zˆ E z (2.18)
H 0 = HT + zˆ H z ,
where ET and HT only have transverse components. By substituting these expressions
into Maxwell’s equations and making use of vector identities it can be shown that
jβ g j ωµ
ET = − (∇T E z ) + ( zˆ ∧ ∇T H z ) (2.19)
βC2 βC2
and, similarly,

j ωε jβ
HT = − ( zˆ ∧ ∇T E z ) − 2g (∇T H z ) . (2.20)
βC2
βC
This pair of equations shows that the transverse field components can be determined
if the axial components are known. Furthermore they are expressed as the sum of
two terms, one of which depends only on the axial electric field, and the other only
on the axial magnetic field. It can be shown that a waveguide supports an infinite
set of normal modes of propagation which are orthogonal to one another so that
any general configuration of the fields can be expressed as a sum of normal modes.
For every waveguide mode it can be shown that the electric and magnetic fields are
mutually orthogonal. The time-​average power flow can be found by integrating the
longitudinal component of the Poynting vector over a plane normal to the z axis
1
(E ∧ H )z dS.
2 ∫∫
P= (2.21)

Three distinct cases are considered in the sections which follow.

45

Waveguide Theory 45

2.2.1 The Transverse Electric and Magnetic Mode

If βC = 0 then (2.19) and (2.20) permit the existence of non-​zero solutions for the
transverse field components when the longitudinal field components are both zero.
The wave is then a plane electromagnetic wave which satisfies the boundary condi-
tions and whose phase velocity is v p0 . This is the transverse electric and magnetic
(TEM) mode shown by the dashed line in Figure  2.1. It can be shown that the
transverse electric field can be found from the gradient of a scalar electric potential
which is a solution of Laplace’s equation in two dimensions. Since Laplace’s equa-
tion can only have a non-​zero solution in a closed region in which there are at least
two independent boundaries it follows that TEM waves can only be supported by
waveguides comprising two or more conductors, for example coaxial lines. The fact
that the cut-​off frequency is zero means that the waveguide must be able to support
an electric field under DC conditions. This confirms the requirement that it must
have at least two independent conductors. Figure 2.2(a) shows the field pattern of
the TEM wave in a coaxial line.
The properties of each mode of a uniform waveguide can be represented by a
lumped-​element equivalent circuit constructed by representing displacement cur-
rents by capacitors, and conduction currents by inductors. Equivalent circuits, con-
structed in this way, are also valuable aids to understanding the properties of cavity
resonators and slow-​wave structures. The displacement current in a TEM line is in
the transverse direction. The conduction current is normal to the direction of the
tangential magnetic field at the wall and is, therefore, in the longitudinal direction.
Thus the equivalent Tee circuit of a short length dz of the line can be constructed as
shown in Figure 2.2(b) where C and L are, respectively, the capacitance and induc­
tance per unit length. Analysis of this circuit shows that the propagation constant is

β g = ± ω LC . (2.22)

But, from (2.11), we see that β g = βw, the propagation constant for plane waves in
the material filling the space between the conductors, so the phase velocity is
1 1
vp = = = vp0 . (2.23)
εµ LC

Figure 2.2:  Coaxial line TEM mode: (a) field pattern, and (b) equivalent circuit for the mode.
46

46 Waveguides

The line is non-​dispersive because the phase velocity is constant. The characteristic
impedance is defined as
2
V
Zc = (2.24)
2P
where V is the magnitude of the potential difference between the conductors, and P is
the power flow. The value of Zc can be calculated from any pair of: the magnitude of
the voltage; the magnitude of the current; and the power flow, yielding identical results.
In terms of the equivalent circuit

L
Zc = . (2.25)
C
The capacitance per unit length of any TEM line can be found from the solution of
the two-​dimensional Laplace equation, and the inductance per unit length can then
be calculated from (2.22).

2.2.2 Transverse Electric Modes

Modes in which the longitudinal component of the electric field is zero are known
as Transverse Electric (or TE) modes. They are also sometimes known as H modes
because they have a longitudinal component of the magnetic field. The longitudinal
magnetic field is a solution of

∇T2 H z + βC2 H z = 0, (2.26)

subject to the condition that ∇T H z = 0 on the conducting boundaries. Each solu-

tion comprises an eigenfunction H z of the transverse coordinates, which is inde-
pendent of frequency, with an associated eigenvalue βC . Once these are known the
corresponding transverse field components can be found directly from (2.19) and
(2.20). Figure 2.3 shows the arrangement of a rectangular waveguide. The field pat-
terns of some TE modes are shown in Figure 2.4. The transverse components of the

Figure 2.3:  Arrangement of a rectangular waveguide.

47

Waveguide Theory 47

Figure 2.4:  Field patterns of TE modes in a rectangular waveguide

(reproduced, with permission from [5]).
48

48 Waveguides

electric and magnetic fields are always at right angles to one another. The ratio of
their amplitudes is the wave impedance which is found from (2.19) and (2.20) to be
ET ωµ βw
ZTE = = = Zw , (2.27)
HT βg βg

where Zw = µ ε is the wave impedance of plane waves in the material filling the wave-
guide. Equation (2.27) is valid for all TE modes and it has been assumed that the mater-
ial filling the guide, if any, is non-​magnetic. It can be seen that ZTE → Zw as ω → ∞.
It is common to use transmission line methods to solve waveguide problems with
the guide wavelength defined by
2π
λg = . (2.28)
βg

The characteristic impedance is not unique, as it is for TEM waves. This arises
because the electric and magnetic fields are not uniform over the surface of the wave-
guide so that the voltage and current can be defined in different ways [1]. Typically:
• The voltage V is defined as the maximum voltage between opposite sides of the
waveguide.
• The current I is defined as the amplitude of the total axial current in the walls of
the waveguide.
However, the power calculated from the voltage and current defined in this way is
not equal the power flow ( P ) calculated from the Poynting vector. Thus the charac-
teristic impedance can be defined by
V2
ZPV = , (2.29)
2P

2P
ZPI = , (2.30)
I2
or
V
ZVI = (2.31)
I
and these equations give different results. To make matters more complicated, the
relationship between them is not fixed but depends upon the shape of the wave-
guide. For problems where only one propagating mode is concerned (e.g. a change
in the height of a rectangular waveguide) it does not matter which definition is used
because the scaling factors are the same for waveguides having the same shape.
There is no consensus in the literature about which definition of the characteristic
impedance should be used for any given problem.
From Figure 2.4 it can be seen that the simplest TE mode (TE10 ) has displacement
current in the vertical direction. Axial and tangential conduction currents flow in
the walls, corresponding to the transverse and axial components of the magnetic
49

Waveguide Theory 49

Figure 2.5:  Equivalent circuit of the TE10 mode in a rectangular waveguide.

field. (The notation used to describe the modes is explained in Section 2.3.2.) The
equivalent circuit is therefore as shown in Figure 2.5. From analysis of the circuit
for a section of waveguide which is short compared with the wavelength we find
that the dispersion equation is
L2 2
β2g = (ω L1C − 1) . (2.32)
L1

Comparing (2.32) with (2.11) shows that

βw2 = ω 2 L2C (2.33)

and
L2
βC2 = (2.34)
L1

so that the cut-​off frequency is

1
ωC = . (2.35)
L1C

The iterative impedance of the network is

βw L2
Zg = . (2.36)
βg C

It is important to understand that the component values in this equivalent circuit

are not the same as those which would be found from static analysis; though, in
some cases, they can be calculated from the dimensions of the waveguide in a simi-
lar manner. Moreover, they are not unique but depend upon the way in which the
impedance is defined. The properties of the waveguide are completely determined
by any two of the parameters since the third is then fixed.

2.2.3 Transverse Magnetic Modes

In transverse magnetic (TM) modes, also known as E modes, the longitudinal com-
ponent of the magnetic field is zero. The longitudinal electric field is a solution of

∇T2 E z + βC2 E z = 0, (2.37)

50

50 Waveguides

subject to the condition that E z = 0 on the conducting boundaries. The full solution
for each mode can then be obtained as before. Figure 2.6 shows examples of the
field patterns of TM modes in a rectangular waveguide. The tangential electric field
and the normal magnetic field are zero at the walls to satisfy the boundary condi-
tions. The tangential magnetic field at the wall only has transverse components and
the conduction current in the walls is longitudinal. The wave impedance for TM
modes can be derived in the same way as for the TE modes with the result

ET βg βg
ZTM = = = Zw . (2.38)
HT ωε βw

As in the case of TE waves, ZTM → Zw as ω → ∞. We also note from (2.27) and

(2.38) that

ZTE ZTM = Zw2 . (2.39)

The equivalent circuit of the TM11 mode is shown in Figure  2.7. It should be
noted that the lower line represents the wall of the waveguide and the upper line
the centre of the waveguide. From analysis of this circuit the dispersion equation is

C1 2
β2g = (ω LC2 − 1) . (2.40)
C2

Comparing this equation with (2.11) we see that

βw2 = ω 2 LC1 (2.41)

and

C1
βC2 = (2.42)
C2

so that the cut-​off frequency is

1
ωC = . (2.43)
LC2

The iterative impedance is

βg L
Zg = . (2.44)
βw C1

In this case the potential difference between opposite sides of the guide is zero and
it is necessary to define the potential between the wall and the axis. But, because
the axial current flow in the walls is non-​zero, it is more natural to choose ZPI as the
definition of characteristic impedance.
51

Waveguide Theory 51

Figure 2.6:  Field patterns of TM modes in a rectangular waveguide

(reproduced, with permission, from [5]).
52

52 Waveguides

Figure 2.7:  Equivalent circuit of the TM11  mode in a rectangular waveguide.

2.3 Practical Waveguides

Waveguides of all kinds are normally operated in the frequency band between the
lowest two cut-​off frequencies so that only one mode can propagate. When, excep-
tionally, the frequency is such that there are two, or more, modes which are not
cut-​off, then the waveguide is said to be overmoded. We shall see that the effect of
any discontinuity in a waveguide is to scatter power into modes other than the one
which has been excited. It follows that, when an overmoded guide is used it is neces-
sary to take great care to avoid this possibility.
The maximum power which can be handled by a waveguide is determined by elec-
tric breakdown in air, or dielectric materials, and by multipactor discharges in vac-
uum (see Section 18.8). Air-​filled waveguides are normally operated under a small
positive pressure of dry air, or dry nitrogen, to prevent condensation and to exclude
dust, both of which lead to breakdown at lower powers [6, 7]. The power handling
capability outside the vacuum envelope can be increased by pressurisation of the
waveguide (see Section 18.7.2). Increasing the pressure requires the waveguide to be
gas-​tight and reinforced to avoid distortion by the gas pressure. It may also require
pressure vessel rules to be followed. The breakdown voltage can also be increased
by filling the waveguide with an electron attaching gas such as sulphur hexafluoride
(SF6). This gas is chemically inert but when it is decomposed by an electrical dis-
charge the products are corrosive and highly toxic [8]. The type and size of the wave-
guide external to the tube is generally specified by the system in which it is installed.
Dielectric waveguide windows are required to separate sections of waveguide at dif-
ferent pressures, especially between the vacuum inside a tube and the atmospheric,
or higher, pressure outside and these are discussed in Section 2.8. Knowledge of the
properties of waveguides is important for understanding the properties of cavity
resonators and slow-​wave structures (see Chapters 3 and 4). It is also necessary for
the design of the input and output coupling structures of tubes. We shall assume
that waveguides are evacuated or air-​filled unless otherwise stated.

2.3.1 Coaxial Lines
The geometry of a coaxial line is shown in Figure 2.8(a). The space between the
conductors may be filled with dielectric material to locate and support the inner
53

Practical Waveguides 53

Figure 2.8:  Arrangement of a coaxial line with (a) a uniform dielectric, and (b) dielectric
spacers.

conductor. For high powers it is usual to use air-​or vacuum-​spaced lines to mini­
mise the dielectric losses. The inner conductor is then supported by dielectric
spacers as shown in Figure  2.8(b) [9]. Standard coaxial cables in small sizes are
commonly used for the RF input connections of high power amplifiers. The high
power output connection is frequently a coaxial line, either directly, or as a means
of coupling RF power to some other kind of waveguide.
The lowest mode of a coaxial transmission line is the TEM mode in which the
electric field is radial and the magnetic field azimuthal, as shown in Figure 2.2(a).
Since this field pattern must be the same at all frequencies it can be derived by static
analysis [10]. The results are
V0 1
Er = ⋅ ⋅ exp  j (ωt − β g z ) , (2.45)
ln ( b a ) r

where V0 is the magnitude of the RF potential difference between the inner and
outer conductors, and

ε
Hθ = Er . (2.46)
µ

The capacitance per unit length is

2 πε
C= (2.47)
ln ( b a )

and the inductance per unit length is

µ
L= ⋅ ln ( b a ) . (2.48)
2π
54

54 Waveguides

Then the phase velocity is

1 1
vp = = (2.49)
LC εµ

and the characteristic impedance is

L 1 µ
Zc = = ln ( b a ) . (2.50)
C 2π ε

Coaxial lines for high-​power transmission are available with 50 Ω and 75 Ω charac-
teristic impedances. The former represents a compromise between breakdown field
strength and power handling capacity, and the latter is chosen for minimum attenu-
ation. The ratio b a is fixed by the characteristic impedance of the line and the rela-
tive permittivity of the medium between the conductors. It is easy to see from the
equivalent circuit and equations (2.49) and (2.50) what the effects of changes in the
dimensions and dielectric material will be.
The wave on an air-​spaced coaxial line loses power through conduction losses in
the conductors so that its amplitude decays as exp ( −α z ) where α is the loss param-
eter is given by

1  Ra Rb 
α=  + . (2.51)
2Z0 ln ( b a )  a b

Here Ra and Rb are the surface resistances of the inner and outer conductors respec­
tively (see Section 3.3) and Z0 = µ 0 ε 0 is the wave impedance of free space [6]. It
can be seen that the greater part of the loss is contributed by the inner conductor
because the current density is higher there than on the outer conductor. One of the
problems with the use of coaxial lines at high power levels is the need to remove heat
from the inner conductor. Large coaxial lines for high-​power RF transmission may
use copper for the inner conductor and aluminium for the outer conductor to save
weight. The use of aluminium as the outer conductor increases the transmission
losses by about 10%. The average power being carried on a coaxial line is given by

1 V02
P= . (2.52)
2 Zc

Now from (2.45) the magnitude of the RF electric field is

V0 1
Er = ⋅ , (2.53)
ln ( b a ) r

which is maximum when r = a . Substituting in (2.52) from (2.50) and (2.53) gives
the relationship between the power and the maximum electric field
πa 2 Ea2
Pmax = ln ( b a ) . (2.54)
Zw
5

Practical Waveguides 55

It is customary to assume that the peak electric field in an air-​filled line is the break-
down electric field strength in dry air at standard pressure and temperature (3 MV/​m).
It must be remembered that this figure provides no margin of safety, and that allow-
ances must be made for changes with altitude, humidity, and the presence of dust par-
ticles in the air. For this reason the breakdown voltage is often an order of magnitude,
and the maximum power two orders of magnitude, below the theoretical limit [11].
When the line is evacuated, breakdown is caused by a multipactor discharge.
It should be noted that this does not apply only within the vacuum envelope of
the tube but is also the cause of breakdown in waveguides used in space [12]. The
maximum power can be increased by the use of anti-​multipactor coatings and
by the application of a DC bias between the inner and outer conductors [13, 14].
Multipactor discharges are discussed in Section 18.8.
Higher-​order TE and TM modes can propagate in a coaxial line at higher fre-
quencies. In general they are to be avoided because mode conversion from the TEM
mode to TE or TM modes represents a source of power loss. The cut-​off frequen-
cies for these modes are found from (2.26) and (2.37). In polar co-​ordinates the
transverse vector operator is
1 ∂  ∂  1 ∂2
∇T2 = r  + . (2.55)
r ∂r  ∂r  r 2 ∂θ 2

For TE modes the variables can be separated by writing

H z ( r, θ ) = R ( r ) cos ( mθ ) , (2.56)

where m = 0,1, 2 . Substituting in (2.26) we find that

1 ∂  ∂R   2 m 2 
r  +  βC − 2  R = 0. (2.57)
r ∂r  ∂r   r

This form of Bessel’s equation has the general solution

R = AJ m (βC r ) + BYm (βC r ) , (2.58)

where J m andYm are Bessel functions of the first and second kinds [15]. The value of βC
is found by applying the boundary conditions ∂R ∂r = 0 when r = a and r = b so that

J m′ (βC a )Ym′ (βC b ) − J m′ (βC b )Ym′ (βC a ) = 0, (2.59)

where the prime represents differentiation with respect to the argument. This equa-
tion has a doubly infinite set of solutions βC = β m,n where n = 1, 2, is the number
of zeroes of R between the conductors. The corresponding modes are denoted by
TE m n . Figures  2.9 and 2.10 show examples of the field patterns of TE and TM
modes in a coaxial line. The lowest mode is the TE11 mode whose cut-​off wavelength
is given very nearly by
2π
λC = = π ( a + b ) (2.60)
βC
56

56 Waveguides

Figure 2.9:  Field patterns of TE modes is coaxial waveguide

(reproduced, with permission, from [5]).
57

Practical Waveguides 57

Figure 2.10:  Field patterns of TM modes is coaxial waveguide

(reproduced, with permission, from [5]).
58

58 Waveguides

over the range 1 < a b < 4. Thus cut-​off occurs when there is a standing wave in
the azimuthal direction. The cut-​off wavelengths for other TE modes and for TM
modes can be found from (2.59). Detailed information about the higher-order
modes in a coaxial line, including diagrams of the field patterns and the equations
for the field components, can be found in [1, 5].

2.3.2 Rectangular Waveguides

Rectangular waveguides are the usually used for transmission of RF power at 300
MHz and above where their power handling and attenuation are better than those
of coaxial lines [6]. Figure 2.3 shows the arrangement of a rectangular waveguide in
which a and b are the internal width and height. Standard rectangular waveguides
have aspect ratios that are close to 2:1 but reduced height waveguides are sometimes
used for special purposes.
The cut-​off frequencies for the modes are found from (2.26) and (2.37). In rectan-
gular Cartesian co-​ordinates the transverse vector operator is
∂2 ∂2
∇T2 = + . (2.61)
∂x 2 ∂y 2

For TE modes the variables can be separated by writing

 mπ x   nπ y 
H z ( x, y ) = H 0 z cos  cos  , (2.62)
 a   b 

where m, n = 0,1, 2. It should be noted that this choice automatically satisfies
the boundary conditions on the walls of the waveguide. Substituting in (2.26) we
find that
2 2
 mπ   n π 
βC2 =  + . (2.63)
 a   b 

The modes are denoted by TE m n and the lowest mode when a > b is the TE10 mode in
which the electric field is normal to the broad wall and the magnetic field is parallel to
the broad wall as shown in Figure 2.4. This field pattern can be generated by superim-
posing plane TEM waves whose directions of propagation are at equal positive and
negative angles with respect to the z axis [16]. At cut-​off this angle is π 2 so that there is
a standing wave across the guide and no power propagates along it. The equations for
the components of the electric and magnetic fields can be found by substituting (2.62)
in (2.19) and (2.20). Table 2.1 shows the cut-​off frequencies of the lowest modes of a
waveguide for which a = 2b, normalised to the frequency of the TE10 mode.
The power flow in the guide is found by integrating the z component of the
Poynting vector across a transverse plane with the result
2
ab β g E x
P= ⋅ ⋅ , (2.64)
4 βw Zw
59

Practical Waveguides 59

Table 2.1:  Relative cut-​off frequencies of TE modes in

a rectangular waveguide (a = 2b)

m n fc

1 0 1.000
0 1 2.000
2 0 2.000
1 1 2.236
2 1 2.828
3 0 3.000
3 1 3.606
0 2 4.000

where |Ex| is the magnitude of the electric field at the centre of the guide. The char-
acteristic impedance is conveniently defined by (2.29) where V = bE x. Substituting
for P from (2.64) gives
2b βw
ZPV = ⋅ Zw , (2.65)
a βg

which can be seen to have the same form as (2.36). The impedance of an empty
waveguide having a 2:1 aspect ratio ranges from around 630Ω at the bottom of the
useful frequency band to 440Ω at the top. Assuming that b = a/​2 the attenuation
constant of an empty rectangular waveguide is found to be [1]

2Rs [1 + ( ωC ω )2 ]
α= , (2.66)
a µ 0 ε 0 [1 − ( ωC ω )2 ]
0.5

where Rs is the surface resistance (see Section 3.3). The theoretical maximum power
in an air, or gas, filled waveguide is
E max
2
ab
Pmax = 1 − ( ωC ω )2 . (2.67)
4 µ0 ε0

where E max is the breakdown field.

The values of the components in the equivalent circuit of the TE10 mode of an
empty waveguide which are consistent with the definition of impedance in terms of
power and voltage ( ZPV ) can be found by comparing (2.65) with (2.36). The result is

L2 2b µ 0 (2.68)
= .
C a ε0

Then from (2.33), (2.34) and (2.68)

a
C= ε0 (2.69)
2b
60

60 Waveguides

2ab
L1 = µ0 (2.70)
π2

2b
L2 = µ 0 . (2.71)
a
We recall that it is not possible to determine these component values using static
field analysis as in the case of the TEM mode. Instead they are chosen so that the
network has the correct values of the cut-​off frequency, the high-​frequency asymp-
tote for the phase velocity and the characteristic impedance [16].
Higher-​order waveguide modes determine the useful bandwidth of rectangu-
lar waveguide. For the conventional choice of a 2 to 1 aspect ratio in the trans-
verse dimensions, the first higher-​order mode is the TE20 mode (H 20 mode), for
which the cut-​off frequency is just twice the cut-​off frequency of the TE10 mode
(see Table 2.1). The useable frequency band of a rectangular waveguide is usu-
ally taken to be from 1.25 fc to 1.90 fc. Reference to Figure 2.1 shows that below
the minimum frequency the dispersion of the waveguide increases rapidly. At
the maximum frequency the attenuation of the next higher-order mode is found
from (2.17) to be 18 dB in one free-​space wavelength. The wavelength of the
propagating TE10 mode at a frequency 1.90 fc is 1.18 times the free-​space wave-
length so that the attenuation of the TE20 mode in one guide wavelength of the
TE10 mode is 21 dB. Thus any higher-​order mode fields generated by discontinui-
ties in the waveguide are of short range. The equations for the field components
of the TE modes are given in [1, 5]. Figure 2.4 shows examples of the field pat-
terns TE modes.
The cut-​off frequencies for the TM modes can be determined from (2.37) in a
similar manner by assuming that

 mπ x   nπ y 
E z ( x, y ) = E0 sin  sin  . (2.72)
 a   b 

This choice ensures that the boundary conditions are satisfied. The cut-​off condition
is again given by (2.63) but we note that modes in which either m or n is zero cannot
exist because then E z = 0 everywhere. Thus the lowest TM mode is the TM11 mode.
The equations for the field components of these modes are to be found in [1, 5].
Figure 2.6 shows examples of the field patterns of TM modes. The dimensions and
properties of standard rectangular waveguides can be found in reference books [4]
and are also available from manufacturers [17].

2.3.3 Ridged Waveguides

In the previous section we saw that the useful bandwidth of a rectangular waveguide
is limited to around 1.5:1. For some purposes this is inconvenient. In order to increase
the useful bandwidth it is necessary to increase the separation between the cut-​off
frequencies of the TE10 and TE20 modes. The equivalent circuit shown in Figure 2.5
61

Practical Waveguides 61

suggests a possible way. If a ridge is added to the centre of one, or both, of the
broad walls of the waveguide, as shown in Figure 2.11, then the effect on the TE10
mode is to increase the capacitance C. As a result, the cut-​off frequency of this
mode is depressed. The TE20 mode, however, is little affected because the ridges are
in regions of weak electric field for that mode. The price paid for the increase in
bandwidth is a reduction in power handling capability because of the concentration
of electric field produced by the ridges. At low power levels coaxial lines provide
a more compact means of broad-​band signal transmission, but ridged waveguides
are useful where a combination of broad-​bandwidth and moderately high power is
required.
Because the shunt capacitance is increased by the addition of the ridges the char-
acteristic impedance is reduced, as can be seen from (2.36). This is useful for match-
ing between coaxial lines and waveguides because the characteristic impedance is
intermediate between those of the other two types of waveguide. The notation for
modes in ridged waveguide is the same as that used for rectangular waveguides
[2, 4, 18, 19]. Detailed information about ridged waveguides is to be found in [18].
The dimensions and other properties of standard WRD series double-​ridged wave-
guides are available from manufacturers [20].

Figure 2.11:  (a) Single ridged waveguide, (b) double ridged waveguide and

(c) the equivalent circuit.
62

62 Waveguides

The cut-​off frequency of the TE10 mode of a ridged waveguide can be calculated
by recalling that, at cut-​off, there is a standing TEM wave across the guide [19]. The
problem may therefore be considered as comprising two parallel plate transmission
lines having differing characteristic impedances as shown in Figure  2.11(c). It is
easy to show that the capacitance and inductance per unit length of a parallel plate
transmission line having unit width and separation b are
ε0
C= (2.73)
b
and

L = µ 0 b, (2.74)

where fringing fields have been ignored. Thus the phase velocity and characteristic
admittance are
1
vp = =c (2.75)
ε0 µ0

and

1 ε0
Yg = . (2.76)
b µ0

The admittance presented at the step by the outer region of the guide is found using
the equation for the transformation of admittance on a transmission line

YL + jYg1 tan βw z
Y1 = ⋅Yg1 , (2.77)
Yg1 + jYL tan βw z

where βw = ω c . Setting z = ( a − s ) 2 and letting the terminating admittanceYL tend

to infinity gives

w ε0
Y1 = − j cot (βw ( a − s ) 2 ) . (2.78)
b µ0

The admittance presented at the step by the inner region of the guide is found, simi-
larly, by setting z = s 2 and YL = 0

w ε0
Y2 = j tan (βw s 2 ) . (2.79)
d µ0

The effect of the fringing electric field at the step is represented in the equiva-
lent circuit by the lumped susceptance B. If the height of the waveguide is small
compared with the free-​space wavelength, and the fringing field of the step does
not extend to the outer wall of the waveguide, the determination of B may be
treated as a quasi-​static problem. The field between a pair of parallel plates
63

Practical Waveguides 63

with a step can be determined by conformal mapping, with the result shown
in Figure 2.12. The fringing field can be represented by a lumped capacitance.
This is the difference between the capacitance calculated from the field pattern
shown and that calculated if it is assumed that all the field lines go straight
across from one plate to the other. The resulting shunt susceptance for a double-​
ridged guide is [21]:

βw ε 0  d b   1 + d b   4d b  
B= ⋅  +  ln   − 2 ln  . (2.80)
2π µ 0  b d   1 − d b   1 − d 2 b2  

At cut-​off βC is the value of βw for which

Y1 + Y2 + jB = 0. (2.81)

Figure  2.13 shows a plot of λ c a against s a with d b as a parameter when

b a = 0.5. A single-​ridged guide can be considered as the part of a double-​ridged
guide above the mid plane. The normalised cut-​off wavelength is therefore the
same as that of a double-​ridged guide of double the height. The same result can be
obtained by doubling the susceptance given by (2.80).
Hopfer calculated the characteristic impedance of a ridged waveguide by first
finding the transverse voltage distribution at cut-​off, using transmission line theory,
and then computing the stored electric energy including that in the fringing field [18].
The power flow in the waveguide was computed by multiplying the stored energy
by the group velocity, and the characteristic impedance found from the power flow
and the voltage across the centre of the waveguide. More accurate solutions have been

Figure 2.12:  Field pattern around a step in a parallel plate transmission line (d/b = 0.5)
(copyright 1946 IEE; reproduced, with permission, from [21]).
64

64 Waveguides

Figure 2.13:  Normalised cut-​off wavelength of a double-​ridged waveguide (b/a = 0.5)

(copyright 1955 IEEE; reproduced, with permission, from [18]).

obtained using variational methods [22, 23] and finite element methods [24, 25].
Montgomery found that, for the dimensions he examined, Hopfer’s value of the
cut-​off wavelength was accurate to better than 1%, and his figure for the impedance
accurate to better than 5%. Approximate closed form expressions for the param-
eters of ridged waveguides are given in [26]. The properties of ridged waveguides
can be explored using Worksheet 2.1.

2.3.4 Circular Waveguides

The fourth type of waveguide which is important in vacuum tube engineering is
the circular waveguide. The field solutions are obtained using cylindrical polar
coordinates so that, for TE modes, the transverse variation of the axial mag-
netic field is given by (2.56) in the same manner as for a coaxial line. The radial
variation is given by (2.58) but with the difference that the coefficient B must
be zero because the Bessel function of the second kind tends to infinity on the
axis. Thus

R = AJ m (βC r ) . (2.82)
65

Waveguide Discontinuities 65

The radial boundary condition requires that

J m′ (βC a ) = 0, (2.83)

where a is the internal radius of the waveguide. The solutions to this equation are
given in [5]. The notation used for the modes is the same as that in coaxial lines.
Figure 2.14 shows the field patterns for examples of TE modes [5]. The mode with
the lowest cut-​off frequency in a circular waveguide is the TE11 ( H11 ) mode. This
mode resembles the TE10 mode in a rectangular waveguide, and it is used in pill-​box
windows (see Section 2.8.2). The TE modes, especially the higher-order ones, are
important because of their use in gyrotrons (see Chapter 17). Modes in which the
plane of polarisation is rotating can be formed by combining modes with planes of
polarisation at right angles to one another with a phase difference of 90° between
them. For example the azimuthal variation of a rotating TE11 mode is given by

cos φ cos ωt + sin φ sin ωt = cos ( φ − ωt ) , (2.84)

so that the direction of the electric field is rotating with angular velocity ω .
The TM modes in circular waveguides are important because the longitudinal
component of the electric field is non-​zero and can be used to interact with a longi-
tudinal stream of electrons. These modes are the foundation for many of the cavity
resonators and slow-​wave structures which are discussed in Chapters 3 and 4. These
are used in slow-​wave tubes such as klystrons and TWTs. The TM modes satisfy
(2.37), subject to the boundary condition E z = 0 when r = a . This requires that

J m (βC a ) = 0. (2.85)

The solutions to this equation are tabulated in [15] and examples of the field pat-
terns are shown in Figure 2.15 [5]. The characteristic impedances of circular wave-
guide modes can be calculated in the same way as those for rectangular waveguides
(see Table 2.1).

2.3.5 Summary of Waveguide Impedances

Table 2.2 shows the characteristic impedances for the lowest modes in some com-
mon waveguides.

2.4 Waveguide Discontinuities

Coupling between waveguides involves some kind of discontinuity in the wave-​

guiding structure leading to reflection of microwave power. This section examines
the cause of this reflection and considers ways of modelling a few examples of dis-
continuities in rectangular waveguides. A general discussion of the problem of the
junctions between two waveguides propagating TE modes is given by Farmer [27].
It is shown that, if the field patterns in the two guides are similar to one another, it
6

66 Waveguides

Figure 2.14:  Field patterns for TE modes in circular waveguide

(reproduced, with permission, from [5]).
67

Waveguide Discontinuities 67

Figure 2.15:  Field patterns for TM modes in circular waveguide

(reproduced, with permission, from [5]).
68

68 Waveguides

Table 2.2:  Characteristic impedances of waveguide modes

Waveguide Mode Z PV ZVI Z PI

Coaxial TEM ln (b a ) ln (b a ) ln (b a )
Zw Zw Zw
2π 2π 2π
Rectangular TE 01 b π b π2 b
2 ZTE ⋅ ZTE ⋅ ZTE
a 2 a 8 a
TM11 π 2  ab 
  ZTM
64  a 2 + b 2 

Circular TE11 2.018 ZTE 1.376 ZTE 0.938 ZTE

TM 01 0.0796 ZTM

is possible to represent the waveguide junction by an abrupt transition between two

transmission lines, together with a lumped susceptance at the transition. In order
for this to be possible it is necessary for the characteristic impedances of the two
transmission lines to be in the ratio

Zg ′ λ g ′ ∫∫ Ed′ 2 dS ′ ∫∫ Ed2 dA (2.86)

= ⋅ ⋅ ,
Zg λ g ∫∫ Ed2 dS ∫∫ Ed′ 2 dA

where the unprimed and primed variables refer to the two waveguides, Ed is the
transverse electric field of the dominant mode in a waveguide, S is the cross-​section
of a waveguide and A is the area of the aperture coupling the two guides. This def-
inition is not necessarily consistent with the ratio calculated using any of the usual
definitions of impedance, though it may be so in certain cases.
Figure 2.16 shows a thin iris which partially obstructs the width of a waveguide.
At the plane of the iris the fields must satisfy the boundary conditions. The electric
field distribution at the plane of the iris is approximately a compressed TE10 mode,
as shown by the solid line in Figure 2.17(a) with E y = 0 outside the gap. The field in
the waveguide adjacent to the diaphragm must match that at the diaphragm at every
point. The field of the incident TE10 wave shown by the solid line in Figure 2.17(b)
plainly does not obey this requirement. In order to match the boundary condi-
tions correctly it is necessary for higher-order modes to be excited. Because the
diaphragm is symmetrical the next mode which can be excited is the TE30 mode
shown by the dashed line in Figure 2.17(b). When these two modes are superim-
posed with amplitudes 0.65 and 0.35 the result is the dashed line in Figure 2.17(b).
This is much closer to satisfying the boundary conditions than the original TE10
mode. The addition of further higher-order modes (TE50 , TE70 , etc.) produces a
better match, but an infinite series is needed to match the fields exactly at the plane
of the diaphragm [28]. Within the normal working band of the waveguide all these
higher-order modes are cut off and their fields are confined to a short distance
69

Waveguide Discontinuities 69

Figure 2.16:  Symmetrical inductive iris in a rectangular waveguide.

Figure 2.17:  Electric field amplitudes at an inductive iris in rectangular waveguide: (a) E y in

the plane of the iris (A) and a two mode approximation (B); (b) transverse variation of E y
for the TE10 mode (C) and the TE30 mode (D).
70

70 Waveguides

either side of the diaphragm. These evanescent modes store energy and their effect
can therefore be represented by a lumped reactance. Bearing in mind the earlier
discussion of equivalent circuits, we note that the iris has a greater effect on the
conduction current than on the displacement current. The discontinuity can there-
fore be represented by a lumped inductance. It is also possible to create an inductive
discontinuity by placing one or two posts between the broad walls of the waveguide.
If the iris changes the height of the guide, instead of its width, as shown in
Figure 2.18, a different set of higher-order modes is excited. Once again these are
normally cut off. The discontinuity mainly affects the electric field distribution so
the shunt reactance is capacitive. An alternative way of introducing a shunt capaci-
tance is by inserting a screw through the broad wall of the waveguide. This arrange-
ment is useful for matching purposes because the capacitance is easily adjusted,
but is not suitable for use at high power levels. In fact capacitive obstacles are best
avoided in high-​power waveguides because they introduce concentrations of elec-
tric field and reduce the peak power which can be carried. Formulae and graphs for
the susceptances of a wide range of capacitive and inductive discontinuities in rec-
tangular, circular, and coaxial waveguides are given in [5, 21]. These susceptances
depend on both the dimensions and the guide wavelength. Although this data is
useful for design purposes it has been superseded to some extent by the availability
of computer codes for modelling problems in electromagnetics. Additional insight
is provided by equivalent circuit models as illustrated in the sections which follow.

2.4.1 Height Step in a Rectangular Waveguide

Figure 2.19 shows a symmetrical height step in a rectangular waveguide. This can be
modelled by the junction between two transmission lines having the same propagation
constant but differing characteristic impedances, together with a lumped susceptance
at the junction. The related problem of a step change in the height of a parallel plate
transmission line was considered in Section 2.3.3. The electrostatic field in the vicin-
ity of the junction is shown in Figure 2.12. It can be seen that the field only differs
from that in the unperturbed transmission lines over a small region close to the junc-
tion. If the higher-order modes excited by a waveguide step are strongly cut off, their
fields are, similarly, confined to a small region close to the junction. Therefore, in the
first approximation, they can be regarded as quasi-​static fields which are solutions to
Laplace’s equation on the mid-​plane of the waveguide. This enables the susceptance
of the waveguide step to be deduced directly from (2.80). If the width of the transmis-
sion line is a then, from (2.76), the characteristic admittance of the transmission line is
a
Yg = Y0 (2.87)
b
and the susceptance of a symmetrical transmission line step can be written

βw b  d b   1 + d b   4d b  
B= Yg ⋅  +  ln   − 2 ln  . (2.88)
2π  b d   1 − d b   1 − d 2 b2  
71

Waveguide Discontinuities 71

Figure 2.18:  Symmetrical capacitive iris in a rectangular waveguide.

Figure 2.19:  Symmetrical height step in a rectangular waveguide.

The susceptance of the symmetrical waveguide step is found by substituting β g for

βw and YPV for Yg [21]. The result, in normalised form, is

B λg  d b  1+ d b  4d b 
⋅ =  +  ln − 2 ln  . (2.89)
YPV b  b d   1 − d b   1 − d 2 b2 

We note that the expression on the right-​hand side of this equation depends only
on the dimensions of the step and is independent of frequency. Close to cut-​off
this expression is exact. A more detailed analysis includes additional terms on the
right-​hand side which are functions of b λ g [5]. However, b λ g < 0.4 throughout
the working band of a standard rectangular waveguide, and the error introduced
by ignoring this correction is less than 7.5% for all values of d b. If the change in
height is achieved by a step in one broad wall (an asymmetrical step) then the sus-
ceptance is twice that given by (2.89).
To develop a simple equivalent circuit model let us suppose that the reactive
effects of the higher-order modes excited by the step can be represented by a short
length δ of a waveguide having height d where δ depends only upon the geometry of
the problem. The step is therefore represented by the equivalent circuit in Figure 2.5
with shunt capacitance Cs and shunt inductance and Ls . Since we expect δ to be
72

72 Waveguides

small we will neglect the effect of the series inductance L2. The susceptance of this
network is

 1 
B = ω Cs δ 1 − 2 . (2.90)
 ω LsCs 

Both sections of the waveguide have the same width and, therefore, both have the
same cut-​off frequency. Hence

Cs δ 2 β2

B= (ω − ωc2 ) = cCs δ g , (2.91)
ω βw

where βw = ω c and c is the velocity of light. The susceptance added by the step is
found by subtracting the susceptance of a section of waveguide having the same
length for which d = b and shunt capacitance is C so that

C  β2g
Bs = cC δ  s − 1 . (2.92)
C  βw

It is reasonable to suppose that the capacitances are inversely proportional to the

heights of the guides, so that

 b  β2g
Bs = cC δ  − 1 (2.93)
 d  βw

and, from (2.36), the characteristic admittance of the waveguide can be written

βg
Yg = cC . (2.94)
βw

Thus, the normalised susceptance of the step is

Bs λ g b δ
⋅ = 2 π  − 1 . (2.95)
Yg b d  b

This equation has the same form as (2.89) so that δ b can be calculated as a func-
tion of d b, as shown in Figure 2.20. It can be seen that δ b ≤ 0.125 confirming the
initial assumption that this quantity is small. We note that the shunt capacitance
and inductance in this model depend only upon the dimensions, so that the depend-
ence of the susceptance on frequency arises from the topology of the network.

2.4.2 Capacitive Iris in a Rectangular Waveguide

The susceptance of the symmetric capacitive iris shown in Figure  2.18 can be
deduced in a similar manner. The electrostatic field around a capacitive iris in a
parallel plate transmission line, obtained by conformal mapping, is shown in
73

Waveguide Discontinuities 73

Figure 2.20:  Normalised susceptance of a symmetrical capacitive step in a rectangular

waveguide (with the susceptance of a symmetrical capacitive iris for comparison), and the
normalised length of an equivalent section of reduced height waveguide.

Figure 2.21:  Field pattern around a thin iris in a parallel-​plate transmission line ( d b = 0.5)

(copyright IEE 1946; reproduced, with permission, from [21]).

Figure 2.21. The normalised susceptance of the waveguide iris is given, in the first
approximation, by [21]

Bi λ g   πd   (2.96)
⋅ = − 4 ln  sin    .
Yg b   2b  

Comparison between Figures  2.12 and 2.21 shows that the fringing fields of the
step, and of the iris, are very similar to one another. Thus the susceptance of a
symmetrical iris is approximately twice that of a symmetrical step having the same
height. The comparison between the two results in Figure 2.20 shows that result
74

74 Waveguides

for the iris can be used as an approximation to that of a step. More exact formulae
for the susceptance of a capacitive iris in a waveguide are given in [5]. The suscep-
tance of an iris having finite thickness can be estimated using the equivalent circuit
method in the previous section by adding the thickness of the iris to δ . The proper-
ties of a capacitive iris can be explored using Worksheet 2.2.

2.4.3 Inductive Iris in a Rectangular Waveguide

The normalised susceptance of the symmetrical inductive iris shown in Figure 2.16
has been found, to a first approximation, to be [21, 30]

B a  πw 
⋅ = − cot 2  . (2.97)
Yg λ g  2a 

A more accurate expression is given by Marcuvitz [5]. This iris can also be modelled
using an equivalent circuit. In this case the shunt susceptance is

βi2
B = cCi δ , (2.98)
βw

where Ci is the capacitance per unit length of a waveguide of height b and width w
and βi = π w is the cut-​off propagation constant of the iris. As before, we subtract
the shunt susceptance of the length δ of the waveguide to give the susceptance
added by the iris

β2g  Ci βi2 
Bi = cC δ −1 . (2.99)
βw  C β2g 

Normalising to the characteristic admittance of the waveguide, and assuming that

Ci C = w a , gives

Bi δ   2 π2   π2   (2.100)
=  w  βw − 2  − a  βw2 − 2   .
Yg β g a   w   a 

By neglecting the series inductance we have implicitly assumed that the waveguide
is close to cut-​off so that βw → π a  and

Bi a π δw a (2.101)
⋅ = ⋅  − .
Yg λ g 2 a  a w 

We note that the susceptance is negative and, therefore, inductive. Now the fraction
of the longitudinal current in the waveguide that is intercepted by the aperture is

 πw 
ka = sin  . (2.102)
 2a 
75

Matching Techniques 75

But the representation of the iris by a shunt susceptance across an equivalent trans-
mission line means that susceptance in (2.101) must be increased by dividing it by
ka. Hence, finally,

Bi a π δw a  πw 
⋅ = ⋅  −  cosec  . (2.103)

Yg λ g 2 a a w   2a 

The normalised susceptance depends only on the dimensions of the iris and the
value of δ can be found by comparison with (2.97). It is found that, to a good
approximation

δ 4   w 
2

= 2 1 −    . (2.104)
a π   a 

The properties of an inductive iris can be explored using Worksheet 2.3. The related
problem of a step change in the width of a waveguide can be treated in the same
way but it is no longer the case that the dominant electric field patterns in the two
waveguides are similar and the use of a two-​port equivalent circuit model is only
approximate [5, 27].

2.5 Matching Techniques

The input and output connections of a vacuum tube amplifier must be well-​matched
at all frequencies within the operating band of the tube. This is important because
reflected power:
• Reduces the useful power output of the tube.
• Leads the periodic variation with frequency in the gain, and output power,
because of the effects of multiple reflections.
• Increases the risk of voltage breakdown, and failure of windows, through the
increased peak electric field in a standing wave.
• Can cause a tube to become unstable.
It may also be important to consider the effects of mismatches at harmonic fre-
quencies since these can also affect the behaviour of the tube. These issues are dis-
cussed further in the chapters dealing with particular types of tube. There are many
well-​established matching techniques in microwave engineering [3]. The basic meth-
ods which are used singly, or in combination, are reviewed briefly below.

2.5.1 Stub Matching

Reactive elements (stubs) can be arranged so that the signal reflected by them more
or less cancels out the reflection from the component to be matched. We recall that,
when a waveguide having characteristic admittance Yg is terminated by a load YL
76

76 Waveguides

the impedance perceived at a plane which is closer to the source by a distance l is

given by

YL ′ YL + jYg tan (β g l ) (2.105)

= ,
Yg Yg + jYL tan (β g l )

where β g is the propagation constant in the waveguide. This equation has been
quoted in terms of admittances because the matching stubs normally used in a
waveguide present shunt reactances, as in the examples in the previous section.
Because the losses in waveguides are small it is usual to treat them as lossless when
carrying out matching calculations so that the characteristic impedance is real. It is
often useful to visualise the transformation of impedances using a Smith chart, and
Vector Network Analysers offer this as a display option.
The simplest matching technique uses (2.105) to find a plane in the waveguide at
which the real part of YL ′ is equal to Yg . The insertion of a stub at that plane, whose
susceptance is equal and opposite to the imaginary part of YL ′, cancels the susceptance
of the load. This ensures that the combination of load and stub appears matched at
all planes which are closer to the source. However, because β g depends on frequency,
the electrical length (β g l ) of the section of waveguide between the load and the stub
is not constant. Thus, this technique can only give a good match at a single frequency
(or, strictly speaking, the set of frequencies at which tan (β g l ) has the same value).

2.5.2 Broad-​band Matching

To increase the band of frequencies over which there is a good match it is neces-
sary to use a number of stubs at intervals along the waveguide. If the stubs are
at least a quarter of a guide wavelength apart the evanescent higher-order modes
excited by one stub have negligible effects at a neighbouring stub. It is therefore
possible to treat the stubs as lumped reactances. The properties of a system com-
prising a number of matching elements can be illustrated simply by considering
the arrangement shown in Figure 2.22. Four reactive elements are equally spaced
on a transmission line with a matched termination. The reflection coefficients
are ρ1 and ρ2 as shown, and the electrical length of the line between them is φ.
We will assume that the reflection coefficients are small so that the amplitude of
the incident wave is hardly affected by the reflections. Now suppose that D is the
obstacle to be matched and A, B and C are the matching elements. The apparent
reflection coefficient at A is

ρ = ρ1 + ρ2 e −2 j φ + ρ2 e −4 j φ + ρ1e −6 j φ
= 2ρ1e −3 j φ cos 3φ + 2ρ2 e −3 j φ cos 3φ. (2.106)

In practical terms it is usually the magnitude of the reflection which is significant,

that is

ρ = 8ρ1 cos3 φ + 2 (ρ2 − 3ρ1 ) cos φ . (2.107)

7

Matching Techniques 77

Figure 2.22:  Arrangement of four regularly-​spaced discontinuities on a transmission line.

which may be written

ρ = 8ρ1x3 + 2 (ρ2 − 3ρ1 ) x , (2.108)

where, for convenience, x = cos φ. Now φ is a function of frequency and, therefore,

so is x. Equation (2.108) describes the variation of the reflection coefficient with
frequency. Clearly the shape of this cubic curve can be changed by making different
choices of ρ2. One possibility is to set ρ2 = 3ρ1. The result is

ρ = 8ρ1x3 , (2.109)

shown as the continuous curve in Figure 2.23. The curve is normalised to the reflec-
tion coefficient when x = 1. A polynomial, in which only the coefficient of the high-
est power of x is non-​zero, has the property that the highest possible number of
derivatives of the function is zero at the origin. The curve is therefore as flat as
possible close to the origin. A match having this frequency dependence is called a
maximally flat or Butterworth response. It is the best possible match over a narrow
band using a given number of matching elements.
For a given maximum reflection coefficient ρm the bandwidth ( ∆x ) is the distance
between the two points on the curve at which ρ = ρm so that, from (2.109)
1 1
 ρ 3 ρ 3
∆x = 2 x = 2  m  =  m  . (2.110)
 8ρ1   ρ1 

This shows that, as one would expect, it is possible to achieve a greater bandwidth
by accepting a poorer match. For broad-​band matching it is better to sacrifice
some of the excellence of the match close to the origin in order to gain greater
bandwidth. The broken curve in Figure 2.23 shows such a possibility. There is, of
course, an infinite number of possible polynomials but of special interest are the
Tchebychev1 polynomials [3, 31]. The first six polynomials are listed in Table 2.3.

Other spellings of this name in the Latin alphabet include Chebyshev, Tchebycheff, Teschebyscheff, and
1

Tchebichef.
78

78 Waveguides

Table 2.3:  Tchebychev polynomials

n Tn ( x )

0 1
1 x
2 2x2 − 1
3 4 x 3 − 3x
4 8x 4 − 8x 2 + 1
5 16 x5 − 20 x3 + 5x

Figure 2.23:  Third-​order Butterworth (maximally flat) and Tchebychev (equal-​ripple)

reflection characteristics.

These polynomials have the property that, over the range −1 < x < 1, the magni-
tude is always less than or equal to 1. The Tchebychev polynomial has the broadest
bandwidth of all possible polynomials of a given order and specified in-​band rip-
ple. To design a broad band match using Tchebychev polynomials we introduce the
scaling factors ρm and xm. The third-order polynomial can then be written
3
 ρ  x  x
 ρ  = 4  x  − 3  x  . (2.111)
m m m

The response curve has an in-​band ripple equal to ρm as shown in Figure 2.23.

Equating coefficients of x3 in (2.108) and (2.111) we obtain

4ρm xm3 = 8ρ1 (2.112)

so that

ρm
xm3 = . (2.113)
2ρ1
79

Coupling without Change of Mode 79

Now ρ1 is the reflection coefficient of the termination to be matched and ρm is the

maximum acceptable in-​band match. Therefore, from (2.113), the bandwidth which
can be achieved in this way using three matching elements is
1 1
 ρ 3 2
ρ 3
∆x = 2 xm = 2  m  = 2 3  m  . (2.114)
 2ρ1   ρ1 

Comparing (2.114) with (2.110) we see that, when three matching elements are
used, the bandwidth of the Tchebychev match is wider than that of the Butterworth
match by a factor of 22 3 = 1.59 .
To complete the design of the Tchebychev matching network the coefficients of
x in (2.108) and (2.111) are equated to give

2 (ρ2 − 3ρ1 ) = − 3ρm xm (2.115)

so that
ρ2 = 3ρ1 − 3ρm 2 xm . (2.116)

Since ρ1, ρm and xm are all known ρ2 can be calculated.

2.5.3 Stepped Impedance Transformers

Another basic matching technique employs sections of waveguide having different
characteristic impedances. If the length of a section of line is one quarter of a guide
wavelength (β g l = π 2 ) then from (2.105) the admittance at the entrance to the line is

Yg2
YL ′ = . (2.117)
YL

Thus, if the admittance of the load is real, it can be matched to a source whose
admittance is also real by using a section of waveguide (a quarter-​wave transformer)
whose characteristic impedance is Yg = YLYL′ . The method can be extended to
match loads which have non-​zero susceptance by first using a section of line to trans-
form the admittance to a plane where it is real. The match achieved by a quarter-​
wave transformer is only exact at frequencies where the length is an odd number
of quarter wavelengths. Broad-​band matching is achieved by using a number of
sections in series to give either a Butterworth or a Tchebychev response. This simple
theory ignores the susceptances of the steps. It is possible to compensate for these
by making slight changes to the lengths of each section of waveguide. Further infor-
mation and tables for the design of multi-​step transformers can be found in [32–​34].

2.6 Coupling without Change of Mode

Microwave circuits involve bends and junctions in the waveguides that introduce
discontinuities into the system. Some of these junctions do not involve any change
80

80 Waveguides

in the mode of propagation. These are discussed here, and those in which a change
of mode is involved are discussed in the next section.
One of the simplest waveguide components is the bend. Provided that the bend is
gradual the mismatch is slight and no problems arise. This is normally the case with
coaxial lines. Sometimes, however, it is necessary to arrange a change of direction
in a small space and the design of the bend then becomes important.
Waveguide bends are often made in the form of circular arcs, as shown in
Figure  2.24a. The direction of the bend can either be in the E plane or in the
H plane. It has been shown that the minimum radius for a satisfactory match
is 1.5 times the width of the waveguide in the plane of the bend [4]. Thus for the
E-​plane bend shown in Figure  2.24(a) the minimum value of R is 1.5b. Because
very accurate manufacturing is necessary to avoid increased reflections, tight bends
are usually electro-​formed. More gentle bends can be made by bending a straight
waveguide using special equipment.
Radiused waveguide bends are expensive so it is sometimes better to use fabri-
cated mitred bends instead. Figure 2.24(b) shows a simple right-​angle bend. Clearly
the field patterns within the bend do not match those in the connecting waveguides
so an appreciable reactive mismatch can be expected. Equivalent circuits can be
developed for right-​angled bends in a manner similar to that used above for wave-
guide discontinuities [5]. An improvement on the simple right-​angled bend, shown
in Figure 2.24(c), chamfers the outside of the bend to make the effective height of
the waveguide the same as that of the connecting waveguides. Better still, if space
allows, is to use a double-​mitred bend as shown in Figure 2.24(d). If the separation
(L) between the two joints is chosen correctly, then the mismatches caused by the
junctions cancel each other out at one frequency and produce near cancellation
over a useful band of frequencies.

Figure 2.24:  E-​plane waveguide bends: (a) radiused bend, (b) mitred bend, (c) chamfered

bend, and (d) double mitred bend.
81

Coupling with Change of Mode 81

It is common in microwave tube engineering to use reduced-​height waveguides to

make connections into cavity resonators and coupled-​cavity slow-​wave structures.
The most straightforward way to make a transition to a reduced-​height waveguide
is to use a linear taper [35]. However, the taper must be around three guide wave-
lengths long to achieve a broad-​band match having a return loss greater than
30 dB. An alternative method is to use one, or more, quarter-​wave transformers
[32, 33, 36, 37]. A very good broad-​band match can be achieved in this way, using
three or four steps, and such a transformer is considerably shorter than a linear
taper having the same performance. To get the best results it is necessary to adjust
the lengths of the steps slightly to compensate for their shunt capacitances. Good
results can also be obtained using non-​uniform transitions [38, 39]. For further
information see [3, 4].

2.7 Coupling with Change of Mode

In the previous section we considered examples of coupling between waveguides

which were propagating the same mode. Very often it is necessary to couple power
from one mode to another. Examples of this are transitions from a coaxial line
to a waveguide, and from one waveguide mode to another. The general principle
is to match the field patterns of the modes in the two waveguides to one another.
In many cases this cannot be done exactly and it is necessary to find some way of
matching the field patterns as closely as possible. In that way a high proportion of
the power is coupled into the desired mode, and relatively little into other modes.
An example of this is the transition from the TE10 rectangular waveguide mode
to the TM 01 circular waveguide mode. Figure 2.25(a) shows the field patterns in
the two waveguides. It is immediately evident that the magnetic field patterns are
similar if the waveguides are arranged at right angles to each other as shown.
The electric fields are not so well matched because the removal of part of the
broad wall of the rectangular guide means that the field lines have to be radi-
cally redistributed. One solution is to put a post at the centre of the transition as
shown in Figure 2.25(b). Additional matching elements may be needed to match
the transition completely over a band of frequencies. This transition is known as
a ‘Door-​Knob’ transition. Further information about mode transducers can be
found in [4, 40].
As a second example we will consider the problems of making a good broad-​
band match from a coaxial line to a waveguide. A standard coaxial line has a char-
acteristic impedance of 50 ohms. The characteristic impedance of a rectangular
waveguide in terms of voltage and power is given by (2.65). Standard waveguides
have an aspect ratio ( a b ) of about 2:1 and are used over a range of ( λ g λ 0 ) of
about 1.7 to 1.2. The impedance to be matched therefore ranges from 630 ohms to
440 ohms, roughly ten times that of the coaxial line. A component providing a well
matched connection is called a coaxial to waveguide transformer.
82

82 Waveguides

Figure 2.25:  TE10 rectangular waveguide to TM 01 circular waveguide junction:

(a) arrangements of the fields in the waveguides, and (b) a cross-​sectional view of a
‘door-​knob’ transition showing the field patterns.

Consideration of the field patterns in the two waveguides suggests that a door-​
knob transition, similar to that shown in Figure 2.25(b), might be used. The field
patterns are shown in Figure 2.26(a). A quarter wavelength of waveguide behind
the transition is used to transform the short circuit at the end of the waveguide
into an open circuit at the plane of the transition. Provided that the diameter of
the door-​knob is not less than 0.15a it is possible to neglect its series inductance so
that the impedance of the coaxial line is presented directly to the waveguide [41].
Cohn found that this type of junction is best modelled by using ZVI as the charac-
teristic impedance of the waveguide but other authors have used ZPI [42]. One of
the difficulties of the door-​knob transition is that it has to cope with the change
from a low-​voltage, high-​current wave in the coaxial line to a high-​voltage, low-​
current wave in the waveguide. The existence of a direct current path makes this
difficult. Two solutions to this problem suggest themselves: one is to make use of
a lower impedance waveguide; the other is to avoid a direct current path.
To reduce the waveguide impedance we must reduce its height. But the reduction
to, perhaps, one tenth of the height of a standard waveguide presents new problems
for the transformation between the two. A better solution is to use a ridged wave-
guide which can have an impedance of 50 ohms without needing such narrow gaps.
Figure 2.26(b) shows a transition to a coaxial line using a ridged waveguide, with a
83

Windows 83

Figure 2.26:  Coaxial line to waveguide transformers using: (a) a door-​knob, (b) a ridged

waveguide, and (c) a probe.

quarter wave section of short circuited waveguide behind it, and a stepped imped-
ance transformer to a standard waveguide [41–​43].
The third approach is to couple into the waveguide using displacement current
rather than conduction current. Figure 2.26(c) shows how this is achieved. The free
end of the centre conductor is, in effect, an antenna radiating into the waveguide.
The use of an expanded end to the centre conductor provides the necessary imped-
ance transformation. Many other kinds of transition have been invented, some of
which are in common use [4, 40]. The three described here have been included to
illustrate the principles involved in coupling between different modes.

2.8 Windows

Dielectric windows form an essential part of any microwave tube because of the
need for RF power to pass into, and out of, the vacuum envelope. It is obviously
important that they should have low insertion loss and be well-​matched at all fre-
quencies of importance. In addition the output window is commonly one of the
most critical components in a tube because it is subject to high thermal and electri-
cal stresses. If a window fails then the tube loses its vacuum and is useless [44–​47].
Electrical stresses are caused by high RF electric fields in the region of the win-
dow, and these may be increased by the presence of a standing wave in the out-
put waveguide. The RF electric fields can also cause single-​surface multipactor
discharges on the vacuum side of the window (see Section 18.8) [48]. The bom-
bardment by multipacting electrons causes heat dissipation and increased thermal
84

84 Waveguides

stresses. A multipactor discharge also generates RF noise and changes the match of
the window. It is therefore important to reduce the risk of multipactor when design-
ing the window and it is common to use anti-​multipactor coatings such as titanium
nitride. Window failure can be caused by a waveguide arc on the air side of the win-
dow, either because of the increased standing wave, or because of the direct impact
of the arc on the window surface [49]. In some tubes the window may be subject to
bombardment by stray electrons from the interaction region, and by X-​rays gener-
ated within the tube. Even in the absence of any other electrical causes the dielectric
material of the window is heated directly by dielectric losses. These dielectric losses
commonly increase with temperature, leading to the possibility of thermal run-​
away. The output window of a high power tube can be protected by monitoring its
temperature, or by the use of an arc detector, so that the tube can be switched off
rapidly in the event of abnormal heating, or the presence of an arc [47].
The mechanical failure of a window may be caused directly by mechanical
stresses, including those caused by thermal expansion, or by dielectric breakdown
caused by the build-​up of surface charge [44]. The window is subject to mechani-
cal stresses in any case because of the pressure differential across it and these are
increased if the external waveguide is pressurised. In addition, the tube must with-
stand the temperature changes which occur during processing and during use. The
coefficients of thermal expansion of the dielectric materials commonly used in
windows differ from those of the metals used in tube construction. Thus appreci-
able thermal stresses can be caused by changes in temperature and the mechanical
design of a window is as important as the electrical design [44].
Various dielectric materials have been used in microwave windows. Glass was
used in the earliest tubes but has now generally been supplanted by ceramics. The
commonest of these is high purity alumina which has low dielectric losses and is
readily metallised for brazing to metal components. Beryllia has sometimes been
used for windows to carry high average powers because its thermal conductivity is
an order of magnitude higher than that of alumina. However, it is a highly toxic
material when in powder form and is now generally avoided. A much better mate­
rial is chemical-​vapour-​deposited (CVD) diamond which has excellent electrical
and mechanical properties. Table 2.4 shows a comparison between the properties
of a number of dielectric materials for use in windows [46].
The simplest form of window is a sheet of dielectric material inserted in a wave-
guide having uniform cross-​section, as shown in Figure 2.27. The figure shows a
rectangular waveguide for convenience, but the discussion which follows applies
to a waveguide of any cross-​section. The same mode can propagate in both the
empty waveguide and the window, though the cut-​off frequency in the window
is less than that of the waveguide by a factor of ε r where ε r is the relative per-
mittivity of the dielectric. Because there is no change of mode, the match of the
window can be calculated by using the wave impedances of the waveguide and the
window. The window always introduces a mismatch except when it is an integral
number of half-​wavelengths thick. This condition only holds at specific frequencies,
and a half-​wavelength window typically has a bandwidth of around 10 to 15% [47,
85

Windows 85

Table 2.4:  Comparison of typical dielectric material properties at 25°C [46, 50]

Property 99.5% Alumina Beryllia CVD Diamond

Density (g cm−3) 3.90 2.90 3.51

Specific heat (J g−1 K−1) 0.88 1.02 0.517
Thermal conductivity (W m K ) −1 −1
35 250 1800
Coefficient of thermal expansion 6.9 × 10−6 6.4 × 10–​6 1.5 × 10–​6
Young’s modulus (GPa) 370 344 1220
Poisson ratio 0.22 0.21 0.2
Tensile strength (MPa) 262 75−150 200−500
Relative permittivity 9.4 6.7 5.67
Loss tangent 9 × 10−5 4 × 10–​4 2 × 10–​5

Figure 2.27:  A dielectric window in a uniform waveguide.

51]. The bandwidth decreases as the relative permittivity of the dielectric increases.
Although this type of window can handle high mean powers it has not been widely
used except in gyrotrons. It is useful at millimetre wavelengths where other types of
window become too thin to have good mechanical strength [52].
If the window is less than a half-​wavelength thick it will always cause a mismatch
which becomes smaller as the thickness of the window is reduced. It is therefore desir-
able to make the window as thin as possible consistent with sufficient mechanical
strength [53]. A good match can be achieved by adding reactive elements to provide
either a maximally-​flat or equal ripple response. The bandwidth of half-​wavelength
windows can be increased in the same way [52]. Other techniques which have been
suggested include the use of a self-​matched pair of windows, tapered or stepped die-
lectrics, and multi-​layer dielectrics [50, 54]. The design of broad-​band windows is
discussed further in the following sections.
Because the phase velocity of electromagnetic waves is reduced in the dielectric
it is possible for the cut-​off frequencies of higher-​order modes to fall within the
86

86 Waveguides

working band of the window. These modes are cut-​off in the adjacent waveguide
and their fields are therefore confined to the region close to the window and are
hard to detect. They are therefore known as ghost or trapped modes [55]. In an
ideal window they would not be coupled to the propagating mode, but slight asym-
metries in the construction of the window, or of the properties of the dielectric, are
sufficient to cause coupling of power into them. Ghost modes have high Q factors
and it is easy for the fields associated with them to become strong enough to cause
failure of the window. Thus it is important to ensure that no ghost modes fall within
the working frequency band of the window. They can be divided into even and odd
modes depending upon whether their fields are symmetrical or anti-​symmetrical
about the mid-​plane of the window. For windows having the arrangement shown in
Figure 2.27 the ghost mode resonances are the solutions of
 β ′gn j β gn 
tan (β ′gn L 2 ) =   (2.118)
ε r j β gn β ′gn 
for modes whose tangential electric field is zero at the mid-​plane, and

 β ′gn j β gn 
cot (β ′gn L 2 ) =   (2.119)
 − ε r j β gn β ′gn 

for modes whose tangential magnetic field is zero at the mid-​plane. In these equa-
tions the upper and lower expressions within the braces refer to TE and TM modes
respectively, the primed variables refer to the region within the dielectric, and the
propagation constants for the nth mode are determined from (2.11) [55]. Ghost
modes can be a problem in any kind of window but they are most troublesome in
half-​wavelength windows.

2.8.1 Windows in Coaxial Lines

The characteristic impedances of the TEM mode of empty, and dielectric-​filled, sec-
tions of a coaxial line can be calculated from (2.50). If the impedance of the dielectric-​
filled section is made equal to that of the air-​filled section, the ratio of the diameters
of the inner and outer conductors must be increased. Figure 2.28(a) shows a typical
arrangement in which the diameter of the inner conductor has been decreased [45, 56].
The window can be modelled using the equivalent circuit, shown in Figure 2.28(b),
comprising sections of transmission line in series, together with lumped capacitances
to represent the reactances of step changes in the diameter of the inner conductor
[57]. In an alternative design the diameters of both conductors differ from those of
the connecting coaxial line and inductive chokes are used to improve the match [58].
Windows of the kind illustrated in Figure 2.28(a) have a partial standing wave
within the dielectric window because the design has focused on the overall match,
rather than the detailed internal fields. The electric field in the window can be
reduced by using external reactive elements to match the impedances at the sur-
faces of the dielectric [59]. It has been shown that the use of a travelling-wave design
can reduce the maximum field in the dielectric by a factor of four, compared with
87

Windows 87

Figure 2.28:  A dielectric window in coaxial line: (a) typical arrangement, and (b) equivalent

circuit.

a half-​wave window. A travelling-wave coaxial window in which the electric field in

the dielectric was only 25% of that in the external coaxial line, successfully carried a
peak power of 1 MW at 1.3 GHz for several months [60]. After initial conditioning
it showed no signs of multipactor discharges.

2.8.2 Windows in Rectangular Waveguide

Although rectangular windows have been used in rectangular waveguide, they suf-
fer from the disadvantage of being difficult to seal and of having stress concen-
trations at the corners. There is, therefore, a general preference for using circular
windows in connection with rectangular waveguides.
The simplest technique is to use a circular window set in a transverse metal wall,
as shown in Figure 2.29. The TE11 circular waveguide mode is excited in the window
and its diameter is chosen so that it is resonant at the desired centre frequency. The
diameter of the window can be less than the height of the waveguide if the rela-
tive permittivity of the dielectric is great enough. Otherwise the window may take
the form of a disk whose top and bottom have been cropped to fit the waveguide.
Because this kind of window is resonant it is inherently narrow band, but the band-
width can be increased by adding reactive elements [61, 62]. When the resonant
diameter of the window is greater than the height of the waveguide it is possible
to use a circular disk if the matching elements take the form of inductive chokes
similar to those used in waveguide choke flanges [44, 45]. It is also possible to design
self-​resonant windows with a pure travelling wave in the dielectric [63–​65].
Self-​resonant waveguide windows suffer from the disadvantage that the diameter
of the window is set by the resonant condition. Thus it is not possible to reduce the
power density in the window by increasing its diameter. This problem is overcome by
setting the window in a section of circular waveguide carrying the TE11 mode, with
mode transformers at either end to connect it to the TE10 mode in the rectangular
8

88 Waveguides

Figure 2.29:  Arrangement of a resonant waveguide window.

Figure 2.30:  Arrangement of a pill-​box window.

waveguide. The simplest form is the widely used pill-​box window shown in Figure 2.30
which has a bandwidth of around 30% [47, 66]. The design of pill-​box windows has
been described in [45, 67, 68]. The electrical length of the circular section is chosen
to be a half-​wavelength when the effects of the dielectric are included. The diameter
of the circular waveguide must be greater than the diagonal of the rectangular wave-
guide, and it is usual to make its cut-​off frequency close to that of the rectangular
waveguide. This type of window has ghost mode resonances associated with reflec-
tions between the transitions between the waveguides. The diameter of the circular
guide is adjusted to ensure that none of these resonances falls within the working
frequency band. The equivalent circuit of this window is shown in Figure 2.28(b),
but some care is needed in defining the characteristic impedances in the circular
section. The use of this equivalent circuit is justified by the close similarity between
the electric field distributions over the aperture connecting the two waveguides. The
correct relationship between their characteristic impedances can be found using the
method described by Farmer [27]. An electromagnetic analysis of a pill-​box window
that includes full dimensions of the window is given in [69]. Pill-​box windows can
be designed to have travelling-​wave fields in the dielectric by the use of inductive
irises at the waveguide transitions [70]. Alternatively the transitions can be made
using tapers, or other wide-​band mode converters [50, 71]. Windows have also been
described which use other modes in the circular waveguide [72, 73].
89

References 89

References

[1]‌ S. Ramo et al., Fields and Waves in Communication Electronics. New York: Wiley, 1965.

[2]‌ J. Feinstein, ‘Passive microwave components’, in D. G. Fink and D. Christiansen, eds,
Electronic Engineers’ Handbook. New York: McGraw-​Hill, pp. 9.2–​9.27, 1982.
[3]‌ R. E. Collin, Foundations for Microwave Engineering. New York: McGraw-​Hill, 1966.
[4]‌ A. F. Harvey, Microwave Engineering. New York: Academic Press, 1963.
[5]‌ N. Marcuvitz, Waveguide Handbook. New York: McGraw-​Hill, 1951.
[6]‌ R. Cooper and R. G. Carter, ‘High power RF transmission’, in J. Miles, ed., CERN
Accelerator School: Radio Frequency Engineering. Geneva: CERN, pp. 210–​235, 2005.
[7]‌ ‘Selecting a transmission line for your broadcast system’, vol. SP50115. Orland Park,
IL: Andrew Corporation, 1998.
[8]‌ S. Choroba, ‘RF power transport’, in R. Bailey, ed., CERN Accelerator School: RF for
accelerators. Geneva: CERN, pp. 117–​123, 2011.
[9]‌ C. Kubota et al., ‘Parts of the 203D coaxial waveguide improved at KEK’, presented at
the LINAC 2002, Gyeongju, Korea, 2002.
[10] R. G. Carter, Electromagnetism for Electronic Engineers. Available at:  http://​book-
boon.com/​en/​electromagnetism-​for-​electronic-​engineers-​ebook (accessed 3 October
2017) 2009.
[11] G. Faillon et al., ‘Microwave tubes’, in J. A. Eichmeier and M. K. Thumm, eds, Vacuum
Electronics: Components and Devices. Berlin: Springer-​Verlag, pp. 1–​84, 2008.
[12] A. M. Perez et al., ‘Prediction of multipactor breakdown thresholds in coaxial trans-
mission lines for traveling, standing, and mixed waves’, IEEE Transactions on Plasma
Science, vol. 37, pp. 2031–​2040, 2009.
[13] J. Lorkiewicz et al., ‘Surface TiN coating of TESLA couplers at DESY as an antimultipac-
tor remedy’, in 10th Workshop on RF Superconductivity, Tsukuba, Japan, pp. 448–​452, 2001.
[14] G. Burt et  al., ‘Multipactor simulations of the SPL power coupler’, in LINAC 2010,
Tsukuba, Japan, pp. 878–​880, 2010.
[15] N. W. McLachlan, Bessel Functions for Engineers. Oxford University Press, 1954.
[16] R. G. Carter, Electromagnetic Waves:  Microwave Components and Devices. London:
Chapman & Hall, 1990.
[17] ‘Rectangular waveguide specifications’, WAVEGUIDE COMPONENT: Specifications
and Design Handbook (7th edn) section DS-​501 Exeter, NH: Cobham Defense Electronic
Systems, 2006.
[18] S. Hopfer, ‘The design of ridged waveguides’, IRE Transactions on Microwave Theory
and Techniques, vol. 3, pp. 20–​29, 1955.
[19] S. B. Cohn, ‘Properties of ridge wave guide’, Proceedings of the IRE, vol. 35,
pp. 783–​788, 1947.
[20] ‘Double ridge waveguide specifications’, WAVEGUIDE COMPONENT: Specifications
and Design Handbook (7th edn) section DS-​503 Exeter, NH: Cobham Defense Electronic
Systems, 2006.
[21] G. G. MacFarlane, ‘Quasi-​stationary field theory and its application to diaphragms and
junctions in transmission lines and wave guides’, Journal of the Institution of Electrical
Engineers –​Part IIIA: Radiolocation, vol. 93, pp. 703–​719, 1946.
[22] J. P. Montgomery, ‘On the complete eigenvalue solution of ridged waveguide’, IEEE
Transactions on Microwave Theory and Techniques, vol. 19, pp. 547–​555, 1971.
[23] Y. Utsumi, ‘Variational analysis of ridged waveguide modes’, IEEE Transactions on
Microwave Theory and Techniques, vol. 33, pp. 111–​120, 1985.
90

90 Waveguides

[24] J. Helszajn and M. McKay, ‘Voltage-​current definition of impedance of double ridge

waveguide using the finite element method’, Proceedings of the IEE:  Microwaves,
Antennas and Propagation, vol. 145, pp. 39–​44, 1998.
[25] M. McKay and J. Helszajn, ‘Voltage-​current definition of impedance of single-​ridge
waveguide’, IEEE Microwave and Guided Wave Letters, vol. 9, pp. 66–​68, 1999.
[26] W. J.  R. Hoefer and M. N. Burton, ‘Closed-​form expressions for the parameters of
finned and ridged waveguides’, IEEE Transactions on Microwave Theory and Techniques,
vol. 30, pp. 2190–​2194, 1982.
[27] E. D. Farmer, ‘Junction admittance between waveguides of arbitrary cross-​sections’,
Proceedings of the IEE –​Part C: Monographs, vol. 103, pp. 145–​152, 1956.
[28] M. Guglielmi and C. Newport, ‘Rigorous, multimode equivalent network representation
of inductive discontinuities’, IEEE Transactions on Microwave Theory and Techniques,
vol. 38, pp. 1651–​1659, 1990.
[29] S. B. Cohn, ‘Microwave coupling by large apertures’, Proceedings of the IRE, vol. 40,
pp. 696–​699, 1952.
[30] W. H. Watson, The Physical Principles of Wave Guide Transmission and Antenna
Systems. Oxford: Clarendon Press, 1947.
[31] F. Lösch, Jahnke-​Emde-​Lösch:  Tables of Higher Functions. Stuttgart:  B.G. Teubner
Verlagsgesellschaft, 1960.
[32] G. L. Matthaei et al., Microwave Filters, Impedance-​Matching Networks and Coupling
Structures. Dedham, MA: Artech House Books, 1980.
[33] L. Young, ‘Tables for cascaded homogeneous quarter-​ wave transformers’, IRE
Transactions on Microwave Theory and Techniques, vol. 7, pp. 233–​237, 1959.
[34] L. Young, ‘Optimum quarter-​ wave transformers’, IRE Transactions on Microwave
Theory and Techniques, vol. 8, pp. 478–​482, 1960.
[35] R. C. Johnson, ‘Design of linear double tapers in rectangular waveguides’, IRE
Transactions on Microwave Theory and Techniques, vol. 7, pp. 374–​378, 1959.
[36] S. B. Cohn, ‘Optimum design of stepped transmission-​ line transformers’, IRE
Transactions on Microwave Theory and Techniques, vol. 3, pp. 16–​20, 1955.
[37] L. Young, ‘Tables for cascaded homogeneous quarter-​wave transformers (Correction)’,
IRE Transactions on Microwave Theory and Techniques, vol. 8, pp. 243–​244, 1960.
[38] R. E. Collin, ‘The optimum tapered transmission line matching section’, Proceedings of
the IRE, vol. 44, pp. 539–​548, 1956.
[39] R. W. Klopfenstein, ‘A transmission line taper of improved design’, Proceedings of the
IRE, vol. 44, pp. 31–​35, 1956.
[40] G. L. Ragan, Microwave Transmission Circuits. New York: McGraw-​Hill, 1948.
[41] S. B. Cohn, ‘Design of simple broad-​ band waveguide to coaxial line junctions’,
Proceedings of the IRE, vol. 35, pp. 920–​926, 1947.
[42] Y. Zhou et al., ‘Design of millimeter wave wideband transition from double-​ridge wave-
guide to coaxial line’, Journal of Infrared, Millimeter and Terahertz Waves, vol. 32,
pp. 26–​33, 2011.
[43] R. Nie et al., ‘Simulation and design of 18–​40 GHz ridge waveguide to coaxial transi-
tion’, in 2011 IEEE International Conference on Microwave Technology & Computational
Electromagnetics (ICMTCE), pp. 183–​185, 2011.
[44] J. F. Gittins, Power Travelling-​Wave Tubes. London: English Universities Press, 1965.
[45] M. J. Smith and G. Phillips, Power Klystrons Today. Taunton, UK: Research Studies
Press, 1995.
91

References 91

[46] D. K. Abe and J. P. Calame, ‘Advanced material technologies’, in R. J. Barker et al., eds,
Modern Microwave and Millimetre-​Wave Power Electronics. Piscataway, NJ: IEEE Press,
pp. 649–​689, 2005.
[47] A. Staprans et al., ‘High-​power linear-​beam tubes’, Proceedings of the IEEE, vol. 61,
pp. 299–​330, 1973.
[48] A. Neuber et  al., ‘Window breakdown caused by high-​ power microwaves’, IEEE
Transactions on Plasma Science, vol. 26, pp. 296–​303, 1998.
[49] Varian, ‘Technical Manual: Installation, Operation, Maintenance, Care and Handling
Instructions, General:  Microwave Tubes, Magnetron Tubes, Electron Tubes’, 1
October 1979.
[50] D. B. Churchill, ‘Problems associated with waveguide pressure windows for microwave
tubes’, in 7th National Conference on Tube Technology, pp. 371–​389, 1964.
[51] S. Chel et  al., ‘Coaxial disc windows for a high power superconducting cavity input
coupler’, in Proceedings of the 1999 Particle Accelerator Conference (PAC 99), vol. 2,
pp. 916–​918, 1999.
[52] S. Liu, ‘A fast computational technique for RF window in millimetre wave tubes’,
International Journal of Infrared and Millimetre Waves, vol. 15, pp. 857–​860, 1994.
[53] A. S. Gilmour Jr., Klystrons, Traveling Wave Tubes, Magnetrons, Crossed-​Field Amplifiers
and Gyrotrons. Norwood, MA: Artech House, 2011.
[54] M. C. Lin and D. S. Chuu, ‘A novel wide-​band high-​transmission window for high-​
frequency microwave tubes’, in Third IEEE International Vacuum Electronics Conference
(IVEC 2002), pp. 216–​217, 2002.
[55] M. P. Forrer and E. T. Jaynes, ‘Resonant modes in waveguide windows’, IRE Transactions
on Microwave Theory and Techniques, vol. 8, pp. 147–​150, 1960.
[56] K. Cummings et  al., ‘Results and lessons learned from conditioning 1 MW CW 350
MHz coaxial vacuum windows’, in LINAC 98, Chicago, IL, pp. 938–​940, 1998.
[57] J. Wang and R. G. Carter, ‘Design of a coaxial window for high power CW magne-
tron’, in International Conference on Microwave and Millimeter Wave Technology 1998,
pp. 744–​747, 1998.
[58] Y. Kang et al., ‘Electromagnetic simulations and properties of the fundamental power
couplers for the SNS superconducting cavities’, in Proceedings of the 2001 Particle
Accelerator Conference (PAC 2001), vol. 2, pp. 1122–​1124, 2001.
[59] X. Hanus and A. Mosnier, ‘Coaxial TW window for power couplers and multipactor
considerations’, in 1995 Workshop on RF Superconductivity, Gif-​sur-​Yvette, France, pp.
701–​705, 1995.
[60] C. Travier et  al., ‘Design and test of a 1.3 GHz travelling wave window’, in 1999
Workshop on RF Superconductivity, Santa Fe, NM, pp. 427–​436, 1999.
[61] R. Z. Gerlack, ‘Small ceramic windows for large waveguides’, Microwave Journal, vol.
37, pp. 110–​114, 1994.
[62] E. Chojnacki et al., ‘Design of a high average power waveguide window’, in Proceedings
of the 1997 Particle Accelerator Conference (PAC 1997), vol. 3., pp. 3177–​3179, 1997.
[63] W. R. Fowkes et  al., ‘Large diameter reduced field TE01 traveling wave window for
X-​band’, in Proceedings of the 1999 Particle Accelerator Conference (PAC 1999), vol. 2,
pp. 783–​785, 1999.
[64] N. M. Kroll et al., ‘Design of traveling wave windows for the PEP-​II RF coupling net-
work’, in Proceedings of the 1995 Particle Accelerator Conference (PAC 95), vol. 3, pp.
1741–​1743, 1995.
92

92 Waveguides

[65] W. R. Fowkes et al., ‘Reduced field TE01 X-​Band travelling wave window’, in Proceedings
of the 1995 Particle Accelerator Conference (PAC 1995), pp. 1587–​1589, 1995.
[66] R. S. Symons, ‘Sealed wave guide window’, US Patent 2,958,834, 1960.
[67] O. S. Lamba et al., ‘Design and optimization of asymmetric RF window for 6 MW pulse
power S-​band klystron’, in International Conference on Recent Advances in Microwave
Theory and Applications (MICROWAVE 2008), pp. 44–​46, 2008.
[68] V. V. P. Singh et al., ‘On the electrical design of pill-​box type high power microwave win-
dow’, Journal of the IETE, vol. 39, pp. 351–​359, 1993.
[69] H. Arai et  al., ‘An analysis of a vacuum window for lower hybrid heating’, IEEE
Transactions on Plasma Science, vol. 14, pp. 947–​954, 1986.
[70] S. Michizono et  al., ‘High-​power test of pill-​box and TW-​in-​ceramic type S-​band rf
windows’, in LINAC 94, Tsukuba, Japan, pp. 457–​459, 1994.
[71] Y. Otake et al., ‘Design and high-​power test of a TE11 mode X-​band RF window with
taper transitions’, in Particle Accelerator Conference (PAC 95), Dallas, TX, vol. 3,
pp. 1590–​1592, 1995.
[72] A. E. Barrington and J. T. Hyman, ‘A non-​resonant waveguide window’, Proceedings of
the IEE –​Part B: Radio and Electronic Engineering, vol. 104, pp. 35–​38, 1957.
[73] R. L. Ives and J. Neilson, ‘TE01 windows for high power klystron applications’, in
International Electron Devices Meeting (IEDM ‘93) Technical Digest, pp. 161–​164, 1993.
93

3 Resonators

3.1 Introduction

Resonant cavities are important components in many microwave tubes because of

their frequency selective properties and because they store electromagnetic energy.
The purpose of this chapter is to describe the features which are common to all
types of microwave resonator and to examine the properties of a few important
types in more detail.
Any closed metal cavity supports an infinite number of electromagnetic reso-
nances. In each of these the fields satisfy Maxwell’s equations and the boundary
conditions on the metal surface. If resistive losses associated with currents in the
walls of the cavity are neglected then the electric and magnetic fields are in phase
quadrature. The losses in metallic cavity resonators are usually small so that the
fields differ very little from those in a loss-​less cavity. The losses can then be cal-
culated to a good approximation from the currents flowing in the walls when the
losses are neglected.
The cavity resonators used in microwave tubes have simple shapes and are
commonly cylindrically symmetrical around the axis of the electron beam. The
modes which are chiefly of interest are those having a strong axial component of
the electric field in the region of the beam. The simplest example is the pill-​box
cavity whose TM 010 resonance is illustrated in Figure 3.1. This mode can be derived
from the TM 01 mode of a circular waveguide by inserting a pair of conducting
planes normal to the axis. The resonant frequency for the lowest mode is the cut-​
off frequency of the waveguide where there is no axial variation of the electric
field. Figure 3.1 shows, schematically, the fields, charges and currents in the cavity
at intervals of a quarter of the resonant period T0. The properties of this cavity are
discussed in detail in Section 3.3.
For many purposes it is convenient to represent the properties of a cavity resona­
tor by its equivalent circuit. This can be derived by representing conduction current
paths by inductors, and displacement current paths by capacitors, as described in
Chapter 2. Figure 3.2a shows the derivation of the equivalent circuit for the TM 010
mode of a pill-​box cavity. The series resistor r represents the conduction losses in the
cavity walls. The analysis of this equivalent circuit is discussed in the next section.
94

94 Resonators

Figure 3.1:  Fields, charges and currents in a pill-​box cavity resonator.

Figure 3.2:  Equivalent circuits of a cavity resonator with conduction losses represented by

(a) a series resistor, and (b) a parallel resistor.

The properties of pill-​box cavity resonators and rectangular cavity resonators

are discussed in Sections 3.3 and 3.4. Cylindrical re-​entrant cavities, which are
commonly used in vacuum tubes, are considered in Section 3.5. Section 3.6 dis-
cusses external coupling to cavity resonators from coaxial lines and waveguides.
The last section of the chapter describes methods for measuring the properties of
cavities.

3.2 Resonant Circuits

The properties of parallel resonant circuits are well-​known and they are only sum-
marised here [1]. We shall concentrate on those issues which are important when
using equivalent circuits to model cavity resonators in microwave tubes.
95

Resonant Circuits 95

3.2.1 The Properties of Resonant Circuits

The equivalent circuit shown in Figure 3.2(a) is derived from the physics of the res-
onator. Analysis of this circuit shows that the input impedance is given by

 j ω0 
rQU 1 −
 QU ω 
Z= , (3.1)
1  ω ω0 
+ j −
QU  ω 0 ω 
where the resonant frequency is
1
ω0 = , (3.2)
LC
the unloaded Q factor is
1 ω L
QU = = 0 , (3.3)
ω 0 rC r
and

L
rQU = . (3.4)
C
Microwave resonators commonly have unloaded Q factors of at least 1000. They
can be made with values as high as 30,000 by careful design and manufacture. The
second term in the numerator of (3.1) is therefore small compared with unity and
can be neglected without serious error. The equation which results is then identical
to that for the parallel resonant circuit shown in Figure 3.2(b).
Rc QU
Z= , (3.5)
1  ω ω0 
+ j −
QU  ω 0 ω 

where the ‘R over Q’ ( R Q ) is defined by

Rc L
= = rQU (3.6)
QU C
and
Rc
QU = = ω 0 RcC . (3.7)
ω0 L
The parallel resonant circuit is generally used in preference to that in Figure 3.2(a)
because it is simpler to analyse. At microwave frequencies it is difficult to assign
meanings to the circuit parameters C, L and Rc and it is usual to use the parameters
ω 0, Rc QU and QU to describe the resonator. Equations (3.2), (3.6) and (3.7) enable
96

96 Resonators

these parameters to be calculated from the circuit parameters and vice versa. It
should be noted that QU is the only parameter which depends upon the properties
of the material from which the cavity is constructed. The resonant frequency and
the R Q of a cavity resonator therefore depend only upon its geometry.
The amplitude of the cavity impedance Z is readily derived from (3.5) as

Z 1
= . (3.8)
Rc  ω ω0 
2

1 + QU2  −
 ω 0 ω 

Figure 3.3 shows the amplitude of the normalised impedance Z Rc in decibels

as a function of normalised frequency and QU (see Worksheet 3.1). When the fre-
quency of the signal applied to the circuit is equal to the resonant frequency the
impedance is purely resistive and equal to Rc. It is clear from (3.8) that Z < Rc at
all other frequencies and that the width of the curve decreases as QU increases. This
relationship is usually expressed in terms of the width of the curve at the points
which are 3 dB below the peak amplitude. Then found that

ω0
QU = , (3.9)
2∆ω
where Z = R 2 when ω = ω 0 ± ∆ω provided that ∆ω  ω 0. Modern test equipment
can display the response curve with the vertical scale in decibels as shown in Figure 3.3.
This makes it easy to measure the unloaded Q factor of a resonator (see also Section 3.7).
From (3.7) we see that to obtain a high unloaded Q factor the circuit must have
a high shunt resistance (i.e low loss) and that leads to a useful alternative definition
of QU. When an alternating voltage V = V0 cos ωt is applied to the terminals of the
circuit shown in Figure 3.2(b) the maximum energy stored in the capacitor is

1
W = CV02 . (3.10)
2

Figure 3.3:  Amplitude of the impedance of a parallel resonant circuit as a function of

normalised frequency and QU.
97

Resonant Circuits 97

This energy is transferred backwards and forwards between the capacitor and the
inductor during each cycle and the total stored energy remains constant, provided
that the amplitude of the voltage at the terminals is held constant. The mean rate
of dissipation of energy through resistive losses is

V02
PL = . (3.11)
2Rc

Eliminating V0 between (3.10) and (3.11) and substituting for RcC in (3.7) gives

ω 0W 2 πW
QU = = , (3.12)
PL ∆W

where ∆W is the energy dissipated per cycle. The rate of dissipation of energy is
therefore

dW ∆W ω
=− = − 0 W , (3.13)
dt T0 QU

where T0 is the resonant period of the circuit. Equation (3.13) can readily be inte-
grated to show that, when the external excitation is removed, the stored energy
decays exponentially with a time constant τ = QU ω 0 . The physical significance of
the parameter R Q is revealed by substituting for PL from (3.11) into (3.12) to give

 Rc  V02
 Q  = 2ω W . (3.14)
U 0

Thus R Q is a measure of the relationship between the voltage across the terminals
of the circuit and the energy stored in it.
To complete our review of the theory of parallel resonant circuits we must exam-
ine the phase of Z given by

 ω ω 
∠Z = arctan QU  0 −   . (3.15)
  ω ω0  

Figure 3.4 shows how the phase of the cavity impedance varies with normalised
frequency and QU. At resonance the impedance is purely resistive and the phase
of Z is zero. At lower frequencies the behaviour of the circuit is dominated by the
reactance of the inductor and ∠Z → 90. At higher frequencies the capacitor has
the greater effect and ∠Z → −90. At the 3 dB points

ω0 ω 1
− =± (3.16)
ω ω0 QU
and ∠Z = 45 .
98

98 Resonators

Figure 3.4:  Phase of the impedance of a parallel resonant circuit as a function of normalised

frequency and QU.

3.2.2 External Loading of Resonant Circuits

When resonant structures are used in microwave tubes their properties are usually
modified by the presence of electrons within them, by connections to external wave-
guides, or both. The ways in which these effects influence the behaviour of a tube
differ somewhat between different types of tubes. We will therefore defer a detailed
examination of electron loading and external loading effects to the discussions of
particular types of tube. For our present purposes it is sufficient to consider the
effect on a resonator when it is connected to an external circuit comprising a resis-
tor and a reactor in parallel as shown in Figure 3.5. This circuit can represent the
combined effects of electron loading and external loading, including any imped-
ance transformer between the resonator and the external circuit. The admittance of
this circuit, measured at its terminals, is

1 1 1
Y= + j ωC + + + jBE . (3.17)
Rc j ωL RE
At resonance the sum of the susceptances is zero. Thus the susceptance jBE of
the load modifies the resonant frequency. This effect is usually quite small and it can
generally be compensated for, if necessary, by provision of an adjustable tuning ele-
ment in the resonator. It manifests itself in klystrons as a small shift in the frequency
of the cavities when the beam is turned on, and in magnetrons as frequency pushing
and pulling (see Chapters  13 and 15). The effect of the load resistance is usually
more significant because it may alter the Q of the circuit considerably. If we assume
that the resonant frequency is unchanged then the Q factor is now

1  1 1  1 1
= ω0 L  +  = + , (3.18)
QL  Rc RE  QU QE
where QL is the loaded Q of the resonator, and QU is the unloaded Q given by (3.7).
QE , known as the external Q, is the Q factor which would be measured if the only
9

Resonant Circuits 99

Figure 3.5:  Parallel resonant circuit with an external load.

loss were that provided by the external resistance. If the external resistance is equal
the resistance of the circuit at resonance then QL = QU 2.
We may define a coupling factor K by

Rc QU
K= = (3.19)
RE QE

so that

QU = QL (1 + K ) . (3.20)

This effect is important in klystrons because both the beam loading and the exter-
nal loading increase the bandwidths of the cavities and, therefore, of the tube. In
addition we shall see that variation of the external match presented to the output
of a klystron has an important effect upon the output gap voltage and on the effi-
ciency of the tube. In magnetrons, changes in the match of the load lead to changes
in the frequency of oscillation, and of the output power, through frequency pulling.

3.2.3 Excitation of Resonant Circuits

Figure 3.6 shows the equivalent circuit of a resonator connected to a source whose
resistance is RS . The rate of change with time of the current drawn from the source is

dI d 2V 1 dV V
=C 2 + + , (3.21)
dt dt R ′ dt L
where R ′ represents the parallel combination of Rc with RS so that

1 1 1 1+ K
= + = , (3.22)
R ′ Rc RS KRS
where K is defined by (3.19). The voltage across the terminals of the resonant circuit
therefore obeys the differential equation

d 2V ω 0 dV ω R′
+ + ω 20V = − 0 I 0 ω sin ( ωt ) (3.23)
dt 2
QL dt QL
in which the equivalent circuit parameters have been replaced by their microwave
equivalents. The general solution of this equation for the case where the source
10

100 Resonators

Figure 3.6:  Equivalent circuit of a resonator connected to a source.

frequency is the resonant frequency of the circuit, and the source is connected to
the circuit at t = 0, is

 I R′   t
V =  − I 0 R ′ cos ( ω 0 t ) + 0 sin ( ω 0 t ) exp  −  + I 0 R ′ cos ( ω 0 t ) , (3.24)
 2QL   τ

where

2QU 2QL
τ= = . (3.25)
ω 0 (1 + K ) ω 0
The derivation of this equation has assumed that QL2  1. The loaded Q fac-
tors of microwave resonators are typically greater than 100 so that (3.24) can be
approximated as

  t
V = I 0 R ′ 1 − exp  −   cos ( ω 0 t ) . (3.26)
  τ 

We note that, if QL  1, the voltage is in phase with the current.

The energy stored in the circuit is equal to the maximum energy stored in the
capacitor, namely
2
1 1 2   t
W (t ) = C V = C ( I 0 R ′ ) 1 − exp  −   . (3.27)
2

2 2   τ 

When the source is delivering power into a matched load, the forward power is

1 2
P+ = I 0 RS , (3.28)
8
so that, making use of (3.7), (3.19) and (3.28), we can write (3.27) as
2
Q 4K   t
W (t ) = P+ U 2 
1 − exp  −   . (3.29)
ω 0 (1 + K )   τ 

This expression is maximum when K = 1 (i.e. Rc = RS ) so that the resonator is

matched to the source. It is then said to be critically coupled to the source and the
final stored energy is

QU
W0 = P+ . (3.30)
ω0
10

Resonant Circuits 101

Then (3.29) can be written

2
4K   t
W (t ) = W0 1 − exp  − τ   . (3.31)
(1 + K )2  
When K < 1 the resonator is under-​coupled and when K > 1 it is over-​coupled. The
time taken to accumulate the energy in the resonator, known as the filling time, can
be estimated by noting that when t = QLT0 the stored energy is 91.5% of its final value.
When K is either much greater than, or much less than, unity the stored energy is low
because most of the power supplied by the source is reflected back into it. Figure 3.7
shows the normalised stored energyW (t ) W0 as a function of time for three values of
K (see Worksheet 3.1). These results are relevant both when a resonator is excited by
an external waveguide and when it is excited by a modulated electron beam.
As the stored energy builds up in the resonator the impedance presented to the
source changes. When the source is connected to a matched load the magnitude of
the load voltage in the steady state is
1
V0 = I 0 RS . (3.32)
2
The magnitude of the voltage in the general case is found from (3.22), (3.26) and
(3.32) to be

V 2K   t
= 1 − exp  − τ   . (3.33)
V0 1 + K  
Since the voltage and current are in phase with each other at resonance, the volt-
age reflection coefficient of the resonator is given by

V − V0 2K   t
S11 = = 1 − exp  − τ   − 1. (3.34)
V0 1+ K  

Figure 3.7:  Normalised energy stored in a cavity resonator as a function of time for three
coupling factors.
102

102 Resonators

Figure 3.8:  Input match of a resonant circuit as a function of time for three coupling factors.

Figure 3.8 shows how S11 varies with time for three values of K. When the output
power from a pulsed microwave tube is supplied to a resonant cavity or structure
then a time-​varying mismatch is seen by the tube. As the reflected power may mod-
ify the behaviour of the tube it is necessary to place a circulator between the tube
and the resonator to prevent this. When the source frequency differs from the res-
onant frequency the voltage and current are no longer in phase with each other and
the analysis above must be modified accordingly.

3.2.4 Coupled Resonators

The bandwidth of a cavity resonator can be increased by coating its surfaces with
a lossy material, by connecting it to an external resistive load, or by increasing
the electron loading. These methods of reducing Q are often unsuitable because
they increase the RF losses. An alternative technique is to couple two resonators
together. This method is used, for example, to achieve the bandwidth required in the
inductive output tubes (IOTs) used for television broadcasting (see Section 12.6).
The properties of coupled-​cavity slow-​wave structures are considered in Section
4.6. Here we restrict our attention to a pair of identical parallel resonant circuits
which are coupled by a mutual inductance, as shown in Figure 3.9. The impedance
matrix for this network is

V1   1 j ωC − 1 j ωC 0   i1 
 0  =  − 1 j ωC Z − j ωkL  i2  , (3.35)
  
 0   0 − j ωkL Z  i3 

where

1
Z= + r + j ωL. (3.36)
j ωC
103

Resonant Circuits 103

From the last line of this equation

j ωkL i2 = Z i3 . (3.37)

Substituting for i3 in (3.35) gives

V1   1 j ωC − 1 j ωC   i1 
 0  =  − 1 j ωC Z ′  i2 
, (3.38)
  
where

ω 2 k 2 L2
Z′ = Z + . (3.39)
Z
When the frequency is equal to the resonant frequency of either resonator

 ω 2 k 2 L2 
Z ′ = r 1 + 0 2  = r (1 + k 2QU2 ) . (3.40)
 r 
If kQU = 1 the second circuit is matched to the first and the circuits are said to be
critically coupled. The input impedance and phase of this circuit can be computed
as a function of frequency and of the coupling factor k (see Worksheet 3.1).
Figure 3.10 shows the magnitude of the input impedance of coupled resonators
whose uncoupled Q is 1000 as a function of frequency and coupling factor. When

Figure 3.9:  Parallel resonant circuits coupled by a mutual inductance.

Figure 3.10:  Amplitude response of two identical resonators coupled by a mutual inductance.

104

104 Resonators

Figure 3.11:  Phase response of two identical resonators coupled by a mutual inductance.

k = 0 the response is identical to that shown in Figure 3.3, as would be expected.

As k is increased the peak of the curve first flattens and then divides into two
peaks. When the resonators are critically coupled the curve resembles a Tchebychev
response. The bandwidth is then increased in comparison with a single resonator
having the same maximum impedance. The phase of the input impedance of the
coupled resonators is shown in Figure 3.11. When k is close to zero the curve is very
similar to that for a single resonator with a single frequency at which the phase is
zero. This continues as k increases until there is a point of inflection when k = 1 QU .
For greater values of k there are three points at which the phase is zero. The behav-
iour of resonators which are coupled in other ways is similar to that shown here.

3.3 Pill-​Box Cavity Resonators

The resonant circuits used at frequencies greater than about 100 MHz are invariably
cavity resonators. These resonators can take many forms but all support an infinite
number of resonant modes. The simplest resonant cavities are formed by placing a
pair of metallic walls across a metal waveguide having a uniform cross-​section. These
reflect the propagating modes in the waveguide and create resonances whenever there
is a standing wave. This is illustrated in this section for cavities based on circular
waveguides and in the following section for those based on rectangular waveguides.
Figure 3.1 shows schematically the TM 010 mode of a resonator based on a cir-
cular waveguide. The possible resonant modes are derived from the TE and TM
modes of the waveguide by applying the appropriate boundary conditions in the z
direction. The TM 0,n modes are of greatest interest because they have a longitudinal
component of the electric field on the axis that can interact with an electron beam.
The tangential component of the electric field must be zero on all the metal surfaces
and the resonant condition is therefore

β m,n h = p π, (3.41)
105

Pill-Box Cavity Resonators 105

where p = 0, 1, 2  and β m,n is the propagation constant of the TM m,n mode of the
circular waveguide found using (2.11) and (2.85). The resonant mode which satisfies
(3.41) is the TM m,n, p mode of the cavity. The z component of the magnetic field is
zero for all these modes and the field patterns can be deduced from Figure 2.15 by
placing the end walls of the cavity at planes where the transverse component of the
electric field is zero.
The lowest mode is the TM 010 mode where the frequency is equal to the cut-​off
frequency of the TM 01 circular waveguide mode, and the electric field has only a z
component that does not vary with z. The field components are

E z = E0 J 0 (βC r ) exp ( jωt ) (3.42)

and, from (2.20),

ε0
Hθ = j E0 J1 (βC r ) exp ( jωt ) , (3.43)
µ0

where βC a = 2.405 . Thus the resonant frequency is

c
ω 0 = 2.405 . (3.44)
a
The stored energy in the cavity can be calculated from the maximum values of
either the electric or magnetic field. Thus
a
ε0
2 ∫0
W= 2 πrhE z2 dr. (3.45)

Substituting for E z from (3.42) gives

a
W = πhε 0 E02 ∫ J 0 (βC r ) r dr. (3.46)
2

0

Evaluating the integral gives [2]

π
W = ( ε 0 ha 2 E02 ) ⋅ J12 ( 2.405) = 0.423ε 0 ha 2 E02 . (3.47)
2
If the terminals of the equivalent circuit are chosen to be at the centres of the two
flat faces then the voltage across the cavity is

V0 = E0 h, (3.48)

so that, from (3.14), (3.44), (3.47), and (3.48)

 Rc  h µ0 h
 Q  = 0.491 a ε = 185 a Ω. (3.49)
U 0
106

106 Resonators

The theoretical value of the shunt resistance of a high Q cavity can be calculated by
assuming that the current density in the walls of the cavity is the same as in a loss-​
less cavity, and that the surface resistance is given by

1 ωµ 0
Rs = = , (3.50)
σδ 2σ
where σ is the conductivity of the walls of the cavity and δ is the skin depth [3].
Some investigators have reported measurements of anomalous surface resistance
at millimetre wavelengths that cause the surface resistance to be appreciably higher
than that obtained from (3.50). However, a careful review of the experimental evi-
dence concluded that these results are probably erroneous [4]. At frequencies up
to several THz the surface resistance can be calculated correctly from the classical
relaxation-​effect (or Drude dispersion) model [5]

ωµ 0  
1 + ( ωτ ) − ωτ  (3.51)
2
Rs =
2σ  
where τ is the relaxation time of the metal. Using the material constants for copper
( σ = 5.959 × 107 S m −1 and τ = 25.018 fs [4]) it is found that the error in using (3.50)
is less than 1% at 100 GHz.
The current density is equal to the tangential magnetic field on the cavity wall
and at right angles to it. Thus the current flow is radial in the flat ends of the cavity
and axial in the curved walls. The power dissipated in the walls is
a
PL = πahRs H θ ( a ) + 2 πRs ∫ H θ ( r ) r dr. (3.52)
2 2

0

Substituting for H θ from (3.43) gives

ε0  
a

ahJ1 ( 2.405) + 2∫ J1 (βC r ) r dr  (3.53)

2 2
PL = πRs E02
µ0  0 
and, evaluating the integral,

ε0
a ( a + h ) J1 ( 2.405) . (3.54)
2
PL = πRs E02
µ0
Then, from (3.12), (3.44), (3.47), and (3.54)

2.405 µ 0 h
QU = . (3.55)
2 ε 0 Rs ( a + h )
It is helpful to illustrate these equations by numerical examples. Table 3.1 shows
the theoretical parameters of copper cavities for which a = h . It can be seen that the
skin depth is small compared with the radius of the cavity in every case, so the effect
of the finite conductivity on the resonant frequency is negligible compared with
107

Pill-Box Cavity Resonators 107

Table 3.1:  Theoretical parameters of copper pill-​box cavity resonators (Worksheet 3.2)  (a = h , σ = 5.959 × 107 S m−1 )

Frequency (GHz) 1.0 3.0 10.0 30.0

a (mm) 114.7 38.25 11.47 3.82

δ (μm) 2.06 1.19 0.65 0.38
Rc/​QU (Ω) 185 185 185 185
QU 27,800 16,100 8,800 5,100
Rc (MΩ) 5.15 2.97 1.63 0.94

manufacturing tolerances and the effects of thermal expansion. The values of QU

are high enough to permit the representation of the cavity losses by a shunt resistor.

3.3.1 Effects of Surface Roughness

The Q factors achieved in practice are commonly less than the theoretical values
in Table 3.1 because the effective lengths of the current paths are increased by sur-
face roughness. This problem has been studied theoretically by considering surfaces
in which there are periodic grooves normal to the current flow [6, 7]. If the skin
depth is small compared with the surface roughness then the resistance is increased
by the ratio of the actual path length to the ideal path length [7]. When the skin
depth is comparable to, or greater than, the surface roughness the resistance is less
than that calculated from the path length and tends to the theoretical resistance as
the skin depth increases. The resistance depends on the shape and the spacing of
the grooves. Similar results have been obtained by modelling surfaces with random
roughness [8]. It was found that the additional loss depends upon the RMS rough-
ness, the correlation length and the correlation function. Grooves parallel to the
current flow have a similar, but somewhat smaller, effect to those normal to the cur-
rent flow. The resistance rises rapidly as the skin depth decreases and then reaches
a value which is approximately constant [7].
An empirical formula for the effect of surface roughness is

Rr 2   ∆2 
= 1 + arctan 1.4    , (3.56)
Rs π   δ  
where Rs is given by (3.50), δ is the skin depth, and Rr is the surface resistance of
a surface having RMS roughness ∆[8–​10]. Figure 3.12 shows the variation of nor-
malised surface resistance with ∆ δ given by (3.56). While this figure indicates in
general how the surface resistance varies with the surface roughness it is clear from
the discussion above that it cannot represent all possible surface conditions.
Experimental measurements of the losses in copper waveguides at 24 GHz
showed normalised loss ranging from 1.09 for a machined surface to 1.8 for an elec-
troplated surface [11]. It was shown that the loss could be increased by increasing
108

108 Resonators

Figure 3.12:  Empirical relationship between surface resistance and surface roughness.

the roughness of the surface. Waveguide measurements at 9.4 GHz were made by
Benson and his co-​workers, who also made measurements of the surface roughness
of the experimental samples [12–​14]. It was found that the effects of surface rough-
ness could be reduced by electro-​polishing or chemical polishing. Table 3.2 shows
the normalised surface resistance of copper waveguides. Formulae were derived
for waveguide losses which included the effects of surface roughness defined by
the ratio of the actual length of the surface to the ideal length. Three different fac-
tors were included to account for the differences in the roughness of the walls of
the waveguide normal to each component of the current flow. The values of these
factors were measured for the waveguides studied and it was concluded that the
increased loss could be explained solely by the increased path length when the skin
depth was small compared with the roughness. For example a 3 GHz brass wave-
guide in which the surface roughness was about an order of magnitude greater than
the skin depth was found to have normalised path lengths in the range 1.10 to 1.73
with an average value of 1.34.
Measurements at different frequencies made using copper waveguides are sum-
marised in Table 3.3 [13, 15]. The theoretical surface resistance was calculated from
measured values of the bulk DC conductivity. The effects of surface roughness
were calculated and it was concluded that the additional loss could be explained
in this way at 10 GHz but not at 35 GHz. It seems likely that the difference is to be
explained by the neglect of factors other than the additional path length in the cal-
culated values. It was found that the surface resistance could be reduced by anneal-
ing and this was explained in terms of the modification of the conductivity of a
surface layer of material which had been work-​hardened during manufacture [15].
Measurements of conduction losses have been made using resonators with arti-
ficially roughened surfaces at 35 GHz [16], and between 0.4 and 0.85 THz [17]. The
variation of surface resistance with RMS surface roughness can be described by
(3.56) if Δ/δ ≥ 1. The asymptotic value corresponds to the increase in the surface
area as expected. When Δ/δ < 1 the agreement with theory is poor. It was suggested
109

Pill-Box Cavity Resonators 109

Table 3.2:  Ratio of measured to calculated surface resistance

for copper waveguides at about 10 GHz [12]

Copper Rr Rs

Bright electroplate 1.001

Electro-​polished 1.002
Chemically polished 1.003
Precision drawn 1.012

Table 3.3:  Ratio of measured to calculated surface resistance

for drawn copper waveguides [11, 13, 15]

Frequency (GHz) Rr /Rs

9.375 1.034
24 1.37
35 1.55–​1.57
70 1.7–​2.5
140 2.1–​2.5

in [16] that the difference could be explained by an anomalous skin effect but this
conclusion has been challenged by Lucyszyn [4].
From the preceding discussion it can be concluded that the effects of surface
roughness may be represented by a modified surface resistance. The nature of the
surface roughness and the direction of scratches and machining marks are impor­
tant and it is not easy to calculate the effective surface resistance in general. When
the surface roughness is appreciably greater than the skin depth the effective resis­
tance can be calculated if the effective path length is known. The unloaded Q fac-
tor of a cavity resonator therefore depends on the manufacturing process as well
as on the material from which it is made. In the case of the pill-​box cavity azi-
muthal machining marks caused by turning the cavity parts on a lathe would have
the greatest effect. The reduction in QU caused by surface roughness is not easy to
quantify on theoretical grounds and estimates must be made from previous experi-
ence. In one case where the surface roughness was comparable with the skin depth
at 3.2 GHz it was found that the measured Q was 72% of the theoretical value [18].

3.3.2 Higher-Order Modes
In addition to the fundamental resonance a pill-​box cavity has an infinite set of
higher-order resonances. The circularly symmetrical TM 0 n 0 modes correspond to
the solutions of J 0 (βC a ) = 0 which occur at βC a = 2.405, 5.520, 8.654,  [2]. When
m > 0 the radial variation of the electric field is as J m (βC r ) and the resonances of
10

110 Resonators

the TM mn0 modes are the zeroes of J m (βC a ). The electric field of these modes is
weak close to the axis and therefore they do not interact strongly with an electron
beam located there. Other higher-order modes (TM mnp and TE mnp with p ≥ 1) exist
in which E has components other than E z but these modes are not usually excited.
However, because a modulated electron beam has currents which are harmonics
of the signal frequency it is necessary to check that the cavities do not have higher-
order resonances at these frequencies.

3.4 Rectangular Cavity Resonators

A rectangular cavity resonator can be developed from a rectangular waveguide as

shown in Figure 3.13. The lowest TM mode in a rectangular waveguide is the TM11
mode shown in Figure 2.6. The z component of the electric field of the TM110 res-
onance of the cavity is

 πx   πy
E z = E0 sin   sin   exp ( jωt ) . (3.57)
 a   b 
Substituting this expression in (2.36) we obtain
2 2
 π  π
βC2 =   +   . (3.58)
 a  b
The components of the magnetic field in an empty waveguide are found from
(2.20) to be

ε0 a  πx   πy
Hx = j E0 sin   cos   exp ( jωt ) (3.59)
µ0 a +b
2 2  a   b 

ε0 b  πx   πy
Hy = − j E0 cos   sin   exp ( jωt ) . (3.60)
µ0 a +b
2 2  a   b 

Figure 3.13:  Arrangement of a rectangular cavity resonator.

1

Re-entrant Cavities 111

The stored energy is

a
⌠ b 
h 2⌠ 2 
πx  2  π y  
W = ε 0 E0  sin   sin   dy dx, (3.61)
2 ⌡ a b 

⌡ 0 
0

which can be integrated to give

ε0
W= abhE02 . (3.62)
8
Then, since V0 = E0 h

R 4 h µ0
= ⋅ . (3.63)
Q π a + b2
2 ε0
The Q of the cavity can be found in the same way as that for the pill-​box cavity [3].
The result in our notation is

 2 h (a 2 + b 2 ) 
32
π µ0
Q=   . (3.64)
4Rs ε0  ab (a + b ) + 2h (a + b ) 
2 2 3 3

Rectangular cavities support higher-​order modes similar to those of the pill-​box cavity.

3.5 Re-​entrant Cavities

Pill-​box cavities are not commonly used in microwave tubes. The reason for this can
be understood by considering the time (t = g u0 ) taken for an electron travelling
with velocity u0 to cross a cavity gap of length g. The change in the phase of the
electric field in the cavity during this time is ωt and it is desirable that this should
have a value of less than, say, π 3 to ensure that the field seen by each electron is
approximately constant as it crosses the cavity. Thus

ω π
≤ . (3.65)
u0 3 g
Combining this equation with (3.44) for a pill-​box cavity where g = h , we find that,
in order to keep the transit angle within the limits specified,

h u
≤ 0.435 0 . (3.66)
a c
Most linear-​beam tubes operate in the region from 5 to 100 kV and the corre-
sponding range of u0 c is 0.1 to 0.5. Table 3.4 shows the theoretical parameters of
a copper pill-​box cavity at 3 GHz for a range of beam voltages. The cavity radius
is 38.2 mm and the skin depth is 1.22 μm in each case. It can be seen that the low
12

112 Resonators

Table 3.4:  Theoretical parameters of copper pill-​box cavities at 3 GHz

u0 c 0.1 0.3 0.5

V0 (kV) 2.6 25 80
ha 0.044 0.131 0.218
a (mm) 38.3 38.3 38.3
h (mm) 1.68 5.01 8.34
QU 1354 3722 5751
Rc/​QU (Ω) 8.14 24.2 40.3
Rc (kΩ) 11.0 90.2 232.0

values of h a required by (3.66) leads to very low unloaded Q factors and shunt
impedances. For this reason simple cavities are only used in high-​power (forward-​
fundamental) coupled-​cavity TWTs, where the cavity performance is acceptable,
and in high frequency (millimetre wave) tubes where constructional difficulties rule
out the use of cavities having other shapes.
The cavities used in microwave tubes usually have the re-​entrant cylindrical shape
shown in Figure 3.14. Consider the comparison between a pill-​box cavity and a re-​
entrant cavity which have the same resonant frequency and interaction gap length.
The capacitance across the interaction gap is smaller in the re-​entrant cavity than
it is in the pill-​box cavity. Therefore the inductance of the re-​entrant cavity must be
greater than that of the pill-​box cavity to keep the resonant frequency constant. As
a result the R Q of the re-​entrant cavity is greater than that of the pill-​box cavity.

3.5.1 Method of Moments Model of Re-​entrant Cavities

The properties of a re-​entrant cavity having the general shape shown in Figure 3.14
can be calculated with great accuracy using the method of moments. The outline
of the method is described in [19] and further information is given in [20, 21]. The
cavity is divided into three concentric regions ( I , II and III ), each having constant
axial length, whose outer radii are a, a’ and A respectively. The axial component of
the electric field and the azimuthal component of the magnetic field are expanded
in Fourier series of basis functions on r = a and r = a ′ so that

Ez  ∞
 emII   mπ z
 = ∑  II  cos  (3.67)
H θ  m = 0 hm   g 

in region II and similarly for the other regions, with the number of terms in the sum-
mations chosen independently for each region. The (unknown) amplitudes are writ-
ten as e, h on r = a and ee, hh on r = a ′. The requirement that the fields should satisfy
Maxwell’s equations and the boundary conditions in region I can be expressed as

[ hI ] = G I ( k ) [e I ], (3.68)

13

Re-entrant Cavities 113

Figure 3.14:  Arrangement of a cylindrical re-​entrant cavity resonator.

where k = ω c and the diagonal matrix GI is defined in [20]. Similarly in region III

[ hhIII ] = G III ( k ) [ee III ]. (3.69)

The constants at the inner and outer boundaries of region II are related to one
another because the fields are solutions of Maxwell’s equations and this can be
expressed as

 hmII   emII 
 hh II  = U m ( k ) ee II  . (3.70)
 m  m

Equation (3.70) can be rearranged as a partitioned matrix

 h II  U1,1 ( k )  U1,2 ( k )   e II 
  =  

   …  . (3.71)
  
 hh II  U 2,1 ( k )  U 2,2 ( k ) ee II 

Expansion of Ez in region II in terms of the basis functions in regions I and III yields

[e I ] = [ P1 ][e II ] (3.72)
and

[ee III ] = [ P2 ][ee II ]. (3.73)

Expansion of H θ in regions I and III in terms of the basis functions in region
II yields

[ hII ] = [Q1 ][ hI ] (3.74)

14

114 Resonators

and

[ hhII ] = [Q2 ][ hhIII ]. (3.75)

The forms of the connection matrices [ P ] and [Q ] are given in [20]. Using (3.68),
(3.69), (3.71), and (3.72) to (3.75) all the coefficients except eII and eeII can be elimi-
nated to give

 e II 
W ( k )    = 0, (3.76)
ee II 

so that the resonant frequency is given by the solution of

W ( k ) = 0. (3.77)
The eigenvector containing the electric field amplitudes in region II can also be
found and, from them, the electric and magnetic fields throughout the cavity. Once
these are known the stored energy and power dissipation can be calculated for any
given gap voltage and, hence, the QU and Rc QU can be found. This method is
implemented in Worksheet 3.3. The termination of the beam hole by a conducting
boundary has no effect on the results if the depth of the beam hole is large enough.
It can be shown that, if the number of terms in the series in region II is finite,
while the number of terms in regions I and III tend to infinity, then the solution is
an upper bound on the frequency. The reverse procedure leads to a lower bound
so that the accuracy of the solution is known. It is also found that, if the numbers
of terms in the series are chosen so that the smallest wavelength in each region is
approximately the same, then the result converges very rapidly with increasing num-
bers of terms to a very accurate figure. When this method was used with 8 terms in
the series representation of the fields within the gap it was found that the resonant
frequency, Rc QU and QU could be computed to accuracy better than 0.01% for a
wide range of cavity shapes [20]. When the same cavities were modelled using a
commercial electromagnetics code it was found that considerable care, and much
greater computational time, were needed to achieve the same accuracy. Thus the
method of moments is valuable for rapid computation of the properties of cavi-
ties with the shape shown in Figure 3.14, and for benchmarking calculations using
computational electromagnetics.
Figure 3.15 shows the results obtained when Worksheet 3.3 was used to investi-
gate the properties of re-​entrant cavities having a resonant frequency of 3 GHz and
an interaction gap of 5 mm. These can be compared with the results for a pill-​box
cavity with u0 c = 0.3 in the third column of Table 3.4. The inner and outer radii of
the drift tube were chosen to be 5 mm and 7 mm respectively. These values are typi-
cal of those encountered in practice. The height of the cavity (h) was increased in
steps and the outer radius ( A) adjusted so that the correct frequency was obtained.
Because the increase in height increases the inductance more than it decreases the
capacitance it is found that the outer radius decreases steadily. From Figure 3.15
15

Re-entrant Cavities 115

Figure 3.15:  Variation in the theoretical properties of a re-​entrant cavity resonator with h g

normalised to those of a pill-​box cavity ( h = g ).

we see that, as the height of the cavity increases, Rc QU increases initially as the
capacitance is reduced, and the inductance increased, to maintain the correct fre-
quency. However, when the normalised height of the cavity is greater than 5, Rc QU
decreases because the side wall is close enough to the drift tube to increase the
capacitance. This is not desirable because the increased radial component of the
electric field reduces the axial component available for interaction with electrons
passing through the cavity. The unloaded Q and the shunt resistance of the cavity
show similar behaviour with an initial increase followed by a decrease. Figure 3.16
shows the electric field distribution in one quarter of a transverse plane of a cavity
for which h = 4 g .
Table 3.5 shows the comparison between the properties of a pill-​box cavity with,
and without, a beam hole, and a re-​entrant cavity whose height is four times the
interaction gap. These results need to be interpreted with a little caution. In the first
place they only illustrate the behaviour of cavities having the dimensions chosen.
Results similar to these but with differences in detail may be expected for cavities
designed for other electron velocities and other frequencies. Secondly, no account
has been taken of the effects of surface roughness which will reduce both QU and
Rc in practice.

3.5.2 Fujisawa’s Model of Re-​entrant Cavities

The design of klystrons and inductive output tubes requires the design of cavity
resonators which have a given frequency, interaction gap length, and drift tube
radius. As can be seen from Figure 3.15 the height, or the radius of the cavity, may
be chosen freely. The remaining parameter is then fixed by the resonant frequency.
The method of moments model described in the preceding section provides a quick
way of calculating the properties of a cavity. However, for parametric studies it
may be faster to use an approximate equivalent circuit model [22]. In this method
a doubly re-​entrant cavity without a beam hole is modelled by the circuit shown in
16

116 Resonators

Figure 3.16:  Electric field in a re-​entrant cavity resonator for which h = 4 g .

Table 3.5:  Theoretical properties of copper cavities at 3 GHz (u0 /c = 0.3)

Pill-​box Pill-​box with beam hole Re-​entrant

a ( mm ) –​ 5.00 5.00
a ′ ( mm ) –​ –​ 7.00
A ( mm ) 38.25 38.61 26.11
g ( mm ) 5.00 5.00 5.00
h ( mm ) 5.00 5.00 20.00
QU 3715 3712 7959
Rc/​QU (Ω) 24.2 23.5 103.3
Rc (kΩ) 89.9 87.2 822.1

Figure 3.2b. The capacitor is made up of three capacitors in parallel (CI ,CII ,CIII )
which represent the contribution to the capacitance from charge on the inside, the
end, and the outside of the drift tube, respectively.
The capacitance associated with charge on the inside of the drift tube can be cal-
culated to sufficient accuracy by quasi-​static analysis. To do this we assume that the
axial component of the electric field is constant in the gap when r = a . This field can
be expressed in terms of an infinite set of components having sinusoidal variation
exp ( j βz ) in the axial direction by taking the Fourier transform of the field at r = a
g
⌠ 2 Vg  sin (β g 2 ) 
Γ (β ) =  exp ( − j βz ) dz = Vg  , (3.78)
⌡− g g  β g 2 
2

where Vg is the gap voltage. Since we have assumed that the electric field inside
the drift tube satisfies Laplace’s equation we know that the radial variation of E z
is as I 0 (βr ) so that, taking the inverse Fourier Transform, we find that, within the
drift tube
∞
Vg ⌠ I 0 (βr ) sin (β g 2 )
E z ( r, z ) = ⋅ exp ( − j βz ) d β. (3.79)
2π 
⌡ I 0 (βa ) βg 2
−∞
17

Re-entrant Cavities 117

The total charge within one drift tube can be found by finding the total charge on a
conducting sheet at z = 0 for r ≤ a

a
Q = ε 0 ∫ 2 π r E z ( r, 0 ) dr. (3.80)
0

Substituting for E z from (3.79) and performing the integration we find that the
capacitance in region I is

∞
⌠ aI1 (βa ) sin (β g 2 )
CI = ε 0  ⋅ d β. (3.81)
⌡ βI 0 (βa ) βg 2
−∞

It can be shown that this capacitance is the same as that given in Figure 19 of [22].
Since we have assumed that the field in the gap is uniform the capacitance in region
II is simply

π (a ′ 2 − a 2 )
CII = ε 0 . (3.82)
g

The capacitance in region III is [22]

e ( A − a ′ )2 + ( h 2)2 
CIII = 2 ε 0 a ′ ln  . (3.83)
 g 

Thus the total capacitance is

C = CI + CII + CIII . (3.84)

The inductance is calculated by assuming that the azimuthal magnetic field is gen-
erated by a uniform axial current I at r = a ′, and is negligible when r < a ′, so that

I
Hθ (r ) = . (3.85)
2 πr
Then the self-​inductance is

µ0 h  A 
L= ln   . (3.86)
2π  a ′ 

The resonant frequency and the R Q of the cavity can then be calculated using (3.2)
and (3.6).
The stored energy is estimated from the magnetic field

µ0 h 2  A 
W= I ln   . (3.87)
4π  a′ 
18

118 Resonators

The power dissipated on the surface of the cavity is also calculated from the
magnetic field

1 2  h (h − g )  A
PL = I Rs  + + 2 ln    , (3.88)
4π A a′  a′  

so that the unloaded Q can be calculated using (3.12).

Table 3.6 shows a comparison between the theoretical properties of re-​entrant
copper cavities with, and without, a beam hole calculated using the method of
moments and Fujisawa’s method (see Worksheet 3.4). It can be seen from these
examples that, using Fujisawa’s method, the accuracy of the frequency is of the
order of 1% and, of the other parameters, approximately 5%. A detailed study of
a wide range of cavities without beam holes showed that the accuracy of the fre-
quency computed using Fujisawa’s method is better than 5% provided that A a′ < 5
and z3 z2 < 12 . For cavities whose shapes are typical of those used in microwave
tubes the accuracy can be expected to be better than this as indicated by Table 3.6.

3.5.3 The Interaction Field

An electron beam passing along the axis of a circularly symmetrical cavity interacts
with the fringing electric field of the gap in the drift tube. In the previous section it
was assumed, for simplicity, that the axial component of the electric field at r = a
was constant. However, it is clear from Figure  3.16 that this is not true because
there are field concentrations close to the drift tube noses. The field distribution
depends upon the shapes of the drift tube noses as well as upon the gap length and
drift tube radius. However, in order to avoid interception of electrons on the drift
tube, the radius of the electron beam is normally not greater than 2a 3. Within that
radius it is found that the variation of the axial component of the electric field does

Table 3.6:  Theoretical properties of re-​entrant copper cavities at 3 GHz (v0 c = 0.3)

Without beam hole With beam hole

Method of moments Fujisawa (error %) Method of moments Fujisawa (error %)

a ( mm ) 5.0 0 5.0 5.0

a ′ ( mm ) 7.0 7.0 7.0 7.0
A ( mm ) 24.39 24.39 26.11 26.11
g ( mm ) 5.0 5.0 5.0 5.0
h ( mm ) 20.0 20.0 20.0 20.0
f (GHz) 3.000 2.987 (−0.4) 3.000 2.964 (−​1.2)
Rc/​QU (Ω) 99.7 93.7 (−6.4) 103.3 98.1 (−​5.3)
QU 7499 7666 (+2.2) 7959 7934 (−​0.3)
Rc (kΩ) 747.7 718.3 (−4.1) 822.1 778.1 (−​5.7)
19

Re-entrant Cavities 119

not depend strongly on the field profile at r = a . It is therefore possible to draw use-
ful conclusions about the interaction field by comparing a number of approximate
field profiles. In addition to the uniform field discussed above two other profiles
may be considered [23]. If the drift tube noses are knife edged then

 Vg
 z <g 2
E z ( a, z ) =  π ( g 2)2 − z 2 , (3.89)

 0 z ≥g 2

where Vg is the gap voltage. A useful approximate profile is

 k cosh ( kz )
 Vg z <g 2
E z ( a, z ) =  2 sinh ( kg 2 ) , (3.90)
 0 z ≥g 2

where k is chosen to match the field to that determined for the actual gap by com-
putation or measurement. This field tends to the uniform field when k → 0 and it
leads to an interaction field which is very close to that given by knife-​edged drift
tubes when k = 4 g. The dependence of the interaction field on the field profile and
the normalised gap length can be investigated using Worksheet 3.5. Because quasi-​
static fields have been assumed it is possible to normalise all dimensions to the drift
tube radius. Figure 3.17 compares some typical results when g a = 1 for a uniform
field at r = a , a knife-​edge field, and one which is intermediate between them. It can
be seen that, in the region of space occupied by the electron beam, there is very
little difference between the fields resulting from the different choices of profile.
At r = a 2 the axial field at the centre of the gap is slightly greater than that on the
axis, and it falls off a little more rapidly with increasing z. When the normalised
gap length is g a = 0.5 it is found that the differences between the results using the
three profiles are negligible, but that there is a greater difference between the fields
on the axis and at r = a 2. When the normalised gap length is 2.0 there is a greater

Figure 3.17:  Axial variation of E z in a gap in a drift tube when g = a for three different
field profiles at r = a : (A) uniform field, (B) hyperbolic cosine field ( k = 2 g ), and
(C) knife-​edge field.
120

120 Resonators

difference between the fields for the different profiles but smaller radial variation.
Overall, setting g a = 1 gives a good compromise between field uniformity in the
radial direction and insensitivity to the field profile in the gap.

3.5.4 Practical Re-​entrant Cavities

In the cavities used in practical tubes it is common for all, or part, of the outside
of the drift tube to be conical, and for the tips of the drift tubes to be rounded to
reduce the peak electric field on the surface. The gap length is often approximately
equal to the inner radius of the drift tube but the actual dimensions are chosen to
achieve a compromise between:
• the electrical properties of the cavity;
• the axial space available;
• avoiding voltage breakdown in the gap and multipactor discharges;
• thermal conduction and mechanical strength of the cavity.
The multipactor discharge is discussed in Section 18.8. The risk of gap breakdown
may be further reduced by changes to the shape of the drift tube noses and by the
application of coatings having low secondary electron emission coefficients [24].
If it is necessary to be able to change the frequency of a cavity this can be done
by adding a tuner. This typically takes the form of a metal plate parallel to the
drift tube whose position can be changed so that the capacitance between it and
the drift tubes can be varied as shown in Figure 3.18(a) [25]. In some UHF kly­
strons and inductive output tubes the drift tube is surrounded by a ceramic cylinder,
which forms the vacuum envelope of the tube as shown in Figure 3.18(b). The outer
part of the cavity is then formed in two halves, which are bolted around the tube.
Because the outer part of the cavity is at atmospheric pressure it is possible to make
parts of the cavity wall moveable with spring fingers to maintain good electrical
contact. This type of cavity is called an external cavity.

3.6 External Coupling to Cavities

The external connections to cavity resonators used in tubes can be made using
either coaxial lines or metallic waveguides. The strength of the coupling is adjusted
to achieve the desired external Q.
A coaxial line may be terminated in either an electric, or a magnetic, antenna.
An electric antenna is made by removing a short section of the outer conductor
and the insulation so that the exposed central conductor couples to the electric
field in the cavity. A  magnetic antenna is made in a similar manner by forming
the exposed central conductor into a loop which may be connected to the outer
conductor (see Figure 3.19(a) and (b)). The loop is inductively coupled to the com-
ponent of the magnetic field in the cavity that is normal to the plane of the loop.
Both of these techniques are used to make probes for experimental measurements
12

External Coupling to Cavities 121

Figure 3.18:  The arrangement of cavity tuners in: (a) internal cavities and (b) external
cavities.

Figure 3.19:  Loop-​coupled cavity: (a) arrangement with a closed loop, (b) arrangement with

an open loop, and (c) equivalent circuit.

on cavity resonators. The strength of the coupling between the antenna and the
mode is affected by
• the size of the antenna;
• the depth of insertion of the antenna into the cavity;
• the angle between the plane of the loop of a magnetic antenna, and the direction
of the local magnetic field.
12

122 Resonators

Electric antennas are unsuitable for coupling into the fundamental TM010 mode
of a cavity in a tube because the region of greatest electric field is inaccessible, and
because the electric field concentration on the antenna could cause electric break-
down. Coupling using a loop (magnetic antenna) is discussed in Section 3.6.1.
When the external connection to a cavity is made using a waveguide it is arranged
so that the field of the fundamental mode in the waveguide couples strongly to the
cavity mode which is to be excited. Thus the broad walls of a rectangular waveguide
are normally perpendicular to the axis of the cavity (see Figure 3.22). The strength
of the coupling is adjusted by changing the size of the aperture (iris) where the
waveguide meets the cavity. Iris coupling is discussed in Section 3.6.2.

3.6.1 Loop Coupling

Figure 3.19 shows the general arrangement of a cavity with a coupling loop. The loop
may either be closed as shown in Figure 3.19(a) or open as shown in Figure 3.19(b).
The loop is coupled inductively to the magnetic field in the cavity and the cou­
pling may be adjusted by rotating the loop around the axis of the coaxial line. The
equivalent circuit of a loop-​coupled cavity is shown in Figure  3.19(c) where the
parameters of the cavity have suffix c. The mutual inductance between the cavity
and the loop is kLc where k is the fraction of the magnetic flux within the cavity
which is coupled to the loop. The self inductance of the loop is LL . Analysis of the
equivalent circuit shows that the input impedance is

k 2 Rc
Zin = j ωLL + , (3.91)
 ω ω0 
1 + jQU  −
 ω 0 ω 

where, from (3.6), Rc = QU2 rc. Comparison between this result and (3.5) shows that,
apart from the effect of the loop inductance, the coupling can be represented by an
ideal transformer having ratio k :1.
The coupling factor k can be calculated to good accuracy, if the distribution of
the magnetic field in the cavity is known, by finding the fraction of the flux which is
linked to the loop [26]. The self inductance of the loop can be calculated by static
analysis if the dimensions of the loop are small compared with the free-​space wave-
length. Formulae for the self inductance of a loop in free space have been given
for a rectangular loop [26] and for a circular loop [3]. The self-​inductance can also
be calculated by treating the loop as a short-​circuited two-​wire line. The results
obtained in these ways agree with one another to around 20%. The self inductance
of a loop can be measured by noting that the second term in (3.91) is negligible at
a frequency remote from the resonant frequency of the cavity. It is found that the
inductance measured in this way is up to an order of magnitude smaller than the
free-​space inductance. This is because the magnetic field of the loop is confined
within the cavity which increases the reluctance of the flux path and reduces the
flux generated by unit current. A very crude estimate of this effect can be made by
123

External Coupling to Cavities 123

supposing that the flux line passing through the centre of a circular loop forms a
circle having the same radius as the loop. The flux density at the centre of the loop
is then reduced by a factor of π compared with that of the same loop in free space.
The model described above is implemented in Worksheet 3.4 which can be used to
explore the properties of re-​entrant cavities with loop coupling. Figure 3.20 shows
the reflection coefficient S11 for a typical cavity if the loop inductance is neglected,
and when it is reduced by a factor of π from the value calculated by treating the
loop as a short-​circuited two-​wire line. This estimate of the loop inductance gave
good agreement with experimental results in a particular case. It can be seen that
the effect of the loop inductance is to shift the resonant frequency of the cavity
slightly and also to change the magnitude of Zin at resonance. To investigate this
effect we assume that the unloaded Q of the cavity is high so that the bandwidth is
narrow, and the loop reactance can be assumed to be constant. Figure 3.21 shows
the equivalent circuit redrawn with an ideal transformer and constant loop react-
ance X = ω 0 LL connected to a source with impedance RS . This circuit can be put in
the form shown in Figure 3.5 by writing

1 k2 R − jX 2
+ jBE = = S2 k . (3.92)
RE (RS + jX ) (RS + X 2 )

Figure 3.20:  Reflection coefficient of a loop-​coupled cavity (QU = 7800,QE = 330 ):

(a) ignoring the loop inductance, and (b) including the loop inductance.

Figure 3.21:  Equivalent circuit of a loop-​coupled cavity with an external resistance.

124

124 Resonators

so that

RE =
(RS2 + X 2 ) (3.93)
k 2 RS
and

k2X
BE = − . (3.94)
(RS2 + X 2 )
The external Q of the cavity and the coupling factor K can then be found from
(3.18) and (3.19). The condition for resonance is that the sum of the reactances
should be zero. Thus the resonant frequency of the cavity with external loading is
the solution of

1 k2X
ωCc − − 2 = 0. (3.95)
ωLc (RS + X 2 )

3.6.2 Iris Coupling

Figure 3.22 shows the arrangement of a cavity coupled to a rectangular waveguide
through an iris. The width of the waveguide is normally that of a standard wave-
guide, but the height may be reduced. The coupling factor is generally adjusted by
changing the width of the aperture to obtain the desired external Q.
The coupling between a waveguide and a cavity can be modelled using the the-
ory of coupling by small holes [27–​30]. In this theory the effect of a small hole is
represented by an electric dipole whose magnitude is proportional to the normal
component of the electric field on the iris, and by two magnetic dipoles whose mag-
nitudes are proportional to the transverse components of the magnetic field. It is

Figure 3.22:  Arrangement of an iris-​coupled cavity.

125

External Coupling to Cavities 125

convenient to consider the case in which power is transferred from the cavity to the
waveguide through the aperture. In the case of the TM 010 mode in the cavity shown
in Figure 3.22 the only non-​zero field component at the iris is the magnetic field
parallel to the broad wall of the waveguide. The magnetic dipole moment is then

M = α m H1 , (3.96)

where H1 is the tangential magnetic field of the cavity at the centre of the aperture
and α m is the magnetic polarisability of the aperture that depends only upon its
shape and size. Since the hole has been assumed to be small the polarisability can
be determined by quasi-​static analysis. Analytical expressions for α m are available
for circular and elliptical apertures [30], and graphs for apertures of other shapes
have been found experimentally [31]. The amplitude of the transverse magnetic
field induced in the waveguide is given by

2β g
H0 = α m H1 (3.97)
ab
if the coupling hole is aligned with the centre of the waveguide [30]. The amplitude
of the voltage between the centres of the broad walls of the waveguide is

βw
V0 = E0 b = Zw H 0 b (3.98)
βg

from (2.38) where βw = ω c and Zw = µ 0 ε 0 . Now the coupling factor can be

written

PE V02
K= = , (3.99)
PL 2Z g PL

where PE and PL are the powers dissipated in the external resistance and in the cav-
ity walls. Then, substituting from (3.97) and (3.98) we obtain

K=
2  Zw βw 
2
(α m H1 )2 . (3.100)
 
Zg a  PL

Since we are working in terms of power and voltage the appropriate waveguide
impedance is ZPV given by

b β
ZPV = 2 ⋅ w Zw (3.101)
a βg

so that

β g βw H12 Zw 2
K= ⋅ α m . (3.102)
ab PL
126

126 Resonators

Substituting for the power loss in the cavity from (3.12)

2β g µ 0 H12QU 2
K= ⋅ ⋅ α m . (3.103)
ab 2W
The coupling factor is seen to be the product of three terms: the first depends only
on the properties of the waveguide; the second only on the properties of the cavity;
and the third depends only on the properties of the aperture. The first term can be
written as 4 π abλ g which is just 4 π divided by the volume of the waveguide in one
guide wavelength. The second term relates the magnitude of the tangential mag-
netic field at the aperture to the stored energy in the cavity. For a pill-​box cavity H1
and W are given by (3.43) and (3.47) so that

 µ 0 H12  1
  = 2 , (3.104)
2W Pill box πrC h

where rC is the radius of the cavity. We note that the right-​hand side of (3.104) is
the reciprocal of the volume of the cavity. Similarly, for a rectangular cavity whose
width is d x and length is d y ,H1 and W are given by (3.59) and (3.62), and

 µ 0 H12  4  d x2 
  = . (3.105)
2W Rectangular d x d y h  d x2 + d y2 

When the cavity is square the right-​hand side of (3.105) reduces to 2 divided by the
volume of the cavity. In general we note that the ratio 2W µ 0 H12 is the volume in
which the stored magnetic energy of a uniform field H1 is equal to the stored energy
in the cavity [32].
For a small circular aperture of radius rA

4 3
αm = rA , (3.106)
3
which is the volume of a sphere which fills the aperture divided by π [28]. Thus the
coupling factor can be expressed as

 4 π 4 3   µ 0 H12 4 3 
K = ⋅ rA  ⋅  ⋅ rA  QU . (3.107)
 abλ g 3   2W 3 

The first bracket can be interpreted as the shunt reactance of a circular iris in a rect-
angular waveguide and the second as the coupling coefficient when two identical
resonators are coupled though a circular iris. Both brackets are expressed as ratios
of effective volumes [32].
It is helpful to consider a numerical example. Suppose that a pill-​box cavity
whose resonant frequency is 3 GHz, having the properties given in Table 3.1, is cou-
pled to standard WR284 waveguide by a circular hole. The internal dimensions of
the waveguide are 72 mm × 34 mm. The hole is taken to have radius 7.2 mm, which
127

External Coupling to Cavities 127

is about as large is it can be while satisfying the assumption that it is small enough
for the tangential magnetic field to be constant over it. The guide wavelength is
138 mm if the slight detuning of the cavity by the coupling is neglected. Hence
2
λ g ab πrC2 h  3 
QE = ⋅ ⋅ = 18700. (3.108)
2π 2  4rA3 
Now QU = 16100 so the coupling factor is 0.86. The use of this method to calculate
the coupling coefficient between a cavity and a waveguide in the range 0 to 1.1 is
described in [29]. The input and output couplings used in microwave tubes are gen-
erally strongly over-​coupled so that the external Q is much less than the unloaded
Q. For output coupling this is necessary to minimise the proportion of the output
power which is dissipated in the cavity losses. Thus the coupling apertures which
are required cannot be regarded as small. In particular their properties are modified
from those of ideal, isolated, apertures by the proximity of the walls of the cavity
and the waveguide and by the finite time taken for waves to propagate across them.
Useful insight can be gained into the properties of a rectangular cavity coupled
to a waveguide by a large inductive iris by studying the arrangement shown in
Figure 3.23(a). Inductive irises are commonly used because the risk of voltage break-
down is less than with capacitive irises. This problem can be modelled by the nor-
malised transmission line circuit shown in Figure 3.23(b). The losses in the cavity are
represented by a normalised lumped conductance g at the centre of the cavity where

Zg
g= . (3.109)
(R Q )QU
It can be shown that this corresponds to the usual definition of R Q if the wave-
guide impedance is taken to be ZPV . The susceptance of the iris is modelled to suf-
ficient accuracy for illustration by (2.97)

2π  πw 
b=− cot 2  . (3.110)
βg a  2a 

Figure 3.23:  (a) Arrangement of a rectangular cavity coupled to a waveguide by an inductive

iris, and (b) transmission line equivalent circuit.
128

128 Resonators

The input impedance of this circuit can be found using standard transmission line
theory and the perturbed resonant frequency is given by the minimum of S11. The
plot of the input impedance on a Smith chart is a perfect circle and the coupling para­
meter and loaded Q can be found using the method described by Kajfez [33]. When
the unloaded Q is computed from these results it is found to be very close to that
set in (3.109) so confirming the self-​consistency of the model (see Worksheet 3.6).
Figure 3.24 shows a typical plot with markers set to the resonant frequency and to
the 3dB points of the loaded resonator.
We saw in Section 2.4.3 that an inductive iris of any width can be modelled by an
equivalent circuit. The normalised shunt reactance of the iris, in ratio of volumes
form, is found from (2.103) and (2.104) to be

1 4π πw 2 b sin ( πw 2a )
x= ⋅ ⋅ . (3.111)
4 abλ g 4 (1 − w 2 a 2 )2 πw 2a

Comparing this expression with the first bracket in (3.107) we see that the polarisa-
bility of a wide aperture which occupies the full height of the waveguide is given by

1 πw 2 b sin ( πw 2a )
αm = ⋅ ⋅ . (3.112)
4 4 (1 − w a )
2 2 2
πw 2a

When w a << 1 this reduces to

1 πw 2 b
αm = ⋅ , (3.113)
4 4

Figure 3.24:  Polar plot (Smith chart) of the reflection coefficient of a rectangular cavity
coupled to a waveguide by an inductive iris showing the resonant frequency (♦) and the
3 dB points (▲).
129

Measurement of Cavity Parameters 129

Figure 3.25:  Equivalent circuit of an iris-​coupled cavity.

agreeing with the formula for a narrow aperture in [32]. Substituting in (3.107) from
(3.105) and (3.112) gives a closed-​form expression for the coupling factor.
2
1  wa  a  a2 
K = βg a  ⋅ sin ( πw 2a ) ⋅  2 2
QU . (3.114)
 2 (1 − w 2 a 2 )  d a +d 
2
2

The coupling factor computed from the distributed circuit model is almost exactly
proportional to (w a ) up to w a = 0.6. The results computed from (3.114) agree
4

with this very closely up to w a = 0.4 (see Model 1 in Worksheet 3.6). In principle
this method can be used to find the loaded Q of any cavity but formulae are not
available, in general, for the polarizability of rectangular apertures taking account
of adjacent boundaries.
The lumped-​element equivalent circuit of an iris-​coupled cavity is shown in
Figure 3.25 where the components with suffix c represent the properties of the cav-
ity. The waveguide has characteristic impedance Z g and is terminated by an exter-
nal load which we shall assume is matched to the waveguide. The iris is represented
by the reactance XA, computed using (3.110). By comparing the input impedance of
this network when ω = ω 0 with that of the distributed circuit in Figure 3.23(b) it can
be shown that, when the cavity is square, k = 1 π. This result is independent of the
reactance of the aperture, or of the choice of characteristic impedance of the wave-
guide. It is tempting to interpret k as the fraction of the circulating current which
is intercepted by the aperture. But considerations of symmetry would then require
k = 1 4 (since the reactance of the aperture has already been adjusted for the frac-
tion of the longitudinal current in the waveguide intercepted by it). It is found that
there is good agreement between this model with k = 1 π (Model 2 in Worksheet
3.6) and the distributed circuit model described above. The equivalent circuit model
is useful because it is valid for any cavity if the values of k and X A can be found.

3.7 Measurement of Cavity Parameters

Although it is possible to compute the properties of any microwave cavity to good

accuracy using computational electromagnetics, it is desirable to confirm the results
by cold test measurements. The parameters to be measured include the resonant
frequency, R/Q, unloaded Q and external Q. In addition it may be useful to map
the electric field intensity and to identify higher-order modes. The measurements
130

130 Resonators

may be made on cavities at full scale or scaled up to a larger size for convenience.
The cavities may be made of copper or brass. If the cavities are bolted together,
rather than being brazed, it is necessary to ensure that there is a good electrical
contact on all joints that intercept the current circulating within the cavity. It is
therefore important to ensure that the mating surfaces are a close fit to one another.
The author has obtained good results using inexpensive aluminium cavities with
conducting grease (obtainable from car accessory shops) on the mating surfaces.
The resonant frequency is measured by coupling into the cavity with small elec-
tric or magnetic probes and examining either the reflection (S11 ) or transmission
(S21 ) using a vector, or scalar, analyser. The transmission measurement is to be pre-
ferred because it is easier to detect the frequency of the resonant peak accurately.
Probes are readily made from prefabricated lengths of semi-​rigid coaxial cable with
SMA connectors at each end. The cable is cut in half and a short length of the outer
conductor and the insulation removed to make an electric probe. A magnetic probe
is made in a similar manner by exposing a greater length of the central conductor,
bending it to form a loop and soldering the free end to the outer conductor of the
line. We saw in Section 3.6.1 that the resonant frequency of the cavity is perturbed
by external coupling. It is therefore very important to make the coupling as weak
as possible to avoid this effect. The positions of the probes are adjusted to give the
greatest response for which there is no measurable perturbation of the frequency. If
necessary the frequency can be measured as the probes are gradually inserted and
the results extrapolated to zero insertion depth.
The R Q can be determined by measuring the change in frequency when a thin
dielectric rod is inserted on the axis of the cavity [34]. This method was originally
proposed by Slater [35]. It is assumed that, if the perturbation is small, then the
fields in the cavity outside the perturbing rod are unchanged. If a pill-​box cavity of
height h is perturbed by a rod having radius r and relative permittivity ε r then the
change in the capacitance of the cavity caused by the rod is

ε 0 ( ε r − 1) πr 2
∆C = . (3.115)
h
The perturbed resonant frequency is then

1  ∆C 
ω= = ω 0 1 −  . (3.116)
L (C + ∆C )  2C 

Now the stored energy in the cavity can be written

1
W = CV 2 , (3.117)
2
where V is the effective voltage across the cavity. Hence the frequency perturbation
can be written

ε 0 ( ε r − 1) πr 2 ω 0 V 2
ω − ω0 = − . (3.118)
h 4W
13

Measurement of Cavity Parameters 131

Making use of the definition of R Q in (3.14)

ε 0 ( ε r − 1) πr 2 ω 20  R 
ω − ω0 = −  Q  , (3.119)
2h
enabling R Q to be found from a measurement of the frequency perturbation. This
measurement requires accurate knowledge of the relative permittivity of the per-
turbing rod. That can be determined by using the rod to perturb a cavity whose
properties are known theoretically. It is obviously important for the frequency shift
to be large enough to be measured accurately. There is, therefore, a conflict between
the requirements for the perturbation to be small enough for the theory to be valid
and for it to be large enough to be measured accurately. Equation (3.118) can be
rewritten as

ω − ω0 = −
( ε r − 1) ω 0 ⋅ πr 2 h ε 0 E z2 (3.120)
2W 2
in which the second term can be recognised as the stored electric energy in the vol-
ume occupied by the rod. More generally, Slater’s perturbation theorem states that
the change in the resonant frequency is proportional to the change in the stored
energy and, thus, to the square of the magnitude of the local electric field. That
makes it possible for the electric field to be mapped within a cavity by measuring
the perturbation caused by a small bead. The electric field components can be mea­
sured individually by using a bead in the shape of a thin rod. A study of the accu­
racy of perturbation methods showed that the assumptions on which they are based
cease to be valid for quite small perturbing objects [36]. However, the errors from
different sources are of opposite signs so that the method gives useful accuracy even
when the rod is not very small. The frequency of a pill-​box cavity perturbed by a
dielectric rod was calculated exactly and compared with the result from (3.119). It
was found that the error in the method was about 0.3% when the ratio of the radius
of the rod to that of the cavity was 0.05 and ε r = 10. The same paper also examined
the perturbation of the frequency of a pill-​box cavity by a small dielectric sphere.
Higher-order modes can be identified by probing the field of the cavity with a
small dielectric rod inserted through holes drilled in the cavity wall. The strength of
the perturbation produced by a given depth of insertion of the rod provides an indi-
cation of the local strength of the electric field. It is therefore possible to detect the
zeroes and the maxima of the magnitude of the electric field, and hence to identify
the mode for each resonant frequency.
The loaded and unloaded Q of a cavity resonator can be measured accurately
using the methods described by Kajfez [33]. The input impedance of the cavity is
determined using a vector analyser with a result similar to that shown in Figure 3.24.
The coupling factor is given by

D
K= , (3.121)
2−D
132

132 Resonators

where D ( 0 ≤ D ≤ 2 ) is the diameter of the circle which is the locus of the impedance
on the Smith chart. The frequencies ( ω1,2 ) are found at the two points indicated
by triangular markers in Figure  3.24 which subtend an angle of 90° at the anti-​
resonant point and then

ω0
QL = . (3.122)
ω1 − ω 2
The unloaded Q of the cavity can then be determined using (3.20).

References

[1]‌ R. G. Carter, Electromagnetic Waves:  Microwave Components and Devices. London:

Chapman and Hall, 1990.
[2]‌ N. W. McLachlan, Bessel Functions for Engineers. Oxford University Press, 1954.
[3]‌ S. Ramo et al., Fields and Waves in Communication Electronics. New York: Wiley, 1965.
[4]‌ S. Lucyszyn, ‘Investigation of anomalous room temperature conduction losses in
normal metals at terahertz frequencies’, IEE Proceedings:  Microwaves, Antennas and
Propagation, vol. 151, pp. 321–​329, 2004.
[5]‌ E. Episkopou et  al., ‘Defining material parameters in commercial EM solvers for
arbitrary metal-​based THz structures’, IEEE Transactions on Terahertz Science and
Technology, vol. 2, pp. 513–​524, 2012.
[6]‌ S. P. Morgan, ‘Effect of surface roughness on eddy current losses at microwave frequen-
cies’, Journal of Applied Physics, vol. 20, pp. 352–​362, 1949.
[7]‌ C. L. Holloway and E. F. Kuester, ‘Power loss associated with conducting and supercon-
ducting rough interfaces’, IEEE Transactions on Microwave Theory and Techniques, vol.
48, pp. 1601–​1610, 2000.
[8]‌ L. Tsang et al., ‘Effects of random rough surface on absorption by conductors at micro-
wave frequencies’, IEEE Microwave and Wireless Components Letters, vol. 16, pp.
221–​223, 2006.
[9]‌ E. O. Hammerstad and F. Bekkedal, Microstrip Handbook. Trondheim, Norway:
University of Trondheim, 1975.
[10] T. Edwards, Foundations for Microstrip Circuit Design, 2nd ed. Chichester, UK: Wiley,
1992.
[11] E. Maxwell, ‘Conductivity of metallic surfaces at microwave frequencies’, Journal of
Applied Physics, vol. 18, pp. 629–​638, 1947.
[12] J. Allison and F. Benson, ‘Surface roughness and attenuation of precision-​drawn, chem-
ically polished, electropolished, electroplated and electroformed waveguides’, Proceedings
of the IEE-​Part B: Radio and Electronic Engineering, vol. 102, pp. 251–​259, 1955.
[13] F. Benson, ‘Waveguide attenuation and its correlation with surface roughness’,
Proceedings of the IEE-​Part III: Radio and Communication Engineering, vol. 100, pp.
85–​90, 1953.
[14] F. Benson, ‘Attenuation and surface roughness of electroplated waveguides’,
Proceedings of the IEE-​Part III: Radio and Communication Engineering, vol. 100, pp.
213–​216, 1953.
[15] F. Benson and D. Steven, ‘Rectangular-​waveguide attenuation at millimetre wavelengths’,
Proceedings of the Institution of Electrical Engineers, vol. 110, pp. 1008–​1014, 1963.
13

References 133

[16] F. J. Tischer, ‘Excess conduction losses at millimeter wavelengths’, IEEE Transactions on

Microwave Theory and Techniques, vol. 24, pp. 853–​858, 1976.
[17] M. P. Kirley and J. H. Booske, ‘Terahertz conductivity of copper surfaces’, IEEE
Transactions on Terahertz Science and Technology, vol. 5, pp. 1012–​1020, 2015.
[18] F. Horner et  al., ‘Resonance methods of dielectric measurement at centimetre wave-
lengths’, Journal of the Institution of Electrical Engineers  –​Part III:  Communication
Engineering, including the Proceedings of the Wireless Section of the Institution, vol. 93,
pp. 53–​68, 1946.
[19] R. G. Carter et al., ‘Rapid calculation of the properties of klystron cavities’, in 2008
IEEE International Vacuum Electronics Conference, pp. 142–​143, 2008.
[20] R. G. Carter et al., ‘Calculation of the properties of reentrant cylindrical cavity resona-
tors’, IEEE Transactions on Microwave Theory and Techniques, vol. 55, pp. 2531–​2538,
December 2007.
[21] R. Taylor, ‘Calculation of resonant frequencies of re-​entrant cylindrical electromagnetic
cavities’, Journal of Nuclear Energy, Part C: Plasma Physics, vol. 3, pp. 129–​134, 1961.
[22] K. Fujisawa, ‘General treatment of klystron resonant cavities’, IRE Transactions on
Microwave Theory and Techniques, vol. MTT-​6, pp. 344–​358, 1958.
[23] H. G. Kosmahl and G. M. Branch, Jr., ‘Generalized representation of electric fields in
interaction gaps of klystrons and traveling-​wave tubes’, IEEE Transactions on Electron
Devices, vol. 20, pp. 621–​629, 1973.
[24] C. Hill and R. G. Carter, ‘Investigation of possible multipactor discharge in a klystron
input cavity’, in 2006 IEEE International Vacuum Electronics Conference held Jointly
with 2006 IEEE International Vacuum Electron Sources, Monterey, CA, pp. 81–​82, 2006.
[25] M. J. Smith and G. Phillips, Power Klystrons Today. Taunton, UK: Research Studies
Press, 1995.
[26] K. R. Engala et al., ‘Simple computation of the coupling coefficient for loop-​coupled
resonant cavities’, Microwave and Optical Technology Letters, vol. 27, pp. 400–​404, 2000.
[27] H. A. Bethe, ‘Theory of diffraction by small holes’, Physical Review, vol.  66,  pp.
163–​182, 1944.
[28] R. E. Collin, Foundations for Microwave Engineering. New York: McGraw-​Hill, 1966.
[29] J. Gao, ‘Analytical formula for the coupling coefficient β of a cavity-​waveguide coupling
system’, Nuclear Instruments and Methods in Physics Research Section A: Accelerators,
Spectrometers, Detectors and Associated Equipment, vol. 309, pp. 5–​10, 1991.
[30] J. Gao, ‘Analytical formulae for the coupling coefficient β between a waveguide and a
travelling wave structure’, in Proceedings of the 1993 Particle Accelerator Conference
(PAC 1993), pp. 868–​870, 1993.
[31] S. B. Cohn, ‘Determination of aperture parameters by electrolytic-​tank measurements’,
Proceedings of the IRE, vol. 39, pp. 1416–​1421, 1951.
[32] H. A. Wheeler, ‘Coupling holes between resonant cavities or waveguides evaluated in
terms of volume ratios’, IEEE Transactions on Microwave Theory and Techniques, vol.
12, pp. 231–​244, 1964.
[33] D. Kajfez and E. J. Hwan, ‘Q-​factor measurement with network analyzer’, IEEE
Transactions on Microwave Theory and Techniques, vol. 32, pp. 666–​670, 1984.
[34] J. F. Gittins, Power Travelling-​Wave Tubes. London: English Universities Press, 1965.
[35] L. C. Maier, Jr. and J. C. Slater, ‘Field strength measurement in resonant cavities’,
Journal of Applied Physics, vol. 31, pp. 68–​77, 1952.
[36] R. G. Carter, ‘Accuracy of microwave cavity perturbation measurements’, IEEE
Transactions on Microwave Theory and Techniques, vol. 49, pp. 918–​923, May 2001.
134

4 Slow-​Wave Structures

4.1 Introduction

The useful bandwidth of tubes based on the resonators described in the previous
chapter is limited to a few percent. The electric field which interacts with the elec-
tron beam is essentially a standing wave restricted to a small region of space. The
R Q of the resonators is typically of the order of 100 Ω and the shunt impedance
is proportional to Q. Thus if Q is reduced to increase the bandwidth, the gap volt-
age induced by a given RF beam current decreases. It eventually falls to a point
where the interaction with the beam is of no practical value. When wide band-
width is required it is necessary to adopt a different approach and arrange for the
interacting electric field to be in the form of a travelling wave whose phase velocity
is approximately synchronous with the electron velocity. The interaction field is
smaller than is the case with a resonator but a strong interaction can be achieved
by maintaining synchronism over a much longer distance that the interaction gap
of a cavity resonator. This chapter deals with electromagnetic structures that sup-
port electromagnetic waves having the properties required for interaction with an
electron beam. These are known as slow-​wave structures. Many types of slow-​wave
structure are used in microwave tubes and many more types have been considered
for potential use [1, 2]. There is a very extensive literature on the subject: these two
papers cite over 300 references. It is therefore only possible to consider the general
principles, and some of the more important types of structure, in the space avail-
able in this book. The basic principles of uniform and periodic structures are dis-
cussed in the subsections of this section. Practical slow-​wave structures of various
types are considered in Sections 4.2 to 4.6. The final section deals with methods for
measuring the properties of slow-​wave structures.

4.1.1 Uniform Slow-​Wave Structures

The essential property of any useful slow-​wave structure is that it should support
an electromagnetic wave having a longitudinal component of the electric field and
a phase velocity rather smaller than the velocity of light. Figure 4.1 shows the gen-
eral arrangement of a cylindrical slow-​wave structure. The diagram is divided into
two concentric regions: the inner region r < a is empty and allows for the passage
of an electron beam; the outer region r ≥ a supports the propagation of a slow
135

Introduction 135

Figure 4.1:  Generic slow-​wave structure.

electromagnetic wave having a z-​component of electric field that is non-​zero on

the axis. These requirements are satisfied if the structure supports a TM 01 mode.
The z component of the electric field in the inner region must then satisfy the wave
equation

1 ∂2 E z
∇2 E z − = 0. (4.1)
c 2 ∂t 2
In cylindrical polar coordinates this becomes

1 ∂  ∂E z  ∂ 2 E z 1 ∂ 2 E z
r + − = 0 (4.2)
r ∂r  ∂r  ∂z 2 c 2 ∂t 2
when azimuthal symmetry is assumed. Now let the electric field on the axis be
given by

E z = E0 exp j ( ωt − β0 z ) , (4.3)

where β0 = ω v p and v p is the phase velocity of the wave. Substituting for E z in

(4.2) gives

1 ∂  ∂E z   2 ω 2 
r  −  β0 − 2  E z = 0. (4.4)
r ∂r  ∂r   c
This is a form of Bessel’s equation [3] whose solutions are modified Bessel func-
tions. Thus the electric field within the inner region is given by

E z = E0 I 0 ( γ 0 r ) exp j ( ωt − β0 z ) , (4.5)

where γ 0 = β02 − ω 2 c 2 is a real quantity because β0 > ω c for a slow wave. The Bessel
function I 0 increases rapidly as its argument increases, as shown in Figure 4.2. It is
therefore necessary for the electron beam to be as close as possible to the surface
of the slow-​wave structure to obtain a strong interaction. When the electron beam
is a solid cylinder which is coaxial with the structure it is usual to choose γ 0 a ≤ 1 to
minimise the variation of the electric field over the cross-​sectional area of the beam.
It can be seen from (4.5) that E z is non-​zero when r = a and, therefore, the structure
in the outer region must be able to support a longitudinal electric field at its surface.
136

136 Slow-Wave Structures

Figure 4.2:  The Bessel function I0(z).

Similar conclusions are obtained when the structure has a planar geometry and the
transverse variation of the electric field is a hyperbolic cosine, hyperbolic sine, or
exponential function, depending upon the boundary conditions.
Further insight into the general properties of slow-​wave structures can be
obtained from an equivalent uniform transmission line having series induct-
ance, and shunt capacitance, as shown in Figure 2.2(b). At a given frequency
the inductance and capacitance can be chosen so that the phase velocity and
the characteristic impedance have the desired values. This is the equivalent
circuit of a two-​wire line propagating a TEM wave whose phase velocity is
given by

1 c
vp = = , (4.6)
LC εr
where L and C are the inductance and capacitance per unit length and ε r is the rela-
tive permittivity of the dielectric separating the conductors. Although (4.6) shows
that this line can have a phase velocity less than the velocity of light it is not useful
as a slow-​wave structure because the longitudinal electric field is zero everywhere.
The phase velocity can be reduced from that given by (4.6) by increasing the induct-
ance per unit length. This is achieved most simply by making the inner conductor
of a coaxial line in the form of a helical wire. The properties of this important
type of slow-​wave structure are considered in detail in Section 4.3. For our present
purposes it is sufficient to note that all methods of increasing the series inductance
lead to structures in which the current flow has other components besides the axial
component. The magnetic field is then no longer purely transverse and the mode is
not a TEM  mode.
The characteristic impedance of the modified line is
2
V
Zc = , (4.7)
2P
137

Introduction 137

where V is the voltage between the conductors and P is the power flow along the
line. In general, the inductance and capacitance per unit length of the modified line
depend on frequency so that the line is dispersive. In that case
2
V
Zc = , (4.8)
2Wvg

where W is the stored energy per unit length and vg is the group velocity. The char-
acteristic impedance can then be written

vp V2 vp L vp
Zc = ⋅ = = Zc 0 , (4.9)
vg 2Wv p vg C vg

where v p is the phase velocity and Zc0 is the characteristic impedance of a non-​
dispersive line having the same inductance and capacitance per unit length. From
(4.9) it can be seen that it is desirable for the characteristic impedance to be as high
as possible because that gives a larger voltage for a given power flow. Hence it is
best to modify the line by increasing the inductance while keeping the capacitance
small. That, in turn, means that the relative permittivity of the dielectric should
also be small.
It is commonly stated that the axial electric field on the surface of the slow-​wave
structure is given by [4, 5]

∂V
E z (a ) = − = jβ0V . (4.10)
∂z
However, this equation cannot be correct as it stands, because the axial component
of the electric field is zero in the unmodified structure. The reason is that (4.10) does
not include the contribution to the electric field from the time-​varying magnetic
field. When this is taken in to account we find that

∂V ∂Az
E z (a ) = − − , (4.11)
∂z ∂t
where Az is the magnetic vector potential which is related to the axial compo-
nent of the current [6]. The magnetic vector potential also satisfies the Lorentz
condition

∂V
∇ ⋅ A = − εµ . (4.12)
∂t
Now the components of A are associated with components of the current density in
the same directions. If the radial component of the current density is zero and the
mode of propagation is azimuthally symmetric then (4.12) becomes

∂Az ∂V
= − εµ . (4.13)
∂z ∂t
138

138 Slow-Wave Structures

Making use of the variation with z and t defined in (4.3)

ω
Az = V , (4.14)
β0 v 2p 0

where v p0 is the phase velocity of the TEM wave on the unmodified structure.
Substituting for Az in (4.11) gives

 v 2p 
E z ( a ) = j β0 1 − 2  V . (4.15)
 vp0 
It is usually the case that v p  v p 0 and then (4.10) is a satisfactory approximation.
We can define an impedance which relates the strength of the electric field to the
power flow in the structure ( P ) by

E z (a )
2 2
 v 2p 
ZS = = 1 −
 v 2  Zc . (4.16)
2β2 P  p0 

This impedance, which we will call the surface impedance, is the impedance pre-
sented to the current that is induced in the slow-​wave structure by the beam. When
v p  v p 0 the surface impedance is equal to the characteristic impedance of the struc-
ture. An indication of the strength of the interaction between the wave on the struc-
ture and an electron beam is provided by the Pierce Impedance defined in terms of
the electric field on the axis by

E z (0)
2
ZS
ZP = = . (4.17)
2β20 P I0 (γ 0a)
2

A simple slow-​wave structure supporting the TM 01 mode can be made by lining a

circular metallic waveguide with dielectric. Then, referring to Figure 4.1, it is neces-
sary for the outer (dielectric) region to support a travelling electromagnetic wave.
Because the velocity of this wave is less than that of light the fields in the inner
(vacuum) region are evanescent and decay towards the axis as shown in Figure 4.2.
Experimental devices incorporating this kind of slow-​wave structure have been con-
structed [7]. The threshold velocity for a growing wave (Cerenkov) interaction is c ε r
so that a device with a relative permittivity of the order of 10 gave maximum gain at
a beam voltage of 160 kV. A device using a multi-​layer photonic band-​gap structure
with a beam voltage of 110 kV is described in [8, 9]. This type of slow-​wave structure
has poor thermal dissipation making it unsuitable for continuous high power oper-
ation. There is also the possibility that static charge may accumulate on the dielectric
surface leading to electric breakdown or disruption of the electron beam.
Another type of uniform slow-​wave structure is the sheath helix in which dielec-
tric material in the outer region supports a helically conducting surface at r = a .
Such a material could be manufactured by arranging insulated metal wires side by
side in a strip and then winding the strip helically around a cylindrical mandrel.
139

Introduction 139

This structure, which is a useful model for the helix slow-​wave structure, is dis-
cussed in detail in Section 4.3.1.

4.1.2 Periodic Slow-​Wave Structures

The previous section has shown that uniform slow-​wave structures are of limited
use. All the types of slow-​wave structure which are used in microwave tubes are
periodic and the interaction space is bounded by sections of metal separated by
gaps as shown in Figure 4.3. The gaps have length g, the period of the structure is
p, and the electric fields in the gaps are determined by the wave propagating in the
structure to which they are coupled. The details of practical slow-​wave structures
are discussed in later sections. Here we restrict our attention to the general prop-
erties of structures of this kind. Thus no assumptions are made about the region
outside the conducting rings other than that it is capable of supporting a travelling
electromagnetic wave with a propagation constant β0 at frequency ω .
Any periodic structure can be represented at a single frequency by a uniform trans-
mission line, having propagation constant β0 and characteristic impedance Zc , which
is tapped at regular intervals p as shown in Figure 4.4(a). A general external connec-
tion to the line can be represented by an ideal transformer whose ratio is 1: N together
with a lumped susceptance Z3 which represents the storage of electromagnetic energy
in the junction [10]. Figure 4.4(b) shows a single junction in this structure in which
the signs of the voltages and currents on the transmission line are chosen so that the
cells can be cascaded. It is convenient to represent this by the hybrid transfer matrix

V2   1 0 0  V1  V1 

 I 2  =  −Y3 
1 N  I1 = [G ]  I1  , (4.18)
 
      
V3   N 0 0   I 3  I3 

where Y3 = 1 Z3 . In this representation the circuit is driven by the voltage and cur-
rent at port 1 and the current induced by the electron beam at port 3. The outputs
are the voltage and current at port 2 and the voltage acting on the beam at port 3.

Figure 4.3:  Arrangement of a periodic slow-​wave structure.

140

140 Slow-Wave Structures

Figure 4.4:  Periodic slow-​wave structure: (a) general equivalent circuit, and (b) a single
junction (after Figure 4.24 in [10]).

The properties of a single cell of the structure, such as that between the planes 1
and 2 in Figure 4.4(a), can be found by noting that the voltage and current on the
line are related to the amplitudes of the forward and backward waves by
V   1 1  V f 
 I  = Y −Yc  Vb 
. (4.19)
 c
The inverse of this equation is
V f  1 1 Zc  V 
V  = 2 1 −Z   I  . (4.20)
 b  c 

Then, from (4.18), (4.19), and (4.20)

V2 f  1 Zc 0  1 0 0 1 1 0  V1 f 

V  = 1 1 −Z 0   −Y3 1 N  Yc −Yc 0  V1b  , (4.21)
 2b  2  c
 
V3  0 0 1   N 0 0   0 0 1   I 3 
which can be evaluated as

V2 f  1 −Y3 Zc 2 −Y3 Zc 2 NZc 2  V1 f  V1 f 

V  =  Y Z 2 1 + Y Z 2 − NZ 2  V  = [T ] V  ⋅
 2b   3 c 3 c c   1b   1b  (4.22)
V 
 3   N N 0  I
  3   I 3 

The matrix representing the propagation of waves on a length p 2 of the transmis-

sion line is
V2 f  exp ( − j β0 p 2 ) 0  V1 f 
V  =  0

exp ( j β0 p 2 ) V1b 
. (4.23)
 2b  
Thus the hybrid wave matrix for a single cell of the structure is
V2 f  e − jβ0 p 2 0 0  e − j β0 p 2 0 0  V1 f 
V  =  0   0  V  .
[ ]
e j β0 p 2
0 T e j β0 p 2
0 (4.24)
 2b     1b 
V3   0 0 1   0 0 1   I 3 
14

Introduction 141

This can be evaluated as

V2 f  (1 −Y3 Zc 2 ) e − jβ0 p −Y3 Zc 2 ( NZc 2) e − jβ p 2  V1 f 

0

V  =  Y3 Zc 2 ( 3 Zc 2) e jβ0 p
1 + Y

− ( NZc 2 ) e jβ p 2  V1b  . (4.25)
 2b  
0
 
V3   Ne − jβ0 p 2 Ne jβ0 p 2 0   I 3 

We note that, if port 3 is open circuit ( I 3 = 0 ) then (4.25) shows that the forward and
backward waves are coupled by the shunt admittanceY3. Furthermore the off-​diagonal
terms (1, 2 ) and ( 2,1) have opposite signs so, from the coupled-​mode theory reviewed
in Section 1.3.8, we recognise this as Case B coupling between the forward and the
backward wave. Thus the presence of a periodic discontinuity in the line leads to the
presence of a stop band, as is well-​known from the theory of filters [10]. In cases where
Y3 = 0 there is no stop band but the periodic coupling to the transmission line remains.
By analogy with the surface impedance of a uniform structure we can define the
total impedance of a periodic structure by
2
V3
ZT = , (4.26)
2P
where P is the power flow on the line. But for a forward wave
2
Vf
P= (4.27)
2Zc
and therefore the turns ratio N is found from the element at the bottom of the first
column of the matrix in (4.25) to be

N = ZT Zc . (4.28)

4.1.3 Space Harmonics

Let us assume, for simplicity, that the field in each gap is uniform (which is permis-
sible if the opposing edges are blunt) and varies sinusoidally with time with frequency
ω. The phases of the fields in adjacent gaps are not independent but are related by

φi +1 = φi exp − j (β0 p + 2 m π ) , (4.29)

where φi is the phase in the ith gap and m = 0, ± 1, ± 2,  The additional term in the
brackets in (4.29) arises because the exact phase relationship is indeterminate to a
multiple of 2π. We also note that, at a fixed time, the field along the wall is a peri-
odic function in space whose wavelength is

λ = 2 π β0 . (4.30)

Figure 4.5 shows an illustration of the way in which the field at r = a varies with
time. It can be seen that the gap fields are samples of a wave which is travelling in
the positive z direction.
142

142 Slow-Wave Structures

Figure 4.5:  Electric field of a periodic slow-​wave structure at r = a for β 0 p = 50 : (a) ωt = 0,

(b) ωt = 60 , and (c) ωt = 120.

If we define propagation constants

 2 mπ 
β m = ±  β0 + (4.31)
 p 
143

Introduction 143

which have periodicity p then, by virtue of Floquet’s Theorem [11], the electric field
along the wall can be expanded as a Fourier series in space

∞
E z ( z, a ) = ∑ E m exp ( − j β m z ), (4.32)
m = −∞

where E m is the amplitude of the mth space harmonic of the field and sinusoidal
time dependency as exp ( j ωt ) is assumed. To find the amplitudes of the individual
harmonics we multiply both sides of (4.32) by exp ( j β n z ) and integrate from − p 2
to p 2. The result is

p 2 ∞  p 2 
∫ E z ( z, a ) exp ( j β n z ) dz = ∑  ∫ Em exp ( − j (βm − βn ) z ) dz . (4.33)
−p 2 m = −∞ − p 2 

Now

 2π 
exp ( − j (β m − β n ) z ) = exp  − j ( m − n) (4.34)
 p 

is zero except when m = n . Then, if E z ( z, a ) = E0 when − g 2 ≤ z ≤ g 2 and

E z ( z, a ) = 0 elsewhere, (4.33) becomes

g 2 p 2

∫ E0 exp ( j β m z ) dz = ∫ E m dz, (4.35)

−g 2 −p 2

which may be integrated to give

E0   β g  β g
Em = exp  j m  − exp  − j m   (4.36)
j β m p   2   2 

and rearranged as

 sin (β m g 2 )  g
E m = E0  ⋅ . (4.37)
 β m g 2  p

If we suppose that the gaps are connected to a dispersionless transmission line,

then the dispersion diagram of the wave in the gaps is as shown in Figure 4.6. The
notation in this figure is consistent with (4.31) but differs from that used by some
other authors. The advantage of the notation used here is that the space harmonics
of the forward and backward waves having the same index ( m ) have the same nor-
malised amplitudes. From Figure 4.6 it is evident that the group velocities ( ∂ω ∂β )
of all the space harmonics are the same at a given frequency but that their phase
velocities ( ω β ) differ. The significance of this will become apparent later on. It is
important to recognise that the amplitudes of the space harmonics are linked by
(4.37) so that they cannot be excited independently of each other. An alternative
way of thinking about space harmonics is to note that an electron passing along
14

144 Slow-Wave Structures

Figure 4.6:  Dispersion diagram for space harmonic waves on a dispersionless periodic slow-​
wave structure.

the wall experiences the instantaneous field in each gap as it passes it. The phase
relationship between the gaps is therefore perceived as that of the space harmonic
whose phase velocity is equal to the velocity of the electron. This is an example of
the aliasing that occurs when waveforms which are periodic in time are sampled
periodically [12].
In Figure  4.6 the range − π p ≤ β ≤ π p is known as the First Brillouin Zone,
while π p ≤ β ≤ 2 π p is the Second Brillouin Zone, and so on. It can be seen that
the dispersion diagram in the range 0 ≤ β ≤ π p is repeated periodically in both
directions. Thus it is often sufficient to plot the dispersion diagram only in this
range. A diagram of this kind is called a reduced zone diagram. We have seen that if
the transmission line has periodic discontinuities with spacing p then the forward
and backward waves are coupled together and stop bands occur. It is easy to see
that these are caused by the cumulative addition of waves reflected from the discon-
tinuities when the electrical length from one discontinuity to the next, and back, is
equal to a whole number of wavelengths. Thus coupling between the modes can
occur when β = ± n π p where n is an integer. Examples of this will be shown in later
sections.
It is common in the literature to find references to ‘forbidden regions’ in the
dispersion diagrams. These occur when a slow-​wave structure is not surrounded
by a conducting wall. It then radiates energy into space when the phase velocity
of a space harmonic exceeds the velocity of light, as it may at some frequencies.
When the structure is contained within a conducting wall, waves that propagate
with phase velocities greater than c do not radiate energy and there are no forbid-
den regions. All the structures to be discussed in this chapter are contained within
metal walls.
145

Planar Slow-Wave Structures 145

4.2 Planar Slow-​Wave Structures

Planar slow-​wave structures find some use in crossed-​field amplifiers, and are of
interest for their potential use in millimetre wave travelling-​wave tubes. We shall see
below that the concepts are helpful for the understanding of helix and helix-​derived
structures. All the planar structures to be discussed are based on a microstrip array,
as shown in the cross-​section in Figure 4.7. At a height h2 above the ground plane
there may be another earthed conducting plane or a symmetry boundary. The strips
have finite thickness, as shown, and the array is assumed to be of infinite extent in
the ( x, z ) plane. If there is a dielectric present it is normally in contact with the strips
to provide mechanical support and a path for the conduction of heat. However, the
theory to be described can also be applied to cases where the dielectric layer does
not make contact with the strips. In many practical structures the dielectric is not
continuous but comprises an array of rods at right angles to the strips. It is found
that such an array of dielectric rods may be replaced, for modelling purposes, by an
equivalent uniform dielectric without serious loss of accuracy.
We wish to investigate the case where the lines propagate a TEM wave in the x
direction such that there is a phase difference φ between adjacent strips. Strictly
speaking, a pure TEM wave cannot exist because of the shear caused by the diffe-
rence between the phase velocities in the dielectric and in free space. However, in
many cases, this effect is small and it is possible to assume that there is a quasi-​TEM
wave [13]. Figure  4.8 shows the self capacitance of one line, the mutual capaci-
tances coupling it to the adjacent lines, and the voltage on each line. Consider, first,
the case where there is no dielectric substrate [14]. The charge per unit length on
the line is
N
q = C0V + ∑ CnV ( 2 − exp ( − jnφ ) − exp ( jnφ )) (4.38)
n =1

so that the effective capacitance per unit length of the line is

N
C ( φ ) = C0 + ∑ 2Cn (1 − cos ( nφ )), (4.39)
n =1

Figure 4.7:  Cross-​section of a microstrip array.

146

146 Slow-Wave Structures

where the summation is taken to include as many terms as are needed to ensure
good accuracy. The coefficients of capacitance can readily be found from finite dif-
ference solutions of Laplace’s equation (see Worksheet 4.1). The use of quasi-​static
analysis is justified because the phase velocity of propagation in the z direction is
normally much less than the velocity of light. In practice it is found that satisfac-
tory results can normally be obtained with N = 2.
The effective inductance per unit length can be found in a similar way. The mag-
netic flux linked to one line is
N
Λ = Ls i + ∑ M n i ( exp ( jnφ ) + exp ( − jnφ )), (4.40)
n =1

where i is the magnitude of the current in each line, Ls is the self-​inductance per unit
length of the line, and M n are the mutual inductances coupling it to the other lines
in the array. Thus the effective inductance per unit length is
N
L ( φ ) = Ls + ∑ 2M n cos ( nφ ). (4.41)
n =1

Now let L0 be the inductance per unit length when φ = 0. Then (4.41) can be written
N
L ( φ ) = L0 − ∑ 2M n (1 − cos ( nφ )). (4.42)
n =1

It is not necessary to find explicit expressions for the coefficients of inductance

because the line supports a TEM wave with phase velocity c so that
1
L (φ) = . (4.43)
c C (φ)
2

However we shall find (4.42) useful in understanding the difference between the
properties of different types of slow-​wave structure.
When there is a dielectric substrate present with relative permittivity ε 2 the effect-
ive capacitance per unit length is
N
CK ( φ ) = CK 0 + ∑ 2CKn (1 − cos ( nφ )), (4.44)
n =1

where the coefficients of capacitance can be found using finite difference calcula-
tions as before. This method can be used when the strip conductors have finite
thickness, whereas analytical methods are restricted to strips having zero thickness
[13, 14]. We note that, approximately

CK 0 = ε 2C0 (4.45)
and
1
CKn = (1 + ε 2 )Cn . (4.46)
2
147

Planar Slow-Wave Structures 147

Figure 4.8:  Equivalent circuit for the capacitance of one line in the microstrip array.

It is convenient to define the effective relative permittivity by

CK (φ)
ε eff ( φ ) = . (4.47)
C (φ)

The effective permittivity calculated in this way with N = 2 agrees closely with that
computed by Weiss [15] if it is assumed that there is a symmetry boundary at y = h2
(see Worksheet 4.2).
The inductance per unit length is not changed by the presence of the dielectric so
the phase velocity of waves on the line with dielectric loading is

1 c
v pe ( φ ) = = , (4.48)
L ( φ )C K ( φ ) ε eff ( φ )

the propagation constant in the x direction is

ω
βx (φ) = , (4.49)
v pe ( φ )

and the characteristic impedance is

L (φ)
Zc ( φ ) = . (4.50)
CK (φ)
The results of this calculation also agree closely with those given by Weiss [15].
This basic formulation can be used to find the properties of several of different
types of planar slow-​wave structures [14]. The procedure is to consider a length
l of the lines in the x direction and to apply boundary conditions as illustrated
in Figure 4.9. In all practical cases the lines must be enclosed by a metal bound-
ary. The region above the line may be terminated by a conducting plane, or two
identical lines may be placed opposite one another. The electrons which interact
148

148 Slow-Wave Structures

Figure 4.9:  Planar slow-​wave structures: (a) Finite array of parallel lines, (b) Ladder line,
(c) Meander line, and (d) Inter-digital line.

with the fields of the line move in the z direction. They may either pass close to
the surface of the line or, if the strips are sufficiently thick, through holes bored
through them.

4.2.1 Ladder Line
A ladder line is created by connecting all the strips together at each end as shown in
Figure 4.9(b). The boundary conditions are

 V1 (l 2 ) = V2 (l 2 ) = 0
 , (4.51)
V1 ( −l 2 ) = V2 ( −l 2 ) = 0
where V1 and V2 are the voltages on two adjacent lines. These conditions require that
there is a standing wave on each line such that

β x ( φ ) l = π (4.52)

for all φ. Then the dispersion equation is

π πc 1
ω= v p (φ) = ⋅ . (4.53)
l l ε eff ( φ )

The short circuits at the ends of the lines can be provided by the side walls of the
enclosing waveguide so no dielectric is necessary to support them. However, the
circuit is resonant at ω = πc l when there is no dielectric loading, so it is then of
little use as a slow-​wave structure. The structure becomes useful if the ladder line is
enclosed by a ridged waveguide (the Karp structure) or by a cruciform waveguide
(the Anti-​Karp structure) [16, 17]. Figure 4.10 shows examples of ladder line slow-​
wave structures and their dispersion curves [1].
149

Planar Slow-Wave Structures 149

Figure 4.10:  Propagation along ladder lines, (a) Cross sections of lines, (b) dispersion curves
with ladder detail
(copyright IEEE 1960, reproduced, with permission, from [1]).

4.2.2 Meander Line
The boundary conditions for the meander line structure shown in Figure 4.9(c), are [15]

 V1 (l 2 ) = V2 (l )
 I l
 1( 2 ) = − I 2 (l 2 )
 . (4.54)
V1 ( −l 2 ) = V2 ( −l 2 ) exp ( j 2φ )
 I1 ( −l 2 ) = − I 2 ( −l 2 ) exp ( j 2φ )

In order to satisfy these conditions four propagating modes are required. Two are
provided by the even mode propagating in the ±x directions with a propagation
constant given by (4.49). The other two waves are those of the odd mode for which
the phase is ( π + φ ) [14, 18]. The voltages on two adjacent lines are given by

V1, 2 = 2 A cos (βe x )  2 jB sin (βo x ) exp ( ± j φ 2 ) , (4.55)

where βe = β x ( φ ) and βo = β x ( φ + π ) are the propagation constants in the x direction

of the even and odd modes, and A and B are the amplitudes of the even and odd
modes whose ratio is
1
B  Zo sin (βe l )  2
= ⋅  , (4.56)
A  Ze sin (βo l ) 
150

150 Slow-Wave Structures

where Ze = Zc ( φ ) and Zo = Zc ( φ + π ). The dispersion equation obtained is

Zo tan (βe l 2 ) tan (βo l 2 )

tan2 ( φ 2 ) =  . (4.57)
Ze cot (βe l 2 ) cot (βo l 2 )
The upper line in (4.57) applies to the first pass-​band of the structure and the lower
line to the second pass-​band. The dispersion diagram for the case presented in
[15, 19] is shown in Figure 4.11. In this figure the lines are calculated from (4.57)
assuming that l is the full length of the line in the x direction. The dashed lines
assume TEM propagation along a folded line whose length is l with phase velocity
determined by the approximate effective relative permittivity (1 + ε 2 ) 2 [13]. The
experimental points are taken from [19] (see Worksheet 4.2). It can be seen that the
periodic discontinuities in the line produce a stop band when φ = β0 p = π. Note that
in [15] the phase ϕ is that for a complete period ( 2 p ) of the structure so that ϕ = 2φ.
To find the coupling impedance of the structure we note that the energy stored
per line in the even modes is
l 2
1
We ( φ ) = CK ( φ ) ∫ 4 A2 cos 2 (βe x ) dx (4.58)
2 −l 2

and in the odd modes is

l 2
1
Wo ( φ ) = CK ( φ + π ) ∫ 4 B 2 sin2 (βo x ) dx. (4.59)
2 −l 2

Hence the power flow in the z direction is

2l A
2
 1  sin βe l  1  B   sin βo l  
2

Pz ( φ ) = vg  1 + β l  + v Z  A  1 − β l   (4.60)

p  ve Ze e o o o 

where the group velocity vg = d ω d β0 is obtained from the dispersion curve. The
magnitude of the electric field in the gaps can be shown to be

4A  B 
E g ( x, φ ) = cos (βe x ) sin ( φ 2 ) + sin (βo x ) cos ( φ 2 ) . (4.61)
g  A 
Figure 4.12 shows how the magnitude of the field on the meander line, whose
dimensions are given in [15], varies with x for various values of φ. The phase of
the field is constant in each gap and the phase difference between the gaps is φ. The
folding of the line means that the curves in Figure 4.12 are reflected about the plane
x = 0 in alternate gaps. This property means that the calculation of the space har-
monics and of the coupling impedance is complicated. For the special case where
x = 0 the magnitude of the fundamental space harmonic is obtained from (4.37) as

 sin (β0 g 2 )  g
E0 ( 0, φ ) = E g ( 0, φ )  ⋅ . (4.62)
 β0 g 2  p
15

Planar Slow-Wave Structures 151

Figure 4.11:  Dispersion diagram of a meander line slow-​wave structure showing propagation

along uncoupled lines (dashed) and experimental points for comparison
(copyright 1974 IEEE, reproduced, with permission, from [15]).

Figure 4.12:  Distribution of the z component of the electric field in a meander line slow-​wave
structure.

where β0 p = φ. The surface impedance of the line at x = 0 is given by

E0 ( 0, φ )
2

ZS = . (4.63)
2β20 Pz

The results obtained from this equation are of the same magnitude as those
in [15] but the curve given there is probably wrong. The results of experi-
mental measurements made on the same structure by covering it with a thin
dielectric sheet are rather different but difficult to reproduce theoretically (see
Worksheet 4.2) [19]. Practical examples of meander line slow-​wave structures
are given in [20–​2 3].
152

152 Slow-Wave Structures

4.2.3 Inter-digital Line
The boundary conditions for the inter-digital line shown in Figure 4.9(d) are [24]

 V1 (l 2 ) = V2 ( −l 2 ) = 0
 . (4.64)
 I1 ( − l 2 ) = I 2 ( l 2 ) = 0

The period of the inter-digital line is 2p like that of the meander line. The disper-
sion equation is the same as that for a meander line if the widths of the strips
and the gaps are interchanged. This is because the two structures are duals of one
another so that the electric field lines in one case correspond to the magnetic field
lines in the other and vice-​versa. This is an application of Babinet’s principle [6].
The interdigital line does not require dielectric support [25, 26].

4.3 Helix Slow-​Wave Structures

The planar structures discussed in the previous section are useful for interaction
with sheet beams, and in crossed-​field devices, but their arrangement is not ideal for
interaction with the cylindrical beams used in most linear-​beam tubes. Also, because
they are periodic structures, they are dispersive and have limited bandwidth. These
problems are avoided in the helix slow-​wave structure shown in Figure 4.13. In this
structure a helical conducting tape is held within a metal shield by three dielectric
support rods. The support rods also provide a path for heat to flow from the helix
to the shield. This structure can be regarded as a coaxial line with a helical inner
conductor. Thus the series inductance is increased and the phase velocity decreased
as discussed in Section 4.1.1. The dielectric loading is kept as small as possible,
consistent with adequate thermal conduction, to avoid increasing the shunt cap-
acitance and thereby reducing the characteristic impedance. The support rods are
usually made of materials such as anisotropic boron nitride and CVD diamond
[27]. Beryllia, which has high thermal conductivity, is toxic and its use is now gener-
ally avoided. The helix is made of tungsten or molybdenum to ensure dimensional

Figure 4.13:  Arrangement of a helix slow-​wave structure.

153

Helix Slow-Wave Structures 153

stability at high temperatures. It is important to maintain good thermal contact

between the support rods and the helix. This can be achieved by using assembly
methods which ensure that the rods exert a compressive force on the helix, or by
brazing the helix to the rods [4, 28]. The thermal analysis of rod supported helix
slow-​wave structures is discussed in [29, 30].
The basic properties of the helix slow-​wave structure can be understood using
the methods of the previous section. Figure 4.14 shows two adjacent strips from a
developed (opened out) helix. From the geometry of the structure the pitch angle
ψ is given by

p
tan ψ = . (4.65)
2 πa
The boundary conditions are now just

V1 (l 2 ) = V2 ( −l 2 )
 (4.66)
 I1 ( l 2 ) = I 2 ( − l 2 )

and the structure supports an azimuthal TEM wave whose phase velocity is deter-
mined by the effective permittivity. If the helix is supported by a material whose
relative permittivity is unity then the axial phase velocity is expected to be

v p = c sin ψ. (4.67)

It is found that this is not quite accurate for reasons which are explored below.

Figure 4.14:  The geometry of the helix slow-​wave structure.

154

154 Slow-Wave Structures

4.3.1 The Sheath Helix Model

Although the helix is geometrically simple it is very difficult to analyse mathem-
atically. The simplest analytical model treats the helix as a thin cylinder in which
conduction is only possible at an angle ψ to transverse planes, where ψ is the pitch
angle of the helix shown, in Figure 4.13 [10]. The region between the cylinder and
the shield is assumed to be filled with uniform dielectric material having relative
permittivity ε 2 . The component of the electric field in the direction in which the
cylinder conducts must be zero and therefore, in region 1 ( r ≤ a ),

Eθ1 cos ψ + E z1 sin ψ = 0 (4.68)

and, in region 2 ( a ≤ r ≤ s ),

Eθ 2 cos ψ + E z 2 sin ψ = 0. (4.69)

The tangential electric field normal to the direction of conduction must be continu-
ous so that

E z1 cos ψ − Eθ1 sin ψ = E z 2 cos ψ − Eθ 2 sin ψ. (4.70)

The current flow in the cylinder is in the direction of conduction and therefore the
tangential magnetic field must be continuous in this direction

H θ1 cos ψ + H z1 sin ψ = H θ 2 cos ψ + H z 2 sin ψ. (4.71)

These four equations define the boundary conditions at r = a . The boundary con-
dition at r = s is Eθ = E z = 0. The axial components of both the electric and mag-
netic field are non-​zero. Therefore the wave propagating on the structure cannot be
a pure TM or TE mode but is a hybrid of the two linked by the conduction angle
of the cylinder. A cylindrically-​symmetrical solution to this problem can be found
by assuming that all quantities vary in the z direction as exp j ( ωt − β0 z ). The radial
propagation constants are then

γ 0 = β20 − ( ω c )
(0 ≤ r ≤ a )
2

(4.72)
γ 2 = β20 − ε 2 ( ω c ) ( a ≤ r ≤ s ) .
2

In practice β20 is normally much greater than ( ω c ) and it is possible to use the
2

approximations γ 0 = γ 2 = β0 without serious error. For the TM  fields

E z1 = A1I 0 ( γ 0 r )
A2 (4.73)
Ez 2 =  I 0 ( γ 2 r ) K 0 ( γ 2 s ) − I 0 ( γ 2 s ) K 0 ( γ 2 r ) ,
K γ s
0 ( 2 )
where A1 and A2 are constants, since the z component of tangential electric field
must be zero when r = s . Similarly for the TE fields

H z1 = B1I1 ( γ 0 r )
(4.74)
H z 2 = B2 I1 ( γ 2 r ) + C2 K1 ( γ 2 r ) ,
15

Helix Slow-Wave Structures 155

where B1, B2 and C2 are constants. The transverse field components H θ and Eθ
can be found from (2.19) and (2.20) expressed in cylindrical polar co-​ordinates.
Applying the boundary condition on Eθ at r = s gives C2 in terms of B2 so that all
the field components can be expressed as functions of r and the four constants A1, A2,
B1 and B2 . Substituting these expressions into the equations for the four boundary
conditions at r = a ((4.68) to (4.71)) leads to the dispersion equation for the struc-
ture. The equations obtained in this way are rather complicated so we will simply
summarise the results.
If the sheath helix is in free space and s → ∞ the dispersion equation is [10]

( γ 0 a )2 tan2 ψ = I1 ( γ 0 a ) K1 ( γ 0 a ) . (4.75)
( ka )2 I0 (γ 0a) K0 (γ 0a)

The effects of the finite shield radius and the supporting dielectric material can
conveniently be expressed using the methods described in [31]. Equation (4.75) is
first expressed in terms of equivalent circuit parameters by

2
µ  β 
L0 =  0  ( I1 ( γ 0 a ) K1 ( γ 0 a ))  0  cot 2 ψ (4.76)
 2π   γ0 

and

2 πε 0
C0 = . (4.77)
I0 (γ 0a) K0 (γ 0a)

Substitution of L0 and C0 into the equation for the propagation constant of a trans-
mission line (2.21) gives the dispersion equation (4.75). It should be noted that the
inductance and capacitance are functions of frequency.
If the sheath helix is surrounded by a shield then the inductance and capacitance
become [31]

 I1 ( γ 0 a ) K1 ( γ 0 s ) 
L = L0 1 −  (4.78)
 I1 ( γ 0 s ) K1 ( γ 0 a ) 

and

−1
 I0 (γ 0a) K0 (γ 0 s) 
C = C0 1 −  (4.79)
 I0 (γ 0 s) K0 (γ 0a) 

The diameter of the shield is normally chosen so that higher-​order modes, espe-
cially the TE11 coaxial waveguide mode, are cut off. If this is not so then the higher-
order modes can provide an unwanted feedback path when the structure is used in
a travelling-​wave tube [32].
156

156 Slow-Wave Structures

The addition of a uniform dielectric supporting medium between the helix and
the shield does not affect the inductance, but it multiplies the capacitance by the
effective relative permittivity [33]

 I1 ( γ 0 a ) K 0 ( γ 0 s ) 
ε eff ( γ 0 a, γ 0 s ) = 1 + ( ε 2 − 1)( γ 0 a ) I 0 ( γ 0 a ) K1 ( γ 0 a ) 1 +  . (4.80)
 I 0 ( γ 0 b ) K1 ( γ 0 a ) 
where it has been assumed that γ 2 = γ 0. Note that this expression differs from that
given in [31] which does not take account of the radial boundary conditions cor-
rectly. The effect of the additional term (in braces) is most important if s a is close
to unity. When the dielectric loading is included (4.79) becomes

−1
 I (γ a) K (γ s) 
C = C0 ε eff ( γ 0 a, γ 0 s ) 1 − 0 0 0 0  (4.81)
 I0 (γ 0 s) K0 (γ 0a) 

In the derivation of these equations it was assumed that the phase velocity was small
enough for the quasi-​static approximation to be valid so that γ 0  β0 everywhere.
The characteristic impedance of the slow-​wave structure is found by substitut-
ing the values of L and C into (4.9). An alternative, and much more complicated,
expression for the characteristic impedance of a helix in free space can be derived
by integrating the Poynting vector to obtain the power flow [34]. The two methods
give identical numerical results.
In practical slow-​wave structures the helix is supported within the shield by
dielectric support rods, as shown in Figure  4.13. If these are wedge-​shaped then
relative permittivity of the equivalent uniform dielectric is given by [31, 35]

Nθ
ε2 = 1 + ( ε r − 1) , (4.82)
2π
where there are N rods having relative permittivity ε r each subtending an angle θ on
the axis. This equation can be derived by considering a coaxial line in which the cen-
tral conductor is supported by wedge-​shaped rods. The problem is a quasi-​static one
in which the equipotentials are concentric circles. The capacitance between the two
conductors is therefore the parallel combination of the capacitances of the vacuum
and the dielectric regions. That leads directly to (4.82). When the rods have some other
shape, such as the ‘T’ shaped rods shown in Figure 4.15(a), the effective relative per-
mittivity can still be determined from the capacitance. It is commonly assumed that
this can be done by separating the region between the helix and the shield into two or
more layers, and then finding the effective relative permittivity for each, using (4.82)
[36]. This method, which assumes that the equipotentials remain concentric circles,
actually leads to an upper bound of the capacitance and, therefore, of the effective
relative permittivity. It can be shown that a lower bound can be found by assuming
that the flux lines remain radial. This is an example of the method of Tubes and Slices
in which upper and lower bounds to the capacitance are calculated from the stored
energy using approximations to the field [37]. An alternative method for calculating
the effective relative permittivity from the stored energy is described in [38].
157

Helix Slow-Wave Structures 157

Figure 4.15:  Different arrangements for helix slow-​wave structures: (a) T-​shaped support

rods, (b) vane loading for dispersion shaping, and (c) theoretical model of vane loading.

Figure  4.16(a) shows typical dispersion curves obtained from the sheath helix
model. It is useful to consider the asymptotic behaviour of these in the limits ω → 0
and ω → ∞. In the low frequency limit it can be shown that the inductance and
capacitance per unit length tend to

µ0  a2 
L00 = 1 −  cot ψ (4.83)
2
4π  s 2 
and

2 πε 0 ε 2
C00 ( ε 2 ) = , (4.84)
ln ( s a )
which is the capacitance per unit length of a coaxial line (see (2.47)). Thus the low-​
frequency limit of the phase velocity is

1 c tan ψ 2 ln ( s a )
v p0 = = ⋅ . (4.85)
L00C00 ε2 1 − a2 s2

In the high frequency limit we note that the modified Bessel functions are given
approximately by [3]

ez
In (z) = (4.86)
2πz
and

π −z
Kn (z) = e . (4.87)
2z
Then from (4.78) and (4.79) we find that L → L0 and C → C0 because the fields are
confined to the region close to the helix and are not dependent on the dimensions
of the shield. Hence the high frequency limits are

µ  1
L0 →  0  cot 2 ψ (4.88)
 4π  γ 0 a
158

158 Slow-Wave Structures

since β0  γ 0 for a slow wave, and

C0 → 4 πε 0 ( γ 0 a ) . (4.89)

From (4.80) we find that the effective relative permittivity tends to

1
ε eff ( γ 0 a, γ 0 s ) → (1 + ε 2 ) , (4.90)
2
which is the average of the values of relative permittivity inside and outside the
helix. Thus the asymptotic value of the phase velocity is

c tan ψ
vp → . (4.91)
(1 + ε 2 ) 2

We note that this differs from the result previously obtained in (4.67). The asymp-
tote of the phase velocity is shown in Figure 4.16(a) The curves of phase velocity
do indeed tend to the asymptote if the plot is extended to higher values of β0 p.
Figure 4.16(b) shows the Pierce impedance, calculated using (4.17), as a function of
the phase shift per turn.

4.3.2 Dispersion Shaping

The dispersion characteristics of a helix can be modified by placing longitudinal
vanes between the support rods, as shown in Figure 4.15(b). The effect of the vanes
is represented by the model shown in Figure 4.15(c). The tips of the vanes define the
radial boundary as far as the electric field is concerned, whilst the magnetic fields
are bounded by the outer shield. Thus s is replaced by v in (4.81). Figure 4.17 shows
the effect of the vane loading on the dispersion and the Pierce impedance of a typi-
cal sheath helix slow-​wave structure. The addition of the vanes changes the shape

Figure 4.16:  Typical properties of a sheath helix slow-​wave structure: (a) normalised

phase velocity, and (b) Pierce impedance, as functions of the phase shift per
turn  ( ψ = 10 ; ε 2 = 2.5) .

159

Helix Slow-Wave Structures 159

Figure 4.17:  The effect of vane loading on a typical sheath helix slow-​wave structure:
(a) normalised phase velocity, and (b) Pierce impedance ( s a = 2; ψ = 10 ; ε 2 = 2.5).

of the dispersion curve so that the phase velocity is constant over a wider range of
phase shift per turn. However, the Pierce impedance falls as the useful bandwidth
is increased. The properties of the sheath helix model can be explored further using
Worksheet 4.3.

4.3.3 Tape Helix Slow-​Wave Structures

When the sheath helix model is used to calculate the properties of a real tape helix
slow-​wave structure it is found that there are appreciable errors. In principle a tape
helix slow-​wave structure could be modelled using the method of moments in a
manner similar to that described in Section 3.5.1. However the tape helix problem
is much more difficult than the re-​entrant cavity, and the efforts of many people
over a period of more than sixty years have not yet produced good agreement with
experimental results. A great many papers have been published on this subject, and
only the more significant ones are reviewed here.
In addition to the fundamental mode described above the sheath helix supports
higher-order modes which satisfy the free space dispersion relation [39]

I n′ ( γ a ) K n′ ( γ a )
=−
( γ 2 a 2 + nβa cot ψ ) . (4.92)
2

In (γ a) Kn (γ a) ( ka )2 ( γ a )2 cot 2 ψ
It is therefore possible to model the fields of a tape helix by superimposing space-​
harmonic waves which satisfy (4.92) and the boundary conditions. It is important
to understand that, unlike the space-​harmonics discussed earlier, these harmonics
have azimuthal variation. In order to make the problem tractable one or more of
the following assumptions are made:
• The tape has negligible thickness, so that it is only necessary to consider the fields
in the inner and outer regions.
• The region between the helix and the shield is filled with a uniform dielectric.
160

160 Slow-Wave Structures

• The current in the helix flows in the direction of the tape and is distributed in a
prescribed fashion.
• The differences between β0 and γ 0 , and between the values of γ in dielectric and in
vacuum, are neglected (this is the quasi-​static approximation).
The earliest work on this subject was that of Sensiper [39, 40] who studied the
properties of a thin helix in free space. Two approximations were considered: nar-
row tapes on which the current density was taken to be constant; and narrow gaps
in which the electric field was taken to be uniform. The dispersion equation for the
case of a narrow tape was

∞
 sin (β m w 2 ) 
∑ (M m + Nm ) 
 β m w 2 
= 0, (4.93)
m = −∞

where w is the width of the tape,

2
  β 
M m =  γ m a −  m m cot ψ   I m ( γ m a ) K m ( γ m a ) , (4.94)
  γm 

where γ 2m = β2m − k 2, and

N m = ( k 2 a 2 cot 2 ψ ) I m′ ( γ m a ) K m′ ( γ m a ) . (4.95)

The dispersion equation for the sheath helix is obtained if only the terms in m = 0
are considered. The same equations were obtained for the case of a wide helix with
narrow gaps except that the last term in (4.93) was replaced by J 0 (β m w 2 ).
The propagation of waves on a narrow tape helix supported within a metal
cylinder by a uniform dielectric was considered by Stark [41]. A dispersion rela-
tion similar to that for the sheath helix (4.75) was obtained with the numerator
and denominator on the right-​hand side multiplied, respectively, by the term in
square brackets in (4.78) and the reciprocal of the term in square brackets in
(4.79). The effects of the tape width were represented by the addition of the term
(sin ψ ) ln (2 p πw ) to both the numerator and the denominator. This approach
does not seem to have been taken up by subsequent authors. The same problem
was considered by Tsutaki [32] who assumed that the current density distribution
on the tape could be described by a hyperbolic cosine function. It was found that
when w p > 0.4 the results converged if only the n = 0, ± 1 space harmonics were
considered. For tapes with w p ~ 0.1 it was necessary to include harmonics up
to n = ±3. It was concluded that in the range 0.4 < w p < 0.8 the solutions were
not sensitive to the current distribution assumed. The phase velocities calculated
were said to be accurate to with ±2 % but there was no detailed comparison with
experiment. The use of a helix as a backward-​wave structure was discussed in
[42]. It should be noted that as a consequence of (4.92) the axial component of
the electric field of the higher space-​harmonics, which have azimuthal variation,
is zero on the axis.
16

Helix Slow-Wave Structures 161

The effect of the finite thickness of the tape was modelled in [43] using a thin
tape at the mean radius of the helix, separated from the dielectric by an air gap
equal to half the thickness of the tape. This followed the work of Swift-​Hook [44]
who carried out a careful theoretical and experimental study of round wire helices
inside glass tubes having different wall thicknesses, but no outer metal boundary.
It was found that the use of an air gap having a thickness equal to one tenth of the
helix radius gave the best results in this case. The justification for the use of the
same technique with tape helices, and with a gap equal to half the tape thickness is
not clear. Nevertheless this method of representing the effects of tape thickness has
been widely used by subsequent authors.
A simplified approach to the modelling of tape helix slow-​wave structures was
described in [45]. This was based upon Sensiper’s dispersion equation for a helix in
free space (4.93). Following the method in [31] the effects of the radial boundary
and dielectric loading were introduced by multiplying the individual terms of the
summation by correction factors similar to (4.78) and (4.79). The effects of dielec-
tric loading were included by assuming that the helix was surrounded by a number
of concentric dielectric layers. The effective permittivity of each layer was deter-
mined from the relative permittivity of the support rods and the fraction of each
layer which was occupied by the rods. This approach was first proposed in [33, 43].
Although there are doubts about the validity of this method, since the boundary
conditions are not correctly maintained, it has been widely used by other authors.
The results of the simplified model of the tape helix were compared with a small
number of experimental results. The agreement was marginally better than that
given by the sheath helix model.
Further developments in the modelling of tape helix slow-​ wave structures
included:  removal of the assumption that the radial and axial propagation con-
stants were equal [46]; calculation of the constants in the field expansions by suc-
cessive approximations [47]; removal of the need to assume the current distribution
on the tape [48–​51]; use of a perturbation technique to evaluate the effects of the
azimuthal distribution of the dielectric loading [52]; and an exact theory which
included the finite thickness of the helix and the shapes of support rods and vanes
by matching radial admittances (but which was only implemented in a simplified
form) [53]. All these methods involve very complicated equations and do not add
much to understanding of these structures. Where comparisons have been made
with experimental results the agreement is generally of the order of 1% for the
phase velocity and of the order of 10% or more for the interaction impedance. On
the whole the accuracy obtained does not justify the complexity of the analysis
involved. Unfortunately only a limited range of experimental data is available in
the literature and some of the measurements are old and of doubtful accuracy.
The problems of making accurate experimental measurements on helix slow-​wave
structures are discussed in Section 4.7.
The alternative to the complicated mathematical analysis discussed above is to
compute the properties of structures using computational electromagnetics [54–​
58]. These methods also present some difficulties because the structure does not
162

162 Slow-Wave Structures

normally conform to the boundaries of a regular mesh, and because a very large
number of mesh elements is required to ensure good accuracy. The accuracies
reported are comparable with those of the best analytical methods.

4.3.4 Equivalent Circuit of Helix Slow-​Wave Structures

In Section 4.3 we briefly considered a model comprising parallel strips. In its simple
form this model does not predict the dispersion correctly [14]. The reason for this is
that in the limit of low frequencies the phase velocity does not correspond to an azi-
muthal TEM wave between the helix and the shield. The presence of an axial magnetic
flux inside the helix causes the inductance per turn to be less than it would be for a
TEM wave and the phase velocity is therefore increased. However, a variation of this
approach does give insight, and useful accuracy, as we shall see. A developed model of
the helix as a set of parallel metal strips supported over a ground plane by dielectric
strips has been used to study the effects of the loading of the structure by vanes simi-
lar to those illustrated in Figure 4.15(b) [59]. Empirical approximations to the capaci-
tances were used. Comparisons with four sets of experimental data showed agreement
to ±1 % for the phase velocity. The use of empirical approximations can be avoided by
using a method similar to that described for planar structures in Section 4.2.
The electrical properties of helix slow-​wave structures can be modelled using the
same equivalent circuit as that for planar structures [60, 61]. The effective capaci-
tance per unit axial distance is given by (4.39) without dielectric loading and (4.44)
with dielectric loading, The direct computation of the capacitances and induct-
ances from the dimensions of the circuit is difficult but this problem can be avoided
by making use of the properties of the sheath helix model. Comparison with the
results of experimental measurements shows that the properties of a tape helix
structure can be calculated to good accuracy if the capacitance and inductance per
unit length for a sheath helix are multiplied, respectively, by constants αC and α L
whose values depend only upon the dimensions, and dielectric loading, of the struc-
ture. The best results are obtained when the radius of the sheath helix is taken to be
the mean radius of the tape helix.
We consider, first, a structure without dielectric loading. The capacitance per
unit length given by (4.39) can be calculated from a two-​dimensional finite diffe-
rence solution for the field between an array of conducting rings and a concentric
conducting cylinder as shown in Figure  4.18. The solution is computed between
the planes A–​A and B–​B (see Worksheet 4.4) for β0 p = 0, π 2 and π to give the first
three terms of the series. The ratio of the capacitance per unit length of the array
of rings, when the phase difference between them is zero, to that of an air-​spaced
coaxial line having the same dimensions is

C (0)
αC 0 = , (4.96)
C00 (1)
where C00 is given by (4.84). Figure 4.19 shows how αC 0 varies with w p for three
values of s p and typical values of a p and t p when C00 is calculated using the
163

Helix Slow-Wave Structures 163

Figure 4.18:  Array of rings used to model the capacitance of a tape helix.

Figure 4.19:  Variation of the capacitance between the helix and the shield with tape width,
and shield radius, for typical values of helix radius and tape width (a p = 1; t p = 0.1).

outer radius of the tape helix. As the tape width is increased the capacitance tends
to that of the coaxial line as would be expected.
The value of α L could be calculated from a numerical solution for the low-​
frequency inductance per unit length of the tape helix surrounded by a shield.
However, this can be avoided by noting that when β0 p = π a helical TEM wave
propagates with phase velocity c. The value of α L can then be calculated directly
since the phase velocity is given by

1
vp = = c sin ψ , (4.97)
α L αC 0 LC
where L and C are given by (4.78) and (4.79) using the mean radius of the helix. It
can be shown that the values of C are close to those computed from (4.39) except
close to β0 p = π.
164

164 Slow-Wave Structures

The Pierce impedance is calculated by assuming that the potential difference

between adjacent turns is the potential difference between the points A and B in
Figure 4.14. Thus

VAB = Vg exp ( j β0 p2 2 ) − exp ( − j β0 p2 2 ) ,

(4.98)
= 2Vg sin (β0 p2 2 )

where Vg is the magnitude of the voltage between the helix and the shield. It is
assumed that the magnitude of the electric field in the gap is constant so that

VAB
Ez = cos ψ
g1
(4.99)
2Vg sin (β0 p2 2 )
= cos ψ.
g1

Then the magnitude of the fundamental space-​harmonic is from (4.37)

2Vg sin (β0 p2 2 )  sin (β0 g 2 )  g

E0 =
g  β g 2  ⋅ p ⋅ cos ψ. (4.100)
1 0

This can be written as

 sin (β0 p2 2 )   sin (β0 g 2 ) 

E0 = β0Vg  cos 2 ψ (4.101)
 β0 p2 2   β0 g 2 

and, from (4.17), the Pierce impedance is

 sin (β0 p2 2 )   sin (β0 g 2 )  cos 2 ψ 

2

ZP =      Zc , (4.102)
 β0 p2 2   β0 g 2  I 0 ( γ a1 ) 
where a1 is the inner radius of the helix and Zc is the characteristic impedance of the
tape helix slow-​wave structure given by (4.9) so that

vp αL L
Zc = . (4.103)
vg αC 0C

This model is found to agree with experimental data within the limits of experimen-
tal error. Figure  4.20 shows a comparison between the calculated and measured
values of the phase velocity and the Pierce impedance (measured by three different
methods) for a typical structure (see Worksheet 4.5).
Tape helix slow-​wave structures with dielectric loading can be modelled by con-
sidering the effect of the dielectric rods on the shunt capacitance. The structure is
first modelled without the dielectric as described above. It is assumed that the value
of α L is unchanged by the addition of the dielectric. The effect of dielectric loading
on the shunt capacitance cannot be computed by multiplying C ( 0 ) by ε 2 because
the presence of the dielectric causes the charges on the surfaces to be redistributed.
165

Helix Slow-Wave Structures 165

Figure 4.20:  Comparisons between calculated and measured properties of a typical tape helix
slow-​wave structure without dielectric loading: (a) phase velocity, and (b) Pierce impedance.
(a = 4.7 mm, p = 6.0 mm, s = 14.0 mm, w = 2.4 mm, t = 1 mm).

Instead, the shunt capacitance per unit length (CK (β0 p )), given by (4.44), can be cal-
culated using finite difference solutions at β0 p = 0, π 2 and π to find the first three
terms of the series (see Worksheet 4.4). Wedge shaped support rods are treated as a
uniform dielectric of effective permittivity ε 2 given by (4.82). Values of capacitance
calculated in this way are unsatisfactory close to β0 p = π where the quasi-​static
approximation starts to break down. Thus it is better to use the capacitance of the
sheath helix given by (4.81) with a correcting factor αC given by

CK (0 )
αC = . (4.104)
C00 ( ε 2 )
The effective permittivity is given by (4.47) and the phase velocity and Pierce
impedance are calculated as before. Figure 4.21 shows the comparison between the
results of calculations and experimental data for SWS-​2 in [47]. This structure had
rectangular support rods that may only been in contact with the helix along narrow
lines. The best agreement between calculations and measurements was obtained by
assuming wedge-​shaped rods occupying the same fraction of the perimeter of the
shield as the rectangular rods. The accuracy obtained using this model is compar-
able with that given by the much more complicated analytical method in [47]. Thus
the equivalent circuit gives insight into the properties of helix slow-​wave structure,
and a simple method for calculating them with useful accuracy. For example, it can
be shown that, if the support rods are not in good contact with the surface of the
helix, there can be a marked change in the phase velocity and the impedance. In one
case an air gap between the helix and the dielectric equal to 5% of the helix radius
was found to reduce the shunt capacitance by a factor of 0.75.

4.3.5 Couplers and Attenuators

When a helix slow-​wave structure is used in a travelling-​wave tube it is important
that the terminations of the structure are correctly matched. The transition between
16

166 Slow-Wave Structures

Figure 4.21:  Comparisons between calculated and measured properties of a tape helix slow-​
wave structure with dielectric loading: (a) phase velocity, and (b) Pierce impedance.

the helix and an external waveguide is normally made through the direct connec-
tion of the helix to a coaxial line. At low power levels this is connected directly to
an external coaxial line. At high power levels it may be coupled outside the vacuum
envelope to a rectangular, or ridge, waveguide. The characteristic impedance of
the helix slow-​wave structure can be computed from (4.103), if the effects of the
support rods are included, and this can be used to design a broad-​band transition
to the coaxial line [60, 61]. A method for the detailed design and optimisation of
coaxial couplers is discussed in [62].
Where the helix is divided (severed) into two or more sections (see Chapter 14)
it is necessary to achieve a good match at each sever. This is achieved by placing a
graduated attenuating coating on the support rods [63–​67]. The helix may be sev-
ered physically but this is not necessary if sufficient attenuation can be introduced
to prevent the passage of signals through the attenuator.

4.4 Ring-​Bar and Ring-​Loop Structures

The helix slow-​wave structure is not suitable for use at high beam voltages because
the interaction impedance falls rapidly as the pitch angle is increased. There is
also an increased tendency to backward-​wave oscillation. Better impedance can
be obtained, at the expense of reduced bandwidth, using helix-​derived structures
such as the ring-​bar structure shown in Figure 4.22. This structure evolved from
a contra-​wound bifilar helix [68, 69]. It comprises a series of rings which are con-
nected by bars placed alternately at diametrically opposed positions, and can readily
be constructed by making transverse cuts in a metal tube. The structure is sup-
ported within a concentric metal tube by two, three or four dielectric support rods
[18, 70, 71]. The interaction impedance of the ring-​bar structure in the fundamental
mode is greater than that of a single helix, while that in the space harmonics is less.
167

Ring-Bar and Ring-Loop Structures 167

Figure 4.22:  The ring-​bar structure.

Thus the tendency to backward-​wave oscillations in the space-​harmonic modes is

reduced. The thermal performance of this structure is similar to that of the helix
structure so its chief use is in pulsed tubes.
Most analytical models of the ring-​bar structure have been based on the methods
for tape helix slow-​wave structures described above [70–​72]. Like the helix models
from which they are derived these models involve considerable mathematical com-
plexity but do not achieve close agreement with experimental results. The dielectric
support rods are replaced by an equivalent uniform dielectric as before. A different
approach, described by Ash et al. [18], is based on an analysis of a developed struc-
ture in which parallel metal strips are supported on dielectric bars, at right angles
to them, in a manner similar to that used for the tape helix by Onodera and Raub
[59]. This model showed useful agreement with experimental results but it does not
seem to have been pursued by other authors. The ring-​bar structure has also been
modelled successfully using electromagnetics software [70, 73].
The main mode of propagation of the ring-​bar structure is symmetrical about
the longitudinal plane passing through the centres of the bars. The current passing
along a bar is divided equally between the two halves of the structure which are
therefore in parallel with one another [68]. A footnote in [18] observes that each
half of the structure can be regarded as a meander line, but this approach does
not seem to have been used elsewhere. It does, however, provide a simple method
for modelling this structure using the theory of Section 4.2.2. The coefficients of
capacitance in (4.39) and (4.44) can be found using a finite-​difference model of
isolated rings, both with and without uniform dielectric loading, in the same man-
ner as for the tape helix (see Worksheet 4.4). The capacitance per unit axial length
is converted to capacitance per unit azimuthal distance by multiplying the results
of the finite-​difference calculations by p 2 πa where p is the pitch of the rings and
a their mean radius. The dispersion diagram of the structure can then be com-
puted directly from the equations in Section 4.2.2 with results similar to those in
Figure 4.11 (see Worksheet 4.6). Figure 4.23(a) shows a comparison between the
168

168 Slow-Wave Structures

phase velocity computed in this way and the experimental results reported in [70].
In this calculation it has been assumed that the effective azimuthal length of each
half ring is πa. The agreement between calculation and measurement is comparable
with the analytical results in [70, 71].
To compute the Pierce impedance of this structure we regard it as a pair of mean-
der lines connected in parallel. The power flow in each line is given by (4.60) where
the lengths of the lines are given by l = πa and the ratio B A is given by (4.56). The
electric field in the gaps between the lines is given by (4.61). This may be expressed
as a set of azimuthal harmonics of which only the fundamental has non-​zero field
on the axis. The mean amplitude of the field in the gaps is then
πa
2
1
E g 0 (φ) = ∫π E g ( x, φ ) dx. (4.105)
πa a
−
2

The magnitude of the fundamental space harmonic field on the axis is obtained
using (4.37) as

E g 0 ( φ ) sin (β0 g 2 ) g
E0 (φ) = ⋅ ⋅ , (4.106)
I0 (γ 0a) β0 g 2 p
where γ 0 = β20 − ω 2 c 2 . Finally, from (4.17) the Pierce impedance is

E02
ZP = , (4.107)
2β20 Pz
where Pz is the total power flow in the structure. Figure 4.23(b) shows a comparison
between the Pierce impedance calculated in this way with the experimental results
given in [70]. Further experimental results are given in [18, 74].
Although this simple model gives good agreement with experimental results, it
does not take account of the widths of the bars and the capacitance between the
bars and the shield. That capacitance is affected by the positions of the support
rods relative to the bars resulting in changes in the Pierce impedance of the struc-
ture [18]. The properties of the structure can also be changed by changing the shape
of the bars as in the ring-​loop structure, shown in Figure 4.24. This structure has
increased interaction impedance in the fundamental mode, while that of the back-
ward wave mode decreases or even vanishes [75, 76]. It can be manufactured by
making the rings and connecting loops from sheet and then folding them using
a mandrel to ensure alignment [77]. The method outlined above can be used to
model this structure if the short-​circuiting bars are replaced by series inductances
[75]. Other, related, structures are the stub-​supported ring-​bar structure [68] and
the ring-​plane structure [78]. In both cases the rings are supported by metal stubs
or planes which are connected to the shield. When stubs are used they are chosen
to have an electrical length of 90° so that they present an open circuit to the rings.
Compared with the ring-​bar structure these structures have improved thermal per-
formance at the expense of greater dispersion.
169

Waveguide Slow-Wave Structures 169

Figure 4.23:  Computed properties of a ring-​bar slow-​wave structure both with and without
dielectric loading compared with the experimental results in [70]: (a) dispersion curves, and
(b) Pierce impedance.

Figure 4.24:  The ring-​loop slow-​wave structure.

4.5 Waveguide Slow-​Wave Structures

The structures which have been described so far are not suitable for use at high aver-
age power levels because of their poor thermal performance. Structures which are
based on metallic waveguides are much better from that point of view, even though
they have greater dispersion than helix and helix-​derived structures. The ladder line
structure discussed in Section 4.2.1 can be considered to fall into this category if the
bars are thick and the electron beam passes through holes in them.

4.5.1 The Folded Waveguide Structure

A slow-​wave structure can be made using a serpentine rectangular waveguide,
as shown in Figure 4.25 [79]. A simple analysis of this structure supposes that a
TE01 rectangular waveguide mode propagates along the mid-​line of the waveguide
170

170 Slow-Wave Structures

Figure 4.25:  Arrangement of a serpentine (folded) waveguide slow-​wave structure with

swept bends.

(shown by the chain dotted line). At the cut-​off frequency of the waveguide the
phase is the same everywhere. However, the folding of the waveguide means that
the phases in successive gaps are in anti-​phase with one another. Thus the phase
shift between successive gaps is π plus the phase shift along the serpentine path
at the same frequency. The length of the serpentine path between adjacent gaps
is ( h + π p 2 ) so that the phase shift perceived by an electron moving along the
axis is

β n p = β g ( h + π p 2 ) + ( 2 n + 1 ) π, (4.108)

where β g is the propagation constant for the guide and n is an integer. Making use
of (2.11) we have
1
 ω2 2
β n p = βC  2 − 1 ( h + π p 2 ) + ( 2 n + 1)π, (4.109)
 ωC 
where βC = ωC c = π w . Figure 4.26 shows typical dispersion curves of the space
harmonics of this structure in the first four Brillouin zones. It can be seen that
the n = 0 curve represents a backward wave in the first Brillouin zone. Thus the
electron velocity must be chosen to be synchronous with the wave on the structure
in the second Brillouin zone for forward-​wave interaction. The total impedance
of this structure is equal to the impedance ZPV of the waveguide which can be
calculated using (2.64), noting that a in that equation is the width of the wave-
guide denoted here by w. The ratio of the height of the waveguide to the pitch of
the structure ( b p ) can be chosen to maximise the impedance of the first space
harmonic.
The derivation of (4.109) assumed that the bends do not present mismatches
to the wave. In practice both the bends, and the beam holes, introduce small mis-
matches that cause stop bands to occur wherever the dispersion curves cross [79].
17

Waveguide Slow-Wave Structures 171

Figure 4.26:  Dispersion curves of the serpentine (folded) waveguide structure shown in

Figure 4.25.

At the edges of a stop band there is a standing wave in the serpentine waveguide
and the group velocity is zero. At one edge of the stop band the electric field is
maximum on the axis, and at the other edge it is zero. Thus the total impedance is
infinite at one band edge and zero at the other.
An alternative form of the folded waveguide structure in which the height of the
waveguide can be different in different sections is shown in Figure  4.27(a). This
can be modelled using the equivalent circuit in Figure 4.27(b) which represents the
structure between the centres of adjacent gaps. The straight lengths of waveguide
are represented by transmission lines having characteristic impedances Z g1, g 2 and
electrical lengths θ1,2. The bends are represented by an equivalent network B [80,
81]. The parameters of this network can be computed simply using quasi-​static
analysis. The basic properties of the structure can be illustrated using a model that
ignores the impedances of the bends. Then the electrical lengths of the two sections
of waveguide are

θ1 = β g h 2 and θ2 = β g p. (4.110)

The value of β g is the same in each case because we have assumed that both sections
have the same width (w ). The transfer matrix for a section of waveguide is

V2   cos θ jZ g sin θ  V1  V 

 I 2  =  jY sin θ    = [T ]  1  . (4.111)
 g cos θ   1  I  I1 

Using this definition we may write the transfer matrix for the section of the struc-
ture between the centres of adjacent gaps as

[T ] = [T1 ][T2 ][T1 ]. (4.112)

172

172 Slow-Wave Structures

Figure 4.27:  Folded waveguide structure with right-​angle bends: (a) cross-​sectional view, and
(b) equivalent circuit.

Figure 4.28:  Characteristics of a typical folded waveguide structure ( b2 b1 = 1.2 ): (a)

dispersion, and (b) total impedance.

The numerical values of the elements of [T ] can be computed from these equa-
tions and the properties of the structure are given by
φ = arccosT11 (4.113)
and

T12
ZT = . (4.114)
T21
Figures 4.28(a) and (b) show the dispersion and impedance curves for a struc-
ture in which the impedance of the longitudinal waveguides is higher than that of
the transverse waveguides ( b2 > b1 ). The impedance of the lowest pass-​band tends
to infinity at the upper cut-​off frequency whilst that of the next higher band tends
to zero at its lower cut-​off. Thus the standing wave maximum is at the centre of
173

Coupled-Cavity Slow-Wave Structures 173

the beam hole at the top of the first pass-​band and at the centre of the longi-
tudinal waveguide at the bottom of the second pass-​band. When the impedance
of the slots is made less than that of the cavities then the situation is reversed.
The properties of folded waveguide slow-​wave structures can be explored using
Worksheet 4.7.
Folded waveguide structures are usually designed so that Z g 2 > Z g1 because this
gives the best interaction impedance in the lowest pass-​band. Such structures are
sometimes referred to as normal structures. When Z g 2 < Z g1 the structure is said to
be inverted. In both cases the cold bandwidth of the lowest mode is similar, and the
usable hot bandwidth is always somewhat less than this. If Z g1 = Z g 2 then, at least
in theory, the stop bands between the modes vanish [82]. Such a structure, called a
coalesced mode structure, would appear to offer the possibility of making a tube with
a much greater bandwidth. But, because of the discontinuities caused by the bends
and the beam holes, and because of the effects of manufacturing tolerances, it is very
difficult to make a structure which does not have any stop bands. Even a very narrow
stop band would tend to cause a dip in the gain of the tube at that frequency with
the risk of oscillation because of the very high interaction impedance at one of the
band edges.

4.5.2 Helical Waveguides

By analogy with the tape helix slow-​wave structure it would appear that a circular
waveguide having a helical groove on its inner surface should be a wide-​band high-​
power slow-​wave structure. Structures of this type have been studied by a number of
authors (see for example [83–​86] and the references therein). Helical waveguides are
employed in gyro-​TWTs but do not appear to have been used in conventional TWTs.

4.6 Coupled-​Cavity Slow-​Wave Structures

The rectangular folded waveguide structure shown in Figure 4.27(a) can, alterna-

tively, be regarded as a structure in which rectangular cavity resonators are coupled
together by slots in the walls which separate them. The folded waveguide struc-
ture is chiefly of interest for use at millimetre wavelengths where the dimensions
are very small. At microwave frequencies improved performance can be achieved
using structures having more complex geometries. These can be divided into two
classes: space harmonic structures, and forward fundamental structures. Space har-
monic structures that resemble the folded waveguide structure cannot achieve the
highest power levels because operation in the first space harmonic limits the beam
voltage that can be used. To achieve higher power levels it is necessary to use struc-
tures in which there is an additional phase reversal between the cavities so that the
fundamental space harmonic is forward wave.
The properties of coupled-​cavity structures can be found by measurement as
described in Section 4.7.1, or by modelling using computational electromagnetics
software [87–​90]. Modern software allows the use of periodic boundary conditions
174

174 Slow-Wave Structures

Figure 4.29:  Arrangement of a space-​harmonic slow-​wave structure

(copyright1983 IEE, reproduced, with permission, from [91]).

so that it is sufficient to model a single period of the structure, and its properties can
be determined for any value of the phase change per cavity. Computer modelling also
provides detailed information about the fields within the structure so that the cou-
pling impedance can be calculated. Further insight into the properties of some of the
more important types of coupled-​cavity structure can be obtained from equivalent
circuit models as described below. These circuits may be used for interpolation on
computed, or measured data. They are also useful for rapid parametric calculations.

4.6.1 Space Harmonic Structures

Figure 4.29 shows the arrangement of a typical space harmonic (‘Hughes’) struc-
ture for use at microwave frequencies. The cavities are circular and re-​entrant in
shape and are coupled together by large kidney-​shaped slots. The slots are arranged
alternately on either side of the axis in successive coupling plates. Although this
structure can be considered as a somewhat distorted folded waveguide, the usual
conceptual approach is to consider the effects of coupling on the properties of a
chain of identical resonant cavities.
The space-​harmonic slow-​wave structure can be represented by the equivalent
circuit shown in Figure 4.30 [64, 92]. Ports 1 and 2 are connected to the adjacent
cavities whilst port 3 is connected to the electron beam. The cavity inductance Lc is
divided into three parts so that a fraction k of the current circulating in the cavity is
intercepted by each slot. When the fields in adjacent cavities are in phase with one
another the net current intercepted by the slots is zero and the frequency is equal to
the resonant frequency of an isolated cavity. When the fields in adjacent cavities are
in antiphase a net current is intercepted by the slots. If the frequency is below the
resonant frequency of the slots they present an inductive reactance to the current
and so reduce the frequency of the mode. This is consistent with a backward-​wave
characteristic in the first Brillouin zone.
175

Coupled-Cavity Slow-Wave Structures 175

Figure 4.30:  Equivalent circuit of a space-​harmonic slow-​wave structure

(copyright 1983 IEE, reproduced, with permission, from [91]).

The equivalent circuit in Figure  4.30 models the lowest mode quite accurately
[91] but fails for the higher modes. The reason is easily understood if the figure
is redrawn in the form of sections of the equivalent circuit of the TE01 waveguide
mode in series with each other. These sections are valid only when the phase change
per section is small, and that assumption is less valid as the frequency rises. Analysis
of the equivalent circuit results in the dispersion equation

1   ω  2    ω  
2

cos φ = 1 − 1 −
    1 + ak −  ω   , (4.115)
kak   ω c    s 
where

ωc = 1 LcCc (4.116)

ωs = 1 LsCs (4.117)

ak = 2 k ( Ls Lc ) (4.118)
and the subscripts c and s refer to the cavity and the slots, respectively. The frequen-
cies at the edges of the stop band are the solutions of (4.115) when φ = 0
1
ω = ωc and ω = (1 + ak ) 2 ω s . (4.119)
The total impedance of the slow-​wave structure is given by

2 ( R Q )c ω
ZT = ⋅ 1 + ak − ( ω ω s )2  (4.120)
kak sin φ ω c  

where ( R Q )c is the R Q of the cavity, p is the structure pitch and vg is the group
velocity. Equations (4.115) and (4.120) can be fitted to experimental data.
176

176 Slow-Wave Structures

The parameters of the circuit can be calculated with useful accuracy from the dimen-
sions of the structure. The resonant frequency and R Q of the cavity are obtained
using the method of moments (see Section 3.5.1 and Worksheet 3.3) or, with less accu-
racy, using Fujisawa’s formulae [93] (see Section 3.5.2 and Worksheet 3.4). The effec-
tive slot length is defined as the area of the slot divided by its width (see Figure 4.29)
πr
ls = αR + (4.121)
2
and the slot is assumed to be resonant when it is a half-​wavelength long so that
fs = c 2ls . (4.122)
The slot is treated as a short length of waveguide carrying the TE01 mode. In order
to find the R Q of the slot it is necessary to define its effective thickness. This is
increased beyond the physical thickness by the fringing of the fields around the slot.
It is found that a useful approximation to the effective slot thickness is
ts = p − h + 2r (4.123)
and then

µ 0 2r
( R Q )s = ⋅ . (4.124)
ε 0 πts
The coupling factor is defined by
ls
k= . (4.125)
2 πR
The dispersion curve and the total impedance can then be calculated using (4.115)
and (4.120). Figure 4.31 shows typical dispersion curves, and impedance plots, cal-
culated for two different slot angles with experimental results for comparison (see
Worksheet 4.8). This model is a simplification and an improvement of the one pre-
viously described by the author, but alternative assumptions about the slot length
and thickness are possible [94]. The bandwidth of the structure increases as the slot
angle ( α ) is increased. However, if α is too great the structure becomes inverted and
the coupling impedance is reduced. It might be thought that the coupling between
the cavities could be increased, without inverting the structure, by using two slots
per cavity in place of one. The result, known as the Chodorow-​Nalos structure
when the slots are in line with each other, actually has a narrow bandwidth because
of direct coupling between the slots [95]. This is a form of ladder-​line which is some-
times useful at millimetre wave frequencies. An alternative in which the slots are
staggered by rotating successive coupling plates by 90° has better bandwidth [96].

4.6.2 The Cloverleaf Structure

At very high powers it is desirable to use a structure in which the fundamental mode is
forward wave. One such is the cloverleaf structure in which the cavities have the shape
17

Coupled-Cavity Slow-Wave Structures 177

Figure 4.31:  Typical results for a space-​harmonic slow-​wave structure: (a) dispersion, and

(b) total admittance.

Figure 4.32:  General arrangement of a cloverleaf slow-​wave structure

(copyright 1983 IEE, reproduced, with permission, from [91]).

with four or more noses, shown in Figure 4.32 [97]. Successive cavities are rotated
with respect to each other so that the noses in one cavity correspond to the gaps in the
next. The cavities are coupled by radial slots that lie close to the noses, as shown. If
the cavities are excited in phase with each other, then the magnetic field lines circulate
in the same direction in adjacent cavities, as shown by the solid and dashed arrows.
The associated azimuthal current components are in the same direction, and the
178

178 Slow-Wave Structures

current intercepted by each slot is non zero. The slot inductances are then effectively
in series with the inductance of the cavity, and the resonant frequency is reduced.
When the fields in adjacent cavities are in antiphase, the intercepted currents cancel,
and the resonant frequency is close to that of the cavity. Hence the fundamental mode
of the structure is a forward wave in the first Brillouin zone.
The cloverleaf structure can be represented by the equivalent circuit in Figure 4.33
which is very like that in Figure 4.30 [92]. This time a fraction k of the current in
each cavity is intercepted by all the slots because they are in line with one another.
The crossing of the connections on the right-​hand side of the figure represents the
phase reversal brought about by the geometry. Analysis of this circuit produces

1 − ( ω ω c )2  1 − ( ω ω s )2 
cos ( φ ) = −1 −   
(4.126)
ak 1 − k − ( ω ω c )
 
2
 
and

2 k ( R Q )c ω 1 − (ω ω s )
2

ZT = ⋅ ⋅ . (4.127)
ak sin φ ω c 1 − k − ( ω ω )2  2
 c

The phase change per cavity is π when ω = ω c and ω = ω s . The cavities have proper-
ties similar to those of the pill-​box resonator shown in Figure 2.2. The cavity resonant
frequency is approximately that of a pill-​box whose diameter is 1.1 times that of the
diameter across the noses. The R Q is slightly less than that of a pill-​box cavity because
of the additional energy storage in the lobes of the cavity between the noses. The
slots resonate when their lengths are close to half of a free-​space wavelength. They
are weakly coupled to each other by the cavities so that the slot band is narrow. This
structure has higher-order modes corresponding to the higher-order TM modes of the
cavity. These are weakly coupled to the slots and therefore have narrow bandwidth,
and low group velocity. The TM 02 mode is potentially troublesome because E z is non
zero on the axis, and the mode has high interaction impedance. Because of the shape

Figure 4.33:  Equivalent circuit of a cloverleaf slow-​wave structure

(copyright 1983 IEE, reproduced, with permission, from [91]).
179

Coupled-Cavity Slow-Wave Structures 179

of the cavity the outer maximum of this mode may break up into a set of individual
maxima at the centres of the lobes between the cloverleaf noses [64].

4.6.3 The Centipede Structure

Another forward-​ fundamental structure is the centipede structure shown in
Figure 4.34 [97–​100]. In this structure there is inductive coupling between the loops
and the cavities.
The structure can be represented by the equivalent circuit shown in Figure 4.35
[91]. Note that it is necessary to include direct loop to loop coupling in the model.
The connections on the right-​hand side of the diagram are reversed to account for
the phase reversal produced by the shape of the loops. An alternative equivalent
circuit is described in [101].
Analysis of the circuit in Figure 4.35 results in the dispersion equation

1 ( ω c ω ) − 1 ( ω l ω ) + 2 k2 − 1
 2
 2

cos φ = −1 + ⋅ , (4.128)
2 k12 + k2 ( ω c ω ) − 1
2
 

Figure 4.34:  General arrangement of a centipede slow-​wave structure

(copyright 1983 IEE, reproduced, with permission, from [91]).

Figure 4.35:  Equivalent circuit of a centipede slow-​wave structure

(copyright 1983 IEE, reproduced, with permission, from [91]).
180

180 Slow-Wave Structures

where the subscripts c and l refer to the cavity and the loops, respectively,

k1 = Lc Ll M lc (4.129)

and

k2 = Ll M ll . (4.130)

The band edges at φ = π are related to the cavity and loop resonant frequencies by

ω = ωc and ω = ω l 1 − 2 k2 . (4.131)

The cavity parameters are close to those for a pill-​box cavity having the same dimen-
sions as the cavities of the structure. The parameters of the loops must generally be
determined by fitting the model to experimental data. The dispersion diagram of
the centipede structure is rich in higher-​order modes with varying rotational sym-
metries. The uniform distribution of the loops around the edge of the cavity allows
modes such as the TM11 loop mode to exist. Pairs of cavity and loop modes having
the same rotational symmetry are normal (inverted) depending on whether the lower
cut-​off frequency of the loop mode lies above (below) the corresponding resonant
frequency of the cavities. The total impedance of the centipede structure is given by

k12 ( R Q )c  ω c  3 (ωl ω ) + 2k2 − 1 . (4.132)

2

ZT = − ⋅ 
sin φ  ω  2
( )
2
k1 + k2 ( ω c ω ) − 1 
2
 

Figure  4.36 shows a typical dispersion diagram for a centipede structure cal-
culated using (4.128). It has been assumed that the dispersion diagrams for the
higher-order modes obey the same equation with the appropriate cavity resonant
frequency. The loop frequencies of the TM11 and TM 21 modes were increased from
that for the TM 01 mode by factors of 1.2 and 1.3, respectively, to achieve a better
match with experimental results. The centipede structure is smaller than a clover-
leaf structure designed to operate at the same frequency but its heat transfer prop-
erties are inferior.

4.6.4 Termination of Coupled-​Cavity Slow-​Wave Structures

The input and output connections to coupled-​cavity slow-​wave structures are com-
monly made using coupling loops, or inductive irises, as described in Section 3.6.
Additional elements may be used to achieve a good match over the intended work-
ing band of the tube [4, 64, 102]. The use of transition cavities to obtain a good
match is discussed in [103]. Direct coupling, from reduced height waveguide, is an
obvious choice for folded waveguide structures, space harmonic structures and clo-
verleaf structures [104]. An alternative is to make use of a door-​knob transition
(Figure 2.25) since a circular TM 01 mode is to be launched in the structure [98].
18

Measurement of the Properties of Slow-Wave Structures 181

Figure 4.36:  Typical mode diagram of a centipede slow-​wave structure ( k1 = 0.3, k2 = 0.1).

In coupled-​cavity travelling-​wave tubes it is usual for there to be a break in the

structure at each sever. There may be a transition to an external waveguide, allow-
ing the sever power to be dissipated in waveguide loads either inside, or outside, the
vacuum envelope. This arrangement adds to the size and complexity of the tube.
Alternatively the structure may be terminated internally by adding lossy ceramics
to one or more cavities [4].

4.7 Measurement of the Properties of Slow-​Wave Structures

Although the properties of slow-​wave structures can usually be calculated to good

accuracy using computational electromagnetics, it is still desirable to be able to
check the results by comparison with cold test measurements on actual, or scaled,
slow-​wave structures. It is normally possible to measure the phase velocity with
good accuracy, but accurate measurement of the coupling impedance presents
greater difficulties.

4.7.1 Measurements on Coupled-​Cavity Slow-​Wave Structures

Measurements on coupled-​cavity slow-​wave structures are made using resonant
sections [64, 105]. The short-​circuit planes must coincide with planes of symme-
try. For some structures (e.g. cloverleaf, centipede) the mid-​plane of the cavity
182

182 Slow-Wave Structures

can be used, giving a stack with half-​height end cavities. For space harmonic
structures the symmetry planes are at the centres of the slots. The structure may
be at full-​scale, or a scaled-​up version may be made to make the measurements
easier. Experimental measurements only yield figures for particular points on
the dispersion curve. Five or six cavities are usually sufficient, and the proper-
ties at intermediate points are then found by numerical interpolation. One way
to achieve this is to fit an equivalent circuit model to the experimental data [91].
The method of measurement is the same as that used for single cavities (see
Section 3.7) with the difference that many more resonant modes exist. These
can usually be identified by careful probing of the structure with a dielectric or
metal rod through small holes drilled in the ends, or the sides, of the structure.
If the resonances are displayed using a swept frequency source, the extent of the
perturbation of a resonance by probing at different positions can be observed.
Since the change in frequency is proportional to the stored electric energy it is
possible to find the positions of field maxima and zeroes. The axial variation
of modes for which E z ≠ 0 on the axis can also be determined by observing
the frequency shifts as a small dielectric bead is moved along the axis of the
structure. In some cases it is not possible to measure the band-​edge frequencies
directly and they must be deduced by extrapolation from the data points avail-
able. A suitable formula is

f0  0.1(15 f1 − 6 f2 + f3 ) , (4.133)

where f0 is the frequency at the band edge and f1,2,3 are the frequencies of three
resonances such that all four frequencies are determined at equal intervals in φ. This
equation fits an even, fourth-order, polynomial to the data points.
The interaction impedance is measured experimentally by perturbing the reson-
ant stack with a dielectric rod on its axis. A simple formula can be obtained for the
calculation of the Pierce impedance [64] based upon the change in the capacitance
resulting from the presence of the rod,

4 ∆f p ∂φ
ZP = ⋅ ⋅ ⋅ , (4.134)
ε 0 ( ε r − 1) π d
2 2
f φ 2 ∂f

where d is the rod diameter, ε r its relative permittivity and ∆f the change in the
resonant frequency when the rod is inserted. The last term is inversely proportional
to the group velocity. This equation is only strictly valid in the limit of a vanish-
ingly small rod. The measurement accuracy is limited by the need to use a rod which
is large enough for the change in frequency to be measured accurately. Equation
(4.134) also ignores the effects of space-​harmonic fields which become more impor-
tant as the rod diameter is increased. A more accurate formula which includes these
effects is given in [106, 107]. Similar methods can be used for all other structures,
such as ladder lines and ring-​bar structures, which possess transverse symmetry
planes.
183

Measurement of the Properties of Slow-Wave Structures 183

4.7.2 Measurements on Helix Slow-​Wave Structures

Helix slow-​wave structures do not possess any planes of symmetry and it is there-
fore necessary to use non-​resonant methods. The one end of the structure under
test is connected to an external termination, and the other end is similarly termi-
nated or fitted with an internal load. It is very important that there should be good
matches at both terminations [108]. The input to the helix is connected to a vector
network analyser. A perturbing object is passed through the helix and the rate of
change of the phase of the reflected signal with the position of the object is just
twice the phase shift per unit length of the structure. The perturbing object may be
a bead or rod made from metal or dielectric material, or it may be small helix hav-
ing a small number of turns [109–​112]. The measurements are normally made on a
full-​scale section of helix. The helix diameter is typically a few millimetres or less,
and that makes it more difficult to obtain accurate results. The equipment is nor-
mally automated so that a large number of measurements can be made to increase
the accuracy of the results.
To measure the coupling impedance of the helix it is necessary to measure the
change in the propagation constant produced by perturbation with a dielectric
or metal rod. One way of doing this is to mount a bead, or a small helix, on
a dielectric rod which passes through the whole length of the structure, and to
measure the phase of the reflected signal [109–​112]. The impedance can be calcu-
lated from the difference between the wavelengths measured with, and without,
the rod. To get reasonable accuracy with this method it is necessary to measure
the wavelengths very accurately. Lagerstrom [113] measured the Pierce imped-
ance by observing the change in the phase of the signal transmitted through the
helix when a dielectric rod was inserted in it. The measurement is subject to some
uncertainty because end effects mean that it is not possible to determine the pre-
cise length of the rod in the helix. This difficulty can be overcome by progressively
advancing the rod into the helix, and observing the rate of change of the phase of
the transmitted signal with the position of the rod. It is necessary to correct the
results for the effects of space-​harmonic fields and transverse field components
[108]. Since the diameter of the rod is normally smaller than the internal diameter
of the helix it is very important to ensure that it is accurately aligned with the axis
of the helix. A metal rod was used in place of the dielectric rod by Kino [114]. The
perturbations are larger with this method, but the end effects are more serious.
The phase velocity and interaction impedance can be measured simultaneously,
using this method, by observing the phases of both the reflected and transmitted
signals as functions of the position of the rod. The phase velocity, and interac-
tion impedance, can also be determined from the amplitude and phase of the
signal reflected by a spherical dielectric bead placed within the helix [110]. This
(S-​parameter) method makes use of the fact that the amplitude of the reflected
signal is proportional to the local change in stored electric energy produced by the
bead. The accuracy achieved is comparable with that obtained by other methods
(see Figure 4.20).
184

184 Slow-Wave Structures

References

[1]‌ A. F. Harvey, ‘Periodic and guiding structures at microwave frequencies’, IRE

Transactions on Microwave Theory and Techniques, vol. 8, pp. 30–​61, 1960.
[2]‌ W. Wang et  al., ‘Review of the novel slow-​wave structures for high-​power traveling-​
wave tube’, International Journal of Infrared and Millimeter Waves, vol. 24, pp.
1469–​1484, 2003.
[3]‌ N. W. McLachlan, Bessel Functions for Engineers. Oxford: Oxford University Press, 1954.
[4]‌ A. S. Gilmour, Jr., Principles of Traveling Wave Tubes. Norwood, MA: Artech House,
1994.
[5]‌ J. H. Booske et al., ‘Traveling-​wave tubes’, in R. J. Barker et al., eds, Modern Microwave
and Millimetre-​Wave Power Electronics. Piscataway, NJ: IEEE, pp.  171–245, 2005.
[6]‌ S. Ramo et al., Fields and Waves in Communication Electronics. New York: Wiley, 1965.
[7]‌ H. Kosai et  al., ‘Plasma-​ filled dielectric Cherenkov maser’, IEEE Transactions on
Plasma Science, vol. 18, pp. 1002–​1007, 1990.
[8]‌ E. I. Smirnova et al., ‘Design, fabrication, and low-​power tests of a W-​band omniguide
traveling-​wave tube structure’, IEEE Transactions on Plasma Science, vol. 36, pp.
763–​767, 2008.
[9]‌ D. Y. Shchegolkov et al., ‘Millimeter-​wave gain experiments with a wide-​band omniguide
traveling-​wave tube’, in IEEE Thirteenth International Vacuum Electronics Conference
(IVEC 2012), pp. 297–​298, 2012.
[10] R. E. Collin, Foundations for Microwave Engineering. New York: McGraw-​Hill, 1966.
[11] L. Brillouin, Wave Propagation in Periodic Structures. New York: Dover, 1953.
[12] R. W. Hamming, Numerical Methods for Scientists and Engineers. New  York:
McGraw-​Hill, 1973.
[13] T. Edwards, Foundations for Microstrip Circuit Design, 2nd ed. Chichester, UK: Wiley,
1992.
[14] P. N. Butcher, ‘The coupling impedance of tape structures’, Proceedings of the IEE –​
Part B: Radio and Electronic Engineering, vol. 104, pp. 177–​187, 1957.
[15] J. A. Weiss, ‘Dispersion and field analysis of a microstrip meander-​line slow-​wave structure’,
IEEE Transactions on Microwave Theory and Techniques, vol. 22, pp. 1194–​1201, 1974.
[16] A. Karp, ‘Traveling-​wave tube experiments at millimeter wavelengths with a new, easily
built, space harmonic circuit’, Proceedings of the IRE, vol. 43, pp. 41–​46, 1955.
[17] J. Froom et al., ‘Ridge-​loaded ladder lines’, IEEE Transactions on Electron Devices, vol.
12, pp. 411–​421, 1965.
[18] E. A. Ash et al., ‘Dispersion and impedance of dielectric-​supported ring-​and-​bar slow-​wave
circuits’, Proceedings of the Institution of Electrical Engineers, vol. 111, pp. 629–​641, 1964.
[19] W. E. Courtney, Printed-​Circuit RF-​Keyed Crossed-​Field Amplifier. Lexington, MA:
Lincoln Laboratory, 1975.
[20] C. Chua and S. Aditya, ‘A 3-​ D U-​ shaped meander-​ line slow-​ wave structure for
traveling-​wave-​tube applications’, IEEE Transactions on Electron Devices, vol. 60, pp.
1251–​1256, 2013.
[21] B. R. Potter et al., ‘High-​power printed circuit traveling wave tubes’, in International
Electron Devices Meeting, pp. 521–​524, 1973.
[22] S. Sengele et  al., ‘Microfabrication and characterization of a selectively metallized
W-​band meander-​line TWT circuit’, IEEE Transactions on Electron Devices, vol. 56,
pp. 730–​737, 2009.
185

References 185

[23] M. Sumathy et  al., ‘Design and RF characterization of W-​band meander-​line and
folded-​ waveguide slow-​ wave structures for TWTs’, IEEE Transactions on Electron
Devices, vol. 60, pp. 1769–​1775, 2013.
[24] A. H. W. Beck, Space-​Charge Waves and Slow Electromagnetic Waves. London: Pergamon
Press, 1958.
[25] R. C. Fletcher, ‘A broad-​band interdigital circuit for use in traveling-​wave-​type ampli-
fiers’, Proceedings of the IRE, vol. 40, pp. 951–​958, 1952.
[26] J. Arnaud, ‘Circuits of travelling wave crossed field tubes’, in E. Okress et  al., eds,
Crossed-​Field Microwave Devices, vol. 1. New York: Academic Press, pp. 47–​67, 1961.
[27] K. Xie, ‘CVD diamond rod and its application in the high power TWT’, in Conference
Digest: 25th International Conference on Infrared and Millimeter Waves, pp. 343–​344, 2000.
[28] G. Fleury et al., ‘Average power limits of brazed-​helix TWT’s’, in International Electron
Devices Meeting, pp. 806–​809, 1980.
[29] R. Crivello and R. W. Grow, ‘Thermal analysis of PPM-​focused rod-​supported TWT
helix structures’, IEEE Transactions on Electron Devices, vol. 35, pp. 1701–​1720, 1988.
[30] Y. Han et al., ‘Thermal analysis of a helix TWT slow-​wave structure’, IEEE Transactions
on Electron Devices, vol. 55, pp. 1269–​1272, 2008.
[31] S. F. Paik, ‘Design formulas for helix dispersion shaping’, IEEE Transactions on Electron
Devices, vol. 16, pp. 1010–​1014, 1969.
[32] K. Tsutaki et al., ‘Numerical analysis and design for high-​performance helix traveling-​
wave tubes’, IEEE Transactions on Electron Devices, vol. 32, pp. 1842–​1849, 1985.
[33] B. Basu, ‘Equivalent circuit analysis of a dielectric-​supported helix in a metal shell’,
International Journal of Electronics, vol. 47, pp. 311–​314, 1979.
[34] J. R. Pierce, Traveling-​Wave Tubes. Princeton, NJ: D. van Nostrand, 1950.
[35] L. Loshakov and E. Ol’Derogge, ‘Propagation of slow electromagnetic waves along a
helix with dielectric supports’, Radio Engineering and Electronic Physics, vol. 13, pp.
45–​51, 1968.
[36] P. K. Jain and B. N. Basu, ‘The inhomogeneous dielectric loading effects of practical
helix supports on the interaction impedance of the slow-​wave structure of a TWT’,
IEEE Transactions on Electron Devices, vol. 39, pp. 727–​733, 1992.
[37] P. Hammond and J. K. Sykulski, ‘Tubes and slices:  a new way of teaching the prin-
ciples of electric and magnetic fields’, IEEE Transactions on Education, vol. 35, pp.
300–​306, 1992.
[38] Q. Chen, ‘Characteristic of vane-​loaded helix slow-​wave structure’, in Vakuumelektronik
und Displays, Garmisch-​Partenkirchen, Germany, pp. 27–​31, 1983.
[39] S. Sensiper, ‘Electromagnetic wave propagation on helical structures (a review and sur-
vey of recent progress)’, Proceedings of the IRE, vol. 43, pp. 149–​161, 1955.
[40] S. Sensiper, ‘Electromagnetic wave propagation in helical structures’, ScD, Department
of Electrical Engineering, MIT, 1951.
[41] L. Stark, ‘Lower modes of a concentric line having a helical inner conductor’, Journal
of Applied Physics, vol. 25, pp. 1155–​1162, 1954.
[42] D. A. Watkins and E. A. Ash, ‘The helix as a backward-​wave circuit structure’, Journal
of Applied Physics, vol. 25, pp. 782–​790, 1954.
[43] P. K. Jain et al., ‘Effect of the finite thickness of the helix wire on the characteristics of
the helical slow-​wave structure of a traveling-​wave tube’, IEEE Transactions on Electron
Devices, vol. 34, pp. 1209–​1213, 1987.
[44] D. T. Swift-​Hook, ‘Dispersion curves for a helix in a glass tube’, Proceedings of the
IEE –​Part B: Radio and Electronic Engineering, vol. 105, pp. 747–​755, 1958.
186

186 Slow-Wave Structures

[45] A. K. Sinha et  al., ‘Simplified tape model of arbitrarily-​ loaded helical slow-​ wave
structures of a travelling-​wave tube’, IEE Proceedings H:  Microwaves, Antennas and
Propagation, vol. 139, pp. 347–​350, 1992.
[46] S. Ghosh et al., ‘Rigorous tape analysis of inhomogeneously-​loaded helical slow-​wave
structures’, IEEE Transactions on Electron Devices, vol. 44, pp. 1158–​1168, 1997.
[47] S. D’Agostino et al., ‘Accurate analysis of helix slow-​wave structures’, IEEE Transactions
on Electron Devices, vol. 45, pp. 1605–​1613, 1998.
[48] D. Chernin et al., ‘Exact treatment of the dispersion and beam interaction impedance
of a thin tape helix surrounded by a radially stratified dielectric’, IEEE Transactions on
Electron Devices, vol. 46, pp. 1472–​1483, 1999.
[49] N. Kalyanasundaram and G. N. Babu, ‘Dispersion of electromagnetic waves guided by
an open tape helix I’, Progress In Electromagnetics Research B, vol. 16, pp. 311–​331, 2009.
[50] N. Kalyanasundaram and G. N. Babu, ‘Dispersion of electromagnetic waves guided
by an open tape helix II’, Progress In Electromagnetics Research B, vol. 19, pp.
133–​150, 2010.
[51] N. Kalyanasundaram et  al., ‘On the distribution of current on an open tape helix’,
Progress In Electromagnetics Research M, vol. 12, pp. 81–​93, 2010.
[52] P. Greninger, ‘Tape helix perturbation including 3-​ D dielectrics for TWTs’, IEEE
Transactions on Electron Devices, vol. 48, pp. 12–​23, 2001.
[53] T. Wessel-​Berg, ‘CAD of tape helix circuits for traveling wave tubes by convolution
methods’, IEEE Transactions on Electron Devices, vol. 51, pp. 1918–​1927, 2004.
[54] M. Aloisio and G. Sorbello, ‘One-​third-​of-​pitch reduction technique for the analysis
of ternary azimuthally periodic helical slow-​wave structures’, IEEE Transactions on
Electron Devices, vol. 53, pp. 1467–​1473, 2006.
[55] C. L. Kory, ‘Three-​dimensional simulation of helix traveling-​wave tube cold-​test character-
istics using MAFIA’, IEEE Transactions on Electron Devices, vol. 43, pp. 1317–​1319, 1996.
[56] K. Ward and J. Wlodarczyk, ‘Transmission line modelling of helix slow wave structures’, in
Technical Digest: International Electron Devices Meeting (IEDM’93), pp. 157–​160, 1993.
[57] C. L. Kory and J. A. Dayton, Jr, ‘Accurate cold-​test model of helical TWT slow-​wave
circuits’, IEEE Transactions on Electron Devices, vol. 45, pp. 966–​971, 1998.
[58] M. Aloisio and P. Waller, ‘Analysis of helical slow-​wave structures for space TWTs using
3-​D electromagnetic simulators’, IEEE Transactions on Electron Devices, vol. 52, pp.
749–​754, 2005.
[59] T. Onodera and W. Raub, ‘Phase velocity dispersion of a generalized metal-​segment-​
loaded helix as used in broad-​band traveling-​wave tubes’, IEEE Transactions on Electron
Devices, vol. 35, pp. 533–​538, 1988.
[60] R. G. Carter and P. Wang, ‘Equivalent circuit modelling of helix slow-​wave structures’,
presented at the ESA/​NATO 1997 Workshop on Microwave Tubes for Space, Military
and Commercial Applications ESTEC, Noordwijk, The Netherlands, 1997.
[61] R. G. Carter and P. Wang, ‘An equivalent circuit model of helix slow-​wave structures’,
presented at the Microwaves 94 Workshop on CAE, Modelling and Measurement
Verification, London, pp. 85–90, 1994.
[62] T. K. Ghosh et al., ‘Optimization of coaxial couplers’, IEEE Transactions on Electron
Devices, vol. 54, pp. 1753–​1759, July 2007.
[63] V. Kumar et  al., ‘Experimental evaluation of lossy coatings of different materials on
helix support rods for a high-​efficient TWT’, in 14th International Crimean Conference
on Microwave and Telecommunication Technology (CriMico 2004), 2004, pp. 648–​650.
187

References 187

[64] J. F. Gittins, Power Travelling-​Wave Tubes. London: English Universities Press, 1965.

[65] V. B. Naidu et  al., ‘Three-​dimensional electromagnetic analysis of attenuator-​coated
helix support rods of a traveling-​wave tube’, IEEE Transactions on Electron Devices,
vol. 56, pp. 945–​950, 2009.
[66] D. M. Goebel et  al., ‘Gain stability of traveling wave tubes’, IEEE Transactions on
Electron Devices, vol. 46, pp. 2235–​2244, 1999.
[67] M. K. Alaria et al., ‘Design of tip loss profile on support rod for a helix TWT’, Journal
of Infrared, Millimeter, and Terahertz Waves, vol. 30, pp. 651–​656, 2009.
[68] C. K. Birdsall and T. E. Everhart, ‘Modified contra-​wound helix circuits for high-​power
traveling-​wave tubes’, IRE Transactions on Electron Devices, vol. 3, pp. 190–​204, 1956.
[69] J. E. Nevins, Jr., ‘An investigation and application of the contrawound helix’, IRE
Transactions on Electron Devices, vol. 6, pp. 195–​202, 1959.
[70] D. T. Lopes and C. C. Motta, ‘Characterization of ring-​bar and contrawound helix cir-
cuits for high-​power traveling-​wave tubes’, IEEE Transactions on Electron Devices, vol.
55, pp. 2498–​2504, 2008.
[71] S. K. Datta et  al., ‘Equivalent circuit analysis of a ring-​bar slow-​wave structure for
high-​power traveling-​wave tubes’, IEEE Transactions on Electron Devices, vol. 56,
pp. 3184–​3190, 2009.
[72] W. N. Cain and R. W. Grow, ‘The effects of dielectric and metal loading on the disper-
sion characteristics for contrawound helix circuits used in high-​power traveling-​wave
tubes’, IEEE Transactions on Electron Devices, vol. 37, pp. 1566–​1578, 1990.
[73] S. Sengele et al., ‘Backward-​wave suppression analysis, and design and fabrication of a
prototype millimeter-​wave ring-​bar slow-​wave structure’, IEEE Transactions on Plasma
Science, vol. 42, pp. 3949–​3960, 2014.
[74] D. C. Buck, ‘Traveling-​wave tube interaction circuits’, IEEE Transactions on Electron
Devices, vol. 14, pp. 721–​723, 1967.
[75] S. Liu, ‘The theory of ring-​loop structure’, Scientia Sinica, vol. 20, pp. 679–​695, 1977.
[76] S. Liu, ‘Calculation of the parameters for ring-​loop traveling wave tube in MMW’,
International Journal of Infrared and Millimeter Waves, vol. 21, pp. 1097–​1101, 2000.
[77] B. K. Vancil and E. G. Wintucky, ‘A low cost electrostatically focused TWT’, in Third
IEEE International Vacuum Electronics Conference (IVEC 2002), pp. 180–​181, 2002.
[78] R. M. White et al., ‘Properties of ring-​plane slow-​wave circuits’, IEEE Transactions on
Electron Devices, vol. 11, pp. 247–​261, 1964.
[79] G. Dohler et  al., ‘Serpentine waveguide TWT’, in International Electron Devices
Meeting, pp. 485–​488, 1987.
[80] R. G. Carter, ‘An improved equivalent circuit model of folded-​waveguide slow-​wave
structures’, in Vacuum Electronics and Displays, Garmisch-​Partenkirchen, Germany,
pp. 55–​59, 2001.
[81] J. H. Booske et  al., ‘Accurate parametric modeling of folded waveguide circuits for
millimeter-​wave traveling wave tubes’, IEEE Transactions on Electron Devices, vol. 52,
pp. 685–​694, 2005.
[82] B. G. James and P. Kolda, ‘A ladder circuit coupled-​cavity TWT at 80–​100 GHz’, in
1986 International Electron Devices Meeting, pp. 494–​497, 1986.
[83] K. J. Bunch and R. W. Grow, ‘The helically wrapped circular waveguide’, IEEE
Transactions on Electron Devices, vol. 34, pp. 1873–​1884, 1987.
[84] Y. Wei et  al., ‘An approach to the analysis of arbitrarily shaped helical groove wave-
guides’, IEEE Microwave and Guided Wave Letters, vol. 10, pp. 4–​6, 2000.
18

188 Slow-Wave Structures

[85] W. Wang et al., ‘Study of the ridge-​loaded helical-​groove slow-​wave structure’, IEEE
Transactions on Microwave Theory and Techniques, vol. 45, pp. 1689–​1695, 1997.
[86] G. Yu et  al., ‘Analysis of the coaxial helical-​ groove slow-​ wave structure’, IEEE
Transactions on Microwave Theory and Techniques, vol. 50, pp. 191–​200, 2002.
[87] R. G. Carter et al., ‘Computer modelling of coupled-​cavity slow-​wave structures’, in
Vakuumelektronik und Displays, Garmish-​Partenkirchen, Germany, pp. 69–​74, 1995.
[88] F. Kantrowitz and I. Tammaru, ‘Three-​dimensional simulations of frequency-​phase
measurements of arbitrary coupled-​cavity RF circuits’, IEEE Transactions on Electron
Devices, vol. 35, pp. 2018–​2026, 1988.
[89] C. L. Kory et al., ‘Simulation of cold-​test dispersion and interaction impedances for
coupled-​cavity tube slow-​wave circuits’, in International Electron Devices Meeting
1992 (IEDM ‘92) Technical Digest, pp. 763–​766, 1992.
[90] P. Prince et al., ‘A numerical method for predicting Brillouin plots of coupled cavity
circuits’, in International Electron Devices Meeting, pp. 676–​679, 1984.
[91] R. G. Carter, ‘Representation of coupled-​cavity slow-​wave structures by equivalent cir-
cuits’, IEE Proceedings I: Solid-​State and Electron Devices, vol. 130, pp. 67–​72, 1983.
[92] H. J. Curnow, ‘A general equivalent circuit for coupled-​cavity slow-​wave structures’,
IEEE Transactions on Microwave Theory and Techniques, vol. 13, pp. 671–​675, 1965.
[93] K. Fujisawa, ‘General treatment of klystron resonant cavities’, IRE Transactions on
Microwave Theory and Techniques, vol. MTT-​6, pp. 344–​358, 1958.
[94] R. G. Carter and S. K. Liu, ‘Method for calculating the properties of coupled-​cavity
slow-​wave structures from their dimensions’, IEE Proceedings-​H Microwaves Antennas
and Propagation, vol. 133, pp. 330–​334, October 1986.
[95] M. Chodorow and E. J. Nalos, ‘The design of high-​power traveling-​wave tubes’,
Proceedings of the IRE, vol. 44, pp. 649–​659, 1956.
[96] A. Karp, ‘“Comb-​Quad” millimeter-​wave coupled-​cavity TWT interaction structure’,
in International Electron Devices Meeting, pp. 325–​327, 1981.
[97] M. Chodorow and R. A. Craig, ‘Some new circuits for high-​power traveling-​wave
tubes’, Proceedings of the IRE, vol. 45, pp. 1106–​1118, 1957.
[98] P. S. Matthews et  al., ‘The centipede structure in practical megawatt tubes’, in 8th
European Microwave Conference, Paris, France, pp. 753–​760, 1978.
[99] T. Roumbanis, ‘Centipede Twystron ® amplifiers and traveling wave tubes for broad-​
band high-​ efficiency super-​
power amplification’, IEEE Transactions on Electron
Devices, vol. 16, p. 242, 1969.
[100] A. F. Pearce, ‘A structure, using resonant coupling elements, suitable for a high-​
power travelling-​wave tube’, Proceedings of the IEE  –​Part B:  Radio and Electronic
Engineering, vol. 105, pp. 719–​726, 1958.
[101] T. M. Reeder, ‘An equivalent circuit for the “Centipede” waveguide’, IEEE Transactions
on Microwave Theory and Techniques, vol. 14, pp. 200–​205, 1966.
[102] W. Wiejak, ‘X-​band coupled cavity slow wave structure of the traveling wave tube
for airborne application’, in 18th International Conference on Microwave Radar and
Wireless Communications (MIKON), pp. 1–​4, 2010.
[103] T. Kageyama, ‘The design of the transition region in coupled-​cavity TWT’, in Third
IEEE International Vacuum Electronics Conference (IVEC 2002), pp. 102–​103, 2002.
[104] G. Faillon et al., ‘Microwave tubes’, in J. A. Eichmeier and M. K. Thumm, eds, Vacuum
Electronics: Components and Devices. Berlin: Springer-​Verlag, pp. 1–​84, 2008.
189

References 189

[105] A. W. Horsley and A. Pearson, ‘Measurement of dispersion and interaction impedance

characteristics of slow-​wave structures by resonance methods’, IEEE Transactions on
Electron Devices, vol. 13, pp. 962–​969, 1966.
[106] D. J. Connolly, ‘Correction to “Determination of the interaction impedance of cou-
pled cavity slow wave structures” ’, IEEE Transactions on Electron Devices, vol. 23,
p. 1195, 1976.
[107] D. J. Connolly, ‘Determination of the interaction impedance of coupled cavity slow
wave structures’, IEEE Transactions on Electron Devices, vol. 23, pp. 491–​493, 1976.
[108] P. Wang et al., ‘An improved technique for measuring the Pierce impedance of helix
slow-​wave structures’, in 24th European Microwave Conference, pp. 998–​1003, 1994.
[109] S. J. Rao et  al., ‘Nonresonant perturbation measurements on dispersion and inter-
action impedance characteristics of helical slow-​wave structures’, IEEE Transactions
on Microwave Theory and Techniques, vol. 45, pp. 1585–​1593, 1997.
[110] P. Wang et al., ‘A simple technique for measuring the interaction impedance of helical
slow-​wave structures’, in Vakuumelektronik und Displays, Garmisch-​Partenkirchen,
Germany, pp. 63–​69, 1995.
[111] B. T.  J. Maharaj and E. W. Schumann, ‘Automated measurement methods to char-
acterise travelling-​wave tube slow-​wave structures’, Elektron (South Africa), vol. 6,
pp. 8–​10, 1989.
[112] J. R. Legarra, ‘Measurement of microwave characteristics of helix traveling wave cir-
cuits’, in International Electron Devices Meeting, pp. 408–​411, 1979.
[113] R. P. Lagerstrom, ‘Interaction impedance measurements by perturbation of traveling
waves’, Electronics Research Lab., Stanford University, Stanford, CA, 1957.
[114] G. S. Kino, ‘Normal mode theory in perturbed transmission systems’, Electronics
Research Labs., Stanford University, Stanford, CA, 1955.
190

5 Thermionic Diodes

5.1 Introduction

The simplest vacuum electron device is the plane parallel diode shown in
Figure 5.1(a). The diode comprises a cathode, from which electrons are emitted, and
an anode within the same vacuum envelope. If the anode is positive with respect to
the cathode, and there is no electron emission, the electrostatic potential varies lin-
early between the cathode and the anode as shown by the solid line in Figure 5.1(b).
Now suppose that electrons are emitted from the cathode with negligible velocity
and a current density which can be controlled in some way. The electrons are drawn
towards the anode by the electric field between the electrodes, and the negative
space-​charge associated with the electron current depresses the potential in the
space between the electrodes, as shown by the dashed line in Figure 5.1(b). If the
electric field at the surface of the cathode is negative all the electrons which are
supplied by the cathode can flow to the anode. The current through the diode is
then determined by the properties of the cathode. When the potential difference
between the electrodes is reversed no electrons can reach the anode. Thus the device
is a diode which only permits the flow of electrons from the cathode to the anode.
It should be noted that the conventional current of electric circuit theory assumes
that the charge carriers are positive so that the conventional current flows from the
anode to the cathode.
As the emitted current density is increased the potential depression increases
until the electric field is zero at the cathode surface. The current through the diode
is then determined by the geometry of the device and the potential difference across
it, and not by the current density available at the cathode. The diode is then said to
be space-​charge limited and the current cannot be increased any further by increas-
ing the cathode emission.
This chapter discusses the steady-​state properties of planar, cylindrical and spher-
ical space-​charge limited diodes, including the effects of the initial energies of the
electrons and of relativity. When the potential difference across a diode varies with
time the static relationship continues to hold until the time taken for electrons to
cross the diode becomes comparable with the time in which the voltage is changing.
It is shown that this sets a limit to the maximum frequency of operation of a diode.
Section 5.8 discusses the injection of electrons into a planar diode and Section 5.9
reviews the properties of diodes in which the current flow is two-​dimensional.
19

Introduction 191

Figure 5.1:  Planar vacuum diode showing: (a) general arrangement, and (b) the electrostatic
potential within the diode showing the potential depression caused by space-​charge
(dashed line).

An important form of diode is the electron gun in which the shapes of the elec-
trodes are chosen in such a way that the electrons do not strike the anode but pass
through a hole in it and are projected into the region beyond. The electron gun is
a fundamental component of many microwave tubes and the theory and design of
the principal types are reviewed in Chapter 9.

5.1.1 Dimensional Analysis of Thermionic Diodes

Useful insights into the properties of thermionic diodes are provided by the use of
dimensional analysis (see Section 1.7.1.). The objective is to discover how the cur-
rent flowing through a diode depends upon the applied voltage and the dimensions
of the diode. The shape of the diode may be quite complicated but, for the present,
we will assume that it can be described by the distance between the electrodes and
by the area of the cathode. Since this is a problem concerning the dynamics of
charged particles we expect that it will involve the charge/​mass ratio of the electron
and the primary electric constant ε 0 . When thermal velocities or relativistic effects
are important it is necessary to introduce a reference voltage Vr related to their mag-
nitude. The dimensional quantities required to specify the problem are summarised
in Table 5.1 together with their dimensions.
The problem has seven parameters and four dimensions. The application of
Buckingham’s theorem shows that it can be specified by three dimensionless groups.
Two dimensionless groups can readily be formed as

Va
Π1 = (5.1)
Vr
and

A
Π2 = . (5.2)
d2
192

192 Thermionic Diodes

Table 5.1:  Parameters of a thermionic diode

Quantity Symbol Dimensions

Current I [ML2T−​3V−​1]
Anode voltage Va [V]‌
Reference voltage Vr [V]‌
Electrode separation d [L]‌
Area A [L2]
Primary electric constant ε0 [MLT−​2V−​2]
Charge/​mass ratio of the electron e m0 [L2T−​2V−​1]

The remaining group is found by assuming that

I = Vrα Lβ ε 0γ ( e m ) (5.3)
δ

where the indices are to be determined. Then, examining the powers of the dimen-
sions, we find that

M: 1= γ
L: 2 = β + γ + 2δ
(5.4)
T: − 3 = −2 γ − 2δ
V: − 1 = α − 2 γ − δ.
The solution to these equations is α = 1; β = 0; γ = 3/​2; δ = ½ and therefore the third
dimensionless group is

 I 
Π3 =   . (5.5)
 ε 0 (e m)Vr3 
Any one of the dimensionless groups can then be expressed as a function of the
remaining groups, for example

Π3 = f ( Π1 , Π 2 ) , (5.6)

where f is a function, to be determined, which is the same for all diodes which are
geometrically similar, that is they can be scaled from one another.
If neither thermal velocities nor relativistic effects are important then Π1 is not
required. Then Vr can be replaced by Va in (5.5) so that (5.6) becomes

Π3 = f ( Π 2 ) . (5.7)

For any set of geometrically similar diodes Π 2 is constant so that Π3 is also con-
stant. Equation (5.5) shows that the current in the diode is proportional to the
3/​2 power of the applied voltage. Thus dimensional analysis shows that all diodes
which do not require a reference voltage must obey the 3/​2 power law regardless of
193

Introduction 193

geometry. The same result was obtained in different ways by Langmuir [1] and by
Tsimring [2].
Because many electron tubes obey the 3/​2 power law it is convenient to describe
a diode by its perveance (K) defined by

I
K= , (5.8)
Va3 2
which depends only on the geometry of the diode. It follows that all diodes whose
dimensions can be scaled from one another have the same perveance. Because the
perveance is typically of the order of 10–​6 A V−1.5 it is usual to use micro-​perveance
( µP = µA V −1.5 ). We shall see below that when thermal velocities, or relativistic
effects, are important the description of the problem requires a reference voltage
and the current does not obey the 3/​2 power law.

5.1.2 Current Limitation

Virtually all power vacuum tubes use thermionic cathodes whose emitted current
increases with temperature (see Section 18.5). Figure 5.2 shows a typical I-​V plot
for a thermionic diode. At low voltages the diode is space-​charge limited and the
curve follows the 3/​2 power law. As the current increases, it becomes equal to the
maximum current available from the cathode, as described by the Schottky equa-
tion (18.10). This current depends on the temperature and work function of the
cathode, and only increases slowly with increasing anode voltage. The diode is then
said to be temperature limited.
Vaughan [3] suggested an empirical model for the transition from space-​charge
limitation to temperature limitation of the current:

1 1 1
α
= α + α , (5.9)
J JC JS
where JC is the space-​charge limited current density given by the Child-​Langmuir
equation (5.16), JS is the current density given by the Schottky equation (18.10), and
α is an empirical constant. It was found that the value of α was typically in the range
6–​10 for well-​designed diodes (electron guns). However, for diodes with uneven emis-
sion, caused by surface defects or uneven heating, the value could be in the range 2–​5.
The diodes used in power vacuum tubes are usually designed to be operated in
the space-​charge limited region because it ensures that:
• The current depends only on the cathode voltage, and the geometry of the diode,
and not on the condition of the cathode.
• The cathode surface is not damaged by the impact of positive ions caused by the
presence of a strong electric field there.
If it is suspected that the cathode emission is failing, an I-​V plot can be drawn to
check that the operating point is below the knee of the curve. There is a trade-​off, for
194

194 Thermionic Diodes

Figure 5.2:  Plot of current against voltage for a thermionic diode showing the Child-​
Langmuir (space-​charge limited) and Schottky (temperature limited) regions.

a particular cathode, between the current density and the life of the cathode. Thus
tubes which are required to have very long working lives (such as those for use in
satellites) are designed to have low cathode current density. However, temperature-​
limited diodes are used in gyrotrons so that the electron current can be controlled
by the cathode temperature.

5.2 The Planar Space-​Charge Limited Diode

Consider a diode comprising two plane parallel electrodes separated by distance d

as shown in Figure 5.1. The electrostatic potential (V ) is zero at the cathode at x = 0
and V = Va at the anode at x = d . For the moment we will assume that electrons
are emitted with zero energy from the cathode and that relativistic effects can be
neglected. The theory of this diode was first studied by Child [4] and Langmuir [1].
The velocity of an electron at any point is found from the principle of conservation
of energy to be

u = 2 ( e m0 )V , (5.10)

where V is the potential with respect to the cathode. In the steady state the current
density is constant and given by

J = ρu, (5.11)

where ρ is the charge density. The potential is related to the space-​charge density by
Poisson’s equation in one dimension,

d 2V ρ
= − . (5.12)
dx 2 ε0
195

The Planar Space-Charge Limited Diode 195

Combining equations (5.10) to (5.12) gives

d 2V J
=− . (5.13)
dx 2
ε 0 2 ( e m0 )V

Equation (5.13) can be integrated by multiplying both sides by 2 ( dV dx ), and inte-

grating with respect to x to give
2
 dV  4J
  =− V + C , (5.14)
dx  ε 0 2 ( e m0 )

where C is a constant. Applying the boundary conditions to (5.14) gives

4J
Ea2 − E k2 = − ⋅ Va , (5.15)
ε 0 2 ( e m0 )

where Va is the potential of the anode with respect to the cathode and Ea and E k are,
respectively, the electric field at the anode and the cathode. The current density is
maximum when the left-​hand side of (5.15) is maximum. From Figure 5.1(b) we see
that, as the space-​charge density increases Ea increases and Ek decreases. Thus the
maximum possible current density is obtained when E k = 0 and this is the condition
for space-​charge limitation of the current. When the current is space-​charge limited
C = 0 and (5.14) may be integrated a second time to give
3
4 ε 0 2 ( e m0 ) 32 V2
JC = − ⋅ ⋅V = − K 2 , (5.16)
9 x 2
x
the well-​known Child–​Langmuir law where K = 2.334 × 10 −6 A V −1.5. Note that the
current density in (5.16) is negative because the conventional current flows from
the anode to the cathode in the negative x direction. A simple physical derivation
of this equation which gives insight into the reason for the power law was given by
Umstattd et al. [5].
Equation (5.16) can be rearranged, using dimensionless groups, to be in the form (5.6)
3
 I  4  Va  2  A 
  = −    2  , (5.17)
 ε 0 2 ( e m )Vr3  9  Vr   d 

where I = AJC , A is the cross-​sectional area of the diode, and the factor 2 has been
included in the term on the left-​hand side of the equation. In this case it is conveni-
ent to choose a reference voltage of 1.0 V.
The transit time of electrons in a space-​charge-​limited diode may be calculated
by noting that the potential in the diode is given by
4
 x 3
V = Va   (5.18)
d
196

196 Thermionic Diodes

so that the velocity of an electron which starts from rest at the cathode is
2
2eVa  x  3
u=   . (5.19)
m0  d 
The time taken for this electron to reach the anode is
2
d 3
1 ⌠ d 3d
τ=    dx = , (5.20)
ua ⌡  x  ua
0

where ua = 2eVa m0 is the velocity of the electron when it reaches the anode.
The transit time calculated in (5.20) can be compared with the result for the space-​
charge free case where the acceleration is uniform and the transit time is τ 0 = 2d ua .
Equation (5.20) can also be written as
1 1
 d 3
K3
τ=   . (5.21)
3 2e m0  JC 

5.3 The Planar Diode Including the Effects of Thermal Velocities

The analysis in the previous section is non-​physical, strictly speaking, because no

electrons can leave the cathode when the electric field there is zero, and because the
solution implies that the charge density at the cathode surface is infinite. In practice
the electrons have thermal velocities as they leave the cathode so that they can move
away from it. Two approaches to this problem were discussed by Langmuir [6–​7].
The first method assumes that all the electrons have the same initial velocity
when they leave the cathode. This enables them to move towards the anode in the
presence of an opposing electric field. There must, therefore, be a plane between
the cathode and the anode at which the gradient of the potential is zero as shown
in Figure 5.3. At this plane the electrons come to rest, a virtual cathode exists, and
the charge density is infinite. Thus this model suffers from the same objections as
before. It is assumed that the current density at the cathode is equal to the satura-
tion (Schottky) current density at a given temperature. Then some of the electrons
must be reflected by the virtual cathode, and some pass it to reach the anode. This
analysis can be criticised on the ground that there is no reason why electrons, which
have the same initial velocities, should behave in different ways in the steady state.
There is also the difficulty of knowing how to define the initial velocity in terms of
the cathode temperature [6]. This approach is therefore not valid when the initial
velocities are thermal. It is, however, the basis of the discussion of space-​charge
limitation of the current when a mono-​energetic beam of electrons is injected into
a diode as discussed in Section 5.8.
The second method assumes that the electrons emitted from the cathode have
a Maxwell–​Boltzmann energy spectrum. Consider a source of electrons at zero
197

The Planar Diode Including Thermal Velocities 197

Figure 5.3:  Potential variation in a planar space-​charge-​limited diode with non-​zero emission

velocities.

potential with temperature T which has saturation current density JS. The current
density collected on an electrode at a potential –​V with respect to the source is

 V 
J = JS exp  −  , (5.22)
 VT 
where VT = kT e is the volt equivalent of temperature [6]. The current density emit-
ted from the cathode in the energy range (V , V + dV ) is

JS  V 
dJ = exp  −  ⋅ dV . (5.23)
VT  VT 
When the current is space-​charge-​limited there is a virtual cathode between the
cathode and the anode with potential is –​Vm relative to the cathode. The current
collected by the anode (Ja) comprises those electrons which have sufficient energy
to pass the virtual cathode. Thus

 V 
J a = JS exp  − m  . (5.24)
 VT 
The electron flow in the regions on either side of the virtual cathode must be con-
sidered separately.

5.3.1 Electron Flow between the Potential Minimum and the Anode

Consider electrons launched with initial energy Vi at the potential minimum. At a
plane where the potential relative to the minimum is V they have velocity

ui = 2 ( e m0 ) (V + Vi ). (5.25)
198

198 Thermionic Diodes

From (5.23) the current density of these electrons is

Ja  V 
dJi = exp  − i  ⋅ dVi (5.26)
VT  VT 
so that the contribution to the charge density at this plane is, from (5.11)

dJi Ja  V 
d ρi = = exp  − i  ⋅ dVi . (5.27)
vi VT 2 ( e m0 ) (V + Vi )  VT 

Integrating this expression over all possible values of Vi the charge density at a
plane where the potential relative to the minimum is V is given by
∞
Ja ⌠ 1  V 
ρ (V ) = ⋅ exp  − i  ⋅ dVi . (5.28)
VT 
⌡0 2 ( e m0 ) (V + Vi )  VT 

Equation (5.28) can be written

∞
Ja ⌠ 1
ρ ( υ) =  ⋅ exp ( − υi ) ⋅ d υi , (5.29)
2 ( e m0 )VT ⌡0 υ + υi

where υ = V VT and υi = Vi VT . Substituting for ρ in Poisson’s equation (5.12) gives

∞
d2υ ⌠ 1
= ⋅ exp ( − υi ) ⋅ d υi , (5.30)
dX 2 ⌡0 υ + υi
where the normalised position X is defined by

−Ja
X= ⋅ x. (5.31)
ε 0 2 ( e m0 )VT3

For convenience let

y2 = υ + υi . (5.32)

Then (5.30) may be written

d υ2 ∞
= 2 ⋅ exp ( υ) ⋅ ∫ exp ( − y2 ) ⋅ dy. (5.33)
dX 2 υ

Now the probability function, defined as

2
exp ( −t 2 ) ⋅ dt (5.34)
x
P (x) = ∫
π 0

tends to unity as x → ∞ so that (5.33) may be written

d2υ
dX 2
= π ⋅ 1− P ( ( υ )) ⋅ exp ( υ). (5.35)
19

The Planar Diode Including Thermal Velocities 199

Equation (5.35) can be integrated by multiplying both sides by 2(dU/dX ) to give

2

 dυ 
  = 2 π ∫ 1 − P
dX 
( υ ) ⋅ exp ( υ) ⋅ d υ. (5.36)
Integrating by parts and noting that ( d υ dX ) = 0 when υ = 0 we obtain
2

 dυ 

dX

 = 2 π ⋅ exp ( υ) − 1 − exp ( υ) P

( υ) + 2
π

υ  . (5.37)

This expression agrees with the term in brackets in equation (11) in [6], taking the
lower signs. The constant on the right-​hand side agrees with that in equation (6) in
[6] when the difference in the units is allowed for by setting ε 0 = 1 4 π (see equation
(4) in [6]). Langmuir’s variables are related to those used here by

ξ = 2 π ⋅ X (5.38)
and

η = υ. (5.39)

5.3.2 Electron Flow between the Cathode and the Potential Minimum
In the region between the cathode and the potential minimum we must consider
the flow in two parts. Those electrons with energies greater than or equal to Vm are
able to pass the minimum and reach the anode with a total current density J a . The
charge density due to these electrons at a plane where the potential is V relative to
the minimum is, from (5.29)
∞
Ja ⌠ 1
ρ1 ( υ) =  ⋅ exp ( − υi ) ⋅ d υi , (5.40)
2 ( e m0 )VT ⌡0 υ + υi

where υ = V VT .
Those electrons which start with energies less than Vm are reflected by the field
at the point where their initial energy is equal to the potential. The potential at any
point relative to the cathode is (V − Vm ) so that the velocity of an electron whose
energy at the cathode is Vi is

ui = 2 ( e m0 ) (Vi + V − Vm ). (5.41)
Thus at the virtual cathode, where V = 0, an electron whose initial energy is Vi = Vm
is just brought to rest. Because electrons whose initial energies are less than Vm are
reflected, they contribute twice to the charge density. The contribution to the charge
density at a plane is therefore

2dJi 2JS  V 
dρi = = exp  − i  ⋅ dVi . (5.42)
vi VT 2 ( e m0 ) (Vi + V − Vm )  VT 
20

200 Thermionic Diodes

At any plane this must be integrated to cover all electrons whose initial energy allows
them to reach that plane, but does not permit them to pass the minimum. Thus
Vm
2JS ⌠ 1  V 
ρ2 (V ) =  exp  − i  ⋅ dVi . (5.43)
VT ⌡V −V 2 ( e m0 ) (Vi + V − Vm )  VT 
m

Substituting for JS from (5.24)

0
2J ⌠ 1  V′ 
ρ2 (V ) = a  exp  − i  ⋅ dVi ′, (5.44)
VT 
⌡−V 2 ( e m0 ) V + Vi ′ (
 VT  )
where Vi ′= Vi − Vm . Then (5.44) can be written
0
2Ja ⌠ 1
ρ2 ( υ) =  ⋅ exp ( − υi ) ⋅ d υi , (5.45)
2 ( e m0 )VT ⌡− υ υ + υi

where the prime on the variable of integration can be omitted without loss of gen-
erality. The total charge density at a plane where the normalised potential with
respect to the minimum is υ is then the sum of ρ1 and ρ2. Substituting for the charge
density from (5.40) and (5.45) in Poisson’s equation (5.13) and making use of (5.31)
we obtain
∞ 0
d2υ ⌠ 1 ⌠ 1
= ⋅ exp ( − υi ) ⋅ d υi + 2 ⋅  ⋅ exp ( − υi ) ⋅ d υi . (5.46)
dX 2 ⌡0 υ + υi ⌡− υ υ + υi
Equation (5.46) can be expressed in terms of the probability function as

d 2υ
dX 2
= π ⋅ exp ( υ) 1 + P
 ( υ ) . (5.47)
As before this may be integrated to give

 dυ 

dX 

 = 2 π ⋅ exp ( υ) − 1 + exp ( υ) P

( υ) − 2
π

υ  (5.48)

that agrees with (11) in [6].
The formal solution to (5.37) and (5.48) is

υ
⌠
1
dυ
X = ( 4π )
−
4 ⋅ 1
, (5.49)

⌡0 exp ( υ) − 1 ± exp ( υ) P
 ( υ)  2 υ π

2

which is, essentially, (11) in [6]. In this equation the upper signs are taken between
the cathode and the minimum, and the lower signs between the minimum and the
anode. The distance between the cathode and the minimum is obtained by using the
201

The Planar Diode Including Thermal Velocities 201

upper signs and setting V = Vm. The distance between the minimum and the anode
is obtained by using the lower signs and setting V = Vm + Vak. For comparison equa-
tion (5.16) can be rewritten in the form

− JC x 2 4 32
3
= ⋅ υ (5.50)
9
ε 0 2 ( e m0 )VT2
or

2 34
X= ⋅ υ . (5.51)
3
Figure 5.4 shows the solutions of (5.49) and (5.51). The curves are identical to
those in Figure 42 in [7] apart from the change in the scale of the horizontal axis. In
this figure the cathode is to the left of the origin and the anode to the right.

5.3.3 Numerical Evaluation

It is not possible to evaluate the integral in (5.49) by analytical methods. Tables of
the solutions found using numerical methods are given in [6]. The presence of the
exponential functions can cause numerical overflows for large values of U. In the
region between the potential minimum and the anode (5.49) can be approximated
without loss of accuracy, when U > 50, by
υ
1 ⌠ dυ
X = ( 4π )
−
4 ⋅ . (5.52)
 1
⌡0 2 υ π − 1 2
 

Figure 5.4:  Dependence of normalised potential against normalised position in a space-​

charge-​limited diode with thermal initial velocities showing the Child-​Langmuir curve
(dashed) for comparison
(copyright 1931, The American Physical Society, reproduced, with permission, from [7]).
20

202 Thermionic Diodes

Between the cathode and the minimum X → 1.3565 when υ > 20. The solutions
shown in Figure 5.4 were obtained in this way (see Worksheet 5.1). Alternatively,
equations (5.30) and (5.46) can be integrated directly using the Runge-​ Kutta
method. Both methods give identical results.
To calculate the current-​voltage characteristic of a thermionic diode we note first
that the potential at the minimum is given by (5.22) as

J 
Vm = VT ln  a  . (5.53)
 JS 
For the purposes of illustration we will assume that the cathode temperature is 1300
°K, which is typical of the temperature of the cathodes used in microwave tubes, so
that VT = 0.112 V. When the ratio of the anode current to the saturation current is
10–​6 the potential at the minimum is about 1.5 V which is small compared with typ-
ical potential differences between the electrodes. The distance of the minimum from
the cathode can be computed using (5.49) for typical values of the parameters. The
results are shown in Figure 5.5 where the distance between the minimum and the
cathode is plotted against the normalised anode current. It can be seen that the dis-
tance is much less than 1 mm, except when the saturation current is small and when
the anode current is a small fraction of the saturation current. Hence we expect that
the effects of thermal velocities will be negligible except for small diode spacings, and
low potential differences. Thus in Figure 5.2 d1 is normally much smaller than d2.
Once d1 has been found then d2 is given by

d 2 = d − d1 (5.54)

and (5.49) can be solved for the potential at the anode relative to the minimum as
a function of the anode current. Finally the potential difference between the anode
and the cathode is found by subtracting the potential at the minimum relative to
the cathode.

Figure 5.5:  Distance between the potential minimum and the cathode, in a space-​charge-​
limited diode with thermal velocities, plotted against normalised current density.
203

The Planar Diode Including Thermal Velocities 203

As the anode current drops, the potential difference between the minimum and
the cathode increases and X tends to a constant, as has been noted. But, from
(5.31), we see that x ∝ − J a and therefore the distance between the cathode and
the minimum increases without limit. Thus, in any diode, the minimum will reach
the anode if the anode current is small enough. When the minimum just reaches
the anode the potential of the anode is negative with respect to the cathode and
equal to the potential at the minimum. For smaller anode currents d2 is negative
and the relationship between the anode current and the anode-​cathode voltage is
obtained from (5.22). The results of these calculations for a range of parameter
values are shown in Figure 5.6. The normalised current ( J a JS ) is plotted against
the anode-​cathode voltage, normalised to the voltage at the saturation current for
a diode without thermal velocities, given by (5.16). The graphs are plotted with the
saturation current as a parameter for four different diode spacings. In each case
the curve given by (5.16) is included for comparison. In microwave tubes the satur-
ation current is typically in the range 1.0 to 10 A cm−2 (see Chapter 18). It is clear

Figure 5.6:  Normalised current in a planar space-​charge limited diode with thermal

velocities plotted against the anode voltage, normalised to the voltage at the saturation
current in a diode without thermal velocities.
204

204 Thermionic Diodes

from Figure 5.6 that there is very little error in using (5.16) when the diode spacing
is greater than 1.0 mm. For example, when JS = 1.0 A cm 2 and d = 1 mm the max-
imum anode voltage is 264 V so thermal effects can be neglected if the anode volt-
age is much greater than this. In general, thermal effects are not important in many
power vacuum tubes. The exceptions are gridded tubes, and gridded electron guns,
where the spacing between the control grid and the cathode may be much less than
1 mm, and also vacuum micro-​electronic devices.

5.4 The Planar Diode Including the Effects of Relativity

In many electron tubes the potential difference between the anode and the cathode
is high enough for relativistic effects to be important. In that case (5.10) is replaced
by (1.4)
1
 1 2
u = c 1 −  , (5.55)
 (1 + V VR ) 
2

where VR = ( m0 c 2 e ) = 511kV is the rest energy of an electron in electron volts

[8–​10]. Following the same method as before we find that, for space-​charge limited
flow, (5.14) is replaced by
2
 dV  2J
  =− ⋅ 2VRV + V 2 , (5.56)
dx  ε0c
where the constant C is zero. It is convenient to introduce the normalised variables

V
U= (5.57)
VR
and

−J
X= ⋅ x. (5.58)
2 ε 0 cVR

Note that these are not the same as the normalised variables used in the previous
section. Equation (5.56) then becomes

2
 dU 
 = 4 2U + U , (5.59)
2

dX 
whose solution can be written formally as

1
−
1 U
X= ∫ ( 2U + U 2 ) ⋅ dU . (5.60)
4

2 0
205

The Planar Diode Including the Effects of Relativity 205

This equation cannot be integrated analytically. Acton [8] and Boers and Kelleher
[9] derived solutions as power series. An alternative solution expressed in terms of
elliptic integrals given in [10] was used by Zhang et al. [11] to derive an approximate
equation for the space-​charge limited current density in a relativistic planar diode

(γ )
23 32
2ε m c 2 0 −1
J = 0 20 ⋅ , (5.61)
ed ( )
3 − 1 γ 0−0.392 + 1

where γ 0 = (1 + eVa m0 c 2 ). The error in this equation is less than 1% for any voltage.
The series given in [9] can be written

2 1 2 1⋅ 3 3 n +1 1 ⋅ 3 ⋅ 5 ( 2 n − 5 ) ( 2 n − 3) 
1
X = (U 2 + 2U )  3 U + 21U + 21⋅11U +  ( −1) 21⋅11⋅15…( 4 n − 5) ( 4 n − 1) U 
− n
4
 
(5.62)
Alternatively (5.60) can be evaluated directly using numerical methods. When volt-
age is small the terms in U 2 and higher powers in (5.60) can be neglected, and the
integral evaluated analytically, to give

1 3
X= (2U ) 4 . (5.63)
3
It is easy to show that this equation is identical to (5.16). Figure 5.7 shows a com-
parison between curves calculated using (5.63), the numerical evaluation of (5.60),
and the first three terms of (5.62). It can be seen that, for a given U, the effect of
relativity is to reduce the value of X and, therefore, the current for a given diode
spacing. The series solution is a good approximation to the exact result with an
error of less than 1% in X when U < 1.3 if three terms are taken (see Worksheet 5.1).
For a fixed diode spacing the current is proportional to X 2 . The ratio of the rela-
tivistic current computed from (5.60) to the non-​relativistic current computed from

Figure 5.7:  Normalised potential in a planar space-​charge-​limited diode plotted against

normalised position both with, and without, relativistic effects and using the series
approximation in (5.62).
206

206 Thermionic Diodes

Figure 5.8:  Relativistic correction to the current in a planar space-​charge-​limited diode

plotted against normalised anode voltage.

(5.63) is shown in Figure 5.8. It can be seen that equation (5.63) over-​estimates the
current by around 9% when U = 1.

5.5 The Cylindrical Space-​Charge Limited Diode

The problem of space-​charge limited flow between coaxial cylinders when the ini-
tial electron velocity is zero is discussed in [12] for the non-​relativistic case, and in
[8] for the relativistic case. For symmetrical flow in a cylindrical diode Poisson’s
equation is

1 ∂  ∂V  ρ
r  = − . (5.64)
r ∂r  ∂r  ε0

The total current per unit length IL is independent of radius so the current density is

IL
J= . (5.65)
2 πr
Then, from equations (5.11), (5.55) and (5.64)

1
−
1 ∂  ∂V  IL  1  2
 r  = − 1 −  . (5.66)
r ∂r ∂r 2 πε 0 cr  (1 + V VR ) 
2

Let R = r rc , where rc is the radius of the cathode, and U = V VR so that (5.66) can
be written

d 2U 1 dU k1  U + 1 
+ = , (5.67)
dR 2 R dR R  U 2 + 2U 
207

The Cylindrical Space-Charge Limited Diode 207

where

− I L rc
k1 = . (5.68)
2 πε 0 cVR
Equation (5.67) can be integrated by numerical methods with the initial conditions
U = 0 and dU dR = 0 when R = 1.
A power series solution for the non-​relativistic case can be derived by setting

γ = ln R (5.69)
and assuming that the current obeys the 3/​2 power law so that (in SI units)

3
8πε 0 2 ( e m0 ) V 2
IL = ⋅ 2 . (5.70)
9 (rβ )
Where V is the potential at radius r relative to the cathode, and β is a function of R
[12]. Two series expansions for β have been found:
∞
β = ∑ An γ n (5.71)
n=0

and

 γ ∞
β = exp  −  ⋅ ∑ Bn γ n , (5.72)
 2  n=0
where the coefficients are given in Table 5.2. The second series generally converges
faster than the first except when R < 0.05 when γ is negative and the exponential
function in (5.72) becomes very large.
Acton [8] obtained the following power series solution for the non-​relativistic case:
4
U = a γ 3 (1 + 0.1333γ + 0.02444 γ 2 + 0.0039236 γ 3 + 0.00052966 γ 4 + ) , (5.73)

Table 5.2:  Coefficients for equations (5.71) and (5.72)

n An Bn

0 0.0 0.0
1 +1.0 +1.0
2 −0.40 +0.10
3 +0.916667 +0.016667
4 −0.014242 +0.002424
5 +0.001679 +0.000266
6 −0.000161 +0.000026
208

208 Thermionic Diodes

where
2
81  I L r0  81 2
a3 = = k1 . (5.74)
32  2 πε 0 cVR  32
Equation (5.70) can be written in the same notation as
1
U = a (R 2 β 4 )3 . (5.75)
It can be shown that (5.73) and (5.75) give identical results. Both equations are writ-
ten in a dimensionless form similar to (5.6). We note that, when R is constant, U is
proportional to a and, therefore, the current is proportional to the 3/​2 power of the
voltage. Langmuir showed that the solution in series was identical to that obtained
by direct integration of (5.67).
For the relativistic case Acton [8] obtained the power series solution
4
 2 11 2  8
 1 6 
U = a γ 3 1 + γ + γ +  + a 2 γ 3  + γ +  , (5.76)
 15 450   14 175 
which is valid when U < 2. The first series on the right-​hand side of (5.76) is identi-
cal to the series in (5.73) so it follows that the effects of relativity are represented
by the second series on the right-​hand side. It would appear from (5.76) that, if the
difference between the relativistic and non-​relativistic values of U is divided by a 2,
the result should be a function of R only. Calculations show that this is only true
for a < 0.1 and we conclude that (5.76) is an approximation to a double power series
expansion of U in terms of a and γ . Zhang et al. [11] derived an approximate expres-
sion for this case in the same manner as that for the planar diode

(γ )
23 32
2 ε 0 m0 c 2 0 −1
Jc = ⋅ , (5.77)
2 ( rc − ra )
2
( )
3 − 1 γ 0−0.392 + 1

where γ 0 = (1 + eVa m0 c 2 ), Jc is the current density on the surface of the cathode

and rc and ra are the radii of the cathode and the anode respectively. The relativistic
term is identical to that in (5.61) and the error is less than 5% for all voltages and
less than 1% for mildly relativistic voltages. The accuracy is best when the ratio of
the anode radius to the cathode radius is close to unity.
When the results from (5.76) are compared with those obtained by direct integra-
tion it is found that the agreement is good when a ≤ 1 and R > 1(see Worksheet 5.2).
When R < 1 the agreement is not so good. For example, when R = 0.5 and a = 1 the
value of U given by (5.76) is about 5% greater than that obtained by direct integra-
tion. It appears that the series in (5.76) do not contain enough terms to give good
accuracy. This equation also shows that U is not proportional to a in the relativistic
case and that, therefore, the current does not obey the 3/​2 power law when relativ-
istic effects are important.
Figure 5.9 shows, as an example, comparisons between the relativistic and non-​
relativistic solutions when a = 1 and the anode is outside, and inside, the cathode.
It can be seen that, for a given value of R, the effect of relativity is to increase U.
209

The Spherical Space-Charge Limited Diode 209

Figure 5.9:  Comparisons between the relativistic and non-​relativistic solutions for a

cylindrical space-​charge-​limited diode when a = 1 and the anode is: (a) outside, and
(b) inside, the cathode.

Thus, for a given geometry and applied voltage, the current is reduced by relativistic
effects. It is found that U decreases when a decreases so that relativistic effects only
become important for more extreme values of R.

5.6 The Spherical Space-​Charge Limited Diode

The problem of space-​charge limited flow between concentric spheres when the initial
electron velocity is zero is discussed in [13] for the non-​relativistic case and in [8] for
the relativistic case. For symmetrical flow in a spherical diode Poisson’s equation is

1 ∂  2 ∂V  ρ
 r  = − . (5.78)
r ∂r
2
∂r ε0
The total current I is independent of the radius so the current density is

I
J= . (5.79)
4 πr 2
Then, from equations (5.11), (5.55) and (5.78)
1
−
1 ∂  2 ∂V  I  1  2
 r  = − 1 −  . (5.80)
r 2 ∂r  ∂r  4 πε 0 cr 2  (1 + V VR ) 
2

Again we set R = r rc , where rc is the radius of the cathode, and U = V VR so that

(5.80) can be written

d 2U 2 dU k2 1+U
+ ⋅ = ⋅ , (5.81)
dR 2 R dR R 2 2U + U 2
210

210 Thermionic Diodes

where

−I
k2 = . (5.82)
4 πε 0 cVR
Equation (5.81) can be integrated by numerical methods with the initial conditions
U = 0 and dU dR = 0 when R = 1.
A power series solution for the non-​relativistic case can be found by setting

γ = ln R (5.83)
and assuming that the current obeys the 3/​2 power law so that (in SI units)
3
16 πε 0 2 ( e m0 ) V 2
I= ⋅ 2 , (5.84)
9 α
where α is a function of R [13]. The following series expansion for α was found:

α = γ − 0.3γ 2 + 0.075 γ 3 − 0.0143182 γ 4 + 0.0021609γ 5 − 0.00026791γ 6 +  (5.85)

Equation (5.84) can be written like (5.75) as

4
U = a α 3 , (5.86)

where

81 k22
a3 = . (5.87)
32
Comparison between (5.86) and the results of direct integration shows excellent
agreement when 0.2 < R < 5 (see Worksheet 5.3). For values of R outside this range
Langmuir used direct integration.
Acton [8] obtained power series solutions in a similar manner. For the
non-​relativistic case
4
 2 3 
U = a γ 3 1 − γ + γ 2 +  . (5.88)
 5 25 
And, for the relativistic case when U < 2
4
 2 417  8
 1 5 
U = a γ 3 1 − γ + γ 2 +  + a 2 γ 3  − + γ +  . (5.89)
 5 25 × 144   14 7 × 72 
Here the first term on the right-​hand side is similar, but not identical, to that in
equation (5.88) so it appears that it is not permissible to regard the second term
as a relativistic correction. These series do not contain enough terms to give good
accuracy outside the range 0.5 < R < 2.
Figure 5.10 shows, as an example, comparisons between the relativistic and non-​
relativistic solutions when a = 1 with the anode outside, and inside, the cathode. As
21

Transit-Time Effects in a Planar Diode 211

Figure 5.10:  Comparisons between the relativistic and non-​relativistic solutions for a

spherical space-​charge-​limited diode when a = 1 and the anode is: (a) outside, and
(b) inside, the cathode.

before, the effect of relativity is to increase U. Thus, for a given geometry and
applied voltage, the current is reduced by relativistic effects. It is found that U
decreases when a decreases so that relativistic effects only become important for
more extreme values of R.

5.7 Transit-​Time Effects in a Planar Diode

The analyses of the properties of space-​charge-​limited diodes in the preceding sec-

tions have assumed that the diodes are in the steady state. When the voltage across
a diode varies with time this will no longer be the case if the transit time of an
electron is comparable with the time during which the voltage changes. The high-​
frequency properties of a space-​charge limited diode, when a small high-​frequency
voltage is superimposed upon a large DC voltage, are described by the Llewellyn-​
Peterson equations [14]. This can be important for low-​power diodes but these are
now of limited importance. High power diodes, however, are generally operated in
the large-​signal regime and it is usual to employ their static characteristics in design
calculations. It is therefore important to know the conditions under which the static
characteristics are a valid approximation to the dynamic behaviour.
It can be shown [15] that, when an electron current is flowing in a region with a
number of electrodes,
n
⌠⌠⌠  ∂E 
∑V i k k =   ρv + ε 0
⌡⌡⌡   ⋅ E dv, (5.90)
∂t 
k =1

where Vk is the potential of the kth electrode, ik is the current flowing to it from the
external circuit and the integral is taken over the whole volume occupied by the field
21

212 Thermionic Diodes

of the electrodes. This equation can be understood as a statement of the principle

of conservation of energy. The left-​hand side represents the instantaneous power
flowing into the region and the two terms in the integral represent, respectively, the
rate of change of the kinetic energy of the electrons and of the potential energy
stored in the electric field. When this equation is applied to a planar diode that has
one electrode earthed then

⌠ ∂E 
Vi = A  ρv + ε 0  E dx, (5.91)
⌡ ∂t 
where A is the area of each electrode. This equation shows that, under dynamic con-
ditions, the current flowing to an electrode is not necessarily equal to the electron
current landing on it. In particular, an induced current can start to flow before any
electrons reach the anode. The term in the brackets in (5.91) is the total current whose
divergence is zero. Thus the flux of the total current through any transverse plane
is constant and depends only on time. To find the relationship between the external
current and voltage under large-​signal conditions it is necessary to start from the
equations of motion of the electrons and apply the appropriate boundary conditions.
The large-​signal high-​frequency characteristics of vacuum tubes were studied by
Wang [16] who investigated the behaviour of a diode when subjected to a short,
parabolic, voltage pulse of amplitude Vpk and duration 2t0 given by

t  t
V (t ) =
t0  2 − t  Vpk (0 ≤ t ≤ 2t ). (5.92)
0
0

This pulse approximates to a single cycle of a tube in class B or class C operation.

When t ≤ 0 there are no electrons between the electrodes. When t > 0 electrons are
emitted from the surface of the cathode at a rate sufficient to ensure that the electric
field is zero there. When t > t0 the voltage starts to drop and, at some point the emis-
sion ceases because the field of the electrodes is no longer sufficient to overcome
the retarding field due to space-​charge. The analysis assumes that the electrodes are
close enough together for the propagation delay of the electromagnetic field to be
negligible. Figure 5.11 shows plots of the positions of electrons as a function of
time. Each curve is specified by the time τ at which the electron is emitted. A num-
ber of times which are of interest are marked:
t1 = the time at which an electron emitted at t = 0 reaches the anode;
t2 = the time at which the cathode stops emitting electrons;
t3 = the time at which the last electron reaches the anode;
t4 = the time at which the last electron leaves the space between the electrodes.
Four cases are shown in Figure 5.14:
(a) The transit time is negligible compared with the duration of the voltage pulse
and the behaviour of the diode is quasi-​static.
(b) The transit time is an appreciable fraction of the pulse duration, not all of
the electrons are able to reach the anode, and the current pulse is appreciably
213

Transit-Time Effects in a Planar Diode 213

Figure 5.11.  Electron trajectories in a diode when the applied voltage is given by (5.92)
(copyright 1941, IEEE, reproduced, with permission, from [16]).

shorter than the voltage pulse. Electrons remain in the space between the elec-
trodes for some time after the cathode stops emitting.
(c) The transit time of the electron emitted at t = 0 is a substantial fraction of the
pulse duration and the cathode stops emitting electrons before the first electron
reaches the anode.
( d) The reversal of the voltage is so fast that no electrons are able to reach the anode.
For our purposes it is case (a) which is of interest. When the transit time is small
the diode is completely filled with electrons throughout the pulse. The transit time
is greatest at the start and end of the pulse when the voltage is small. Wang pro-
vides a detailed analysis of this case and shows that, when second-order terms are
included, the convection current at the anode is given by
3
4ε 2 e V (t ) 2 2 ε 0 dV (t )
J (t ) = 0 ⋅ 2 − ⋅ . (5.93)
9 m0 d 5d dt
The first term on the right hand side of this equation is identical to (5.16) and the
second term is 40% of the reactive current flowing in the capacitance of the diode,
but with the sign reversed. To investigate the point at which transit time effects first
become important we consider a diode subjected to a sinusoidal voltage given by

V (t ) = Va sin ( ωt ) . (5.94)
214

214 Thermionic Diodes

The first term in (5.93) is non-​zero only during the positive half-​cycle. This wave-
form can be expanded as a Fourier series. If only the first harmonic term is retained
then substitution in (5.93) gives the approximation

4 ε 0Va u0  9 
J (t )   0.458 sin ( ωt ) − βe d ⋅ cos ( ωt )
9d 2
 10 

ε 0Va u0
 sin ( ωt ) − 2βe d ⋅ cos ( ωt ) , (5.95)
5d2 

where βe d = ωd u0 and u0 = 2eVa m0 . The transit time effects are represented

by the second term in the brackets whose magnitude depends on βe d . This factor
increases with frequency, and with the diode spacing, and decreases as the peak
voltage applied to the diode is increased. Transit time effects therefore increase
the magnitude of the current by a factor of approximately 1 + ( 2βe d )2 and retard
its phase by 2βe d . The condition that transit time effects should be negligible is
therefore βe d  1 2. This can be used as a guide to the spacings required between
electrodes if transit time effects are to be avoided. It is important to note that
(5.95) can only be used to calculate the convection current when transit time effects
are small.

5.8 Injection of Electrons into a Planar Diode

We saw in Section 5.3 that it is not possible to use a single initial velocity to model
the emission from a thermionic cathode. However, there are other cases where this
problem is of interest. Consider, for example, a plane-​parallel diode in which the
electrodes are conducting grids that are perfectly transparent to the electrons. Let a
stream of electrons having uniform current density Ji and velocity u0 normal to the
grids be injected into the space between them as shown in Figure 5.12. This prob-
lem has been studied by many authors (see [17] and the references cited therein).
Here we follow the treatment in [18].
If relativistic effects are neglected, the initial velocity of the electrons is given
by (5.10) as

u0 = 2 ( e m0 )Vk , (5.96)

where the source of electrons is at potential −Vk relative to the electrode at x = 0,

then the velocity at a plane where the potential is V is

u = u0 1 + V Vk . (5.97)

Making use of (5.11) and (5.12) we obtain

d 2V Ji
=− . (5.98)
dx 2
ε 0 u0 1 + V Vk
215

Injection of Electrons into a Planar Diode 215

Figure 5.12:  Injection of electrons into a diode formed by parallel conducting grids.

Multiplication by 2 ( dV dx ) as before and integrating yields

2
 dV  4 Ji Vs
  =− ⋅ 1 + V Vk + C1 , (5.99)
dx  ε 0 u0
where C1 is a constant. It is convenient to introduce normalised variables  U = V Vk  and

−4 Ji
X= ⋅ x. (5.100)
ε 0 u0Vk
Substitution in (5.99) gives
1

dU
dX
=± ( 1+U + α1 ) 2
, (5.101)

where α1 is a dimensionless constant. Setting W = 1 + U and integrating a second

time gives
4 1
(W − 2α1 ) (W + α1 )2 = ± X + α 2 , (5.102)
3
where α 2 is a second constant of integration. We note that (5.101) implies that the
gradient of the potential may be either positive or negative. If the gradient is posi-
tive then the potential will increase monotonically through the diode whereas, if
the gradient is initially negative there will be a potential minimum followed by a
monotonic increase.
a) Monotonic increase (Branch I)
Let X = X a correspond to x = d and let

Wa = 1 + U a = 1 + Va Vk (5.103)
so that W = 1 when x = 0. Substituting these boundary conditions in (5.102) and
taking the plus sign gives
216

216 Thermionic Diodes

4 1
(1 − 2α1 ) (1 + α1 )2 = α 2 (5.104)
3
and

4 1
(Wa − 2α1 ) (Wa + α1 )2 = X a + α 2 . (5.105)
3
The constants α1 and α 2 can then be determined and, hence, the potential distribu-
tion from (5.102) (taking the positive sign).
b) Negative potential minimum (Branch II)
Let the position of the minimum be X = X m. In the range 0 ≤ X ≤ X m

4 1
(W − 2α1 ) (W + α1 )2 = −X + α3 (5.106)
3
and, in the range X ≥ X m

4 1
(W − 2α1 ) (W + α1 )2 = X + α 4 , (5.107)
3
where α 3 and α 4 are new constants of integration. Since W is continuous we find,
from (5.101), that W + α1 = 0 when X = X m so that

α3 = X m and α 4 = −X m . (5.108)

Substituting for α 3 in (5.106) and setting W = 1 when X = 0 gives

4 1
(1 − 2α1 ) (1 + α1 )2 = X m (5.109)
3
and setting W = Wa when X = X a equation (5.107) becomes

4 1
(Wa − 2α1 ) (Wa + α1 )2 = X a − X m . (5.110)
3
The properties of a diode with injected current can be determined from equations
(5.104), (5.105), (5.109) and (5.110) (see Worksheet 5.4). We note that α1 ≥ −1 for
both branches because α 2 and Xm are real. Equation (5.101) requires α1 ≤ 0 for
branch II because otherwise it is not possible for the gradient of the potential to
be zero. The solution has two branches corresponding to the two cases described
above. It is convenient to plot graphs of Xa against α1 with Ua as a parameter, as
shown in Figure 5.13. Branch I solutions are plotted as solid curves and branch II
solutions as dashed curves in this figure.
Solutions for given values of Xa and Ua are found from the intersections between
lines of constant Xa and Ua. It can be seen from the figure that the number of inter-
sections can be zero, one, or two. If the diode spacing is fixed and the diode voltage,
and the injection voltage, are held constant then Ua is constant. When the injected
current is gradually increased then Xa increases from zero. Initially there is only a
217

Injection of Electrons into a Planar Diode 217

Figure 5.13:  Dependence of the normalised diode spacing Xa on α1 with the normalised

potential difference Ua as a parameter
(copyright 1944, The American Physical Society, reproduced, with permission, from [18]).

single intersection with branch I for a large value of α1, and the potential in the
diode increases monotonically with x. As Xa increases a point is reached where two
solutions are possible: one in branch I, and the other in branch II. It can be shown
that the branch II solution has higher potential energy and is therefore unstable.
Eventually at α1 = −1 the space charge becomes sufficient to generate a potential
minimum and there are two solutions on branch II of which the one corresponding
to the more negative value of α1 is stable. Finally the two solutions coalesce at the
maximum of the curve of constant Ua, the current is space-​charge limited and can-
not increase any further. It is straightforward to show that at the maximum

Wa
α1 = − . (5.111)
1 + Wa
Substituting in (5.110) from (5.109) and (5.111) gives the maximum value of X a for
a given value of Wa

4 3
X max = (1 + Wa )2 . (5.112)
3
This equation can be expressed in terms of physical variables in the form

4 ε 0 2 ( e m0 ) 3 2
( ) , (5.113)
3
Ji = − ⋅ Va Vk Va + 1 + Vk Va
9 x2
which can be compared directly with the Child–​Langmuir Law given by (5.16). It
can be seen that the two equations become identical when Vk → 0. For non-​zero
values of the injection potential the last term in the equation is greater than unity
and the maximum current which can flow through the diode is therefore increased
as a result of the finite injection velocity. We note that the potential at the minimum
is always greater than that at the source of electrons, and that there is no plane
within the diode at which the electrons come to rest.
218

218 Thermionic Diodes

When the electrons are injected into the diode, the current is usually deter-
mined externally. It is therefore of interest to enquire what happens if we
attempt to inject a current greater than that given by (5.113). This problem was
studied by Langmuir [7]. The additional space charge depresses the potential
at the minimum so that a virtual cathode is formed at which V = −Vk and the
electrons come to rest. We have to assume that the electrons do not all have
the same injection velocity, but that there is a small velocity spread caused by
their emission from a thermionic cathode. As a consequence some electrons are
reflected by the virtual cathode while others pass onwards and are collected by
the anode. We saw in Section 3.2.1 that, for a small spread of velocities, the
behaviour of the diode is described by the Child–​Langmuir Law. It is therefore
justifiable to assume partial reflection of the electrons while ignoring the vel-
ocity spread. We assume that the injected current density is Ji , that the current
density at the anode is J a , and that the virtual cathode forms at x = xm . In the
region 0 ≤ x ≤ xm the effective current density is ( 2Ji − J a ) since the space-​charge
density is not affected by the direction of motion of the electrons. Then, from
(5.16) we have

4 ε 0 2 ( e m0 ) 3
2 Ji − J a = − ⋅ ⋅ (V + Vk ) 2 . (5.114)
9 ( xm − x ) 2

Equation (5.114) can be written

4 Ja 3
X = Xm − ⋅ ⋅W 2 , (5.115)
3 2 Ji − J a

where

−4 J a
X= ⋅ x (5.116)
ε 0 u0Vk
and

V
W = 1+ . (5.117)
Vk

When X = 0, W = 1 , and therefore

4 Ja
Xm = ⋅ . (5.118)
3 2 Ji − J a

Substituting in (5.115) gives for the region 0 ≤ x ≤ xm

4 Ja  3

X= ⋅ ⋅ 1 − W 2  . (5.119)
3 2 Ji − J a  
219

Diodes with Two-Dimensional Flow of Current 219

Similarly in the region xm ≤ x ≤ d

4 ε 0 2 ( e m0 ) 3
Ja = − ⋅ ⋅ (V + Vk ) 2 , (5.120)
9 ( x − xm ) 2

which becomes

4 3
4 Ja
X= ⋅W 2 + ⋅ . (5.121)
3 3 2 Ji − J a
If there is no reflection the anode current is equal to the injected current and, at
the anode,

4 
3
Xa =  1 + Wa  . (5.122)
2
3 
This is identical to the result obtained by setting α1 = 0 in (5.110). We note that this
is less than the space-​charge limited value of X given by (5.112). If the injected cur-
rent is increased, at constant electron velocity and anode voltage, then the last term
in (5.121) decreases and, therefore, so does the anode current. In particular, it is
observed experimentally that, if the injected current is increased slightly above the
space-​charge limit, then there is an abrupt reduction in the anode current [19–​20].
Thus the behaviour of such a diode is chaotic. This behaviour is exploited in virtual
cathode oscillators [21]. Further information about diodes with injected current is
to be found in [17] and [22].

5.9 Diodes with Two-​Dimensional Flow of Current

In many cases the electron flow in a diode can be regarded as one dimensional.
However, the emitting surface necessarily has a finite area, and it is usual to provide
focusing electrodes at the edges which are designed to compensate for edge effects
to produce near-​uniform current density on the cathode surface. This technique is
discussed in detail in Chapter 9.
If no focusing electrodes are used then the electron flow becomes two, or three,
dimensional according to the geometry. Problems of this nature have been studied
by a number of authors using Particle in Cell (PIC) codes. Two-​dimensional flow
between planar electrodes, when the electrons are emitted from a narrow strip on
the surface of one of them, was studied in [23] using two different simulations.
The results from both simulations showed good agreement. The current density
across the emitting surface was assumed to be uniform and increased until oscil-
latory behaviour was observed. This was taken to define the limiting current in
two-​dimensional flow. The current was limited by the formation of a virtual cath-
ode at the centre of the emitting strip. Simulations were carried out both with,
and without, a uniform magnetic field normal to the electrodes. It was found that
20

220 Thermionic Diodes

the limiting current was almost independent of the intensity of the magnetic field.
The results obtained were generalised as an empirical geometrical correction to the
Child–​Langmuir law (5.16) given by
2
J2D d d
= 0.3145   − 0.0004   , (5.123)
J1D  w  w
where d is the distance between the electrodes, w is the width of the emitting strip,
J 2 D is the limiting current density and J1D is the one-​dimensional Child–​Langmuir
current density given by (5.16). Equation (5.123) was found to be valid to an accur-
acy of 5% in the range 0.1 < w d < 8, and for magnetic fields in the range 0 to 100 T.
Further studies of this problem showed that, when the current density emitted
from the cathode was allowed to be non-​uniform, there was a sharp rise in the
current density at the edges of the strip as shown in Figure 5.14 [24]. This is to be
expected because the absence of space-​charge in the vacuum region immediately
outside the strip reduces the space-​charge potential depression that limits the cur-
rent. When the strip is wide compared with the gap, the current density is given by
the one-​dimensional Child–​Langmuir law over the greater part of the width of the
strip, only rising at the extreme edges. When w = 2d the current density is greater
than that given by the one-​dimensional law, except at the centre of the strip. For
narrower normalised strip widths the current density everywhere exceeds that given
by the one-​dimensional law.
When a step was introduced in the gap between the electrodes, opposite the cen-
tre of the emitting strip, it was found that the emitted current density closely fol-
lowed the square of the local vacuum electric field, except at the edges of the strip.
It was also found that a very slight change in the electric field at the edge of the
strip, such as that produced by a small non-​emitting focus electrode, could have a
significant effect on the emitted current density.

Figure 5.14:  Simulated normalized current density emitted at cathode versus position

(normalized to gap distance) for a variety of emission strip widths (w) and gap distances (d)
in cm; each trace is labeled with its corresponding (w d ) value
(copyright 2001, The American Physical Society, reproduced, with permission, from [24]).
21

Diodes with Two-Dimensional Flow of Current 221

The scaling of the equations governing space-​charge limited emission showed

that the ratio J 2 D J1D is a universal function of w d for zero emission velocity [23].
This universal function was derived in [25] where it was shown that, to first order,

J2D d d
≅ 1+ = 1 + 0.318 . (5.124)
J1D πw w
This expression is very close to the empirical expression in (5.123). The same
method was also used to find the geometrical correction when the electrons are
emitted from a circular patch on the cathode of radius R

J2D d
= 1+ . (5.125)
J1D 4R
It was stressed that neither (5.124) nor (5.125) give the correct limits as W and
R tend to zero. A  review of these results was given in [26]. The method in [25]
was extended in [27] to a number of other cases with plane parallel electrodes
and finite emitting patches of varying shapes including squares, rectangles, and
polygons.
The space-​charge limited current flow between concentric cylinders, when elec-
trons are emitted from a band on one of them, was studied in [28]. The simulations
ignored the self magnetic field of the electron flow and assumed that the current
density was uniform on the surface of the emitter. Both convergent flow (anode
inside the cathode) and divergent flow were considered. It was found that both cases
were described by the same empirical correction factor
2
I 2D  R  R
= 1 + 0.1536   + 0.0183   , (5.126)
I LB  L  L
where R is the radius of the outer electrode, L is the length of the emitting band
and I LB is the current calculated for a length L using (5.70). Equation (5.126) was
found to be valid to an accuracy of 2.5% for R r = 3 and 10 (where r is the radius
of the inner electrode) and 0.2 < L R < 5. The reason why (5.126) does not depend
upon R r is that the value of β in (5.70) rapidly approaches an asymptotic value
when R 3 > 3. When R r < 3 this is no longer the case and (5.126) ceases to be valid.
A theoretical expression for this case takes the form

I 2D 4 (R − r )
= 1+ F ( R r ) , (5.127)
I LB πL
where F ( R r ) is a function (plotted in [27]) that takes different values for conver-
gent and divergent flow. In the limit R r → 1, F → 1 4, which is approximately the
average of the values for convergent and divergent flow for R r > 1. It is shown that
(5.126) and (5.127) give similar results for R r = 3 and 10 when L R > 1 but that they
differ for smaller values of R r where (5.126) does not tend to the planar limit as it
should. Neither method is accurate for L R < 1 because the enhanced emission at
the edges of the emitting band has not been included.
2

222 Thermionic Diodes

References

[1]‌ I. Langmuir, ‘The effect of space charge and residual gases on thermionic currents in
high vacuum’, Physical Review, vol. 2, pp. 450–486, 1913.
[2]‌ S. E. Tsimring, Electron Beams and Microwave Vacuum Electronics. Hoboken, NJ: Wiley,
2007.
[3]‌ R. Vaughan, ‘A synthesis of the Longo and Eng cathode emission models’, IEEE
Transactions on Electron Devices, vol. 33, pp. 1925–​1927, 1986.
[4]‌ C. D. Child, ‘Discharge from hot CaO’, Physical Review (Series I), vol. 32, pp. 492–511,
1911.
[5]‌ R. J. Umstattd et  al., ‘A simple physical derivation of Child-​Langmuir space-​charge-​
limited emission using vacuum capacitance’, American Journal of Physics, vol. 73, pp.
160–​163, February 2005.
[6]‌ I. Langmuir, ‘The effect of space charge and initial velocities on the potential distribu-
tion and thermionic current between parallel plane electrodes’, Physical Review, vol. 21,
pp. 419–435, 1923.
[7]‌ I. Langmuir and K. T. Compton, ‘Electrical discharges in gases Part II: fundamental phe-
nomena in electrical discharges’, Reviews of Modern Physics, vol. 3, pp. 191–257, 1931.
[8]‌ E. W. V. Acton, ‘The space-​charge limited flow of charged particles in planar, cylindrical
and spherical diodes at relativistic velocities’, International Journal of Electronics, vol. 3,
pp. 203–​210, August 1957.
[9]‌ J. E. Boers and D. Kelleher, ‘Exact solution of Poisson’s equation for space-​charge-​
limited flow in a relativistic planar diode’, Journal of Applied Physics, vol. 40, pp.
2409–​2412, 1969.
[10] H. R. Jory and A. W. Trivelpiece, ‘Exact relativistic solution for the one-​dimensional
diode‘, Journal of Applied Physics, vol. 40, pp. 3924–​3926, 1969.
[11] Y. P. Zhang et  al., ‘Simple solutions for relativistic generalizations of the Child–​
Langmuir law and the Langmuir–​Blodgett law’, Physics of Plasmas, vol. 16, p. 044511,
April 2009.
[12] I. Langmuir and K. B. Blodgett, ‘Currents limited by space charge between coaxial
cylinders’, Physical Review, vol. 22, pp. 347–​356, 1923.
[13] I. Langmuir and K. B. Blodgett, ‘Currents limited by space charge between concentric
spheres’, Physical Review, vol. 24, pp. 49–​59, 1924.
[14] F. B. Llewellyn, ‘Vacuum tube electronics at ultra-​high frequencies’, Proceedings of the
Institute of Radio Engineers, vol. 21, pp. 1532–​1573, 1933.
[15] C. K. Jen, ‘On the induced current and energy balance in electronics’, Proceedings of the
I.R.E., vol. 29, pp. 345–​349, 1941.
[16] C. C. Wang, ‘Large-​signal high-​frequency electronics of thermionic vacuum tubes’,
Proc. IRE, pp. 200–​214, April 1941.
[17] P. V. Akimov et  al., ‘The true nature of space-​charge-​limited currents in electron
vacuum diodes:  a Lagrangian revision with corrections’, Physics of Plasmas, vol. 8,
pp. 3788–3798, 2001.
[18] G. Jaffe, ‘On the currents carried by electrons of uniform initial velocity’, Physical
Review, vol. 65, pp. 91–98, 1944.
[19] E. W. B. Gill, ‘XCIX. A space charge effect’, Philosophical Magazine Series 6, vol. 49,
pp. 993–​1005, 1925.
23

References 223

[20] V. I. Kuznetsov and E. A. Ya., ‘Time-​dependent regimes of a Bursian Diode I: Stability

of steady solutions’, Plasma Physics Reports, vol. 36, pp. 226–​235, 2010.
[21] J. Benford et al., High Power Microwaves. Boca Raton, FL: CRC Press, 2007.
[22] H. Kolinsky and H. Schamel, ‘Arbitrary potential drops between collector and emitter
in pure electron diodes’, Journal of Plasma Physics, vol. 57, pp. 403–​423, 1997.
[23] J. W. Luginsland et  al., ‘Two-​ dimensional Child–​ Langmuir law’, Physical Review
Letters, vol. 77, pp. 4688–4670, 1996.
[24] R. J. Umstattd and J. W. Luginsland, ‘Two-​dimensional space-​charge-​limited emis-
sion:  beam-​edge characteristics and applications’, Physical Review Letters, vol. 87,
p. 145002, 2001.
[25] Y. Y. Lau, ‘Simple theory for the two-​dimensional Child–​Langmuir law’, Physical
Review Letters, vol. 87, p. 278301, 2001.
[26] J. W. Luginsland et al., ‘Beyond the Child–​Langmuir law: a review of recent results on
multidimensional space-​charge-​limited flow’, Physics of Plasmas, vol. 9, pp. 2371–​2376,
May 2002.
[27] W. S. Koh et  al., ‘Three-​dimensional Child–​Langmuir law for uniform hot electron
emission’, Physics of Plasmas, vol. 12, p. 053107, 2005.
[28] K. G. Kostov and J. J. Barroso, ‘Space-​charge-​limited current in cylindrical diodes with
finite-​length emitter’, Physics of Plasmas, vol. 9, p. 1039, 2002.
24

6 Triodes and Tetrodes

6.1 Introduction

A triode is constructed by placing a conducting control grid between the cath-

ode and the anode of a vacuum diode, as shown schematically in Figure 6.1(a).
The current drawn from the cathode depends on the potentials of the grid and
the anode with respect to the cathode. The anode is known as the plate in the
USA. The grid potential can therefore be used to control the current flowing to the
anode. The electron flow between the cathode and the grid is space-​charge limited,
and it is found experimentally that the properties of many tubes can be described
by the equation
3
I a = K (Vg + Va µ ) 2 , (6.1)

where Ia is the anode current, Vg and Va are the potentials of the grid and the anode
with respect to the cathode, and K and µ are constants. In some tubes the power of
the term in the brackets is greater than 1.5 when the grid voltage is negative, and
it may be as high as 2.5 for reasons which are explained later. It is also found that
K and µ are not precisely constant but may vary somewhat with the conditions of
operation. The parameter K is the perveance by analogy with the perveance of a
diode and µ is known as the amplification factor from its role in the small-​signal
theory of triode amplifiers in which [1]

 ∂V 
µ =  a  . (6.2)
 ∂Vg  I
a

This parameter is typically in the range from 5 to 200 depending upon the pur-
pose for which the tube has been designed [2]. The penetration factor, defined by
D = 1 µ is more useful when the physics of the device are under consideration. The
symbol ( D ) comes from the German for penetration (Durchgriff), Other param-
eters used in small-​signal theory are the mutual conductance (or transconduct-
ance) defined by

 ∂I 
g m =  a  (6.3)
 ∂Vg V
a
25

Electrostatic Models of Triodes 225

Figure 6.1:  Schematic diagrams of (a) the triode, and (b) the tetrode.

and the dynamic anode resistance (or anode slope resistance) defined by

 ∂V 
ra =  a  . (6.4)
 ∂I a V
g

From these definitions it follows that

µ = g m ra . (6.5)
It must be emphasised that the values of these parameters depend on the quiescent
point chosen for operation of the tube. They are therefore of limited value in dealing
with the operation of triodes under large-​signal conditions. They are mentioned here
for completeness because their values are sometimes given in manufacturers’ catalogues.
Figure 6.1(b) shows a schematic diagram of a tetrode in which a second grid (the
screen grid) has been placed between the control grid and the anode. The screen
grid is maintained at a constant positive potential so that it screens the cathode
from the varying voltage of the anode. The characteristics of a tetrode are given
approximately by
3
 V V 2
I a = K Vg + s + a  , (6.6)
 µs µa 
where Vs is the potential of the screen grid with respect to the cathode. Typically
µ s ∼ 5 − 10 and µ a ∼ 100 − 200 . The basic theory of triodes and tetrodes is dis-
cussed in this chapter whilst practical devices are dealt with in Chapter 12.

6.2 Electrostatic Models of Triodes

The accurate calculation of the performance of triodes and tetrodes requires

detailed modelling of the electron trajectories under the influence of the electrode
26

226 Triodes and Tetrodes

potentials, space-​charge, and secondary electron emission. This can be achieved

with a particle in cell (PIC) code [3]. The objective, here, is to provide the reader
with simple models which provide insight into the properties of gridded tubes and
the means to make conceptual design calculations. This is approached through the
use of electrostatic models from which the large-​signal behaviour, including the
effects of space-​charge, can be inferred.
Figure 6.2 shows the arrangement of a planar triode whose geometry is defined by
the parameters d1, d2, r, a. The geometrical screening factor is defined by S = 2r a.
In the absence of space-​charge, the normal component of the electric field on the
surface of the cathode is linearly dependent on the potentials of the grid and the
anode with respect to the cathode. Thus we may write

Vg Va
E x ( 0, y ) = − − , (6.7)
d g ( y) da ( y)

where d g ( y ) and d a ( y ) are unknown functions of y which have the dimensions of

distance. Equation (6.7) can be rewritten as

Vg + D ( y )Va
E x ( 0, y ) = − , (6.8)
d g ( y)

where D ( y ) is the penetration factor which, in general, varies over the surface of the
cathode. Now consider the case where the grid is thin and biased to the potential
which would exist at that plane if the grid were removed. Thus

d1
Vg = Va . (6.9)
d1 + d2
Then, substituting in (6.8)

d1 + D ( y ) ( d1 + d 2 ) Va
E x ( 0, y ) = − ⋅ . (6.10)
d g ( y) d1 + d 2

Figure 6.2:  Arrangement of a planar triode.

27

Electrostatic Models of Triodes 227

But, in the absence of the grid, the electric field on the cathode surface is uniform and

Va
Ex (0) = . (6.11)
d1 + d2
Thus

d g ( y ) = d1 + D ( y ) ( d1 + d 2 ) (6.12)
and, substituting in (6.8),

Vg + D ( y )Va
E x ( 0, y ) = − . (6.13)
d1 + D ( y ) ( d1 + d 2 )
In many cases, as we shall see, the electric field is uniform over the surface of the
cathode and then (6.13) reduces to

Vg + DVa
Ex (0) = − , (6.14)
d1 + D ( d1 + d 2 )
where D is a constant. The same argument can be used to derive the corresponding
equations for cylindrical triodes. These are not required here, however, because the
distances between the electrodes in coaxial high power triodes are small compared
with their diameters so that the expressions for planar geometry may be used.
It is useful to investigate the relationship between the expressions derived, and
the inter-​electrode capacitances shown in Figure 6.3. It should be noted that these
capacitances are not normally the same as those measured at the external terminals
of the tube because of the capacitances of the connections between the internal
electrodes and the terminals. The charge per unit length on the section of the cath-
ode associated with one grid wire is given by

qc = − (CcgVg + CcaVa ) . (6.15)

If the charge density on the cathode is uniform then electric field normal to the
cathode is

qc 1
Ec =
aε 0
=−
aε 0
(CcgVg + CcaVa ) , (6.16)

Figure 6.3:  Inter-​electrode capacitances in a triode.

28

228 Triodes and Tetrodes

where a is the width of the strip of cathode between two adjacent grid wires.
Comparing (6.16) with (6.14) we see that

aε 0
Ccg = (6.17)
d1 + D ( d1 + d 2 )
and

Daε 0
Cca = = DCcg (6.18)
d1 + D ( d1 + d 2 )
so that

D = Cca Ccg . (6.19)

Equation (6.17) can be rewritten as

aε 0
Ccg = . (6.20)
  1 1 
d1 1 + d 2 D  +  
  d1 d 2  

This can be interpreted as the capacitance of a parallel plate capacitor whose spacing
is modified by the term in brackets which represents the difference between the grid
and a metal sheet. The modification should be independent of the position of the
grid, provided that the distances between it and the other electrodes are greater than
a. Thus the quantity d 2 D should depend only upon the dimensions of the grid. We
shall see later that this conclusion is confirmed by field theory. Similarly, the capaci-
tance between the grid and the anode is obtained by interchanging d1 and d 2 to give

aε 0 d1
C ga = = Ccg . (6.21)
  1 1  d2
d2 1 + d2 D  +  
  d1 d 2  

The conclusions drawn from this analysis are valid as long as the grid wires are
small compared with the other dimensions.

6.3 Penetration Factor in a Planar Triode

The potential distribution in a planar triode can be found by placing a line charge
at the centre of each grid wire. This problem was first solved by Maxwell using con-
formal mapping [4]. The solution is derived from that of a transformed problem in
which two line charges are placed within a conducting cylinder [5]. The potential of
a grid of equally spaced wires is given by

qg   2 πx 2π y  
V ( x, y ) = − ln 2  cosh − cos  + C , (6.22)
4 πε 0   a a  
29

Penetration Factor in a Planar Triode 229

where qg is the charge per unit length on each wire and C is a constant. If the grid is
located at x = d1 and a second grid carrying charge −qg is located at x = −d1 then the
potential on the plane x = 0 is zero and that plane can be taken to be the cathode.
Adding a linear variation of potential, for the effect of the anode, gives

1
V ( x, y ) =
aε 0
( fg ( x, y) qg + xqa ) , (6.23)
where

a  cosh 2 π ( x + d1 ) a − cos 2 π y a 
f g ( x, y ) = ln   (6.24)
4 π  cosh 2 π ( x − d1 ) a − cos 2 π y a 

and qa is the charge per unit length on a strip of the anode of width a. On the grid
Vg = V ( d1 , r ) so that

1
Vg =
aε 0
( fg (d1 , r ) qg + d1 qa ). (6.25)
The anode must be an equipotential surface and, therefore, when x = d1 + d 2 the
hyperbolic cosines must be much greater than unity. This condition is satisfied if
d 2 a ≥ 1 (when cosh ( 2 πd 2 a ) ≥ 268) and then, to a good approximation,

a  exp 2 π ( 2d1 + d2 ) a 
f g ( d1 + d 2 , y ) = ln   = d1 . (6.26)
4π  exp 2 πd 2 a 

Thus

1
Va =
aε 0
(d1qg + (d1 + d2 ) qa ). (6.27)
Equations (6.27) and (6.25) can be written as

Vg  1  f g ( d1 , r ) d1  qg 
V  = aε  d 
d1 + d 2   qa 
. (6.28)
 a 0  1

Inverting this equation we obtain

 qg  d1 + d 2 −d1  Vg 

 q  = C1  −d f g ( d1 , r ) Va 
. (6.29)
 a  1

where

aε 0
C1 = . (6.30)
(d1 + d2 ) fg (d1 , r ) − d12
230

230 Triodes and Tetrodes

6.3.1 A Triode with Uniform Electric Field on the Cathode

If d1 a ≥ 1 the potential is independent of y close to the cathode and, to a good
approximation,

a  exp ( 4 πd1 a ) 
f g ( d1 , r ) = ln  
4 π  2 − 2 cos ( 2 πr a ) 
(6.31)
a   πr  
= d1 − ln 2 sin   
2π   a 

and f g ( x, y ) → x as x → 0. The normal component of the electric field on the cath-

ode is given by

E x ( 0, y ) = −
(q g + qa )
=−
d 2Vg + ( f g ( d1 , r ) − d1 )Va
. (6.32)
aε 0 (d1 + d2 ) fg (d1 , r ) − d12
Comparing (6.32) with (6.14) we see that

d 2 D = f g ( d1 , r ) − d1 (6.33)

so, from (6.31),

a   πr  
D=− ln 2 sin    (6.34)
2 πd 2   a 

and d 2 D depends only on r and a as required by (6.20).

The validity of (6.34) has been confirmed experimentally when r a = 0.025,
d1 a = 0.97 and 0.86 < d 2 a < 8.8 [5]. Substituting in (6.30) we find that

aε 0
C1 = . (6.35)
d 2 ( d1 + D ( d1 + d 2 ))

The potential at any point within a planar triode may be calculated using
(6.23), (6.24) and (6.29). The potential contours for a typical geometry are
shown in Figure 6.4 for Va = 100V and a range of grid voltages (see Worksheet
6.1). The triode is close to cut off when Vg = −12 V. The electric field on the
surface of the cathode is independent of position in every case. The equipoten-
tial surfaces close to the grid wires are approximately circles centred on the
line charges representing the grid. Thus the model represents the potentials
correctly if r a is small. However, as the radius of the equipotential surface
increases, it is found that it is no longer centred on the line charge. A  num-
ber of authors have presented formulae for calculating the penetration factor
for larger screening factors. These formulae are based on the addition of line,
dipole, and multipole, charges to ensure that the surfaces of the grids wire are
 231

Figure 6.4:  Equipotential contours in a planar triode for various grid voltages and Va = 100V ( d1 a = 1, d 2 a = 1.5, r a = 0.05, S = 0.1)
(copyright 1948, McGraw-​Hill Education, reproduced, with permission, from [8]).
23

232 Triodes and Tetrodes

equipotential surfaces. Equation (6.34) is valid if r/a ≤ 0.05. For r/a ≤ 0.08 it
has been shown [6] that

  2 πr  
ln  coth 
  a  
D= . (6.36)
2 πd 2   2 πr  
− ln  cosh 
a   a  

We note that d 2 D is not now a function of r and a only because the diameters of
the grid wires are no longer small compared with the other dimensions. Further
formulae which are valid up to r a = 0.16 are given in [7, 8]. These expressions are
somewhat complicated and do not give additional insight into the properties of
triodes. In addition, they are of limited relevance to modern high power triodes in
which the grid ‘wires’ are usually rectangular in cross-​section.

6.3.2 A Triode with Non-​Uniform Electric Field on the Cathode

When d1 a < 1 the x component of the electric field is found by differentiating (6.23)

E z ( x, y ) = −
∂V
∂x
=−
1
aε 0
( )
f g ′ ( x, y ) qg + qa , (6.37)

where

1 sinh 2 π ( x + d1 ) a 1 sinh 2 π ( x − d1 ) a
f g ′ ( x, y ) = − . (6.38)
2 cosh 2 π ( x + d1 ) a − cos 2 π y a 2 cosh 2 π ( x − d1 ) a − cos 2 π y a
In the limit x → 0

sinh ( 2 πd1 a )
f g ′ ( 0, y ) = (6.39)
cosh ( 2 πd1 a ) − ( cos 2 π y a )
so that, on the surface of the cathode

1
E z ( 0, y ) = −
aε 0
( fg′ (0, y) qg + qa ). (6.40)
Substituting for qg and qa from (6.29)

C1 d1 + d 2 −d1  Vg 

E z ( 0, y ) = −  f g′ ( 0, y ) 1  . (6.41)
aε 0  −d1 f g ( d1 , r ) Va 

The electric field depends linearly on the grid and anode potentials and therefore,
following Fremlin, we define the electrostatic penetration factor by

 ∂E ∂E z 
Dg ( y ) =  z . (6.42)
 ∂Va ∂Vg  x =0
23

Penetration Factor in a Planar Triode 233

From (6.41)

∂E z C
= − 1 ( −d1 f g′ ( 0, y ) + f g ( d1r )) . (6.43)
∂Va aε 0
Similarly

∂E c C
= − 1 (( d1 + d 2 ) f g′ ( 0, y ) − d1 ) (6.44)
∂Vg aε 0

so that

f g ( d1 , r ) − d1 f g′ ( 0, y )
Dg ( y ) = , (6.45)
(d1 + d2 ) fg ′ (0, y) − d1
which is valid when r a ≤ 0.05. In the limit d1  a f g′ ( 0, y ) → 1 and Dg ( y ) → D as
expected. It can be shown, by substitution, that

Vg + Dg ( y )Va
E z ( 0, y ) = − , (6.46)
d1 + Dg ( y ) ( d1 + d 2 )

which was previously derived as (6.13).

Figure 6.5 shows how the normalised penetration factor (Dg ( y ) D ) varies with y
for a number of grid-​cathode spacings. This figure confirms that the emission from

Figure 6.5:  Variation of normalised penetration factor with position

( d 2 a = 5, r a = 1 14, Va = 100 V , Vg = −2 V, D = 0.026 ).
234

234 Triodes and Tetrodes

the cathode is uniform when d1 a ≥ 1 (see Worksheet 6.1). A different method for
calculating the potential when d1 a ≥ 0.4 using theta functions gives essentially the
same results [9].

6.3.3 Calculation of Penetration Factors Using Numerical Methods

We have seen that the analytical solutions for the penetration factor are restricted
to grids composed of circular wires whose diameter is small compared with their
spacing. For wires of other shapes and for general dimensions it is necessary to use
numerical methods. It is sufficient to calculate the potentials in two cases:
i) Vg = 1, Va = 0
ii) Vg = 0, Va = 1,
since all other cases can be found by superposition. The penetration factor is then
the ratio of the electric field at the surface of the cathode computed in (ii) to that
computed in (i).

6.4 Static Characteristics of Triodes

The accurate calculation of the static characteristic curves of a triode requires

the simultaneous solution of the equations of motion of the electrons, and
Poisson’s equation for the electrostatic potential in the presence of space charge.
This is a difficult task, and a number of authors have offered expressions for
the current based on equivalent diodes derived from the electrostatic solution.
When the electric field at the cathode is uniform and given by equation (6.14)
we can write

Ve
E x ( 0, y ) = , (6.47)
de
where the equivalent voltage is

Ve = A (Vg + DVa ) , (6.48)

and the spacing of the equivalent diode is

d e = A ( d1 + D ( d1 + d 2 )) , (6.49)

where A is a constant whose value is to be chosen so that the current dens-

ity in the triode is equal to the space-​charge-​limited current in the equivalent
diode. Thus
3
V2
J = K e2 (6.50)
de
235

Static Characteristics of Triodes 235

or
3

J=
K
⋅
(V g + DVa ) 2
. (6.51)
A (d
1 + D ( d1 + d 2 ))
2

To find an expression for A we consider two limiting cases:

i) The distance between the anode and the grid is increased ( d 2 a → ∞ ) keeping all
other parameters constant. Then
3

K Vg2
J→ ⋅ 2
(6.52)
A  a   πr   
 d1 − 2 π ln 2 sin  a  
 
from (6.34) since D → 0 as 1 d 2 . We note that the denominator in (6.52) is con-
stant and slightly greater then d1 reflecting the fact that there is some penetration
of the electric field through the array of grid wires. If we now reduce the spac-
ing between the grid wires, while keeping r a constant, then the current density
should tend to that of a diode in which there is a continuous electrode at the
plane of the grid. This requires  A → 1 as D → 0 .
ii) The normalised grid wire radius r a is small, so that the current intercepted by
the grid is negligible. Then, from (6.34), D → ∞ as r → 0 and

3
K Va2
J→ ⋅ . (6.53)
AD (d1 + d2 )2
This must be the current density in a diode with voltage Va and spacing ( d1 + d 2 )
which requires  AD → 1 as D → ∞.

These two limits are satisfied correctly if A = 1 (1 + D ). Then

Vg + DVa
Ve = (6.54)
1+ D
and

d1 + D ( d1 + d 2 ) Dd 2
de = = d1 + . (6.55)
1+ D 1+ D
It can be seen from (6.55) that this definition of the equivalent diode places the
equivalent anode at a plane whose distance from the grid is independent of d1 [10].
Then the current density in the triode is given by
3

J = K 1+ D ⋅
(V g + DVa ) 2
. (6.56)
(d
1 + D ( d1 + d 2 ))
2
236

236 Triodes and Tetrodes

This equation was proposed by Walker [11] but does not appear to have been gen-
erally adopted.
Two other definitions of the equivalent diode appear in the literature:
• The equivalent voltage at the plane of the grid is given by (6.14) [9, 10] so that the
current density is given by
3
K  Vg + DVa  2
J= 2  1 + D (1 + d d )  . (6.57)
d1  2 1 

• The numerator of (6.14) is taken to be the equivalent voltage, and the denomin-
ator to be the equivalent spacing [8, 9] with the result that

(V + DVa )
1.5
g
J=K . (6.58)
(d1 + D ( d1 + d 2 ))
2

Neither of these expressions has the correct behaviour at both the limits discussed
above. It has been argued that (6.57) should be modified so that the potential gradi-
ent close to the grid corresponds to that in a space-​charge-​limited diode [10, 12–​14]
to give
3

K  Vg + DVa 2
J= 2  . (6.59)
d1  1 + D (1 + ( 4d 2 ) (3d1 )) 

The adoption of this expression by industry suggests that it has been found useful
in tube design [15]. However, it fails to tend to the correct limit as D → ∞.
The equivalent diode approach has been criticised on the grounds that the
assumptions made are somewhat arbitrary [5]. As an alternative, it was proposed
that the perveance of the triode should be determined by setting Vg to the poten-
tial which would exist in the resulting space-​charge-​limited diode if the grid were
removed [5, 16] with the result that
3
 2
K  Vg + DVa
J= 2  . (6.60)
d1 
( 2 1 ) 
4
 1 + D 1 + d d 3

It appears, therefore, that (6.56) is the only equation that can be fully justified.
However (6.59) has been widely used, presumably because it was found to give use-
ful agreement with experimental results. It has been shown that (6.60) agrees with
experiment for a limited range of cases. Figure 6.6 shows a comparison between
the values of Jd12 K calculated from these three equations, normalised to the value
when D → 0. For values of D which are of practical importance there is good agree-
ment between (6.56) and (6.59). As a further test Figure 6.7 shows a comparison
between the experimental points from Figure 7 in [5], and a curve calculated using
237

Static Characteristics of Triodes 237

Figure 6.6:  Comparison between normalised values of perveance calculated from (6.56),

(6.59) and (6.60) when d 2 d1 = 10.

Figure 6.7:  Variation of current density with grid potential at an anode potential

of 200 volts. Experimental points from Figure 7 of [5] and curve calculated using
(6.56). d1 = 1.7 mm, d 2 = 5.03 mm, r = 0.0445 mm, a = 1.75 mm.

(6.56) (see Worksheet 6.2). There is excellent agreement between the theoretical and
experimental curves confirming that (6.56) is correct.
The effective diode voltage may be affected by the thermal velocities of the elec-
trons, and by the electrode contact potentials. Fremlin included a correction of
about 1.2 V for thermal velocities when comparing his calculated curve with exper-
imental results. In this case the effect of thermal velocities, calculated using the
methods described in Chapter 3, can be represented by adding 0.1 V to the equiva-
lent voltage. It therefore appears that the correction applied by Fremlin was too big.
The contact potential depends on the materials from which the electrodes are made
and upon surface contamination, and must therefore be determined experimentally.
The potential may be positive or negative but is usually less than 0.5 V [11]. Both
238

238 Triodes and Tetrodes

these effects are small compared with the voltages used in large power tubes, and
they may normally be neglected.

6.4.1 Grid Current

When the grid potential is positive with respect to the cathode then some of the
primary electrons are collected by the grid, so reducing the anode current. The situ-
ation is complicated by the possibility of secondary electron emission, and thermi-
onic emission, from the grid. Thus any calculations based on primary grid current
can only be approximate. However, as the energies of secondary and thermionic
electrons are small, the primary current is useful in calculating the power deposited
on the grid. This problem is discussed in [8, 11, 17].
If the effects of space-​charge are ignored, then the trajectories of non-​relativistic
electrons remain unchanged when the ratio of the anode voltage to the grid volt-
age is constant. Since the trajectories depend on the velocities of the electrons, we
expect that the ratio I g I a is a function only of Vg Va . It is found experimentally
in many cases that when Va ≥ Vg the ratio of the anode current to the grid current is
given to useful accuracy by

Ia V
= δ a , (6.61)
Ig Vg

where δ is the current division factor which is a constant of the tube [8]. The value
of δ can be found by considering the electron trajectories when Va = Vg. The trajec-
tory which just grazes the surface of the grid provides a boundary between those
electrons which are collected by the grid, and those which are collected by the
anode. The approximate primary current division factor is [17]

a
δ= − 1. (6.62)
a r  4ed1 
⋅ ln   + 2r
π (1 + D ) 2d1  r 
An alternative approach is described in [11, 12]. An electron which just grazes the
grid wire starts from the cathode at a distance x1 from the centre line of a grid wire.
The electron enters the deflecting field around the grid wire with energy given by an
equivalent grid voltage obtained from (6.14) as

Veg = −d1E x ( 0 ) , (6.63)

which can be expressed in terms of Vg Va for a given anode voltage. The angular
momentum of the electron is conserved as it moves through the deflecting field
so that

Vg
x1 = r . (6.64)
Veg
239

Static Characteristics of Triodes 239

The ratio of the anode current to the grid current is then

Ia a Veg
= − 1. (6.65)
I g 2r Vg

This equation assumes that the grid-​cathode spacing is large enough for the current
density on the cathode surface to be uniform. The effects of space charge, and ther-
mal velocities, are neglected. An improved expression, based on a more accurate
representation of the field between the grid and the cathode, is [11]

Ia a Veg  2Veg ln ( a 2 πr ) 
=   − 1. (6.66)
I g 2r Vg  2Veg ln ( a 2 πr ) + (Vg − Veg ) 

A third way of calculating the grid current uses numerical integration of the equa-
tions of motion of the electrons in the electrostatic field of the triode. The effects of
space charge are neglected. Figure 6.8 shows the trajectories corresponding to the
potential maps in Figure 6.4. The grid current can be computed by finding the tra-
jectory which just touches the edge of the grid (see Worksheet 6.1). Figure 6.9 shows
a plot of I g I a against Vg Va obtained in this way with curves given by (6.61), (6.65)
and (6.66) for comparison. The computed grid current is zero for small positive grid
voltages because the electrons are steered away from the grid by the field. It can be
seen that (6.61) only gives a very rough guide to the magnitude of the grid current
and that the assumption of a constant current division factor is not justified. The
best agreement is given by (6.66).

Figure 6.8:  Electron trajectories without space charge corresponding to the potential

distributions shown in Figure 6.4 (d1 a = 1, d 2 a = 1.5, r a = 0.05, Va = 100 V ) .
240

240 Triodes and Tetrodes

Figure 6.9:  Dependence of grid current fraction on the normalised grid

voltage (d1 a = 1, d 2 a = 1.5, r a = 0.05, Va = 100 V ).

6.4.2 Triodes with Island Formation

We saw in Section 6.3.2 that, when r a < 1, the electric field at the surface of the
cathode depends on the transverse position in the absence of electrons. In particu-
lar, as the grid is made more negative with respect to the cathode, the polarity of
the electric field is reversed on the section of the cathode immediately opposite the
grid. In that case only the part of the cathode on which the vacuum electric field is
accelerating can emit electrons, the current density is non-​uniform and parts of the
cathode may cease to contribute to the current. This condition is known as island
formation [5, 9, 18, 19].
The current in a triode, when there is island formation, can be calculated by sum-
ming the currents contributed by elementary strips of the cathode. From (6.46) and
(6.56) the average current density is given by
a 3

K
⌠
J =  1 + Dg ( y ) ⋅
(Vg + Dg ( y)Va )2 ⋅ dy. (6.67)
a (d1 + Dg ( y) (d1 + d2 ))
2

⌡0
where the integrand is zero if the effective voltage in the numerator is negative. The
current is cut off when the effective voltage is less than or equal to zero for all values
of y. It can be seen from Figure  6.5 that the penetration factor is greatest when
y = a 2, thus the grid voltage at cut-​off is given by

Vg 0 = −Dg ( a 2 )Va . (6.68)

The magnitude of this voltage increases as d1 is reduced. As the grid voltage

is increased a point is reached at which electrons are emitted from the whole
cathode surface. For higher grid voltages the effects of non-​uniform cathode
241

Static Characteristics of Triodes 241

emission tend to average out and it may be expected that the current will obey
the 3/​2 power law. This can be revealed by plotting J 2 3 against the grid voltage,
as shown in Figure  6.10, for a range of values of d1 (see Worksheet 6.1). The
effects of island formation are shown in the curvature when the grid voltage is
negative. This curvature is not present when d1 = a and is barely perceptible, for
the parameters illustrated, when d1 = 0.5a . For smaller spacings, the grid volt-
age required to reduce the current to zero is increasingly negative and the rate
of change of current with grid voltage is reduced. Some authors have suggested
that, when there is island formation, the behaviour of the triode can still be rep-
resented by (6.56) if the power law is increased [9, 16]. This model is only appro-
priate when the grid voltage is negative because it does not represent the 3/​2
power law behaviour when the grid voltage is positive. It is therefore preferable to
regard the 3/​2 power law as the norm and consider the effects of island formation
as a modification of it for negative grid voltages and small electrode spacings.
As the grid voltage increases all the curves tend to the straight lines calculated
from (6.56). The agreement is exact for d1 ≥ 0.5a and is a good approximation for
d1 = 0.2a . Figure 6.11 shows the variation of the current density across the cath-
ode for a variety of grid voltages when d1 = 0.2a . The current density is normally
far from uniform. It is least under the grid wires for negative grid voltages, and
greatest under the grid wires for positive grid voltages. This non-​uniform emis-
sion means that the grid current is increased. There is also a possibility that the
emission saturates in regions of high current density, so that the electron flow
ceases to be space-​charge limited everywhere. Saturation would be revealed, in a
plot like that in Figure 6.10, by current densities below the asymptotic straight
line at high positive grid voltages.

Figure 6.10:  Variation of cathode current density to the power 2/​3 with grid
voltage showing the effects of island formation in triodes with a range of values
of d1  (a = 1 mm, r = 0.05 mm, d 2 = 5 mm, Va = 100 V ).
24

242 Triodes and Tetrodes

Figure 6.11:  Variation of current density across the cathode surface for a range of grid voltages
showing island formation (a = 1 mm, r = 0.05 mm, d1 = 0.2 mm, d 2 = 5 mm, Va = 100 V ).

6.5 Electrostatic Models of Tetrodes

Figure 6.12 shows the arrangement of a planar tetrode. It has been assumed, for
convenience, that the spacings of both grids are the same. The electrostatic analysis
of the tetrode proceeds in exactly the same manner as that for the triode. By ana-
logy with (6.7) the electric field on the surface of the cathode is linearly dependent
on the potentials of the electrodes with respect to the cathode

Vg Vs V
E x ( 0, y ) = − − − a . (6.69)
d g ( y) d s ( y) da ( y)
This equation can be re-​written as

Vg + D1Vs + D2Va
E x ( 0, y ) = − , (6.70)
dg
where the dependence on y has been omitted for simplicity, and

dg dg
D1 ( y ) = and D2 ( y ) = . (6.71)
ds da
Now, assume that the grids are thin and their potentials are set to

d1 d1 + d 2
Vg = Va and Vs = Va . (6.72)
d1 + d 2 + d3 d1 + d 2 + d3
Then, substituting for Vg and Vs and noting that E x ( 0, y ) = −Va ( d1 + d 2 + d3 ), we
obtain

d g ( y ) = d1 + D1 ( d1 + d 2 ) + D2 ( d1 + d 2 + d3 ) (6.73)
243

Penetration Factors in Tetrodes 243

Figure 6.12:  Arrangement of a planar tetrode.

so that, substituting in (6.70),

Vg + D1Vs + D2Va
E x ( 0, y ) = − . (6.74)
d1 + D1 ( d1 + d 2 ) + D2 ( d1 + d 2 + d3 )

6.6 Penetration Factors in Tetrodes

To find the penetration factors in a planar tetrode, we use the principle of super-
position to add the potential of the screen grid to the expression for the potential
previously developed for the triode [20]. The potential within the tetrode is then,
from (6.23)

1
V ( x, y ) =
aε 0
( fg ( x, y) qg + fs ( x, y) qs + xqa ) , (6.75)
where qs is the charge per unit length on a screen grid wire, f g ( x, y ) is defined by
equation (6.24), the equivalent expression for the screen grid is

a  cosh 2 π ( x + d1 + d 2 ) a − cos 2 π y a 
fs ( x, y ) = ln   , (6.76)
4 π  cosh 2 π ( x − d1 − d2 ) a − cos 2 π y a 

and it has been assumed that both grids have the same spacing. We will also
assume, for simplicity, that the distances between the electrodes are all greater
than or equal to a so that the potential at the plane of each electrode is inde-
pendent of y. Then

a   πr  
Ds = − ln 2 sin  s   (6.77)
2 πd 3   a 
24

244 Triodes and Tetrodes

Vg =
1
aε 0
(f g (d1 , rg ) qg + d1qs + d1qa )
1
Vs =
aε 0
(d1qg + fs (d1 + d2 , rs ) qs + (d1 + d2 ) qa ) (6.78)
1
Va =
aε 0
(d1qg + (d1 + d2 ) qs + (d1 + d2 + d3 ) qa ) ,
where, from (6.33)

f g (d1 , rg ) = d1 + d 2 Dg , (6.79)

and

fs ( d1 + d 2 , rs ) = d1 + d 2 + d3 Ds , (6.80)

with

a   πrg  
Dg = − ln 2 sin   . (6.81)
2 πd 2   a  

Substituting in (6.78) gives

Vg  d1 + d 2 Dg d1 d1   qg 
V  = 1  d1 d1 + d 2 + d3 Ds d1 + d 2   qs  . (6.82)
 s  aε 0 
Va   d1 d1 + d 2 d1 + d 2 + d3   qa 

This equation can be inverted to give

  d1 + d 2 +  
 d3  − d1d3 − d1d3 Ds 
  Ds (d1 + d 2 + d3 )

q g   Vg 
 qs  = C 
  2 − d1d3 ( ) (
d1 d 2 + d3 + d 2 Dg d1 + d 2 + d3 ) (
− d 2 d1 + Dg (d1 + d 2 ))  V 
 s
q
 a  

− d1d3 Ds ( (
− d 2 d1 + Dg d1 + d 2 )) d 2 d1 + Dg (d1 + d 2 + d3 Ds )   a 

V

 

(6.83)
where

aε 0
C2 = . (6.84)
d3 d1 ( d 2 + Ds ( d 2 + d3 )) + d 2 Dg ( d1 + d 2 + Ds ( d1 + d 2 + d3 ))

Now the electric field at the surface of the cathode is given by

E x ( 0, y ) = −
(q g + qs + qa )
(6.85)
aε 0
245

Static Characteristics of Tetrodes 245

so that

E x ( 0, y ) =
(d 2 + Ds ( d 2 + d3 ))Vg + d 2 Dg (Vs + DsVa )
, (6.86)
d1 ( d 2 + Ds ( d 2 + d3 )) + d 2 Dg ( d1 + d 2 + Ds ( d1 + d 2 + d3 ))
which can be rewritten as

Vg + Dg′ (Vs + DsVa )

E x ( 0, y ) = , (6.87)
d1 + Dg′ ( d1 + d 2 + Ds ( d1 + d 2 + d3 ))

where

Dg
Dg′ = . (6.88)
1 + Ds (1 + d3 d 2 )
Comparing (6.87) with (6.74) we see that

D1 = Dg′ and D2 = Dg′ Ds . (6.89)

Equation (6.87) can also be written

Vg + DgVe
E x ( 0, y ) = − (6.90)
d1 + Dg d e

in which the tetrode has been replaced by an equivalent triode whose anode
potential is

Vs + DsVa
Ve = (6.91)
1 + Ds (1 + d3 d 2 )
and the position of the anode is

d1 + d 2 + Ds ( d1 + d 2 + d3 )
de = . (6.92)
1 + Ds (1 + d3 d 2 )
This is similar, but not identical, to the expression for reducing a triode to an
equivalent diode (see equation (6.55).

6.7 Static Characteristics of Tetrodes

The static characteristics of a tetrode may be calculated from the equivalent diode
in exactly the same way as those of triodes. Thus from (6.90) we obtain
3

J = K 1 + Dg ⋅
(V g + DgVe ) 2
(6.93)
(d + Dg d e )
2
1

by analogy with (6.56).

246

246 Triodes and Tetrodes

Island formation occurs when d1 a < 1. Since, normally, d 2 a ≥ 1 the variation

of the vacuum electric field on the cathode surface depends only upon the control
grid. Then the characteristics of the tetrode can be determined from the equivalent
triode using (6.45) and (6.46) with

DE ( y ) =
( )
d1 1 − f g ′ ( 0, y ) + d e Dg
, (6.94)
(d1 + de ) fg ′ (0, y) − d1
where f g ′ ( 0, y ) is given by (6.39) and

Vg + DE ( y )Ve
E z ( 0, y ) = − . (6.95)
d1 + DE ( y ) d e
Then the characteristic curves can be computed using
a 3

K
⌠
J =  1 + DE ( y ) ⋅
(Vg + DE ( y)Ve )2 ⋅ dy. (6.96)
a
⌡0 (d1 + DE ( y) de )2

6.7.1 Grid Currents in Tetrodes

The current intercepted by the control grid in a tetrode may be estimated in exactly
the same way as for a triode. The current is zero if the grid is negative and increases
with increasingly positive voltages. Theoretically, the screen grid should not inter-
cept any electrons because the wires are aligned with those of the control grid.
Hence the fields should focus the electron trajectories so that they pass between
the screen grid wires [21]. However, at high cathode current and low anode voltage,
the accumulation of space charge between the screen grid and the anode causes the
electron trajectories to move outwards so that they may be intercepted by the wires
of the screen grid. There is no simple way of estimating the current intercepted by
the screen grid but it is commonly a small fraction of the cathode current except
when the anode voltage is low and the anode current is high. For a model of a pla-
nar tetrode see Worksheet (6.3).

6.7.2 Effect of Space-​Charge between the Screen Grid and the Anode

The tetrode behaves approximately like an electron source whose current is con-
trolled by the control grid voltage and whose energy is controlled by the screen
grid voltage. Thus the flow of current in the space between the screen grid and
the anode approximates to that in a diode with injected current as described in
Section 5.8. The potential of the anode must always be greater than that of the
screen grid so that any secondary electrons emitted by the anode are returned to
the anode. Therefore we are only interested in conditions in which there is a mono-
tonic increase in the potential between the screen grid and the anode, or where
247

Static Characteristics of Tetrodes 247

there is a potential minimum which is not deep enough to form a virtual cathode.
Using equations (5.96) and (5.100), and recalling that the potentials in Section 5.8
are referred to the potential at the injection plane (in this case the screen grid), the
normalised spacing between the screen grid and the anode is

−4 J a
Xa = ⋅ d3 , (6.97)
ε 0 2 e m0Vs1.5

where Vs is the potential of the screen grid with respect to the cathode. The normal-
ised anode potential is

Va
Wa = . (6.98)
Vs
The parameter α1 can be then found by numerical solution of the equation

4 4
Xa = (Wa − 2α1 ) (Wa + α1 )0.5  (1 − 2α1 ) (1 + α1 )0.5 , (6.99)
3 3
derived from (5.104) and (5.105) by eliminating α 2 , where the negative sign is taken
if there is no potential minimum between the screen grid and the anode. Then, from
(3.90), the normalised potential gradient at the screen grid is

dU
= ± (1 + α1 ) . (6.100)
0.5

dX
We assume that this potential gradient is produced by an equivalent vacuum anode
voltage Va′ so that

dU  Va′ − Vs  1
= ⋅ . (6.101)
dX  Vs  X a
Since the effective anode voltage has been reduced, it is necessary for the control
grid voltage to be increased to maintain constant anode current by

∆Vg = Dg′ Ds (Va − Va′) . (6.102)

As the space charge increases, the electric field at the screen grid is reduced, Va′
becomes smaller and, eventually, becomes negative as a potential minimum is
formed. Thus this model predicts that the effects of space-​charge are to increase the
control grid voltage required to produce a given current. The space-​charge effects
are greatest when the current is maximum and the anode voltage is least, that is
when Wa = 1. Then, taking the positive sign in (6.99) we find that the maximum
possible value of X a is 3.77 when α1 = − 1 2 [22]. Thus, once the screen grid voltage
and the maximum current have been specified there is a maximum permissible value
of the spacing between the screen grid and the anode beyond which the current is
space-​charge limited. It is shown in Chapter 12 that this simple model is in quali-
tative agreement with the behaviour of real tetrodes. Equation (6.97) shows that
248

248 Triodes and Tetrodes

the effect of increasing the spacing between the screen grid and the anode is simi-
lar to that of increasing the injected current. Thus the space-​charge effects can be
reduced by decreasing that spacing. Equation (6.98) shows that the effects of space-​
charge at a given anode voltage are increased as the screen grid voltage is increased.
However, the range of screen grid voltages over which a tube is operated is usually
small, so this is less important.

References

[1]‌ B. I. Bleaney and B. Bleaney, Electricity and Magnetism. London: Oxford University

Press, 1957.
[2]‌ EIMAC, Care and Feeding of Power Grid Tubes, 5th ed. San Carlos, CA:  CPI Inc.
Eimac Division, 2003.
[3]‌ S. G. McNees, ‘Two megawatts RF power tetrode’, IEEE Transactions on Nuclear
Science, vol. NS-​20, pp. 422–​423, 1973.
[4]‌ J. C. Maxwell, Electricity and Magnetism, 3rd ed., vol. 1. Oxford: Clarendon Press, 1892.
[5]‌ J. H. Fremlin, ‘LXVIII. Calculation of triode constants’, Philosophical Magazine Series
7, vol. 27, pp. 709–​741, 1939.
[6]‌ F. B. Vogdes and F. R. Elder, ‘Formulas for the amplification constant for three-​element
tubes in which the diameter of grid wires is large compared to the spacing’, Physical
Review, vol. 24, p. 683, 1924.
[7]‌ F. Ollendorf, ‘Berechnung des Durchgriffes durch enger Steggitter’, Elektotech. und
Machinenbau, vol. 21, pp. 59–​64, 1934.
[8]‌ K. R. Spangenberg, Vacuum Tubes. New York: McGraw-​Hill, 1948.
[9]‌ W. Bennett and L. Peterson, ‘The electrostatic field in vacuum tubes with arbitrarily
spaced elements’, Bell System Technical Journal, vol. 28, pp. 303–314, 1949.
[10] B. J. Thompson, ‘Space-​current flow in vacuum-​tube structures’, Proceedings of the
IRE, vol. 31, pp. 485–​491, 1943.
[11] A. H. W. Beck, Thermionic Valves: Their Theory and Design. Cambridge: Cambridge
University Press, 1953.
[12] B. D. H. Tellegen, ‘The calculation of the emitted current in a triode’, Physica, vol. 5e,
pp. 301–​315, 1925.
[13] W. Dahlke, ‘Gittereffektivpotential un Kathodenstromdichte einer ebenen Triode unter
Berucksichtigung der Inselbildung’, Telefunken Zeitung, vol. 24, pp. 213–​222, 1951.
[14] S. Magdo, ‘The optimum grid to cathode spacing in planar triodes’, IEEE Transactions
on Electron Devices, vol. 12, pp. 273–​280, 1965.
[15] L. C. Scholz, ‘Calculation of fields and currents’, in R. S. Burnap, ed., Electron Tube
Design. Harrison, NJ: Radio Corporation of America, pp. 159–​201, 1962.
[16] J. W. Gewartowski and H. A. Watson, Principles of Electron Tubes. Princeton, NJ: D. van
Nostrand Co. Inc., 1965.
[17] K. R. Spangenberg, ‘Current division in plane-​electrode triodes’, Proceedings of the
I.R.E., vol. 28, pp. 226–​236, 1940.
249

References 249

[18] W. Dahlke, ‘Gittereffektivpotential un Kathodenstromdichte einer ebenen Triode unter

Berucksichtigung der Inselbildung’, Telefunken Zeitung, vol. 24, pp. 213–​222, 1951.
[19] W. Dahlke, ‘Statische Kennwerte der nichtidealen Triode’, Telefunken Zeitung, vol. 27,
pp. 172–​186, September 1954.
[20] S. Deb, ‘Amplification factors and mutual conductance of a beam power valve’,
Proceedings of the IEE  –​Part B:  Radio and Electronic Engineering, vol. 102, pp.
469–​474, 1955.
[21] C. S. Bull, ‘The alignment of grids in thermionic valves’, Journal of the Institution of
Electrical Engineers  –​Part III:  Radio and Communication Engineering, vol. 92, pp.
86–​92, 1945.
[22] G. Jaffe, ‘On the currents carried by electrons of uniform initial velocity’, Physical
Review, vol. 65, pp. 91–98, 1944.
250

7 Linear Electron Beams

7.1 Introduction

In many vacuum electron devices the interaction between the electrons and RF
electric fields takes place over an extended linear region. In order for this to be
possible it is necessary to control the paths of the electrons to ensure that they are
held in the correct relationship with the RF fields. This means that the electrons
must be formed into an electron beam having a cross-​section which is approxi-
mately constant. The charge density in the electron beams used in microwave tubes
is high enough to cause them to spread out rapidly because of the mutual repulsion
of the electrons. If this were not controlled in some way, the electrons would be
intercepted on the tube body. They would then be lost to the interaction, and would
cause damaging dissipation of heat.
The greater part of this chapter deals with methods of controlling the cylindrical
electron beams used in the majority of microwave tubes. The effects of external elec-
tric and magnetic fields are considered both with, and without, space-​charge forces.
The properties of simple electric and magnetic lenses are discussed. There is an exten-
sive treatment of the properties of cylindrical electron beams in the presence of a
uniform axial magnetic field (solenoid focusing). The conditions for stable electron
flow are discussed together with periodic variations of the beam radius (scalloping),
and the changes in beam radius produced by modulation of the charge density (beam
stiffness). This is followed by consideration of the spreading of electron beams, in the
absence of a magnetic field, caused by space-​charge, and centrifugal, forces. Next it
is shown that stable electron flow can be achieved using periodic permanent magnets
(PPM focusing), and periodic electrostatic fields. A  brief summary is given of the
properties of sheet beams and annular beams. Finally, there is a review of the imper-
fections introduced by thermal electron velocities and trapped ions.

7.2 Cylindrical Electron Beams

The majority of linear-​beam tubes employ electron beams having circular symmetry
about the axis of propagation. If the beam has radius b and is located concentric-
ally within a cylindrical conducting drift tube of radius a as shown in Figure 7.1
then the electric field produced by the electron space charge is everywhere radial.
251

Cylindrical Electron Beams 251

Figure 7.1:  A cylindrical electron beam.

If it is assumed that the charge density is uniform and the electrons are moving
with uniform axial velocity u0 then the current carried by the beam is
I 0 = πb2 ρu0 , (7.1)
where ρ is the charge density, and u0 represents the axial component of the electron
velocity, including the effects of relativity, unless otherwise stated. A simple application
of Gauss’ Theorem shows that the radial component of the electric field, when r ≤ b, is
ρ I0
Er = r= r (7.2)
2ε 0 2 πb2 ε 0 u0
and, when r > b

I0 1
Er = ⋅ . (7.3)
2 πε 0 u0 r
The electrostatic potential within these two regions is found by integrating (7.2) and
(7.3) with respect to r to give
I0
V =− r 2 + C1 (7.4)
4 πb2 ε 0 u0
for r ≤ b, and
I0
V =− ln r + C2 (7.5)
2 πε 0 u0
for r > b, where C1 and C2 are constants. If the potential is zero when r = a then (7.5)
becomes
I0
V =− ln ( r a ) . (7.6)
2 πε 0 u0
Equating the values of the potential at r = b  gives
I0
C1 = 1 − 2 ln ( b a ) (7.7)
4 πε 0 u0 
so that, within the beam

I0  r2 
V= 1 − 2 ln ( b a ) − . (7.8)
2 πε 0 u0  b2 
25

252 Linear Electron Beams

Since b < a the logarithm is negative and the term within the brackets is always
positive. The conventional current in the direction of electron flow is nega-
tive and so the potential within the beam is negative with respect to the sur-
rounding drift tube. The properties of electron beams can be calculated using
Worksheet 7.1.
The effect of this space-​charge potential depression is best illustrated by an
example. Suppose that Va = 10 kV, I 0 = −1 A, b = 5 mm and a = 10 mm. Then the
electron velocity at zero potential is 58.5 × 106 m s −1 . Substitution of these figures
into (7.8) shows that the potential is −213 V with respect to the drift tube at the
beam edge, and −367 V on the axis. Strictly speaking, the small change in the elec-
tron velocity produced by this difference in potential should be used to recalculate
the potential depression iteratively until mutually consistent values are found. In
practice it is either ignored, or used to make a small correction to the velocity of
all the electrons. The situation is complicated by the effect of the rotation of the
electrons about the axis caused by the magnetic focusing field which is discussed
in Section 7.4. In practical tubes the tunnel is not continuous, but interrupted by
a series of gaps. The effect of the gaps can be represented by an effective tunnel
radius, if necessary [1].

7.3 Electron Optics without Space-​Charge

Consider an electron which is moving in fields which are cylindrically symmet-

rical so that the tangential component of the electric field is zero. The equations of
motion of an electron in cylindrical polar co-​ordinates are from (1.6)

d  dr 
(

 γ m0  = −e Er + rθBz − zB
dt  dt
)
 θ + rθ 2 (7.9)

d
dt
( )
γ m0 r 2θ = er ( rB  r ) (7.10)
 z − zB

d  dz 

dt dt
 (
 γ m0  = −eE z + e r θBr − rB
 θ . (7.11) )
We note that, from (1.3) we can write

1 V
γ= = 1+ , (7.12)
1− u c 2 2 VR

where VR = m0 c 2 e and V is the electrostatic potential relative to the cathode.1

In developing the theory of vacuum tubes it is convenient take the cathode as the zero of electrostatic poten-
1

tial. However, it should be noted that, in practice, the body of the tube is normally at earth potential and the
cathode is negative with respect to it.
253

Electron Optics without Space-Charge 253

7.3.1 The Paraxial Ray Equation of Electrostatic Electron Optics

Consider an electron which is moving, without rotation, in a cylindrically symmet-
ric electric field [2]. From (7.9) the radial motion is determined by

d  dr  eEr
γ  + = 0. (7.13)
dt  dt  m0
Substituting for γ from (7.12) in (7.13) gives

d 2 r dV dr eV0 ∂V
(V + VR ) + ⋅ − = 0. (7.14)
dt 2 dt dt m0 ∂r
If the electron is always close to the axis, so that its radial velocity is small com-
pared with its axial velocity, then, to a good approximation,

dz
= u, (7.15)
dt
where u is the electron velocity determined from (1.4). Then we may write

d d
= u (7.16)
dt dz
and

d2 d2 ∂u d
= u 2
+ u ⋅ (7.17)
dt 2
dz 2
∂z dz
so that (7.14) becomes

d 2r ∂u dr ∂V dr ∂V
(V + VR ) u 2 + (V + VR ) u ⋅ + u 2 ⋅ − c2 ⋅ = 0. (7.18)
dz 2 ∂z dz ∂z dz ∂r
If space charge can be neglected, then the electric field satisfies Laplace’s equation

1 ∂  ∂V  ∂ 2V
r + = 0. (7.19)
r ∂r  ∂r  ∂z 2
When the trajectory of an electron lies close to the axis (a paraxial trajectory) the
potential does not differ much from the potential on the axis (VA). Then (7.19) can
be integrated with respect to r, to give the approximate relationship

∂V 1 d 2VA
=− r . (7.20)
∂r 2 dz 2
Equation (1.4) can be written in the form

V (V + 2VR )
u2 = c2 (7.21)
(V + VR )2
254

254 Linear Electron Beams

and differentiated with respect to z to give

du VR2 c 2
u = . (7.22)
dz (V + VR )3

Then, substituting into (7.18) from (7.20) and (7.22) we obtain

VA (VA + 2VR ) d 2 r ∂VA dr 1 ∂ 2VA

⋅ + ⋅ + ⋅ r = 0. (7.23)
(VA + VR ) dz 2 ∂z dz 2 ∂z 2
This is the paraxial ray equation of electrostatic electron optics. Its application to
electrostatic lenses is discussed in the next section.

7.3.2 Thin Electrostatic Lenses

Figure  7.2 shows a short transition region, defined by the planes 1 and 2,
between two regions in which the axial electric field is constant. Since ∇ ⋅ E = 0
the radial component of the electric field must be non-​z ero in this region. When
an electron enters the transition region from the left moving parallel to the axis
at a radius r1 its trajectory is perturbed. It then emerges on a trajectory that
intersects the axis at a distance f2 from the region. The electric field in the tran-
sition region therefore acts as a lens. If the lens is thin then the radial position
of the trajectory can be assumed to be the same on entering and on leaving the
lens ( r1 = r2 = r ).
In order to derive an expression for the focal length of a thin electrostatic lens it
is convenient to express (7.23) in the form

d  dr  1 d 2V
U U  = − (V + VR ) 2 r, (7.24)
dz dz 2 dz
where the subscript A has been dropped because the potential is understood to be
defined on the axis relative to the potential of the cathode, and

U = V (V + 2VR ). (7.25)

Figure 7.2:  Geometry of an electrostatic electron lens.

25

Electron Optics without Space-Charge 255

Because the electric field is constant, except between the two planes, it follows that
d 2V dz 2 is zero outside the transition region. Integration of (7.24) between planes
1 and 2 produces
2
 dr  1 ⌠ z2 V + VR d 2V
U  = −  ⋅ 2 r dz. (7.26)
 dz 1 2 ⌡z1 U dz
Now the gradient of the trajectory at plane 1 is zero and at plane 2 it is − r f2 , so
the focal length of the lens is given by

1 1 ⌠ z2 V + VR d 2V
=  ⋅ 2 dz, (7.27)
f2 2U 2 ⌡z1 U dz
where U 2 is the value of U at plane 2. If V ( z ) is known, the focal length can be calcu-
lated from (7.27). Integration of the right-​hand side of this equation by parts yields
z z2 2
1 1 V + V0 dV  2 1 ⌠ V02  dV 
= ⋅ + 3   dz. (7.28)

f2 2U 2  U dz  z1 2U 2 

⌡z1 U  dz 
Equation (7.28) can be applied to the two simple lenses illustrated in Figure 7.3.
An Einzel lens (Figure  7.3(a)) has a central electrode at one potential within
another electrode at a different potential (commonly zero). The electric field of the
lens is confined between the planes 1 and 2, so that the gradient of the potential is
zero at both planes, and the first term on the right-​hand side of (7.28) is zero. Since
the square of the electric field appears in the integral, the second term is positive for

Figure 7.3:  Electrostatic lenses: (a) an Einzel lens, and (b) an aperture lens.

256

256 Linear Electron Beams

all potentials of the inner electrode and this type of lens is always converging. The
application of this to the theory of periodic electrostatic focusing of electron beams
is discussed in Section 7.6.3.
In an aperture lens (Figure 7.3(b)) the initial and final fields are not zero, and
it is found that the first term in (7.28) is commonly much larger than the second.
The potential does not change very much through the lens so that V1 ≈ V2. The focal
length of the lens is given by

1

Va + VR  dV 
2

=
(Va + VR ) ( E1 − E2 ) , (7.29)
z

 
f2 2Va (Va + 2VR )  dz  z1 2Va (Va + 2VR )

where Va is the potential of the aperture. The focal length may be positive or nega-
tive (converging or diverging) depending upon the signs and the magnitudes of the
fields on the two sides of the aperture. The application of (7.29) in the theory of the
Pierce electron gun is discussed in Section 9.2.1. Further information on electro-
static lenses can be found in [3].

7.3.3 Busch’s Theorem

We now establish an important theorem relating to the motion of an electron
through a cylindrically symmetrical magnetic field. The proof given here is based
on [4] which is more general and rigorous than those published elsewhere. Consider
an electron which is distant r from the axis. The magnetic flux linked to a circle of
radius r is defined by
r
Φ = 2π ∫ rBz dr. (7.30)
0

The rate of change of the flux linked to a circle which travels with the electron is
found by differentiating of (7.30) with respect to time

d Φ ∂Φ ∂Φ  r
∂Bz 
= r + z = 2 π  Bz rr + z ⌠
r dr . (7.31)
dt ∂r ∂z  ⌡ ∂z 
0

Now the axial and radial components of the magnetic field are linked together by
the requirement that ∇ ⋅ B = 0. In cylindrical polar coordinates this becomes

1 ∂ ∂B
(rBr ) + z = 0 (7.32)
r ∂r ∂r
so the integral in (7.31) can be written
r r
⌠ r ∂Bz dr = − ⌠ ∂ rB dr = −rB (7.33)
  ( r) r
⌡ ∂z ⌡ ∂r
0 0
257

Electron Optics without Space-Charge 257

and (7.31) becomes

dΦ
= 2 π ( rrB  r ) . (7.34)
 z − rzB
dt
Comparing (7.34) with (7.10) we observe that the right hand sides are identical
apart from constants and, therefore

e dΦ

d 2
dt
( )
rθ −
2 πγ m0 dt
= 0, (7.35)

where it is assumed that the electrostatic potential is constant, so that the relativistic
factor γ is constant. When (7.35) is integrated with respect to time we obtain the
important result that, for the motion of the electron,

e
r 2θ − Φ = constant. (7.36)
2 πγ m0

Equation (7.36) is known as Busch’s theorem. It is used to calculate the angular

velocity of electrons as they move through an axially symmetric magnetic field that
varies with axial position. It remains valid when the effects of space-​charge are
important.

7.3.4 Magnetostatic Electron Optics without Space Charge

The equation of radial motion for an electron moving in a region of constant elec-
trostatic potential, and a magnetic field with cylindrical symmetry, is obtained from
(7.9) as

d 2r e
= rθ2 − r θ Bz . (7.37)
dt 2
γ m0

It is assumed that the space charge forces, and the magnetic field of the electron
current, are negligible. The relativistic factor γ is constant because the electrostatic
potential is constant. Then, making use of (7.17) and, assuming paraxial motion,
(7.37) becomes

d 2r e
u2 = rθ2 − r θ Bz . (7.38)
dz 2
γ m0

If the electron has originated from a cathode shielded from the magnetic field, then
the constant in (7.36) is zero. Also, since the electron remains close to the axis, we
can write Φ = π r 2 B0, where B0 is the flux density on the axis, so that, from (7.36)

eB0 ω
θ = = L , (7.39)
2 γ m0 γ
258

258 Linear Electron Beams

where ω L is known as the Larmor frequency [5]. The right hand side of (7.38) is

e ω2
r θ2 − r θ Bz = −r 2L . (7.40)
γ m0 γ
Substitution in (7.38) yields the magnetostatic paraxial ray equation

d 2 r β2L
+ r = 0, (7.41)
dz 2 γ 2
where β L = ω L u .

7.3.5 Thin Magnetic Lenses

Magnetic electron lenses commonly take the form shown in Figure 7.4 where the
magnetic field generated by a coil, or a permanent magnet, is confined axially by a
pair of iron pole pieces. The magnetic field is therefore zero outside the lens. If the
lens is thin, r is constant within it, and the change of the slope of the trajectory is
obtained by integrating (7.41) through the lens
2 2
dr  e  z2

dz 1
= −r 
 2 γ m0 u  ∫ z1
B02 dz. (7.42)

The focal length of the lens is given by

2
1  e  z2
=
f2  2 γ m0 u  ∫ z1
B02 dz. (7.43)

Since the integral is always positive, a magnetic lens is always converging. Further
information on magnetic lenses can be found in [3].

Figure 7.4:  A magnetic electron lens.

259

Electron Optics with Space-Charge 259

7.4 Electron Optics with Space-​Charge

An uncontrolled cylindrical electron beam spreads out radially under the influence
of its space charge. In order to create a beam with sufficient length and power for
use in a vacuum tube it is necessary to control it so that the tendency to spread is
eliminated. This is usually achieved by using either a uniform axial magnetic field or
a periodic array of converging magnetic lenses. An alternative which is rarely used
is a periodic array of converging electrostatic lenses.
In order to study the properties of electron beams under the influence of magnetic
fields we make use of equation (7.9) for the radial motion of an electron on the edge
of the beam. We will assume that the axial velocity of the electrons ( u0 ) is constant
to a good approximation. The radial component of the electric field is given by (7.3)
and the azimuthal component of the magnetic field caused by the beam current is

µ0 I0
Bθ = . (7.44)
2 πr
Substituting in (7.9) gives

d 2r eI 0 1 eθBz θ2
=− ⋅ + r+ r, (7.45)
dt 2
2 πu0 ε 0 γ m0 r γ m0
3
γ m0
where u0 is the relativistically correct velocity. This can be written

d 2r ω 2p b2 eθBz θ2
= − ⋅ + r+ r, (7.46)
dt 2
γ 2r γ m0
3
γ m0
where ω p is the electron plasma frequency [6] defined when r = b so that

eρ e I
ω 2p = − =− ⋅ 20 . (7.47)
ε 0 m0 ε 0 m0 πb u0
We will assume that the beam is launched with zero angular velocity and radius rc from
a cathode where the magnetic flux density is Bc. Then, by Busch’s theorem (7.36), the
angular velocity at a position where the radius is r, and the flux density is Bz , is given by

e  rc2 
θ =  Bz − 2 Bc  . (7.48)
2 γ m0 r
Substituting for the angular velocity in (7.46) gives
2 2
d 2 r 1 ω 2p b2  eBc rc2   eBz 
= + 2
r− r. (7.49)
dt 2
2 γ r  2 γ m0 r 
3
 2 γ m0 
If the flux density Bz = B0 f ( z ) then (7.49) can be written

d 2 r 1 ω 2p b2 ω 2L b 4 ω 2L
+ 2 K 3 − 2 f ( z ) r, (7.50)
2
=
dt 2 2 γ 3 r γ r γ
260

260 Linear Electron Beams

where ω L is the Larmor frequency defined in (7.39), and

2
 B r2 
K =  c c2  . (7.51)
 B0 b 

Finally we use (7.17) to change the independent variable from t to z, and introduce
the normalised radius R = r b so that

d 2 R β2p 1 β2L 1 β2L

f ( z ) R, (7.52)
2
= ⋅ + K −
dz 2
2γ R γ
3 2
R 3
γ 2

where β p = ω p u0 and β L = ω L u0 . This is the general differential equation govern-

ing the flow of a cylindrical electron beam through an axial magnetic field. The
non-​relativistic form is found by setting γ = 1 to give

d 2 R β2p 1 1
⋅ + β2L K 3 − β2L f ( z ) R. (7.53)
2
=
dz 2 2 R R
In the following sections the solutions of this equation are examined for a number
of cases of interest. Non-​relativistic analysis will be used because the effects of rela-
tivity can readily be included by adjusting the parameters β p and β L .

7.4.1 Solenoid Focusing

The simplest way to control a cylindrical electron beam is by using a uniform axial
magnetic field ( B0 ). Such a field is usually produced by an electromagnet, though
permanent magnets are sometimes used [7, 8]. The condition for smooth flow is
that the radial electrostatic, magnetic and centrifugal forces should balance. Thus
setting R = 1 and f ( z ) = 1 in (7.52)

β2p = 2 γβ2L (1 − K ) . (7.54)

Since β p, β L and K are all positive this equation can only be satisfied when 0 ≤ K ≤ 1.
The smallest magnetic field for which smooth flow is possible occurs when the
magnetic flux linked to the cathode is zero and, therefore, K = 0. This condition is
known as Brillouin flow [9]. The Brillouin field ( BB ) is given by

1 2 ωp
βL = βp or BB = ⋅ . (7.55)
2γ γ e m0

If the magnetic field is m times the Brillouin field then the relationship between the
Larmor frequency and the plasma frequency is

m
ωL = ω p . (7.56)
2γ
261

Electron Optics with Space-Charge 261

Substituting for wL in (7.54) we find that the relationship between K and m can be
written

1
K = 1− . (7.57)
m2
When m → ∞ then K → 1 so that the flux linked to the cathode is equal to that
linked to the beam. This condition, which is known as confined flow, is useful as a
theoretical concept but would normally require too strong a magnetic field to be
useful in practice. Equation (7.57) is important for the design of linear electron
beams because it is a necessary, though not sufficient, condition for achieving uni-
form electron flow. The injection of the beam into the focusing field is discussed in
Chapter 9. A detailed discussion of the magnetic focusing of relativistic beams is
given in [10, 11].
When the current in a beam is modulated, the radius is perturbed to an extent
that decreases as m increases (see Section 7.4.3). This radial perturbation is unde-
sirable and it is usual to select a value of m greater than unity to ensure adequate
beam stiffness. Thus the beams in practical tubes generally operate in space-​charge
balanced flow in which m > 1. In order to select a suitable value of B0 the Brillouin
field is computed from (7.55), m is chosen to give the desired beam stiffness, and
(7.57) then gives the cathode flux required (see Worksheet 7.1). Alternatively, if B0
and the cathode flux are known, then an iterative calculation employing (7.55) and
(7.57) can be used to find the equilibrium beam radius and the corresponding value
of m. The design of the magnetic field in the beam entry region to achieve the cor-
rect initial conditions is described in Section 9.3.1, and the design of the solenoid
and the magnetic circuit are described in Section 19.5.2.
The angular velocity of the electrons on the surface of a solenoid focused beam
in smooth flow is found from (7.48) to be

ω
( )
θ = L 1 − K , (7.58)
γ
which is in the right-​hand corkscrew direction when B0 is in the positive z direction.
Because a solenoid focused beam is rotating about its axis, the axial velocity of the
electrons is reduced by the transfer of some of the kinetic energy into rotational
motion. If the charge density and the magnetic flux density are uniform across the
beam, at a given plane, then the angular velocity of all the electrons is given by
(7.58). The axial velocity of an electron at radius r is given by

z = u 2 − r 2 θ2 , (7.59)

where u is determined from (1.4) by setting V to the local potential relative to the
cathode, allowing for space-​charge potential depression. Now the rotational kinetic
energy is zero on the axis and increases with r, whilst the potential depression is
greatest on the axis and decreases with r. Thus these two effects compensate for
each other to some extent. For the special case of a beam in Brillouin flow it can be
26

262 Linear Electron Beams

shown that the two effects cancel each other exactly [7, 12]. The axial velocity of all
the electrons is then the same as that of an electron on the axis. When the beam is
not in Brillouin flow ( m > 1) the angular velocity is smaller and the axial velocity of
the electrons varies with radius. For example, if m = 2, then the angular velocity is
only 13% of the Brillouin value, and the variation in the axial velocity is determined
almost entirely by the variation of potential. These corrections to the calculation
of the axial velocity are small but can be important, especially when travelling-wave
tubes (TWTs) are modelled, because the centre frequency is particularly sensitive
to variations in the electron velocity. In many cases it is sufficient to use the veloc-
ity of an electron on the axis as an approximation to the mean axial velocity of the
electron beam.

7.4.2 Scalloping
An electron beam will only flow smoothly through a uniform axial magnetic
field if it has the correct angular velocity, and if the radial velocity is zero at
the equilibrium beam radius ( r = b ). When these conditions are not satisfied
the force on an electron on the surface of the beam is not zero, and the radial
motion of the electron is governed by equation (7.52). In general the solution
of this equation requires the use of numerical methods (see Worksheet 7.2)
but useful insight is provided by examining the approximate equation obtained
by assuming that the radial perturbations are small and neglecting relativistic
effects. To do this we assume that R = 1 + x where x  1. The powers of R are
expanded using the Binomial Theorem, and powers of x higher than the first
are ignored. Then

d 2x  1 2  1 
=  β p + β2L K − β2L  −  β2p + 3β2L K + β2L  x. (7.60)
dz 2  2   2 
From (7.54) we find that the first bracket on the right-​hand side of (7.60) is zero and
that the equation can be rewritten as

d 2x
= −2β2L (1 + K ) x. (7.61)
dz 2
An alternative form of the equation can be found by substituting for β L and K using
(7.56) and (7.57) to give

d 2x
= − (2 m2 − 1) β2p x. (7.62)
dz 2
This is the equation of simple harmonic motion. Therefore the electrons at the edge
of the beam execute a sinusoidal motion in the radial direction, with a wavelength
which is equal to 2 π β p when m = 1. This wavelength decreases as m is increased. This
motion is known as scalloping. We shall see in the Section 7.4.3 that the decrease in
263

Electron Optics with Space-Charge 263

the scalloping wavelength as the magnetic field increases is consistent with increas-
ing stiffness of the beam against radial perturbations. At the level of approximation
we have used in this analysis, a beam which is launched into the focusing field at a
radius ( b − x ) will scallop between radii ( b − x ) and ( b + x ). It should be noted that
a similar argument applies to electrons lying within the beam. It is necessary to
ensure that the launching conditions are correct for all the electrons if the beam is
to be well-​controlled. The behaviour of a beam when the initial conditions are var-
ied can be explored using Worksheet 7.2. The same worksheet can be used to study
the effects of different assumptions about the axial variation of the magnetic field.
If the beam is initially in smooth flow then a step change in the magnetic field pro-
duces scalloping. If, on the other hand, there is a gradual increase in the field then
the beam moves smoothly to a new equilibrium beam radius with little scalloping
even for quite rapid changes in the field.

7.4.3 Beam Stiffness

In a linear-​beam tube the electrons in the beam become bunched axially to produce
a time-​varying current. This bunching upsets the equilibrium of the beam so that
it tends to expand in regions of high charge density, and to contract elsewhere. In
order to estimate the magnitude of this effect we consider the way in which the
equilibrium beam radius varies if the beam current is increased, while keeping the
magnetic field and the cathode flux constant. Since this calculation is only approxi-
mate we can safely ignore second-order effects resulting from changes in the space
potential. We also assume that the wavelength on the beam is great enough for axial
components of the RF electric field to be negligible.
The effect of changes in the local beam current on the behaviour of a solenoid
focused beam can be explored using equation (7.52). We note that the beam current
only affects the first term on the right hand side of the equation and that the square
of the plasma frequency is proportional to the beam current. If we assume that the
current density varies as k ( z ) then the effect of the changes in current density can
be modelled by multiplying the space-​charge term in equation (7.52) by this factor.
The effect of different assumptions about k ( z ) can be examined using Worksheet
7.2. It is found that smooth changes in the beam current cause the beam to move
to a new equilibrium radius with little scalloping and we are therefore justified in
supposing that the effects of density modulation of the beam can be described quite
accurately by the changes in the equilibrium beam radius. It should be noted that
this differs from the approach used by other authors who assume that the modu-
lation of the beam leads to scalloping [7]. We shall see that the analysis presented
here is consistent with the small-​signal properties of magnetically-​focused beams
presented in Chapter 11. The equilibrium equation is obtained from (7.52) as

β2p k 1 β2L 1 β2L

⋅ + K − R = 0, (7.63)
2 γ 3 R γ 2 R3 γ 2
264

264 Linear Electron Beams

where the instantaneous beam current is kI 0 . Multiplying through by R 4 , and sub-

stituting for β p and K using (7.56) and (7.57), gives

m2 R 4 − kR 2 − ( m2 − 1) = 0, (7.64)

which has the solution

R2 =
1
2 m2
( )
k + k 2 + 4 m2 ( m2 − 1) . (7.65)

The solution is trivial when m = 1 because then R 2 = k so that a doubling of the

current increases the beam radius by a factor of 2. Thus, in a Brillouin focused
beam the current density is constant when the beam is bunched, and the modula-
tion takes the form of perturbation of the beam radius. For other values of m it is
simple to compute the variation of R with k and the results are shown in Figure 7.5.
It is evident that the radial perturbation of the beam boundary decreases sharply
as m increases. Thus the stiffness of the beam increases with m, as suggested in the
previous section. This is an important result because the power density in the elec-
tron beam in a microwave tube is high enough to cause damage to the surrounding
metal structures if an appreciable part of the beam current is intercepted by them.
It is therefore necessary to design the focusing system with the objective of redu-
cing the intercepted current as far as possible. The interception could, in theory, be
reduced to zero by making the magnetic field very strong. This is sometimes assumed
for purposes of calculation, but it is not a practical solution because of the size,
weight, and cost of the focusing system that would be required. The use of Brillouin
flow ( m = 1) is also not practical because it generally leads to an unacceptably large
intercepted current. In practice it is usual to use a value of m in the range 1.3 to 3.0
where the electrons are in space-​charge-​balanced flow. To achieve the correct condi-
tions it is necessary to arrange for a portion of the focusing flux to be linked to the
cathode to satisfy (7.57). Sometimes, for simplicity, the beam may be launched from
a shielded cathode into a magnetic field which is stronger than the Brillouin field.

Figure 7.5:  Stiffness of a solenoid-​focused beam.

265

Electron Optics with Space-Charge 265

The beam then adopts the equilibrium radius given by (7.65) and will scallop, unless
it is launched into Brillouin flow, and then compressed by a gradual increase of the
field. Beams launched from non-​shielded cathodes show similar behaviour [13–​17].
The behaviour of such a beam can be investigated using Worksheet 7.2.
In high-​power linear-​beam tubes the peak beam current is of the order 2–​4 times
the DC current. We shall see that it is desirable to keep the diameter of the beam
tunnel small, and the beam filling factor ( b a ) as large as possible, in order to get a
strong RF interaction between the beam and the surrounding structure. But the use
of a small beam diameter requires a stronger magnetic field and, consequently, the
dissipation of more power in the solenoid which supplies it. There is, thus, a trade-​
off between the RF interaction and the size, weight, and cost of the solenoid and
its power supply. Figure 7.5 shows that, with a filling factor of 0.67, there should
be no interception when m is as low as 1.5. In practice the current density profile
in an electron beam does not have a sharp cut-​off at the designed beam radius but
commonly extends a little beyond it. Thus it may be necessary to use a higher value
of m to provide a safety margin. We shall return to this issue when discussing the
design principles for klystrons and TWTs. It is usual to use a computer model of
the electron gun, and a portion of the focusing structure, to check that the beam is
well-​controlled and has the intended diameter [18].
The results shown in Figure 7.5 can be used to estimate the relative proportions
of the RF beam current which are carried by modulation of the charge density
(bulk current) and of the beam radius (surface current). If we assume that the axial
velocity of the electrons is u0 and that the RF current density ( J1 ) is uniform across
the beam then we can write

kI 0 I
J1 = − 0 . (7.66)
π r 2 π b2
The amplitude of the bulk current ( I B ) is taken to be the RF current contained
within the equilibrium beam diameter, so that

 b2 
I B = πb2 J1 =  2 k − 1 I 0 . (7.67)
r 
Now the amplitude of the total RF current I1 is given by

I1 = ( k − 1) I 0 (7.68)

so that the normalised bulk current is

I B  1   b2 
=   k − 1 . (7.69)
I1  k − 1  r 2
The RF surface current ( I S ) is just ( I1 − I B ). Figure 7.6 shows how the normalised
bulk and surface currents vary with m for various values of k. Since the current
density in a Brillouin focused beam is equal to the equilibrium current density,
it follows that the whole of the RF current is surface current when m = 1. As the
26

266 Linear Electron Beams

Figure 7.6:  Bulk and surface currents in a modulated solenoid-​focused beam.

magnetic field is increased the surface current falls and the body current increases
but, even when m = 3 (which is the highest value normally used), there is still an
appreciable surface current, especially at the higher values of k. We shall see in
Section 11.3.1 that, in the limit of small modulation ( k → 0 ), this analysis agrees
with the results of space-​charge wave theory.

7.5 Beam Spreading

An uncontrolled electron beam spreads out through the influence of its own space-​
charge. If, in addition, the electrons are rotating about the axis of the beam then the
rate at which the beam spreads is increased by the centrifugal force. It is necessary
to be able to compute the spreading of an electron beam under these conditions,
both to model its expansion into the collector as it leaves the end of the focusing
system, and to understand the behaviour of focusing systems based on the use of
magnetic or electrostatic lenses.

7.5.1 The Universal Beam-​Spreading Curve

If there is no magnetic field and the angular velocity of the electrons is zero, the
equation of motion of an electron on the edge of a non-​relativistic beam is obtained
from (7.53) as [19]

d 2R 1 2 1
= β p . (7.70)
dz 2 2 R
267

Beam Spreading 267

Multiplying both sides of equation (7.70) by 2 ( dR dz ) and integrating with respect

to z gives
2
 dR 
 = β p ln ( R ) + C . (7.71)
2

dz 
If we choose R = 1 when the electrons are moving parallel to the axis (the beam
waist) the constant of integration C is zero. We recall that β p has been defined at
this plane (see (7.47)). Then

dR
= βp ln ( R ) (7.72)
dz
so that
R
⌠ dR
βpz =  . (7.73)
⌡1 ln ( R )

The integral can be evaluated numerically after first making the substitution
x 2 = ln ( R ) to give

ln( R )
βpz = ∫ 2 exp ( x 2 ) dx. (7.74)
0

The result of the integration is the universal beam spreading curve shown in
Figure 7.7. This curve is symmetrical about the R axis because the rate of conver-
gence is the same as the rate of expansion. An approximate analytical equation for
the curve can be obtained by assuming that R = 1 + x where x  1 and using the
series expansion of the logarithm

x 2 x3
ln (1 + x ) = x − + −  (7.75)
2 3

Figure 7.7:  The universal beam spreading curve.

268

268 Linear Electron Beams

so that (7.73) can be written approximately as

x
⌠ dx
βpz =  , (7.76)
⌡0 x
which can be integrated to give

2
1 
x =  β p z  (7.77)
2 

so that

1
R ≈ 1 + β2p z 2 . (7.78)
4
Figure 7.7 shows a comparison between the universal beam spreading curve and
the approximation given by (7.78). It can be seen that the approximation is useful
when β p z ≤ 2. The spreading of a relativistic beam can be found by replacing β p
with β p γ 1.5 .

7.5.2 Spreading of Rotating Beams

Consider a beam composed of mono-​energetic electrons emerging from a
focusing field into an electron collector. If the electrons are launched from a
cathode which is shielded from magnetic flux then, by Busch’s theorem, they
emerge from the magnetic field with no angular velocity and beam edge within
the collector follows the universal beam spreading curve. In many cases, how-
ever, the beam is designed to be in space-​charge balanced flow with magnetic
flux linked to the cathode. When such a beam emerges into a region which is
free from magnetic field its non-​relativistic equation of motion is obtained
from (7.53) as [7]

d 2R 1 2 1 1
= β p + β2L K 3 . (7.79)
dz 2 2 R R
The derivation now follows the same procedure as that for the universal beam
spreading curve. Equation (7.79) is multiplied by 2 ( dR dz ) and integrated with
respect to z with the initial condition R = 1. The result is
2
 dR  1 2  1 
 = β p ln ( R ) + β L K 1 − 2  . (7.80)
2

dz 2 R
Substituting for β L and K using (7.56) and (7.57) gives
2
 dR   1 2  1 
 = β p ln ( R ) + ( m − 1) 1 − 2   . (7.81)
2

dz   4 R 
269

Periodic Focusing 269

Figure 7.8:  Spreading of rotating beams.

Equation (7.81) reduces to (7.72) when m = 1. As before we make the substitution

x 2 = ln R so that
ln R
⌠
2 x exp ( x 2 )
βpz =  dx. (7.82)
 1
⌡0 x + ( m2 − 1) (1 − exp ( −2 x 2 ))
2
4
Equation (7.82) can be integrated numerically (see Worksheet 7.3) and the results
are shown in Figure 7.8 for a range of values of m. The trajectories spread more
rapidly as m increases as a result of the increasing centrifugal force. An approxi-
mate solution to (7.81) can be found by assuming that, by analogy with (7.78), the
beam radius varies with position as

R ≈ 1 + α 2 z 2 . (7.83)
Substituting in (7.80) yields

1 2
α2 =
4
(β p + K β2L ) (7.84)
so that

1 2
R ≈ 1+
4
(β p + K β2L ) z 2 . (7.85)
This reduces to (7.78) when K = 0. The spreading of relativistic beams can be found
by making relativistic corrections to β p and β L .

7.6 Periodic Focusing

As an alternative to solenoid focusing it is possible to control an electron beam by

passing it through a periodic array of converging lenses, as shown in Figure  7.9
270

270 Linear Electron Beams

Figure 7.9:  Focusing of a beam by an array of converging lenses.

[20–​22]. The beam converges as it leaves a lens in such a way that the space-​charge
forces produce an equal divergence as it enters the next lens. The edge of the beam
therefore has a ripple with the periodicity of the lenses. Systems using magnetic
lenses with permanent magnets, known as periodic permanent magnet (PPM) sys-
tems, are commonly used for TWTs and occasionally for klystrons. Periodic electro-
static focusing is sometimes used, but is usually more difficult to implement.

7.6.1 Periodic Permanent Magnet (PPM) Focusing

Figure  7.10(a) shows the general arrangement of a periodic permanent magnet
system. The permanent magnets, in the form of rings, are magnetised axially and
arranged so that their polarities alternate. The flux is conveyed to the region of the
electron beam by iron pole pieces so that the field on the axis varies approximately
sinusoidally with position, as shown in Figure 7.10(b). This arrangement has the
advantage that the size and weight of the magnets is considerably less than would
be required to produce an equivalent uniform magnetic field. The properties of a
PPM system can be modelled by finding the solutions of (7.53) when the magnetic
field profile is defined by

 2πz 
f ( z ) = cos  = cos (β0 z ) . (7.86)
 L 
The differential equation to be solved is

d 2R 1 2 1 1
= β p + β2L K 3 − β2L cos 2 (β0 z ) R. (7.87)
dz 2
2 R R
If we let θ = β0 z then (7.87) becomes

d 2R 1 1
= β + 2α K 3 − 2α cos 2 (θ ) R, (7.88)
dθ 2
R R
where the magnetic field parameter

1 β2L
α= , (7.89)
2 β20
271

Periodic Focusing 271

Figure 7.10:  Periodic permanent magnet (PPM) focusing system: (a) sectioned elevation, and
(b) the magnetic field profile.

the space-​charge parameter

1 β2p
β= , (7.90)
2 β02
and the cathode flux parameter K are defined at z = 0. This equation cannot be
solved analytically but solutions using analogue computers are given in [22] for
K = 0 and in [23] for K ≥ 0. The solution of (7.88) is readily achieved using numer-
ical methods, as shown in Worksheet 7.4, with results which agree with those in the
literature. Figure 7.11 shows some typical solutions. The beam profile is approxi-
mately sinusoidal with a wavelength which is one half of the wavelength of the
magnetic field. This is because the focusing effect depends upon the square of the
magnetic field and, therefore, there are two converging lenses in each wavelength
of the magnetic field. If the focusing conditions are correct then the electron tra-
jectories are perfectly periodic as shown in Figure 7.11(a). If, however, the focusing
is not correct then scalloping is superimposed upon the PPM ripple as shown in
Figures 7.11(b) and (c) in which β has been altered by ±20%.
When the beam has no scalloping (the minimum ripple solution) it is found
that the value of β is completely determined when α and K are fixed, as shown in
Figure 7.12 [23]. The ripple in the beam edge defined by

rmax − rmin
δ= (7.91)
rmax + rmin
also depends only on α and K, as shown in Figure 7.13 [23].
If α is increased, with a fixed value of K, and β is adjusted to seek for a minimum
ripple solution it is found that, beyond a certain value of α, no stable solutions can
be found. For higher values of α the ripple may show a sinusoidal modulation but,
for a further increase in α, the ripple increases exponentially with z. It has been
shown experimentally that for a shielded cathode ( K = 0 ) α < 0.57 [24]. Theoretical
analyses have suggested other values for this limit [23–​25]. In practice, however, the
27

272 Linear Electron Beams

Figure 7.11:  PPM focused beam profiles ( α = 0.3, K = 0.1): (a) β = 0.187, (b) β = 0.224,
and (c) β = 0.15 .
273

Periodic Focusing 273

Figure 7.12:  Dependence of β on α and K for minimum ripple of a PPM focused beam

(copyright 1955, IEEE, reproduced, with permission, from [23]).

Figure 7.13:  Dependence of ripple on α and K for minimum ripple of a PPM focused beam

(copyright 1955, IEEE, reproduced, with permission, from [23]).

requirement for the ripple to be small in a well-​focused beam means that the value
of α chosen is well below the limiting value.
Approximate analytical solutions to (7.88) have been described using Fourier
expansion or successive approximations [23, 25, 26], but these are mathematically
274

274 Linear Electron Beams

difficult. A  good approximation can be obtained, however, by assuming that the

beam profile is

Rm = 1 + δ cos ( 2β0 z ) , (7.92)

where Rm is the beam radius normalised to the mean radius. Following [26] we
express (7.87) in terms of the mean beam radius as

d 2 Rm 1 2 1 1
= β pm + β2Lm K m 3 − 2β2Lm cos 2 (β0 z ) Rm , (7.93)
dz 2 2 Rm Rm
where the suffix m denotes that the parameters are defined at the plane where the
beam has its mean radius, that is when β0 z = π 4. We note that the magnitude of the
magnetic field at that plane is B0 2 which is the RMS value of the magnetic field.
Substitution for Rm in (7.93) gives

β2pm β2Lm K
−4β20 δ cos ( 2β0 z ) = +
2 (1 + δ cos ( 2β0 z )) (1 + δ cos (2β z ))
0
3
(7.94)
− 2β 2
Lm cos 2
(β z ) (1 + δ cos (2β z )) .
0 0

When β0 z = π 4 (7.94) becomes

β2pm
+ β2Lm K m − β2Lm = 0 (7.95)
2
or

2β2Lm
β2pm = 2β2Lm (1 − K m ) = . (7.96)
m2
Comparing (7.96) with (7.54) we see that the equilibrium conditions for an electron
beam, having a given mean beam radius and cathode flux, are the same for both
solenoid and PPM focusing if the RMS PPM field is equal to the solenoid field.
Note that that (7.96) shows that there is a linear relationship between β pm and β Lm
for a given choice of K m within the range of validity of this analysis.
To find the amplitude of the ripple we set β0 z = π 2 in (7.94) and obtain

β2pm β2 K
4β20 δ = + Lm m3 . (7.97)
2 (1 − δ ) (1 − δ )

Substituting for β pm and rearranging gives

−1
  1 
α m = 2δ (1 − δ ) 1 − K m 1 − 2
 , (7.98)
  (1 − δ )  
where α m , given by

1 β2Lm 1
αm = = α (7.99)
2 β20 2
275

Periodic Focusing 275

has been defined by analogy with (7.89). Now because a = 2am and it can easily be
shown that

K m = 2 K (1 + δ ) (7.100)
4

so we may calculate the dependence of α on δ for different values of K for com-

parison with Figure 7.13. The comparison shows good agreement for α ≤ 0.4 so this
approximate theory is valid for a ripple of up to 15%. Also, since

β2pm
β2p = , (7.101)
(1 + δ )2
we can solve (7.98) numerically for δ in terms of α m , and then calculate β in terms of
α and K using (7.96). When the results are plotted and compared with Figure 7.12
there is, again, good agreement for α ≤ 0.4 (see Worksheet 7.4). This range of
parameters covers the values which are generally used, so the approximate theory
can be used for design calculations.
We have seen, in Figure 7.11, that when a PPM focused beam is launched incor-
rectly, scalloping is superimposed upon the PPM ripple. The stiffness of a PPM
focused beam can be investigated in the same way as that for solenoid focusing
by multiplying the space-​charge coefficient β by a factor k = I I 0 . As before, we
assume that the axial variation of charge density is slow so that it is possible to
consider each part of the beam as being in equilibrium. It is then only necessary
to calculate the equilibrium radius as a function of the space-​charge and magnetic
field parameters.
From (7.93) d 2 Rm d θ2 = 0 when θ = π 4 so the equilibrium radius is the
solution of

1 1
βm + 2α m K m 3 − 2α m Rm = 0, (7.102)
Rm Rm
which is identical in form to (7.63). The stiffness of a PPM focused beam is, there-
fore, the same as that of the equivalent solenoid-​focused beam. Thus, a modulated
PPM focused beam remains in equal ripple flow but with the mean radius deter-
mined locally by the instantaneous beam current. This can be confirmed by numeri-
cal integration of (7.87) in which the current increases linearly with distance (see
Worksheet 7.4).
In some tubes the axial variation of the magnetic field is periodic but not sinusoidal
[27]. The focusing in these cases can be investigated by the use of Worksheet 7.4,
if a suitable function for the field profile is defined. A particular case is the use of
double period PPM focusing in which adjacent pairs of cells have the same polarity.
This is used when it is necessary to fit the periodicity of the focusing to the period-
icity of the RF structure surrounding the beam [28]. Long period focusing in which
the magnetic field is kept approximately constant in each of a series of sections
separated by abrupt field reversals is described in [29, 30].
276

276 Linear Electron Beams

7.6.2 Practical PPM Focusing Systems

PPM focusing systems are sensitive to variations from the ideal field. These may be
caused by:
• variations in the strengths of the magnets caused by manufacturing tolerances or
accidental demagnetisation when the stack is being assembled;
• variations in the strength of the field in the cells near the ends of the stack caused
by end effects [31], or by the omission of magnets to allow input and output wave-
guides to pass through [32];
• variations in the azimuthal symmetry of the fields caused by tilting of the pole-​
pieces, misalignment of the magnets, or omission of magnets as already men-
tioned [32].
It is standard practice to make empirical adjustments to the magnetic field of a
PPM focused tube by the addition of soft iron shunts that bridge the pole-​pieces
on the outside of the stack. In this way both the magnitude and the symmetry of
the field can be adjusted until good beam transmission is obtained. The process is
time-​consuming and it is desirable to design PPM focusing systems in such a way
that empirical adjustments are kept to a minimum.
Problems caused by end effects, or by the need to pass waveguides through the
PPM structure can be dealt with, in principle, by adjusting the design of the cells
near to ends of the stack to compensate for the departures from the ideal field. Kory,
who examined the magnitudes of these effects for a helix TWT, found that the great-
est changes were caused by the omission of magnets to allow for the passage of
waveguides [32]. The asymmetries in the field caused appreciable movement of the
axis of the beam away from the axis of the stack. The effects of a simulated shunt
were small, and those of misalignment of a magnet were found to be negligible.
The effects of random variations in the strengths of the magnets have been dis-
cussed by a number of authors [33–​36]. It was found that such variations cause the
ripple in the beam to grow with distance. The growth can be reduced by increasing
the strength of the magnetic field above the theoretical value. This is in qualita-
tive agreement with experimental observations. The most comprehensive study was
carried out by MacGregor who examined the conditions for optimal beam trans-
mission with random variations in magnet strengths of 2.5% and 5%. Figure 7.14
summarises his results using Harker’s parameters [33, 34]. McGregor also showed
that good beam transmission would be difficult to obtain with values of α greater
than 0.4, when the field variations were 2.5%, and 0.3 when they were 5%.
The design of a PPM focusing system starts from the desired current, voltage,
and mean radius of the beam. It is also necessary to fix the maximum acceptable
ripple in the beam and the value of m which will give adequate stiffness. The design
calculations are:
i) Calculate the Brillouin field using (7.55) and the peak field B0 = 2 mBB. In
order to ensure good beam transmission it is usual to increase the magnetic field
from this theoretical value as described above (see Figure 7.14).
ii) Calculate K m using (7.57).
27

Periodic Focusing 277

Figure 7.14:  Regions of acceptable beam transmission with 2.5% and 5% random variations
of magnet strength compared with the minimum ripple solution (dashed): (a) K = 0,
(b) K = 0.1, and (c) K = 0.2
(copyright 1986, A. MacGregor, reproduced, with permission from [34]).

iii) Calculate α m using (7.98) and check that the value is less than 0.2.
iv) Calculate β0 and then L using (7.99).
v) Design the magnetic field in the beam entry region to achieve the correct initial
conditions as described in Section 9.3.2.
vi) Design the magnet stack as discussed in Section 19.7.3.
Finally, in nearly every case, it is necessary to make empirical adjustments to the
field in each tube by adding shunts to the outside of the PPM stack to minimise the
intercepted current.

7.6.3 Periodic Electrostatic Focusing

A periodic electrostatic focusing system can be constructed as shown in Figure 7.9
using electrostatic lenses [21, 37]. If cylindrical symmetry is assumed then the non-​
relativistic equations of motion are, from (7.9) and (7.11)

d 2r e
=− Er (7.103)
dt 2
m0
278

278 Linear Electron Beams

and

d2z e
=− E z . (7.104)
dt 2 m0
The angular velocity is zero in the absence of a magnetic field. Note that it is no
longer possible to assume that the axial velocity is constant. Let us suppose that the
electrostatic potential on the axis relative to the cathode is given by

V ( 0, z ) = Va + V1 1 − cos (β0 z ) . (7.105)

This potential could be provided by a series of equally spaced plates, with circular
holes for the passage of the beam, that are held alternately at potentials Va and
(Va + 2V1 ) relative to the cathode. From the discussion of the Einzel lens in Section
7.3.2 we expect this potential variation to correspond to a series of converging
lenses. The higher-order terms in the Fourier expansion of the potential at the edge
of the apertures decay rapidly with decreasing radius. Thus the potential variation
close to the axis can be represented by a single term from the series. Substituting in
Laplace’s equation (7.19) we obtain

1 ∂  ∂V 
r  − β0V = 0. (7.106)
2
r ∂r  ∂r 
This is a form of Bessel’s equation which has the general solution

V ( r, z ) = V1I 0 (β0 r ) cos (β0 z ) , (7.107)

where I 0 (β0 r ) is the modified Bessel function of the first kind [38]. If β0 r ≤ 1 the
Bessel function can be written to a good approximation as

I 0 (β 0 r )  1 +
(β0 r )2 . (7.108)
4
Then, in the paraxial ray approximation, the axial electric field is given by

∂V
Ez = − ≈ −β0V1 sin (β0 z ) (7.109)
∂z
and the radial electric field is given by

∂V 1
Er = − ≈ V1β20 r cos (β0 z ) . (7.110)
∂r 2
When the radial field component is added to the radial component of the space-​
charge field from (7.3) and substituted in (7.103) the result can be expressed as

d 2R 1 1
= β − U1R cos ( Z ) , (7.111)
dθ 2
R 4
279

Periodic Focusing 279

where Va is the accelerating potential, R = r b , θ = β0 u0 t, Z = β0 z, β is defined by

(7.90) and the strength of the lenses is represented by the normalised potential
U1 = V1 Va . Similarly (7.104) can be written as

d 2Z 1
= U1 sin ( Z ) . (7.112)
d θ2 2
This pair of equations corresponds to equation (7.88) for PPM focusing. The prop-
erties of this system can be investigated by the simultaneous numerical integra-
tion of (7.111) and (7.112) with the initial conditions R = 1, dR d θ = 0, Z = 0 and
dZ d θ = 1 − U1 . Stable minimum ripple solutions can be found in the same man-
ner as for PPM focusing, as shown in Figure 7.15 (see Worksheet 7.5). In the case
shown, the potentials of alternate electrodes are positive with respect to the poten-
tial at the input and the beam radius is always less than or equal to the initial radius.
If the potentials of those electrodes are made negative by reversing the sign of U1
then minimum ripple solutions can be found in which the minimum radius is equal
to the initial radius.
Figure  7.16 shows the relationship between β and U1 for minimum ripple, and
Figure  7.17 shows the dependence of the ripple on the normalised potential.
Figure  7.17 is similar to the curve for K = 0 in Figure  7.13. However, from the
definition of the normalised potential (U1 ), it can be seen that for 10% ripple
the variation of potential on the axis is nearly 80% of the accelerating voltage. The
voltage required on off-​axis electrodes would be appreciably higher. Comparison
between Figure 7.16 and Figure 7.12 shows that the value of β in a periodic elec-
trostatic system is much less than that in a PPM system with the same ripple. Thus,
in the electrostatic system, the space-​charge in the beam must be much less, and/​
or the spacing between the lenses must be smaller, than in a PPM system. It should
also be noted that the electrostatic system has only one converging lens per period
of the field whereas a PPM system has two. Finally we note that, when alternate
electrodes are positive, the mean velocity of the electrons is appreciably greater than
the initial velocity. For further information see [39, 40]. In practice, the need to use

Figure 7.15:  Example of a minimum ripple solution for periodic electrostatic

focusing: (β = 0.01, U1 = 0.35).
280

280 Linear Electron Beams

Figure 7.16:  Periodic electrostatic focusing parameters for minimum ripple solutions.

Figure 7.17:  Periodic electrostatic focusing: dependence of the beam edge ripple on the

normalised electrostatic potential.

high voltages, and the difficulty of combining the lens electrodes with the radiofre-
quency structure of a tube has, meant that the method has only been used for a few
low power tubes [8, 37].

7.7 Other Forms of Linear Electron Beam

Although most linear beam tubes use solid, cylindrical, electron beams, these
suffer from the disadvantage that the diameter of the beam must be kept small to
minimise the variation in the field of the radio-​frequency structure over the beam.
That, in turn means that the beam current is limited by the need to focus the
beam. Two other arrangements, described below, are sometimes used that address
this issue by concentrating the electrons in a region of strong radio-​frequency
electric field.
281

Imperfections in Electron Beams 281

7.7.1 Sheet Electron Beams

A sheet electron beam provides an alternative way of achieving a high current with
an acceptable current density. This type of beam has become the subject of intense
interest in recent years because of its potential application in devices at frequencies
higher than 100 GHz where the size of the radiofrequency structure becomes very
small. It is then advantageous to employ the interaction between a planar struc-
ture and an electron beam in the form of a thin sheet passing close to its surface.
Unfortunately, this kind of electron beam can become unstable [41] because the
space-​charge fields caused by perturbations of the sheet normal to its plane act to
increase those perturbations. It is found that the edges of the sheet tend to curl so
that the uniformity of the beam is destroyed [42]. This is a particular example of
the diocotron instability which occurs when a thin electron beam is controlled by
crossed electric and magnetic fields [12]. However, it is possible to achieve stable
flow of a sheet beam by arranging that the shape of the beam tunnel is such that
the forces tending to produce instability are minimised [42, 43]. Stable flow has also
been produced by using periodically cusped magnetic fields [44, 45], or with peri-
odic permanent magnet fields [46].

7.7.2 Annular Electron Beams

An ideal annular beam has a uniform current density within an annulus centred on
the axis. This type of beam has the advantage that it can concentrate the electrons
in a region where the radio-​frequency electric field of the surrounding structure is
strong. It therefore strengthens the interaction between them. Like sheet beams,
annular beams have a tendency to become unstable at high current densities [41,
47, 48]. An annular beam can be thought of as equivalent to a sheet beam in which
periodic boundary conditions are applied at the edges. Although annular beams
have been used in high power klystrons [49] the additional complexities involved in
creating them and focusing them are not generally worthwhile. It has been shown
that PPM focused annular beams are inherently unstable [50]. An analysis by Lau
[51], which included relativistic effects, showed that at high beam energies the nega-
tive mass and cyclotron maser instabilities are more important than the diocotron
instability. These are exploited in fast-​wave devices such as the gyrotron where it
is found that stable transport of annular beams is possible when the space-​charge
density is small.

7.8 Imperfections in Electron Beams

The theory and design of microwave linear-​beam tubes is usually based on the
assumption that the electron beam has uniform axial velocity and current den-
sity, and that the electrons flow without scalloping. In practice, all real electron
beams depart from these ideals to some extent, and the imperfections cause the
28

282 Linear Electron Beams

performance of the tube to depart from the predictions of theory. It is generally

possible to produce an electron beam whose properties are close to the theoreti-
cal ideal by careful design of the electron gun and the beam entry conditions, as
described in Chapter 9. Here we consider briefly the effects of thermal velocities,
and ion trapping, which may be significant even when the beam is well collimated.

7.8.1 Thermal Velocities

Since the electrons in the beam are normally obtained from a thermionic cath-
ode, which is operated at a temperature of around 1000 °C, they are not actually
mono-​energetic but have an energy spectrum with a width of rather less than 1 eV.
The variation in the axial velocity is, therefore, very small and it does not normally
have much effect on the operation of the tube other than contributing to the noise
signal at the output. The radial velocity components, though small, tend to cause
the beam to have a ‘halo’ outside the theoretical beam edge. These electrons may
be collected on the RF structure so that there is a small, but non-​zero, body current
even in a tube which should, theoretically, show no such interception [52–​55] (see
Section 9.2.4). Further information can be found in [56].

7.8.2 Trapped Ions
The potential well, caused by the electronic space-​charge, acts as a trap for any
positive ions produced by collisions between the electrons and any residual gas mol-
ecules in the tube. This is not normally a problem in pulsed tubes because there is
insufficient time for ions to accumulate during the pulse. Any ions which have been
formed tend to drift to the tube walls between pulses. In continuous wave (CW)
tubes the accumulation of ions can lead to:
• Disruption of the focusing because the ionic space-​charge tends to neutralise
the electronic space charge. In extreme cases where the background gas pres-
sure is high it has been known for a tube to continue to operate without a
magnetic focusing field. The trapping, and de-​trapping, of ions can lead to
relaxation oscillations which modulate the output of the tube at low frequen-
cies [57–​60].
• Ion plasma oscillations which are the result of a two-​stream interaction between
the electron beam and the stationary cloud of ions. The ion plasma frequency is
much lower than the electron plasma frequency because of the greater masses of
the ions. Ion plasma oscillations typically show themselves as a modulation of the
output of the tube at a frequency of a few megahertz [61].
• Bombardment of the cathode by ions that drift back into the field of the electron
gun and strike it with high energies. The result is usually seen as a damaged spot
in the centre of the cathode. In some tubes a hole is provided in the centre of
the cathode to allow any ions to pass through and be collected harmlessly on an
electrode provided for the purpose. An alternative strategy is to provide an ion
trapping electrode close to the electron gun which provides a potential barrier
283

References 283

to the flow of ions towards the gun. That has the disadvantage that the density
of ions trapped in the beam region will increase resulting in the effects noted in
(a) and (b) above.
On the whole the best solution to the problems caused by ions is to try to reduce their
generation, as far as possible, by maintaining a low background gas pressure in the
tube. This requires careful attention to the outgassing of the tube by baking it for a
sufficient time when it is evacuated. In large tubes it is normal for one, or two, ion
pumps to be permanently attached to the tube to maintain the gas pressure when it is
in storage, and to enable gas bursts to be monitored. In a working tube the electrons
tend to sweep ions into the collector where they become embedded in the surface. For
this reason the gas pressure tends to drop when a tube is operated for a long period.

References

[1]‌ V. Srivastava and R. G. Carter, ‘Effect of boundaries on the space charge potential
in coupled cavity travelling wave tubes’, IEE Proceedings I:  Solid-​State and Electron
Devices, vol. 133, pp. 185–​188, 1986.
[2]‌ J. D. Lawson, The Physics of Charged-​ Particle Beams. Oxford:  Oxford University
Press, 1977.
[3]‌ S. E. Tsimring, Electron Beams and Microwave Vacuum Electronics. Hoboken, NJ: John
Wiley & Sons, 2006.
[4]‌ W. J. Kleen, Electronics of Microwave Tubes. New York: Academic Press, 1958.
[5]‌ B. I. Bleaney and B. Bleaney, Electricity and Magnetism. London: Oxford University
Press, 1957.
[6]‌ S. Ramo et al., Fields and Waves in Communication Electronics. New York: Wiley, 1965.
[7]‌ J. F. Gittins, Power Travelling-​Wave Tubes. London: English Universities Press, 1965.
[8]‌ M. J. Smith and G. Phillips, Power Klystrons Today. Taunton, England:  Research
Studies Press Ltd., 1995.
[9]‌ L. Brillouin, ‘A theorem of Larmor and its importance for electrons in magnetic fields’,
Physical Review, vol. 67, pp. 260–266, 1945.
[10] W. Neugebauer, ‘Equlibrium solutions for partially immersed relativistic electron
beams’, IEEE Transactions on Electron Devices, vol. ED-​14, pp. 686–​693, 1967.
[11] W. Neugebauer, ‘Equlibrium solutions for partially immersed relativistic electron
beams’, IEEE Transactions on Electron Devices, vol. ED-​15, p. 43, 1968.
[12] A. H.  W. Beck, Space-​ Charge Waves and Slow Electromagnetic Waves. London:
Pergamon Press, 1958.
[13] E. A. Ash, ‘Compression and reflection of Brillouin beams’, Journal of Electronics and
Control, vol. 15, pp. 401–​417, 1963.
[14] O. P. Gandhi and N. C. Vaidya, ‘Behavior of electron beams under magnetic compres-
sion’, Proceedings of the IEEE, vol. 52, pp. 1052–​1053, 1964.
[15] N. C. Vaidya and O. P. Gandhi, ‘Performance of magnetically compressed O-​type elec-
tron beams emitted from nonshielded cathodes’, IEEE Transactions on Electron Devices,
vol. 13, pp. 453–​458, 1966.
[16] K. Amboss, ‘Studies of a magnetically compressed electron beam’, IEEE Transactions
on Electron Devices, vol. 16, pp. 897–​904, 1969.
284

284 Linear Electron Beams

[17] J. A. Seeger, ‘Magnetic compression of axially symmetric Brillouin-​focused electron

beams’, IEEE Transactions on Electron Devices, vol. 16, pp. 1–​5, 1969.
[18] W. B. Herrmannsfeldt, ‘Numerical design of electron guns and space charge limited
transport systems’, Nuclear Instruments and Methods in Physics Research, vol. 187,
pp. 245–​253, 1981.
[19] K. R. Spangenberg, Vacuum Tubes. New York: McGraw-​Hill, 1948.
[20] J. R. Pierce, ‘Spatially alternating magnetic fields for focusing low-​voltage electron
beams’, Journal of Applied Physics, vol. 24, p. ​1247, 1953.
[21] A. M. Clogston and H. Heffner, ‘Focusing of an electron beam by periodic fields’,
Journal of Applied Physics, vol. 25, pp. 436–​447, 1954.
[22] J. T. Mendel et  al., ‘Electron beam focusing with periodic permanent magnet fields’,
Proceedings of the I.R.E., vol. 42, pp. 800–​810, 1954.
[23] K. J. Harker, ‘Periodic focusing of beams from partially shielded cathodes’, IRE
Transactions on Electron Devices, vol. ED-​2, pp. 13–​19, October 1955.
[24] J. T. Mendel et  al., ‘Electron beam focusing with periodic permanent magnet fields’,
Proceedings of the Institute of Radio Engineers, vol. 42, pp. 800–​810, 1954.
[25] K. K. N. Chang, ‘Stability of periodic-​field beam focusing’, Journal of Applied Physics,
vol. 27, pp. 1527–​1532, 1956.
[26] K. Ura and M. Terada, ‘Equivalence of periodic magnetic field to uniform magnetic
field in electron beam focusing’, IEEE Transactions on Electron Devices, vol. ED-​13,
pp. 930–​934, 1966.
[27] D. C. Buck, ‘Stability of a cylindrical electron beam in nonsinusoidal periodic magnetic-​
focusing fields’, IRE Transactions on Electron Devices, vol. 4, pp. 44–​49, 1957.
[28] S. K. Datta et al., ‘Stability analysis for electron beam transport in double-​periodic per-
manent magnet focusing structure’, in Eighth IEEE International Vacuum Electronics
Conference, Kitakyushu, Japan, pp. 357–​358, 2007.
[29] P. F. C. Burke, ‘Compensated reversed field focusing of electron beams’, Proceedings of
the IEEE, vol. 51, pp. 1653–​1659, 1963.
[30] W. Siekanowicz and J. Cash Jr, ‘Focusing of Brillouin electron beams by use of long-​
period magnetic fields’, in 1964 International Electron Devices Meeting, pp. 88–​90, 1964.
[31] M. J. Schindler, ‘The magnetic field and flux distributions in a periodic focusing stack
for TWTs’, RCA Review, vol. 21, pp. 414–​436, 1960.
[32] C. L. Kory, ‘Effect of geometric azimuthal asymmetries of PPM stack on electron beam
characteristics [TWTs]’, IEEE Transactions on Electron Devices, vol. 48, pp. 38–​44, 2001.
[33] R. G. Carter and A. MacGregor, ‘Effect of imperfections on periodic permanent magnet
focusing of electron beams’, in 2nd IEEE International Vacuum Electronics Conference,
Noordwijk, The Netherlands, pp. 211–​213, 2001.
[34] A. MacGregor, ‘Periodic permanent magnet focusing of electron beams’, PhD,
Engineering Department, Lancaster University, Lancaster, UK, 1986.
[35] R. R. Moats, ‘Calculations of beam trajectories under non-​ideal conditions of PPM
focusing’, in 1976 International Electron Devices Meeting, pp. 515–519, 1976.
[36] B. Minakovic, ‘Effect on an electron beam of variations in periodic permanent-​magnet
focusing systems’, Electrical Communication, vol. 38, pp. 415–​424, 1963.
[37] P. K. Tien, ‘Focusing of a long cylindrical electron stream by means of periodic electro-
static fields’, Journal of Applied Physics, vol. 25, pp. 1281–​1288, 1954.
[38] N. W. McLachlan, Bessel Functions for Engineers. Oxford University Press, 1954.
285

References 285

[39] P. B. Priestland and H. L. Hartnagel, ‘Theory of periodic electrostatic focusing of elec-

tron beams’, IEEE Transactions on Electron Devices, vol. 15, pp. 915–​935, 1968.
[40] P. B. Priestland and H. L. Hartnagel, ‘A further contribution to the theory of
electrostatic focusing’, IEEE Transactions on Electron Devices, vol. 16, pp.  803–​812,
1969.
[41] C. C. Cutler, ‘Instability in hollow and strip electron beams’, Journal of Applied Physics,
vol. 27, pp. 1028–​1029, 1956.
[42] K. T. Nguyen et al., ‘Intense sheet electron beam transport in a uniform solenoidal mag-
netic field’, IEEE Transactions on Electron Devices, vol. 56, pp. 744–​752, 2009.
[43] J. Pasour et  al., ‘Demonstration of a 100-​ kW solenoidally focused sheet electron
beam for millimeter-​wave amplifiers’, IEEE Transactions on Electron Devices, vol. 58,
pp. 1792–​1797, 2011.
[44] J. H. Booske et  al., ‘Stability and confinement of nonrelativistic sheet electron
beams with periodic cusped magnetic focusing’, Journal of Applied Physics, vol. 73,
pp. 4140–​4155, 1993.
[45] J. H. Booske and M. A. Basten, ‘Demonstration via simulation of stable confinement of
sheet electron beams using periodic magnetic focusing’, IEEE Transactions on Plasma
Science, vol. 27, pp. 134–​135, 1999.
[46] Z. L. Wang et al., ‘The conditions for stable sheet electron beams transport in periodic
permanent magnet fields’, Journal of Infrared Millimeter and TeraHertz Waves, vol. 31,
pp. 649–​658, June 2010.
[47] R. L. Kyhl and H. F. Webster, ‘Breakup of hollow cylindrical electron beams’, IRE
Transactions on Electron Devices, vol. 3, pp. 172–​183, 1956.
[48] J. R. Pierce, ‘Instability of hollow beams’, IRE Transactions on Electron Devices, vol. 3,
pp. 183–​189, 1956.
[49] E. K. Demmel, ‘Some studies on a high-​ perveance hollow-​beam klystron’, IEEE
Transactions on Electron Devices, vol. 11, pp. 66–​73, 1964.
[50] G. Dohler, ‘On the stability of hollow beams in PPM focused TWTs’, IEEE Transactions
on Electron Devices, vol. 28, pp. 602–​604, 1981.
[51] Y. Y. Lau, ‘A unified theory of the diocotron, cyclotron maser, and negative-​mass insta-
bilities’, IEEE Transactions on Electron Devices, vol. ED-​31, pp. 329–337, 1984.
[52] A. Szabo, ‘Thermal velocity effects in magnetically confined beams’, IRE Transactions
on Electron Devices, vol. 5, pp. 183–​185, 1958.
[53] J. L. Palmer and C. Susskind, ‘Effects of transverse velocities in magnetically focused
cylindrical electron beams’, in Third International Congress on Microwave Tubes,
Munich, pp. 456–​460, 1960.
[54] P. T. Kirstein, ‘On the effects of thermal velocities in two-​dimensional and axially sym-
metric beams’, IEEE Transactions on Electron Devices, vol. 10, pp. 69–​80, 1963.
[55] J. R. Hechtel, ‘Magnetic focusing of electron beams in the presence of transverse vel-
ocity components’, IEEE Transactions on Electron Devices, vol. 28, pp. 473–​482, 1981.
[56] G. Herrmann, ‘Optical theory of thermal velocity effects in cylindrical electron beams’,
Journal of Applied Physics, vol. 29, pp. 127–​136, 1958.
[57] R. L. Jepsen, ‘Ion oscillations in electron beam tubes; ion motion and energy transfer’,
Proceedings of the IRE, vol. 45, pp. 1069–​1080, 1957.
[58] A. D. Sutherland, ‘Relaxation instabilities in high-​perveance electron beams’, IRE
Transactions on Electron Devices, vol. 7, pp. 268–​273, 1960.
286

286 Linear Electron Beams

[59] C. B. Thorington, ‘Computer simulation of ion trapping and detrapping in a

PPM focused traveling wave tube’, IEEE Transactions on Electron Devices, vol. 48,
pp. 56–​61, 2001.
[60] W. Tighe et al., ‘Transient ion disturbances in traveling wave tubes’, IEEE Transactions
on Electron Devices, vol. 48, pp. 82–​87, 2001.
[61] R. G. Carter and R. H.  C. Newton, ‘An experimental study of charge densities and
radio-​frequency propagation characteristics in an ion beam/​plasma system’, J. Phys. D,
vol. 7, pp. 1670–​1676, 1974.
287

8 Electron Flow in Crossed Fields

8.1 Introduction

This chapter reviews the theory of electron motion in crossed fields in both planar
and cylindrical geometries. The effects of relativity are neglected because they are
not normally important in commercial crossed-​field devices. Relativistic magne-
trons, which are outside the scope of this book, are discussed in [1].
The equation of motion for a non-​relativistic electron moving in electric and
magnetic fields is

d 2r e  dr 
=−  E + × B . (8.1)
dt 2
m0 dt

The force exerted by the magnetic field is always perpendicular both to the direc-
tion the magnetic field, and to the motion of the electron, as shown by the vector
product. If the term in brackets on the right-​hand side of the equation is zero then
the electron will move in a straight line with constant velocity. This can be achieved
by arranging that the electric field (the sum of externally imposed and space-​charge
fields), the magnetic field, and the direction of motion of the electron, are mutu-
ally perpendicular. This method of controlling the flow of electrons, using crossed
fields, provides an alternative to the use of a magnetic field parallel to the electron
velocity, discussed in the previous chapter.
Electron flow in crossed fields falls into four categories, as shown in Figure 8.1.
It can be seen that the electron motion is essentially two-​dimensional, and that
the beam is a sheet beam. The beam may be injected from an electron gun, similar
to that in a linear beam tube, or emitted from the whole surface of the negative
electrode (cathode). The geometry may either be planar, or cylindrical. In injected
beam tubes, the potential of the negative electrode (sometimes known as the sole)
may differ from that of the cathode from which the electrons are emitted (see
Section 8.2.2). The positive electrode, incorporating the RF structure with which
the electrons interact, is known as the anode (or sometimes the line). This chapter
examines the conditions necessary to achieve uniform, stable, electron flow in both
planar and cylindrical geometries of both kinds. A review of the experimental evi-
dence shows the limitations of the theoretical treatments.
28

288 Electron Flow in Crossed Fields

Figure 8.1:  Types of crossed-​field device: (a) Planar injected beam, (b) Planar emitting

cathode, (c) Cylindrical injected beam, and (d) Cylindrical emitting cathode.

Crossed-​field tubes include magnetron oscillators (see Chapter 15), and crossed-​

field amplifiers and oscillators (see Chapter  16). In the older literature they are
sometimes referred to as ‘M-​type’ tubes, from their description in French as ‘Tubes
a Ondes Magnetique’. All tubes which are of continuing importance are of the
cylindrical, emitting cathode, type. Planar geometry has only been used in a few
experimental devices because cylindrical tubes are smaller, and lighter, for the same
performance. Injected beams, which tend to be unstable, were used in some low
power devices but are now, largely, obsolete.

8.2 Crossed-​Field Electron Flow in Planar Geometry

Electron flow in planar geometry is considered in this section, and the following
one, because it provides useful insights into the properties of electrons in crossed-​
field flow. The cylindrical case, which is of greater practical importance is discussed
in Sections 8.4 and 8.5. Figure 8.2 shows a planar magnetron diode immersed in a
uniform magnetic field normal to the plane of the paper. An electron emitted from
the cathode is accelerated towards the anode by the electric field but the effect of the
magnetic field is to bend the trajectory as shown. As the magnetic field is increased,
at constant anode voltage, a point is reached where the electron can no longer reach
the anode. The current through the diode is then zero and it is said to be cut-​off or
magnetically insulated. Alternatively, the current ceases to flow if the anode voltage
is reduced, at constant magnetic field. Then the electron just grazes the surface of
the anode at the cut-​off potential Vc .
289

Crossed-Field Electron Flow in Planar Geometry 289

Figure 8.2:  Crossed-​field electron trajectory in planar geometry.

8.2.1 Electron Motion without Space-​Charge

Let us suppose that the magnetic field is in the z direction and that the electron is
moving in the ( x, y ) plane. Then from (8.1) the equations of motion in Cartesian
co-​ordinates are

d 2x e dy dy
=− Bz = −ω c (8.2)
dt 2 m0 dt dt
d2y e e dx  dx E y 
=− Ey + Bz = ω c  − , (8.3)
dt 2
m0 m0 dt  dt Bz 

where ω c = eBz m0 is the cyclotron frequency. It is easy to show by substitution that

a solution of these equations, in the absence of an electric field, is

x = −r sin ( ω c t )
(8.4)
y = r cos ( ω c t ) ,

when the initial position of the electron is ( x, y ) = ( 0, r ). Equations (8.4) are just
the parametric form of a circle centred on the origin. The initial vector veloc-
ity is ( x , y ) = ( −rω, 0 ).
When a uniform electric field is added in the y direction equation (8.3) shows that
the centre of rotation, known as the guiding centre, moves with uniform velocity
given by
Ey
u0 = .
Bz (8.5)

By extension of (8.4) the solution of (8.2) and (8.3) is

x = −r sin ( ω c t ) + u0 t
(8.6)
y = r cos ( ω c t ) ,

where the initial conditions are as before, and the initial velocity components are

x = u0 − ω c r,
(8.7)
y = 0.
290

290 Electron Flow in Crossed Fields

The electron trajectories calculated from (8.6) are shown in Figure 8.3 for a number
of different initial velocities. It can be seen that the choice of initial position of the
electron with respect to the guiding centre (which lies on the x axis) determines the
initial velocity and that the motion of the electron is cycloidal. Strictly speaking
the trajectory in Figure 8.3(a) is a lengthened cycloid, 8.3(b) is a cycloid and 8.3(c)
is a shortened cycloid [2]. When r = 0 then the electric and magnetic forces are bal-
anced, x = u0, and the electron moves in a straight line parallel to the x axis. The
solutions for other initial conditions can be derived from (8.6) by changes of origin
(see Worksheet 8.1).

Figure 8.3:  Electron trajectories in crossed fields in Cartesian geometry.

291

Crossed-Field Electron Flow in Planar Geometry 291

Integrating (8.2) with respect to t gives

e
x = − Bz y = −ω c y, (8.8)
m0

where it has been assumed that x = 0 when y = 0. It can be seen that x depends only
on y and Bz . By the principle of conservation of energy the relationship between the
electron velocity and the electric potential with respect to the cathode (V ) is
m0 2
V= (x + y 2 ) . (8.9)
2e
Substituting for x from (8.8) gives
m0 2
V= ( y + ωc2 y2 ) . (8.10)
2e
When y = 0 the electron is moving parallel to the anode and (8.10) becomes
m0 2 2
V= ω c y . (8.11)
2e
In particular, when the diode is just cut-​off y = h, and we obtain the Hull cut-​off
voltage (VH ) for a planar diode [3]
e
VH = Bz2 h2 . (8.12)
2 m0

The derivation of this equation is independent of the space-​charge density.

In order to use crossed fields to control the flow of electrons in a planar device
we envisage an arrangement such as that shown in Figure 8.4. The electrons start
from rest at a cathode at zero potential and are injected into a space in which
there is a uniform electric field in the negative y direction. The field is provided
by a planar anode at potential Va and a parallel electrode at a lower potential. To
achieve crossed-​field flow the magnetic field must be in the negative z direction. It is
immediately obvious that the electron velocity at x = 0 is not constant because the

Figure 8.4:  Injection of a sheet electron beam into crossed-​field flow.

29

292 Electron Flow in Crossed Fields

potential varies with y. In the absence of space-​charge the electric field is uniform
and, from the results of the preceding section, we see that motion in a straight line
is only possible for those electrons which satisfy (8.5) and which therefore enter the
focusing system at a particular value of y. All other electrons must exhibit cycloidal
motion with guiding centre velocity u0 .

8.2.2 Injected Beam with Space-​Charge

When space-​charge is taken into consideration the electric field is no longer inde-
pendent of y. We therefore wish to find whether a sheet electron beam can exist in
which all quantities are independent of x [4–​5]. The x component of the electron
velocity is given by (8.8). If the flow is uniform, the electric and magnetic forces
must balance, so that

 z .
E y = xB (8.13)

Eliminating x between (8.8) and (8.13) we obtain

e 2
Ey = − Bz y. (8.14)
m0

But the electric field must also satisfy the one-​dimensional form of Gauss’ Law
∂E y ρ
= . (8.15)
∂y ε0

Differentiating (8.14) with respect to y and combining it with (8.15) gives

2
eρ  e 
− = Bz , (8.16)
ε 0 m0  m0 

which may be written using (7.47) as

ω 2p = ω c2 . (8.17)

This is the condition for smooth Brillouin flow in a planar crossed-​field diode [6].
Note that, since the magnetic field is constant everywhere, (8.16) shows that the
charge density must be constant within the electron beam. The electron velocity is
not constant but depends upon y. The beam current density is

J x = ρx , (8.18)

where the x component of the velocity is obtained from (8.13) and (8.14) as

e
x = − Bz y, (8.19)
m0
293

Crossed-Field Electron Flow in Planar Geometry 293

hence

eρ
Jx = − Bz y. (8.20)
m0

Substituting for the charge density from (8.16) yields

2
 e 
J x = ε 0   Bz3 y. (8.21)
 m0 

Slater has pointed out that the current density can be substantial for quite
modest values of the constants [4]. For example, if Bz = 0.1 T and y = 10 mm,
then  J x = 277 A cm −2 .
To find the electrostatic potential within the beam we integrate (8.14) with respect
to y to give
e
V= Bz2 y2 + const. (8.22)
2 m0

This is identical to (8.11), which applies to all electrons moving parallel to the anode,
and therefore the constant of integration is zero. Outside the beam the charge den-
sity is zero and, therefore, from (8.15) the electric field is constant. Now the electric
field is continuous at the edges of the beam so that
e 2
Ey = − Bz d1 (0 ≤ y ≤ d1 ) (8.23)
m0

and

e 2
Ey = − Bz d (d ≤ y ≤ h) . (8.24)
m0

Therefore the potential of the anode is given by

e e 2
Va = Bz2 d 2 + Bz d ( h − d )
2 m0 m0

e  2h 
= Bz2 d 2  − 1 . (8.25)
2 m0  d 

We note that this expression is independent of d1 and that it therefore applies to all
cases including d1 = 0. Thus it is theoretically possible to inject an electron beam
into uniform flow in crossed electric and magnetic fields. The method by which
this can be accomplished is discussed in the Section 9.6.1. In order for the electron
velocity to be close to uniform it is necessary for the beam to be thin. However, it is
found that a beam of this kind suffers from an instability known as the diocotron
instability [7–​8]. Injected beam crossed-​field devices are therefore not common, and
tend to be of low power.
294

294 Electron Flow in Crossed Fields

The theory in this section can also be applied to devices with an emitting cathode
(see Figure 8.1(b)). The potential variation within the beam given by (8.22) with
the constant set to zero so that V = 0 when y = 0. Comparing (8.22) with (8.12) we
see that the potential on the upper surface of the beam is equal to the Hull cut-​off
voltage. Thus it is apparently possible for Brillouin flow to exist in a planar crossed-​
field device with a space-​charge layer close to the cathode. This raises the question
of how this condition can be set up, because the electrons emitted from the cathode
do not experience any force in the y direction.

8.3 The Planar Magnetron Diode

In a planar magnetron diode the whole of the zero-​potential electrode forms the
cathode as illustrated in Figure  8.1(b). This resembles the planar space-​charge
limited diode, discussed in Chapter 5, with the addition of a uniform transverse
magnetic field. We will assume that the problem is one-​dimensional so that all
quantities depend only upon y. The properties of the diode should tend to the
Child–​Langmuir solution as the magnetic field tends to zero. The difference is that
the magnetron diode becomes cut-​off for sufficiently high magnetic fields. Two
cases must therefore be examined depending on whether the diode is conducting
or cut off.

8.3.1 The Diode Is Conducting

Poisson’s equation in one dimension is

d 2V ρ (8.26)
=− ,
dy 2
ε0

where the charge density ρ is related to the y component of the current density by

J y = ρy, (8.27)

provided that y ≥ 0 everywhere. Because all quantities only vary with y it follows
from the continuity equation that J y is constant. Substituting for V from (8.10), and
for ρ from (8.27), in (8.26)

d  dy  eJ y
y  y  + ω c2 y = − , (8.28)
dy  dy  ε 0 m0

which may be written

d 2 y eJ y
+ ω c2 y = − . (8.29)
dt 2
ε 0 m0
295

The Planar Magnetron Diode 295

This equation has the particular integral

eJ y
y = − (8.30)
ε 0 m0 ω c2

and the complementary function

y = A sin ( ω c t ) + B cos ( ω c t ) , (8.31)

where A and B are constants. Thus the general solution of (8.29) is

eJ y
y = A sin ( ω c t ) + B cos ( ω c t ) − . (8.32)
ε 0 m0 ω c2

Now y = 0 for an electron which leaves the cathode when t = 0 so

eJ y
B= (8.33)
ε 0 m0 ω c2

and

eJ y
y =
ε 0 m0 ω c2
(cos (ω t ) − 1) + A sin (ω t ).
c c
(8.34)

For space-​charge-​limited flow the normal component of the electric field is zero at
the surface of the cathode so that y = 0 when t = 0. Then A = 0 and

eJ y
y =
ε 0 m0 ω c2
(cos (ω t ) − 1).
c (8.35)

Integrating (8.35) with respect to t we obtain

eJ y
y=
ε 0 m0 ω3c
(sin (ωct ) − ωct ) , (8.36)

where y = 0 when t = 0. In the limit Bz → 0 the trigonometric functions in (8.35) and

(8.36) can be expanded as power series to give

e (−J y )
y→ t3 (8.37)
6 ε 0 m0

and

e (−J y )
y → t 2 . (8.38)
2 ε 0 m0
296

296 Electron Flow in Crossed Fields

Substituting for t in (8.38) and then for y in (8.10) we recover the Child–​Langmuir
Law (5.16) which can be written in the form
3
eJC 2  2e Va  2 (8.39)
= − ⋅ ,
ε 0 m0 ω3c h 9  m0 ω c2 h2 

where JC is the Child–​Langmuir current density in a planar diode in the absence of

a magnetic field and h is the distance between the cathode and the anode. It is not
possible to find an explicit solution of this kind for the magnetron diode. However,
eliminating J y from (8.35) using (8.36), gives

y 1 − cos ( ω c t )
= (8.40)
ω c y ω c t − sin ( ω c t )

and, substituting for ( y ω c y ) from (8.10), we find that

 1 − cos ( ω c ta ) 
2
2e Va (8.41)
⋅ 2 2 =  + 1,
m0 ω c h  ω c ta − sin ( ω c ta ) 

where ta is the transit time. Also, from (8.36) we find that

eJ y 1
=− . (8.42)
ε 0 m0 ω h 3
c ω c ta − sin ( ω c ta )

Thus, from (8.42), (8.39), and (8.41) we can compute the ratio J y JC as a function
of the normalised transit time ω c ta. At the Hull cut-​off condition y = 0 so that, from
(8.40), ω c ta = 2 π when y = h and J y JC → 9 4 π = 0.716 .
The ratio of the magnetic field to the cut-​off field ( BH ) at constant anode poten-
tial (Va ) can be written, using (8.12), as
−1
B 2  2eVa  (8.43)
= ,
BH2  m0 ω c2 h2 

which can be computed as a function of the normalised transit angle. We can there-
fore plot the normalised current against the normalised magnetic field as shown in
Figure 8.5. At cut-​off the curve has infinite slope as the current drops abruptly to
zero [9] (see Worksheet 8.2).
It is also interesting to enquire how the current varies with anode voltage at con-
stant magnetic field. From (8.10)
2
Va  y 
= 1+  a  , (8.44)
VH  ωc h 

so that, using (8.40) we can plot the normalised current against the normalised
voltage, with ω c ta as a parameter, as shown in Figure 8.6. The current is zero until
297

The Planar Magnetron Diode 297

Figure 8.5:  Variation of current with magnetic field in a planar crossed-​field diode

(copyright 1993, IEEE, reproduced, with permission, from [9]).

Figure 8.6:  Variation of current with applied voltage in a planar crossed-​field diode.

the anode voltage rises above the cut-​off voltage. It then increases with increasing
anode voltage and tends to the Child-​Langmuir limit.
To find an electron trajectory we substitute for y in (8.8) using (8.36) to give

eJ y
x = −
ε 0 m0 ω c2
(sin (ωct ) − ωct ). (8.45)

Integrating with respect to t gives

eJ y  1 2 2
x=  cos ( ω c t ) − 1 + ω c t  , (8.46)
ε 0 m0 ω 3
c 2
298

298 Electron Flow in Crossed Fields

Figure 8.7:  Normalised electron trajectory in a space-​charge limited magnetron diode which

is conducting.

where the constant of integration has been chosen so that the initial position of the
electron is at the origin. Normalised x and y co-​ordinates can be defined by

ε 0 m0 ω3c

(X ,Y ) = ( x, y) . (8.47)
eJ y

Figure  8.7 shows the normalised electron trajectory plotted using (8.36) and
(8.46) with normalised transit time ( ω c t ) as a parameter. At the maximum tran-
sit time the diode is just cut off and the trajectory is parallel to the x axis (see
Worksheet 8.2).

8.3.2 The Diode Is Cut-​Off

When the diode is cut-​off the net current in the y direction is zero. It is commonly
assumed that the electrons are in Brillouin flow so that the electrons flow parallel to
the cathode, as discussed in Section 8.2.2. There is then a space-​charge layer in the
range 0 ≤ y ≤ d, where d < h, so that no electrons are able to reach the upper elec-
trode and the diode is cut off. Eliminating Bz between (8.25) and (8.12) produces an
expression for the position of the surface of the space-​charge layer within the diode
in terms of the cut-​off voltage, and the anode voltage
Va d  d
= 2 −  . (8.48)
VH h  h

Equation (8.48) describes a parabola which passes through the origin and has a
maximum when d = h . The equation can be rearranged to give the dependence of
the height of the electron layer on the anode voltage

d V
= 1− 1− a . (8.49)
h VH
29

The Planar Magnetron Diode 299

From equation (8.12) we know that, for fixed h, the cut-​off voltage increases with
increasing magnetic field. It follows that the effect of increasing the magnetic field
at constant anode voltage is to reduce d. All practical crossed field devices are oper-
ated below cut-​off with a ratio d h of the order of 0.5.
Brillouin flow is not the only possible solution to this problem. If it is supposed
that the motion of the electrons is cycloidal, as in the space-​charge-​free case, then
alternative solutions exist. The electron flow is treated as the superposition of elec-
tron currents having equal and opposite y components J + and J − [4]. This is known
as two-​stream flow. Since the space-​charge density is independent of the direction
of motion of the electrons, each current component must be half the current in a
conducting diode. Thus (8.28) becomes

d  dy  2eJ +
y  y  + ω c2 y = − , (8.50)
dy  dy  ε 0 m0

where J + is y component of the outward flow of current. The solution to the prob-
lem can then be derived directly from that in the previous section by replacing J y by
2J + in (8.34) to give
2eJ +
y =
ε 0 m0 ω c2
(cos (ωct ) − 1) + A sin (ωct ) , (8.51)

where we have, again, assumed that the electron starts from rest on the cathode
surface. This type of solution was first discussed by Slater [4] who assumed space-​
charge limited emission from the cathode. Then the normal component of the elec-
tric field is zero at the cathode surface and A = 0. The normalised (Slater) electron
orbits can then be obtained from (8.47) with the substitution of 2J + for J y as shown
in Figure 8.8. The points of inflexion of the curve are branching points at which
the trajectory may continue to move upwards, or follow a downwards path, which
is the mirror image of the initial upward path. There is an infinity of such possi-
ble Slater orbits corresponding to transit angles from the cathode to the anode of

Figure 8.8:  Two-​stream electron orbit in a cut-​off planar magnetron diode.

30

300 Electron Flow in Crossed Fields

Figure 8.9:  Dependence of normalised charge density (ω 2p ω c2 ) on normalised position in a

cut off planar magnetron diode.

2 π, 4 π,  2 nπ,, where n is a positive integer. In each case the maximum value of y

is given by (8.12) so the trajectories become flatter with increasing n.
The charge density is found by substituting for y in (8.27) using (8.35) and the
result can be written as
ω c2
ω 2p = , (8.52)
1 − cos ( ω c t )

which can be compared with (8.17) for Brillouin flow. Using ω c t as a parameter
in (8.36) we can plot the variation in normalised charge density (ω 2p ω c2 ) with Y,
for two-​stream flow, as shown in Figure  8.9. The charge is found to be in layers
(described as striations by Slater [4]) with a minimum charge density half that of
Brillouin flow. The average charge density, found by integrating ρ with respect to
y over a whole number of layers, is identical to that for Brillouin flow. It follows
that the electric field between the top of the space-​charge layer and the anode is the
same in both cases. The fact that the charge density is infinite close to the cathode
resembles the situation in a space-​charge limited diode where it is necessary to sup-
pose that the thermal velocities of the electrons allow them to diffuse away from
the cathode.
The current flowing from the cathode can be calculated in the same way as for a
conducting diode. For the first Slater orbit this current is just one half of that for a
diode which is just conducting (see Figure 8.5). The results for the first few orbits
are given in Table 8.1. Slater suggested that Brillouin flow might be seen as the limit-
ing case when n → ∞ and the current emitted from the cathode falls to zero. Thus,
an infinite set of theoretical solutions exists for the electron flow in a cut-​off planar
magnetron diode, and it is not apparent which of these is correct in any particular
case. We shall see in Section 8.5 that the problems raised by two-​stream flow are
resolved when cylindrical geometry is considered.
301

Crossed-Field Electron Flow in Cylindrical Geometry 301

Table 8.1:  Ratio of the injected current to the

Child–​Langmuir current for the first few Slater orbits

n J+ /​ JC

1 0.358
2 0.179
3 0.119
4 0.090

Figure 8.10:  Arrangement of a cylindrical magnetron diode.

8.4 Crossed-​Field Electron Flow in Cylindrical Geometry

Figure 8.10 shows the arrangement of a cylindrical crossed-​field diode in which the

anode lies outside the cathode. The alternative arrangement with the anode inside
the cathode is not much used and will not be discussed further here. When the diode
is cut off it is assumed that the space-​charge is confined into a cylindrical layer (or
hub) surrounding the cathode.

8.4.1 Electron Motion without Space-​Charge

The analysis of a cylindrical, crossed-​field diode, in which the outer electrode is the
anode, proceeds in the same manner as that for a planar diode. The non-​relativistic
equations of motion of the electrons are (7.11) and (7.12)

d 2r e e 
=− Er + rθ2 − rθBz (8.53)
dt 2
m0 m0

and

dt
( )
d 2
rθ =
e
m0
 z.
rrB
(8.54)
302

302 Electron Flow in Crossed Fields

This equation can be integrated with respect to time to give

1
m0 r 2θ = eBz r 2 + const. (8.55)
2

If it is assumed that the electrons have zero angular velocity as they leave the
cathode then

1  r2 
θ = ω c 1 − c2  , (8.56)
2  r 

where rc is the cathode radius. Using the principle of conservation of energy, the
relationship between the electron velocity and the electric potential with respect to
the cathode is

V=
2e
(
m0 2 2 2
r + r θ .
) (8.57)

Substituting for θ from (8.56) gives

m0  1 2 2  rc2  
2

V=  
r 2
+ r ω 1 − . (8.58)
2e  4  r 2  
c 

When the diode is just cut off the electrons have zero radial velocity as they reach
the anode. Then the relationship between the anode potential at cut-​off (VH ) and
the magnetic field is
2
e  r2 
VH = Bz2 ra2 1 − c2  . (8.59)
8m0  ra 

This is the Hull cut-​off equation for a cylindrical diode (compare (8.12) for the
planar diode). Note that the derivation of this equation is similarly unchanged by
the presence of space charge.
To find the electron trajectories we note that, in the absence of space-​charge,
Va
Er = − . (8.60)
r ln ( ra rc )

Substituting for Er and for θ from (8.56) in (8.53) gives

d 2r e Va 1 2  rc4 
= ⋅ + rω c  4 − 1 . (8.61)
dt 2 m0 r ln ( ra rc ) 4 r 

Now, if we define the normalised variables R = r rc , A = ra rc , α = ω c t and

U = Va VH , then (8.61) and (8.56) can be written
2
d 2 R 1 A2U  1  1 1  1 
= 1 − 2  + R − 1 (8.62)
dα 2
8 ln ( A) A R 4  R4 
30

Crossed-Field Electron Flow in Cylindrical Geometry 303

Figure 8.11:  Typical electron trajectory in a cylindrical magnetron without space-​

charge ( A = 2, U = 0.5).

and
dθ 1  1 
= 1 − 2  . (8.63)
dα 2  R 

This pair of equations can be integrated numerically to find the trajectories of the
electrons. The trajectories are epicycloidal in the counter-​clockwise direction as
shown in Figure 8.11 (see Worksheet 8.1).

8.4.2 Injected Beam with Space-​Charge

We wish to find out whether Brillouin flow, in which all quantities are independent
of θ, can exist in the space between the two electrodes, as shown in Figure 8.10. If
the flow is uniform the radial acceleration of the electrons must be zero. Therefore,
from (8.53)

m0 2
Er = rθ − rθBz . (8.64)
e

Substituting for the angular velocity from (8.56) and collecting terms gives

e  r4 
Er (r ) = − Bz2 r 1 − c4  . (8.65)
4 m0  r 

Now Gauss’ Law in cylindrical co-​ordinates is

304

304 Electron Flow in Crossed Fields

1 ∂ ρ
(rEr ) = . (8.66)
r ∂r ε0

Substitution of Er from (8.65) into (8.66) gives

1 2  rc4 

ω 2p = ωc 1 +  . (8.67)
2  r4 

Comparison between (8.67) and the corresponding equation for a planar magne-
tron (8.17) shows that they are similar but that the charge density is not uniform in
the cylindrical diode.
To find the electrostatic potential within the space-​charge layer (8.65) is inte-
grated with respect to r to give
2
e  r2 
V= Bz2 r 2 1 − c2  , (8.68)
8m0  r 

since V = 0 when r = rc . We note that when r = ra this is identical to the Hull cut-​off
equation (8.59). The radial component of the electric field on the surface of the
space-​charge layer at r = rb is obtained from (8.65) as

e  r4 
Er = − Bz2 rb 1 − c4  . (8.69)
4 m0  rb 

Now, the electric field is continuous at r = rb and ∇.E = 0 in the space between the
surface of the space-​charge layer and the anode. The electric field in this region is,
therefore,

rb
Er = Er ( rb ) (rb ≤ r ≤ ra ) , (8.70)
r

where Er ( rb ) is obtained from (8.65). The potential difference between the surface
of the space-​charge layer and the anode is
ra
rb  ra 
Vab = − ⌠
 Er ( rb ) dr = − Er ( rb ) rb ln   . (8.71)
⌡rb r  rb 

The potential of the anode is, therefore,

2
e  r4  r  e  r2 
Va = Bz2 rb2 1 − c4  ⋅ ln  a  + Bz2 rb2 1 − c2  . (8.72)
4 m0  rb   rb  8m0  rb 

This equation is equivalent to equation (8.25) for the planar diode. Because that
equation is so much simpler than (8.72) it is commonly used instead with the sub-
stitutions d = rb − rc and h = ra − rc . When the results of this substitution are exam-
ined it is found that the difference between the two equations is up to 10% when
305

The Cylindrical Magnetron Diode 305

ra rc = 1.1 and that it increases as ra rc increases. However, ra rc > 1.1 in most crossed-​
field devices, so the use of the equation for planar geometry, as an approximation,
is hardly ever valid.

8.5 The Cylindrical Magnetron Diode

The behaviour of a cylindrical magnetron diode close to cut-​off resembles that of

the planar diode described in Section 8.3, but with some important differences. This
problem was studied in [4, 10–​11]. The equation of motion (8.53) can be written
using (8.56) in terms of normalised parameters as

d 2R 1
= 3 − R + Er′, (8.73)
dτ 2
R

where R = r rc , τ = ω L t, ω L = ω c 2 is the Larmor frequency, and the normalised

electric field Er′ is defined by

e Er
Er′ = . (8.74)
m0 rc ω 2L

When the diode is cut off there are equal and opposite currents J + and J − per unit
surface area of the cathode. Gauss’ Law gives

1d ρ
(rEr ) = , (8.75)
r dr ε0

where, as in Section 8.3.2, half of the charge density is associated with each of the
components of the current. Equation (8.75) can be written

d ρr
(REr ) = c R. (8.76)
dR ε0

The continuity equation is

rc 1 dr
J + = ρ , (8.77)
r 2 dt

which can be written as

1 dR
J+ = ρRrc ω L . (8.78)
2 dτ

Combining (8.76) and (8.78) gives

306

306 Electron Flow in Crossed Fields

d 2J
(REr ) = + , (8.79)
dR ε0 ωL u

where u = dR d τ , or

d 2J
(REr ) = + . (8.80)
dτ ε0 ω L

Then, integrating (8.79) along a trajectory

2J + R

REr − Ec =
ε0 ωL ∫
R =1
d τ, (8.81)

where Ec is the radial electric field on the surface of the cathode. Now, using the
normalised electric field defined in (8.74), equation (8.81) becomes

REr′ − Ec′ = J + ′ τ, (8.82)

where J + ′ is the normalised cathode current density given by

2 ( e m0 ) J +
J+′ = − . (8.83)
ε 0 ω3L rc

Then, substituting for Er′ from (8.82), equation (8.73) becomes

d 2R 1 E ′ J ′τ
= 3 − R + c + + . (8.84)
dτ 2
R R R
This equation can be solved numerically to give J +′ as a function of Ec′ and
Ra = ra rc . It is found that, if the normalised field at the cathode Ec′ = 0, equation

Figure 8.12:  The relationship between the cathode current density and the electric field on
the surface of the cathode in a cylindrical magnetron diode
(copyright 1973, IEE, reproduced, with permission, from [10]).
307

The Cylindrical Magnetron Diode 307

(8.84) has no solutions when Ra < 2.01. This is the condition which normally exists
in magnetron oscillators. Figure 8.12 shows the solutions of (8.84) for first-​order
orbits, that is, those where the radial component of the velocity is only zero on the
surface of the anode (see Worksheet 8.3). Higher-​order solutions can exist to the
left of the dashed envelope curve but these exist only for values of Ra greater than
those usually found in magnetron oscillators [10]. This analysis suggests that the
electrons in the space-​charge layer of a cut-​off cylindrical diode follow first-order
Slater orbits, and that the electric field on the surface of the cathode is greater than
zero. Numerical modelling of magnetrons supports this conclusion [12]. We saw,
in Section 8.3.2, that the average charge density is the same for both zero-order,
and first-order, space-​charge layers in a planar diode. A comparison between the
charge per unit length in the space-​charge hub of a cylindrical diode shows that
they are the same to within ±5% in the range 1.2 ≤ ra rc ≤ 2.0 (see Worksheet 8.3).
The current in a smooth bore magnetron diode is provided by thermionic emis-
sion. In a magnetron oscillator there is usually a combination of thermionic and
secondary electron emission. In either case, provided that the available current is
great enough, we expect that the operating point lies on the dashed envelope shown
in Figure  8.12. It is interesting to compare the current density injected from the
cathode with that in a diode obeying the Child–​Langmuir Law where the current
per unit length is given by (5.69)
3
8πε 0 2 ( e m0 ) V2
IC = ⋅ a2 , (8.85)
9 (raβ )
where β is a function of Ra . Using the same normalisations as before, and noting
that the anode voltage is given by (8.68), we obtain
3
4 R2  1  (8.86)
JC′ = − ⋅ 2a 1 − R 2  .
9 β a

Figure 8.13:  Ratio of the injected current, for first-order Slater orbits, to the Child–​Langmuir
current in a cylindrical magnetron diode which is just cut off.
308

308 Electron Flow in Crossed Fields

Figure 8.13 shows the ratio J + JC , plotted against the normalised anode radius,
for a diode which is just cut off (see Worksheet 8.3). When Ra → 1 the current ratio
tends to the value 0.358 obtained for a planar diode (see Section 8.3.2). It can be
seen that the current ratio varies by less than 10% over the whole range of Ra which
has been plotted. If the diode is more strongly cut off then Ra is replaced by the
normalised radius of the space-​charge layer Rb . When the diode is just conducting
then the current is doubled, as it is in a planar diode, with results very similar to
those in Figure 8.6 [11].
Early numerical simulations by Hartree suggested that two-​stream flow would
tend to Brillouin flow within a short time [7]. More recent simulations show that
two-​stream flow is unstable, when subject to small perturbations, leading to a flow
similar to Brillouin flow but with some turbulence [13]. Agafonov showed that
Brillouin flow could be seen as the limiting case of two-​stream flow [14]. These
conclusions, however, cannot explain key features of the experimental behaviour
of cut-​off magnetron diodes, which are reviewed in the next section. We shall see
that the experimental results mean that the principle of conservation of energy can-
not apply to individual electrons and, therefore, that neither the Brillouin nor the
two-​stream model can be correct. However, it should be noted that these models
are still of value in throwing light on the problem and also because the results are
found to be useful for modelling magnetron oscillators and emitting sole CFAs (see
Chapters 15 and 16).

8.6 Experimental Behaviour of Magnetron Diodes

When a magnetron diode is conducting the current is space-​charge-​limited, provided

that the cathode emission is sufficient, and it decreases as cut-​off is approached, as
predicted by theory [9, 11]. When the diode is cut-​off, however, it is found that
there are major effects which cannot be explained by the theories discussed in the

Figure 8.14:  Dependence of the current in a magnetron diode on the applied voltage, with
the curve from Figure 8.6 (dashed) for comparison.
309

Experimental Behaviour of Magnetron Diodes 309

Figure 8.15:  Characteristic curves of a smooth-​bore magnetron diode with a platinum

cathode (ra rc = 1.58)
(copyright 1951, AIP Publishing, reproduced, with permission, from [15]).

previous sections. Very early in the study of magnetron diodes it was noticed that
the cut-​off characteristic was not sharp as predicted by the theories reviewed above
[3]. Appreciable current continues to flow below the cut-​off voltage, as shown in
Figure 8.14, for a diode in which ra rc = 2.3 [11]. It was found that this effect cannot
be explained by the thermal velocities of the electrons.
All the early experiments employed diodes in which the cathode was a thin wire
and the anode voltage was a few hundred volts. These conditions are not typical of
the dimensions and voltages used in practical tubes. An important series of experi-
ments was carried out by Jepsen and Muller who observed that the currents in mag-
netron oscillators were often greater, by up to two orders of magnitude, than could
be sustained by thermionic emission from the cathode [15]. It was thought that this
was the result of secondary electron emission from the cathode. They constructed
a series of experimental magnetron diodes with smooth anodes whose dimensions
corresponded to those of an existing magnetron oscillator. In these diodes the cath-
ode was not heated but formed of one of a number of pure materials having dif-
ferent secondary electron emission properties. The temperature of the cathode was
not sufficient to allow any appreciable thermionic emission. A  small thermionic
priming cathode was included at one end of the main cathode. The results were
displayed in the form of curves of anode voltage against anode current at constant
magnetic field, as shown in Figure 8.15. These curves are very similar to those for
310

310 Electron Flow in Crossed Fields

an operating magnetron oscillator having similar dimensions. They curves reveal a

number of important features:
• The anode current was tens of amps, even though the anode voltage was always
less than the cut off voltage for the magnetic field used. This current was much
greater than that which could be supplied by the priming cathode, except when
the main cathode was made of carbon (whose secondary electron emission coef-
ficient is less than unity). The current was always much smaller than the space-​
charge-​limited current which would flow in the absence of the magnetic field.
• For each magnetic field it was found that, as the applied voltage was increased,
the current increased until it reached a maximum. If the voltage was increased
further then the current fell abruptly to approximately that available from the
thermionic priming cathode. This is a particular feature of tubes with pure metal
cathodes. The maximum current boundary was approximately a straight line and
depended on the dimensions of the tube and on the material of the cathode.
• The flow of current to the anode was accompanied by heating of the main cath-
ode caused by back-​bombardment.
Figure 8.16 shows typical results for the dependence of the back-​bombardment
power, normalised to the power input to the diode, on the anode current. At low
currents a high proportion of the input power is dissipated on the cathode, but that
this falls as the anode current is increased. It was noted that the back-​bombardment
power was greater in the smooth-​bore magnetron than in the equivalent magnetron
oscillator by a factor of at least 2. The results shown in Figure 8.16 were obtained
using a tube in which the ratio of anode diameter to cathode diameter was 1.58. It
was found that the back-​bombardment power decreased rapidly as this ratio was
increased.
Measurements were made of the currents in the priming cathode, the main
cathode, and the anode as a function of anode voltage, with the results shown in

Figure 8.16:  Back-​bombardment power in a smooth-​bore magnetron with a molybdenum

cathode (ra rc = 1.58).
(copyright 1951, AIP Publishing, reproduced, with permission, from [15]).
31

Experimental Behaviour of Magnetron Diodes 311

Figure 8.17:  Division of current between the priming cathode and the main cathode of a
smooth-​bore magnetron diode.
(copyright 1951, AIP Publishing, reproduced, with permission, from [15]).

Figure 8.17. It can be seen that up to a critical anode voltage all the currents are
small, and the main cathode current is negative, indicating that more electrons are
landing on the cathode than are emitted by it. As the voltage is increased a point is
reached at which the main cathode current changes sign and then increases rapidly.
The zero of the main cathode current was interpreted as being the point at which
the energy of the bombarding electrons corresponded to a secondary electron emis-
sion coefficient of unity. Thus, at higher voltages, the greater part of the anode cur-
rent is from secondary electron emission.
The maximum current boundary was explained qualitatively by observing that
the total current emitted from the cathodes must always be greater than, or equal
to, the anode current, that is

I p + I b ( δ e − 1) ≥ I a , (8.87)

where I p , I b and I a are, respectively, the priming cathode current, the back-​
bombardment current, and the anode current. The effective secondary electron
emission coefficient ( δ e ) takes account of the angles of incidence of the bombard-
ing electrons. The energy of the bombarding electrons increases as the anode volt-
age increases. However, as shown in Section 18.3, the secondary electron emission
coefficient decreases with increasing primary electron energy, and eventually falls
below unity. Thus, at some point, the inequality in (8.87) can no longer be satisfied,
as the anode voltage is increased, and the anode current must fall to around that
emitted by the priming cathode.
The work of Jepsen and Muller showed conclusively that the simple theory of the
magnetron diode was inaccurate. It was noted, on the basis of a very rough calcula-
tion, that electron scattering based on collisions failed to explain the observations
312

312 Electron Flow in Crossed Fields

Figure 8.18:  Typical noise/​voltage curves measured, with uncalibrated probes in a cavity

magnetron operated below oscillation
(copyright 1955, IEE, reproduced, with permission from [16]).

of current striking the cathode by a factor of around 106. Because the experiments
employed a diode in which the thermionic current was deliberately kept small
they are not representative of the situation in most practical devices where ther-
mionic emission plays a larger role. However, the similarity between the curves in
Figure 8.15, and the characteristic curves of a magnetron oscillator, suggest that
there may be similarities between the electron dynamics in the two cases.
An important series of measurements of the noise in cut off magnetrons was
made by Glass et  al. [16]. It was found that noise could be detected in fluctua-
tions of the anode current, or in noise power radiated by the heater leads, when a
magnetron oscillator was operated in its pre-​oscillating state. This noise was found
to contain clear frequency peaks whose amplitude and frequency depended on
the anode voltage. Frequency peaks were detected in 3 GHz tubes in the range
0.5 MHz to 4.0 GHz. Typical results are shown in Figure  8.18. The general
conclusions were:
• The noise, including the peaks, increased as the anode voltage was increased.
• There was always a noise maximum near the oscillating voltage.
• A given peak moved to higher frequencies as the anode voltage was increased.
When the voltages at which the peaks occurred were plotted against frequency it
was found that they lay on a series of curves diverging from the origin. Calculations
suggested that these lines corresponded to azimuthal modes with differing periodic-
ity. The conclusion of this paper was:
31

Experimental Behaviour of Magnetron Diodes 313

Figure 8.19:  Distribution of space-​charge in a cut-​off magnetron diode for various voltages

(copyright 1955, AIP Publishing, reproduced, with permission, from [19]).

‘The experimental results described above have shown the presence of oscillations in the
space-​charge cloud down to extremely low voltages. Thus for all practical purposes the
space-​charge can be considered as being always in a state of oscillation, and a true steady
state therefore never really exists. The question of whether a single-​or double-​stream steady
state is present is thus largely an academic one (my emphasis)’
(copyright 1955, IEE, reproduced, with permission from [16]).

Despite this, authors have continued to argue the claims of these theoretical states
to be the stable state existing in a magnetron diode. Further experimental evidence
on oscillations in magnetron diodes was presented by Osepchuk [17] who noted
that, although violation of cut-​off is not always accompanied by detectable oscilla-
tions, it was possible that oscillations were present under most conditions.
At the same period a number of authors reported measurements of the space-​
charge distribution between the cathode and the anode [18]. A very careful set of
results was obtained by Nedderman [19] who used the radiation from low-​pres-
sure helium within the magnetron to calculate the charge density. A  representa-
tive set of results is shown in Figure 8.19 with the edge of the space-​charge layer
according to the Hull equation indicated by a dotted line for each anode voltage.
The position of the cathode coincides with the left-​hand edge of the figure. The
method of measurement was unable to detect electrons having energies less than
about 20 eV and, for that reason, the space-​charge density appears to be very low
close to the cathode. Nedderman estimated the charge which had been missed by
the measurements and showed that it could mostly be accounted for by continuing
the curves to the left of their peaks with constant charge density. It is evident from
these results that the space-​charge occupies most of the space between the edge
of the theoretical space-​charge layer and the anode. This is particularly noticeable
314

314 Electron Flow in Crossed Fields

when the diode is strongly cut off. Further measurements were made at a reduced
cathode temperature and at voltages too low for secondary emission to occur. It
was found that, though the space-​charge was reduced by this process, it was still
far from zero. The conclusion was drawn that electrons were being trapped by the
field and unable to reach either the anode or the cathode. Such electrons might be
expected to remain there for long periods and gradually drift out of the field under
the influence of large numbers of small perturbations. It was concluded that the
experimental results could not be explained by theories of the magnetron diode
based on self-​consistent fields with unique electron trajectories.
More recent experimental results were reported by Hirsch [20–​22] who studied
strongly cut-​off magnetrons. It was shown that the anode current distribution was
not uniform but was concentrated towards the ends of the anode, suggesting that
any satisfactory theory would need to be three-​dimensional. The effects of back-
ground gas pressure were investigated and it was shown that the anode current
increased because of scattering from gas molecules at pressures greater than 10–​5
Torr. At low currents oscillations were observed at very well-​defined frequencies
with very little noise. The frequency of each mode of oscillation was directly pro-
portional to the magnetic field, and independent of the anode voltage. Most of
these measurements were made on a tube which had a filamentary cathode, and a
large ratio of anode to cathode radius, at voltages of a few hundred volts and cur-
rents of a few microamps.
Direct measurements of the potential distribution in a strongly cut-​off magne-
tron showed the presence of space-​charge outside the theoretical space-​charge layer
[23]. It was shown that the potential curve coincided with the theoretical curve for
Brillouin flow close to the cathode, but diverged at greater distances. The radius at
which the divergence commenced was around 40% of the theoretical radius of the
space-​charge cloud. The results of the measurements were said to be in agreement
with those of Nedderman [19].

8.7 The Magnetron Problem

It is evident that the experimental results obtained with cut-​off magnetron diodes
cannot be explained by the elementary theory developed above. This problem is the
subject of an extensive literature with divided opinions. In the discussion below,
the papers cited are those which appear most important in the development of the
argument. According to the elementary theory it is impossible for any electrons to
reach the anode when the diode is cut off. This is a consequence of the assumption
that the principle of conservation of energy applies to each electron, and to the
assumption that the flow does not vary with time. The electrons on the surface of
the space-​charge layer have no radial acceleration because the electric and magnetic
forces are in balance. If any of these electrons are to be able to reach the anode they
must first lose energy so that the magnetic force is reduced. This can only happen if
they exchange energy with other electrons which are speeded up, and so returned to
315

References 315

the cathode, which they strike with non-​zero energy. This explains the simultaneous
appearance of anode current and back-​bombardment of the cathode. It follows
that while energy and momentum must be conserved for the electron cloud, as a
whole, these quantities cannot be conserved for individual electrons [24]. Thus it is
impossible for any theory based on the conservation of energy by individual elec-
trons to explain the experimental results. Three possible causes of energy exchange
have been proposed in the literature:
• two-​or three-​body collisions between electrons [25–​28];
• cumulative effects of random thermal oscillations [25–​26, 29];
• oscillations of the whole electron cloud [7, 17, 21–​22, 24, 27, 30–​33].
The consensus appears to be that electron collisions are too rare, and random ther-
mal effects too small, to explain observed phenomena. The importance of collective
instabilities is supported by the existence of peaks in the noise spectrum which can
be explained in terms of azimuthal space-​charge modes (see also Section 15.2.3).
It may be noted that those authors who believe that collisions are important also
refer to collective effects. The explanation of the behaviour of cut-​off magnetrons
in terms of space-​charge oscillations has been confirmed by PIC simulations [12,
32–​33] which also showed the importance of non-​stationary secondary electron
emission from the cathode. In a 2D simulation, agreement was obtained with exper-
imental results to within 10% when the spacing between the cathode and the anode
was small compared with their length [15, 33].

References

[1]‌ J. Benford et al., High Power Microwaves. Boca Raton, FL: CRC Press, 2015.
[2]‌ I. N. Bronshtein and K. A. Semendyayev, A Guide-​Book to Mathematics. Frankfurt and
Zurich: Verlag Harry Deutsch, 1971.
[3]‌ A. W. Hull, ‘The effect of a uniform magnetic field on the motion of electrons between
coaxial cylinders’, Physical Review, vol. 18, pp. 31–​57, 1921.
[4]‌ J. C. Slater, Microwave Electronics. New York: D. van Nostrand, 1950.
[5]‌ R. G. E. Hutter, Beam and Wave Electronics in Microwave Tubes. Princeton, NJ: D. van
Nostrand, 1960.
[6]‌ L. Brillouin, ‘Theory of the magnetron. I’, Physical Review, vol. 60, pp. 385–​396, 1941.
[7]‌ O. Buneman, ‘Symmetrical states and their breakup’, in E. Okress, ed., Crossed-​Field
Microwave Devices, vol. 1. New York: Academic Press, pp. 209–​233, 1961.
[8]‌ R. M. Gilgenbach et al., ‘Crossed-​field devices’, in R. J. Barker et al., Modern Microwave
and Millimeter-​wave Power Electronics. Piscataway, NJ: IEEE Press, pp. 289–​342, 2005.
[9]‌ Y. Y. Lau et  al., ‘Limiting current in a crossed-​field gap’, Physics of Fluids B, vol. 5,
pp. 4486–​4489, 1993.
[10] A. H. Falkner, ‘Double stream flow in the smooth-​bore magnetron’, Proceedings of the
Institution of Electrical Engineers, vol. 120, pp. 959–​961, 1973.
[11] J. A. Bradshaw, ‘Cutoff characteristics of the static magnetron diode’, in E. Okress, ed.,
Crossed-​Field Microwave Devices, vol. 1. New York: Academic Press, pp. 261–274, 1961.
316

316 Electron Flow in Crossed Fields

[12] H. L. McDowell, ‘Smooth-​bore magnetron simulations using a moving wavelength

computer code’, IEEE Transactions on Plasma Science, vol. 30, pp. 980–​983, 2002.
[13] P. J. Christenson and Y. Y. Lau, ‘One-​dimensional modulation instability in a crossed-​
field gap’, Physical Review Letters, vol. 76, pp. 3324–​3327, 1996.
[14] A. V. Agafonov and V. S. Voronin, ‘Brillouin flow in a magnetron diode as the kinetic
model limit’, in Proceedings of the 1997 Particle Accelerator Conference, Vancouver, BC,
vol. 1, pp. 1302–1304, 1997.
[15] R. L. Jepsen and M. W. Muller, ‘Enhanced emission from magnetron cathodes’, Journal
of Applied Physics, vol. 22, pp. 1196–​1207, 1951.
[16] R. C. Glass et al., ‘Noise in cut-​off magnetrons’, Proceedings of the IEE –​Part B: Radio
and Electronic Engineering, vol. 102, pp. 81–​86, 1955.
[17] J. M. Osepchuk, ‘Magnetron diode oscillations’, in E. Okress, ed., Crossed-​Field
Microwave Devices, vol. 1. New York: Academic Press, pp. 275–318, 1961.
[18] R. L. Jepsen, ‘Electron interaction in the static magnetron’, in E. Okress, ed., Crossed-​
Field Microwave Devices, vol. 1. New York: Academic Press, pp. 251–​259, 1961.
[19] H. C. Nedderman, ‘Space-​charge distribution in a static magnetron’, Journal of Applied
Physics, vol. 26, pp. 1420–​1430, 1955.
[20] E. H. Hirsch, ‘Cut-​off characteristics of a cylindrical magnetron’, International Journal
of Electronics, vol. 21, pp. 521–533, 1966.
[21] E. H. Hirsch, ‘Electron transport in the strongly cut-​off magnetron’, International
Journal of Electronics, vol. 22, pp. 297–305, 1967.
[22] E. H. Hirsch, ‘Space charge oscillations in the cylindrical magnetron’, International
Journal of Electronics, vol. 23, pp. 497–509, 1967.
[23] V. A. Berbasov and L. M. Groshkov, ‘Measurement of the potential distribution in
a magnetron diode for a strongly transcritical mode’, Radiophysics and Quantum
Electronics, vol. 13, pp. 617–​618, 1970.
[24] G. D. Sims, ‘Preoscillation phenomena in space-​charge clouds below the main oscilla-
tion threshold’, in E. Okress, ed., Crossed-​Field Microwave Devices, vol. 1. New York:
Academic Press, pp. 179–​208, 1961.
[25] G. Hok, ‘A statistical approach to the space-​charge distribution in a cut-​off magnetron’,
Journal of Applied Physics, vol. 23, pp. 983–​989, 1952.
[26] G. Hok, ‘Statistical theory of the magnetron space charge’, in E. Okress, ed., Crossed-​
Field Microwave Devices, vol. 1. New York: Academic Press, pp. 235–​250, 1961.
[27] E. Q. Zhang, ‘An interpretation of magnetron problem’, IEEE Transactions on Electron
Devices, vol. 27, pp. 1280–​1283, 1980.
[28] E. Q. Zhang, ‘An interpretation of the magnetron problem –​Reply’, IEEE Transactions
on Electron Devices, vol. 28, p. 893, 1981.
[29] K. Mouthaan and C. Susskind, ‘Anode current in the smooth-​ bore magnetron’,
Proceedings of the IEEE, vol. 54, pp. 430–​432, 1966.
[30] E. H. Hirsch, ‘Comment on “An interpretation of the magnetron problem”’, IEEE
Transactions on Electron Devices, vol. 28, pp. 892–​893, 1981.
[31] E. Q. Zhang, ‘On the magnetron cathode’, IEEE Transactions on Electron Devices,
vol. 33, pp. 1383–​1384, 1986.
[32] A. V. Agafonov et al., ‘Dynamics of magnetic insulation violation in smooth-​bore mag-
netrons’, in 1997 Particle Accelerator Conference, vol. 1, pp. 1299–​1301, 1997.
[33] A. V. Agafonov et al., ‘Dynamics of magnetic insulation failure and self-​organization of
an electron flow in a magnetron diode’, Technical Physics, vol. 49, pp. 93–​103, 2004.
317

9 Electron Guns

9.1 Introduction

Many high-​power vacuum tubes depend for their operation on the interaction
between an extended electron beam and an electromagnetic structure. In general
the current density required in the beam is greater than that which can be obtained
from the cathode. Thus the area of the cathode must be greater than that of the
beam. Careful design of static electric and magnetic fields is needed to ensure that
the electron flow converges smoothly from the cathode into the beam. The arrange-
ment of electrodes and magnetic materials used to achieve this is known as an elec-
tron gun. The objective of this chapter is to review the main types of electron gun
and the principles of their design. The greater part of the chapter is devoted to
Pierce electron guns which are used in inductive output tubes (IOTs) (Chapter 12),
klystrons (Chapter 13) and travelling-​wave tubes (TWTs) (Chapter 14). The discus-
sion includes the electrostatic design of the Pierce gun, and methods for launching
the beam formed by the gun into uniform and periodic magnetic focusing fields. The
remainder of the chapter introduces Kino guns, used in injected beam crossed-​field
tubes (Chapter 16), and magnetron injection guns, used in gyrotrons (Chapter 17).
Most electron beams are cylindrically symmetrical about the direction of flow being
either solid or hollow cylinders. However beams whose cross-​sections are either flat
sheets or flattened ellipses are of increasing interest. Multiple-​beam tubes incorp-
orate several electron beams in parallel with one another and this raises special
problems in the design of the electron guns.
Many electron guns are space-​charge limited diodes whose perveance is defined
by (5.8)
I0
K= , (9.1)
Va1.5

where I 0 is the beam current and Va the potential of the anode with respect to the
cathode. This ratio is typically in the order of 10 −6 A V −1.5 for electron guns and it
is customary to express K in microperveance. The performance of an electron gun
is limited by:
• The maximum current density available at the cathode. This must always be suf-
ficient to maintain space-​charge-​limited flow over the whole area of the cathode
318

318 Electron Guns

to ensure that the current is not dependent on the properties of the cathode sur-
face or on its temperature. (But note that the magnetron injection guns used in
gyrotrons are normally operated in the temperature-​limited region (see Section
9.6.2).)
• The ratio of the cross-​sectional areas of the electron flow at the cathode and in
the final beam (the Area Convergence of the gun). If this is too great it becomes
impossible to ensure that the electron trajectories flow smoothly into the beam
without crossovers. Most guns have area convergences less than 100:1.
• Electric breakdown between the focus electrode and the anode (see Section
18.7.1).

9.2 The Pierce Electron Gun

The essential feature of the Pierce electron gun is rectilinear electron flow based on
a simple space-​charge limited diode together with the use of electrodes at cathode
and anode potentials to maintain the correct boundary conditions along the edge
of the beam [1]. The commonest form, used in most linear-​beam tubes, is the coni-
cal convergent gun illustrated in Figure 9.1. This is based on the spherical diode,
discussed in Section 5.6, and it is described in detail below. The purpose of the gun
is to create a cylindrical electron beam having current I 0 , velocity u0 , radius b and
uniform current density, in which the electron trajectories do not cross over one
another.

9.2.1 Electrostatic Theory of the Pierce Electron Gun

The electrons flow in a cone having half-​angle θ from the cathode whose radius
of curvature is Rc towards a concentric anode having radius of curvature Ra . For
the present we will assume that the anode takes the form of a perfectly permeable

Figure 9.1:  Arrangement of a Pierce electron gun.

319

The Pierce Electron Gun 319

conducting grid at potential Va with respect to the cathode. The fraction of the sur-
face area of the cathode sphere from which current is drawn is

2 πRc2 (1 − cos θ ) (9.2)

4 πRc2

and the space-​charge-​limited current is then derived from (5.84) as

16 πε 0 2 ( e m0 ) Va1.5
I0 = ⋅ 2 (1 − cos θ ) , (9.3)
18 αa

where α a is the value of the function α of Rc R defined by (5.85) calculated at

R = Ra . This equation neglects relativistic effects which can be included by making
use of (5.89). Equation (9.3) can be written
9α 2a I0
cos θ = 1 − ⋅ , (9.4)
8πε 0 2 ( e m0 ) Va1.5

from which it can be seen that the angle of convergence required to produce a beam
having a given perveance depends only upon α a and therefore upon the ratio Rc Ra .
It follows from (9.4) that all guns which are geometrically similar (i.e. scaled from
one another) have the same perveance.
In practical electron guns it is not possible to use a grid at the anode because it
would quickly be destroyed by the power dissipated by interception of electrons.
Instead there must be an aperture in the anode sufficiently large for the beam to
pass through without interception. Figure 9.2 shows the modification in the field
of a spherical capacitor when a hole is bored through the inner electrode. It can
be seen that the electric field lines penetrate into the hole and that the transverse

Figure 9.2:  Field distortion in a spherical capacitor due to a hole in the inner electrode
(copyright 1956, The IET, reproduced, with permission, from [2]).
320

320 Electron Guns

component of the electric field is no longer zero. The transverse field acts as a
diverging lens so that the slope of the beam edge trajectory, as it leaves the anode
region, is less than the initial angle of convergence [1].
The strength of this lens can be estimated by the using the aperture lens theory
described in Section 7.3.2. The effects of space-​charge may be neglected, in the first
approximation, because the local fields due to the anode are much stronger than the
space-​charge field. The focal length of the anode lens is given by (7.29)

1 (Va + VR ) ( E1 − E2 ) (9.5)
= ,
f0 2Va (Va + 2VR )

where VR = m0 c 2 e and E1 and E2 are the axial electric fields before and after the
anode so that E2 = 0 in the drift region and f0 is positive for a converging lens. The
electrostatic potential between the cathode and the anode can be written, using
(9.3), in the form1
4
 α ( Rc R )  3
V ( R ) = Va   . (9.6)
 α ( Rc Ra ) 

If the electric field at the surface of the ideal spherical anode is taken as an estimate
of E1 then

 ∂V  4Va Rc  d α  4Va Rc α a′
E1 = −  = − ⋅ = − ⋅ (9.7)
 ∂R  R = Ra 3α a Ra2  d ( Rc R )  R = R 3 Ra2 α a
a

so that, substituting in (9.5) the focal length of the lens, normalised to the spherical
radius of the anode, is given by

Ra 2 (Va + VR ) Rc α a′
=− ⋅ ⋅ . (9.8)
f0 3 (Va + 2VR ) Ra α a

In the non-​relativistic limit (9.8) becomes

Ra 1 R α′
= − ⋅ c ⋅ a . (9.9)
f0 3 Ra α a

We note that, at a fixed anode voltage, Ra f0 is a function of Rc Ra  only.

Figure 9.3 shows the geometry of the model of the Pierce electron gun including
the effects of the anode aperture. From the lens equation
ra ra r
= + a , (9.10)
z1 Ra cos θ0 f0

which can be written as

Other authors express (9.6) and (9.7) in terms of α 2 because this function was tabulated by Langmuir in his
1

1924 paper. As it is easy to calculate α directly the use of α 2 is unnecessary.

321

The Pierce Electron Gun 321

Figure 9.3:  Geometry of the Pierce electron gun.

Ra
tan θ1 = tan θ0 + sin θ0 . (9.11)
f0

We note that θ1 is a function of Rc Ra and the perveance. Variations of this equa-

tion appear in the literature. Some authors assume that the angles are small so that
θ ≈ sin θ ≈ tan θ [1, 3, 4] Others have sin θ0 in place of tan θ0 as the first term on the
right-​hand side [5, 6]. This appears to be a consequence of the inconsistent choice
of the position of the lens at the interception of the theoretical anode with the axis,
whilst maintaining the point at which the angle of the trajectory changes, as shown
in Figure  9.3. Others have sine functions throughout for reasons which are not
stated [7, 8]. If the initial angle of convergence is small it is possible for the second
term on the right-​hand side of (9.11) to be greater than the first so that θ1 is negative
and the beam diverges after leaving the lens. Since the beam must be converging as
it leaves the lens this sets a minimum value to θ0.
After the electrons have passed through the anode aperture they move into a
region where the electrostatic potential on the boundaries of the beam tunnel is
constant. The beam then obeys the universal beam spreading curve, discussed in
Section 7.5.1, and has minimum radius b. It is necessary for the initial electron flow
to be convergent so that the beam waist falls at a convenient distance beyond the
anode. The magnetic or electrostatic focusing system for the beam is then designed
to maintain this radius. The slope of the edge of the beam as it leaves the anode lens
can be written, using (7.72) as

tan θ1 = β p b ln ( ra b ), (9.12)

where β p includes the relativistic correction if required. Rearranging this equation

and making use of the geometry of the gun to eliminate ra  gives
2
rc Rc  tan θ1 
= exp   . (9.13)
b Ra  β pb 

We note again that, for given parameters of the electron beam, the ratio rc b
depends only upon Rc Ra and the perveance. Thus, if rc b and the perveance are
32

322 Electron Guns

known, then the value of Rc Ra is fixed by (9.13) and the other parameters of the
gun can be found by substitution.

9.2.2 The Focus Electrode and Anode Nose

In order to maintain the desired conical flow of the electrons it is necessary to
choose the shapes of the focus electrode and of the anode in such a way that the
electrostatic potential outside the beam is continuous with that defined by (9.6).
It is possible to find the theoretical electrode shapes for conical flow Pierce guns
[1, 9]. In practice the presence of the anode aperture distorts the field, as shown
in Figure  9.2, and it is found that electrodes having the general form shown in
Figure 9.4 are required to restore the equipotential surfaces close to the cathode to
the desired spherical form [2, 10, 11]. Because of the concentration of equipotential
surfaces close to the anode nose it is not possible to match the potential distribution
along the whole of the beam edge from the cathode C to the theoretical position
of the anode A. Instead the potentials are matched from C to the intersection of a
surface such as B with the beam edge. Brewer noted that, if B is too close to C, then
the equipotential surfaces close to the cathode will be distorted, and that if B is too
close to A then the fields are strongly distorted at the edge of the beam close to the
anode. He suggested that B should be approximately 0.6 of the distance between C
and A [10]. For typical electron guns the potential at B is between 25% and 35% of
Va and the electron velocity is, therefore, between 50% and 60% of u0 .
Comparison between experimental results, and the simple theory of the Pierce
electron gun presented in the previous section, shows that the strength of the anode
lens is generally greater that that predicted by (9.9), especially as the angle of con-
vergence increases. Danielson et al. investigated this problem and suggested that the
last term in (9.11) should be multiplied by a factor which was calculated to be 1.1
for a 0.33 µPerv gun with Rc Ra = 2.37 [12]. The correction factor was found to be
close to Va Vx where Vx was approximately equal to the potential on the axis at its
intersection with the theoretical anode. Müller used ray tracing through the field of

Figure 9.4:  Equipotential surfaces in a realistic Pierce electron gun

(copyright 1956, the IET, reproduced, with permission, from [2]).
32

The Pierce Electron Gun 323

the pierced vacuum capacitor shown in Figure 9.2 and obtained a correction factor
of 1 cos θ0 for θ0 ≥ 20° [2]. Vaughan suggested an empirical correction factor of
1.25 based the analysis of a number of guns [5]. An alternative approach was used
by Yang et al. who fitted the results for θ0 of a set of 30 electron guns to experi-
mental values by multiplying the last term in (9.4) by an empirical constant [13].
These empirical approaches are not very satisfactory and it is useful to consider the
problem more closely.
In order to model the gun it is convenient to divide it into three regions as shown
in Figure 9.4. Between C and B the equipotential surfaces are spherical and gov-
erned by (9.6) and the current is determined by the space-​charge limited flow. The
equipotential surfaces outside the beam in this region should be very similar to
those determined theoretically [9]. The solution for conical flow is obtained in a
plane containing the axis and expressed in polar coordinates centred on the centre
of curvature of the cathode. The parameter γ defined in (5.83) is generalised as

 R
γ = ln   + j (θ − θ0 ) (9.14)
 Rc 

and the potential is given by

 4

V ( R, θ ) = k Re α ( R, θ ) 3  , (9.15)
 
where k is a constant. The equipotential surfaces calculated in two dimensions
remain solutions of Laplace’s equation when they are rotated about the axis of sym-
metry of the gun. Figure 9.5 shows the shapes of the equipotential surfaces which
are obtained. It should be noted that these curves are plotted for the case where
θ0 = 0 with the cathode at the origin. The shapes for other angles of convergence are

Figure 9.5:  Normalised shapes of the beam-​forming electrodes for a solid convergent

conical beam
(copyright 1982, Elsevier, reproduced, with permission, from [9]).
324

324 Electron Guns

obtained by rotating the figure so that the horizontal axis coincides with the edge of
the beam. The curve corresponding to the focus electrode makes an angle of 67.5°
with the edge of the beam [1]. Comparison between Figures 9.4 and 9.5 shows the
similarity between the equipotential surfaces in this region.
Beyond D the external potential is effectively equal to the anode potential and
the beam drifts under the influence of space-​charge forces alone and follows the
universal beam spreading curve.
Between B and D the electrons continue to accelerate but gradually deviate from
the initial conical flow under the influence of the transverse electric fields in the
region of the anode. As the electron flow is space-​charge limited the electric field at
the cathode is zero and, thus, from Gauss’ Law we know that the flux of D out of
the conical volume bounded by C and A must be equal to the charge enclosed. Now,
the current is continuous, but the electron velocity increases as the electrons move
from the cathode towards the anode. Therefore the flux of D is determined very
largely by the charge close to the cathode. The simple model of the gun discussed
in Section 9.2.1 assumes that the whole of this flux passes transversely through
the beam edge as it passes through the anode. In the region between A and D the
simple model assumes that the transverse electric field is determined by the local
charge density in the beam. Thus the whole of the electric field which influences the
transverse motion of the electrons is determined by the distribution of the space
charge, and of the equipotential boundaries. Because the greater part of the charge
is concentrated close to the cathode, the electric field associated with it is much
greater than that due to the local space-​charge in the region of the anode. An esti-
mate of the contribution from the local space charge shows that it is typically less
than 10% of that from the charge close to the cathode [14]. Thus the effects of the
local space-​charge on the strength of the anode lens can be neglected in the first
approximation [15]. We note that the region between B and D is not thin and that
the electron velocity varies considerably within it. However, the transverse electric
field is strongest in the part of it which is close to A. It is therefore an acceptable
approximation to model the electron motion in this region as:  conical flow from
C to A; a thin electrostatic lens at A; and motion governed by the universal beam
spreading curve beyond A. It is desirable, however, to include the effects of large
angles of convergence and variation of the electron velocity in the model of the
anode lens.

9.2.3 Improved Model of the Anode Lens

We have already noted that the modification of the electron trajectories as they
pass through the anode aperture is a consequence of the transverse component of
the electric field. Following [12] we note that the flux of the electric field within the
beam which would have terminated on the theoretical anode is terminated on the
anode aperture. This flux is

Φ = 2 πRa2 (1 − cos θ0 ) E1 , (9.16)

325

The Pierce Electron Gun 325

Figure 9.6:  Geometry of the anode lens.

where E1 is given by (9.7). Let us suppose that this transverse flux is confined to a
short axial length 2δ of the beam edge trajectory centred on the point at which it
intersects the theoretical anode as shown in Figure 9.6. Thus we are assuming that
the anode lens is thin. This is not strictly correct, but the detailed distribution of
the electric field depends on the shape of the anode and, therefore, it is not possible
to derive a simple general model which includes the spatial variation of the electric
field. The area of conical surface is

A = 4 πRa δ sin θ0 . (9.17)

Then, because the flux must be conserved, the average of the normal component of
the electric field over the element is

Φ Ra (1 − cos θ0 )
En = = ⋅ E1 . (9.18)
A δ 2 sin θ0

Now consider an electron which enters the lens at an angle θ0 and passes through it
with effective velocity ue corresponding to a potential Ve . This is less than u0 because
the penetration of the field into the aperture means that the average potential in the
lens is less than Va. The impulse given to the electron is equal to the change in its
momentum normal to the direction of motion. Thus
m0
∆ur = −e ∫ E n dt. (9.19)
1− ue2 c 2

But we have assumed that E n is constant so (9.19) becomes

2δ 1 − ue2 c 2 e
∆ur = − ⋅ En (9.20)
ue m0

and, substituting for E n from (9.18)

1 − ue2 c 2 (1 − cos θ0 ) e
∆un = − ⋅ ⋅ Ra E1 . (9.21)
ue sin θ0 m0
326

326 Electron Guns

Thus the direction of the trajectory changes by an angle given by

∆un 1 − ue2 c 2 (1 − cos θ0 ) e

tan φ = =− ⋅ ⋅ Ra E1 . (9.22)
ue ue2 sin θ0 m0

Now

1 − ue2 c 2 1 VR (Ve + VR )
= 2⋅ (9.23)
ue2
c Ve (Ve + 2VR )

so that

tan φ = −
(Ve + VR ) ⋅ (1 − cos θ0 ) ⋅ R E (9.24)
Ve (Ve + 2VR )
a 1
sin θ0

and, using (9.5), this may be written as

tan φ = −
(Ve + VR ) ⋅ Va (Va + 2VR ) ⋅ 2 (1 − cos θ0 ) ⋅ Ra . (9.25)
Ve (Ve + 2VR ) (Va + VR ) sin θ0 f0

When relativistic effects are not important this reduces to

Va 2 (1 − cos θ0 ) Ra
tan φ = − ⋅ ⋅ . (9.26)
Ve sin θ0 f0

Hence the angle of the trajectory as it leaves the lens is given by

V 2 (1 − cos θ0 ) Ra 
θ1 = θ0 − tan −1  a ⋅ ⋅ sin θ0  . (9.27)
Ve sin θ0
2
f0 

Comparing (9.27) with (9.11) we see that the estimate of the focal length has been
adjusted by two terms. The first recognises that the electron has not yet reached its
full speed as it passes through the lens, and the second takes account of the finite
angle between the trajectory and the axis. We note also the lens effect is shown
in the addition of angles rather than their tangents. When inner trajectories are
considered the effective voltage varies with the angle of convergence as can be seen
from Figure 9.4. Thus the lens shows spherical aberration in which the focal length
varies with the angle of the trajectory. Spherical aberration causes the current den-
sity in the beam to be non-​uniform. This can be corrected by changing the shape of
the cathode surface so that the spherical aberration is compensated by changes in
the current density [5, 11, 16].
Examination of Figure 9.4 shows that the effective potential Ve depends upon the
shape of the anode and upon the distance between the beam edge trajectory and the
edge of the aperture. It is, therefore, not possible to find a theoretical expression for
Ve Va . However, we may expect that this ratio will have a similar value for many elec-
tron guns and that the computed dimensions of the gun will not depend strongly on
it. This was investigated by applying the revised theory to the data for electron guns
327

The Pierce Electron Gun 327

in [5, 13]. This includes guns with microperveances in the range 0.29 to 3.67 and area
convergences from 5 to 300. In each case the ratio Ve Va was adjusted so that the
angle θ0 was the same as that tabulated. In all but six cases the ratio lay in the range
0.6–​0.8 with a mean value of 0.7. This is consistent with the typical potential distri-
bution shown in Figure 9.4. Figure 9.7 shows how the normalised effective voltage
depended upon the angle of convergence. The scatter may be explained by differ-
ences between the shape and position of the anode nose relative to the beam edge
in different guns. The effective voltage tends to decrease as the angle of convergence
increases. This can also be understood by reference to Figure 9.4 by noting that the
ratio of the diameter of the anode aperture to the distance between the anode and
the cathode increases as θ0 increases. Thus the field penetrates further into the beam
tunnel and the effective potential at the theoretical anode decreases.
The angle θ0 was computed for the same data set, using the fixed ratioVe Va = 0.7 ,
with the results shown in Figure 9.8. It can be seen that there is a very strong correla-
tion between the computed and experimental results. A useful empirical expression

Figure 9.7:  Dependence of the normalised effective voltage on the angle of convergence.

Figure 9.8:  Comparison between computed and experimental angles of convergence.

328

328 Electron Guns

for the relationship between the angle of convergence in degrees, the perveance and
the area convergence has been given by True [17]

θ0 = 12.9 × µP ( rc b ), (9.28)

where µP is the microperveance. This expression is said to be valid for θ < 40, µP < 3
and 2 < rc b < 8. When (9.28) was applied to the subset of the data given in [13] that
satisfies these criteria, the scatter of the points was very similar to that shown in
Figure  9.8. It was found that the mean value of the ratio of the estimated, and
experimental, angles of convergence was unity with a standard deviation of 0.14.

9.2.4 The Effects of Thermal Velocities

Because the electrons in a Pierce gun are obtained by thermionic emission from the
cathode their random thermal velocities are superimposed on the overall motion.
These thermal velocities mean that the edge of the electron beam is not abrupt,
as has been assumed so far, but the current density falls off more gradually. This
problem was studied by Cutler and Hines who determined the effect of thermal
velocities on the current density profile in the beam formed by a Pierce gun [18]. It
was shown that the ratio of the current density at radius r to the idealised current
density J 0 can be expressed as
r σ
Jr ⌠  R   rR   R
e

= exp ( − r 2 2σ 2 )    I 0  2  exp ( − R 2 2σ 2 ) d   (9.29)

J0 ⌡0  σ   σ   σ 

where re is the expected beam radius, and the normalised value of σ is given by

σ 2eVa re r π
= A( −α a ) + 3 e ( B ± C ) ( −α a )2 .
23
(9.30)
Rc kT rc b 2

Here
Rc Ra
⌠ 1 R 
A=  d c (9.31)
⌡1 ( −α ) 23
 R 

(
B = erf ln ra b ) (9.32)

and

(
C = erf ln re b . ) (9.33)

Figure 9.9 shows the variation in current density with radius for a range of param-
eters. The fall-​off in current density takes place over a range r σ ∼ 5 centred on the
value of re σ. Thus thermal velocity spreading of the beam is negligible when re σ is
greater than 100. For a particular case the value of σ can be computed from (9.30)
329

The Pierce Electron Gun 329

Figure 9.9:  Variation in current density with radius from (9.29)

(copyright 1955, IEEE, reproduced, with permission, from [18]).

Figure 9.10:  Variation in the normalised value of σ at the anode and at the beam waist with
Rc Ra from (9.30)
(copyright 1955, IEEE, reproduced, with permission, from [18]).

where the negative sign is taken for points between the anode and the beam waist,
and the positive sign for points beyond the beam waist.
At the anode, and at the beam waist, the right-​hand side of (9.30) depends only
on Rc Ra , and the normalised value of σ on the left-​hand side of (9.30) can be plot-
ted against Rc Ra , as shown in Figure 9.10. This figure differs slightly from Figure 5
in [18] because of the difference in the method of calculating the anode lens effect.
It can be seen that the effects of thermal velocity are more pronounced at the beam
30

330 Electron Guns

waist than at the anode. In any given case the importance of thermal velocity effects
can be computed from this graph or directly from (9.30).
In practice it is important that, in linear-​beam tubes, as little of the beam cur-
rent should be intercepted by the structure of the tube as possible. If we require the
beam transmission to exceed 99% then the envelope enclosing 99% of the beam cur-
rent must be smaller than the diameter of the beam tunnel throughout its length.
The fraction of the beam current within radius r is given by

I ( r σ , re σ )  σ  ⌠  r  J r ( r σ , re σ )  r 
2 r σ

= 2     ⋅ d . (9.34)
I0  re  ⌡0  σ  J0  σ

The beam radius within which any given proportion of the current is contained
can be computed by solving this equation (see Worksheet 9.1). Figure 9.11 shows,
as an example, the normalised beam tunnel radius ( a σ ), for different transmission
percentages, as a function of the normalised expected radius. It can be seen that the
increase in the effective beam radius is greatest when re σ is small.
Thermal velocity effects are greatest at low beam voltages, and for guns which
have low perveance and high area convergence. The beam radius ( r99 ) enclosing
99% of the beam current was computed at the beam waist, with an anode voltage
of 10 kV, for the 30 guns in [13]. It was found that r99 b was more than 1.1 in about
half the cases, more than 1.2 in a third of them, and that the greatest value was
1.6. Similar computations for the six guns in [5] all showed ratios of more than 1.1
and the greatest ratio was 2.6. Thermal effects were dominant in all cases of this

Figure 9.11:  Normalised radius enclosing different percentages of the current

(copyright 1955, IEEE, reproduced, with permission, from [18]).
31

The Pierce Electron Gun 331

set except No. 4 and No. 5. When these two cases were recomputed, with the target
value of the cold beam radius (b) reduced in proportion to the thermal increase,
then r99 was found to be very close to the original target radius. A small increase
of up to about 1° in the angle of convergence was also required. If r99 b is greater
than 1.2 then the position of the beam waist is closer to the anode than is predicted
by calculations which ignore thermal velocities [18]. Thus for thermal increases in
the beam radius greater than about 20% it is necessary to compute the full beam
profile including the effects of thermal velocities in order to find the true position
and radius of the beam waist.

9.2.5 Electrostatic Design of a Pierce Electron Gun

The theory given in the preceding sections is useful for the initial design of Pierce
electron guns with conical flow. The starting point is to choose the beam voltage
and current, the final beam radius, and the cathode current density. The cathode
loading is determined by whether the operation of the gun is pulsed or continu-
ous, by the type of cathode chosen, and the intended lifetime of the cathode (see
Section 18.5). In a typical electron gun it is found that the current density varies
approximately as 1 cos 2 α over the surface of the cathode, where α is the angle from
the centre of curvature to the point on the cathode surface [19]. The value of the
normalised effective voltage at the anode lens can be chosen by reference to the
results presented above, or from data for other electron guns similar to that to be
designed. The scatter in Figure 9.8 suggests that setting this figure to 0.7 is likely to
give a good starting point for the initial design. The problem is then one of find-
ing the value of Rc Ra using the equations above. Various authors have described
the solution of this problem using graphical methods [1, 8], iterative methods [5]
or direct methods [3, 13]. Modern computer software makes it very easy to seek
the solutions of the equations directly by numerical methods. This is illustrated
in Worksheet 9.1 which includes the improved model of the anode lens described
above. Once the value of Rc Ra has been found the values of the other dimensions
in Figure 9.3 follow directly.
Sufficient clearance must be left between the anode tunnel and the beam to
ensure that electrons are not intercepted on it. Typically the diameter of the aper-
ture should be 20–​25% greater than the beam diameter at that plane [5]. The tip
of the anode nose usually lies close to the theoretical spherical anode surface. The
detailed shapes of the focus electrode and of the anode nose are found empirically.
Originally this was achieved using an electrolytic tank in which a strip of dielectric
material was placed along the beam edge. The shapes and positions of the external
electrodes were adjusted until the potential distribution measured along this strip
matched that required [10]. More recently the same task has been achieved using
computer programs which model the whole electron gun [20–​26]. In high-​voltage
guns it is necessary to ensure that the surfaces are designed to avoid voltage break-
down (see Section 18.7.1) [27].
32

332 Electron Guns

Table 9.1:  Parameters of a Pierce electron gun

Frost et al. Vaughan, ­ Calculated Calculated

Figure 9 [11] example 2 [5] Ve /Va = 0.7 Ve /Va = 0.8

b (mm) 1.35  1.35  1.35  1.35

Rc (mm) 12.8 12.5 11.6 12.9
Ra (mm) –​ –​ 4.39  5.24
θ0 (deg) 38.5 40.1 43.6 38.5
θ1 (deg) –​ –​ 12.2 12.7
zm (mm) 26.3 25.9 22.1 23.5

Figure 9.12:  Theoretical and computed beam profiles for an electron gun

(copyright 1966, IEEE, reproduced, with permission, from [24]).

As an example, let us consider the type 4A gun shown in Figure  9 of [11]

which has been used as an example by other authors [5, 24]. This was designed
to have a microperveance of 1.9 and cathode radius 8.0  mm (0.316 in). The
spherical radius of the cathode was increased to improve the uniformity of
the current density in the beam. The measured beam radius was 1.35 mm and
the area convergence approximately 35:1. The leading dimensions of this gun
can be calculated using the method described above using Worksheet 9.1 with
the normalised effective anode voltage set to 0.7Va. Better agreement with the
experimental data is obtained by increasing the effective anode voltage to 0.8Va
Table 9.1 shows a comparison between the dimensions of the gun and the results
of the calculation.
Figure 9.12 shows the results of the calculations with Ve Va = 0.8 superimposed
upon the trajectories computed in [24]. The dashed lines show the positions of the
theoretical anode, the focus electrode, and the anode nose. The choice of the higher
value of Ve Va can be justified by noting that the anode aperture is only about 11%
greater than the computed diameter of the beam at that plane. It can be seen that
3

Magnetic Field Design for a Pierce Electron Gun 333

the tip of the anode nose lies close to the theoretical spherical anode. The plane
of the beam waist lies slightly closer to the cathode than the position determined
experimentally. Overall the comparison shows that the simple calculation provides
a good starting point for subsequent improvement of the design. A gun for a 50
MW pulsed klystron is described in [28] and a detailed description of the design of
a 0.52 μperv gun for a TWT is to be found in [29].

9.3 Magnetic Field Design for a Pierce Electron Gun

We have seen that a Pierce electron gun uses convergent flow of the electrons to
form an electron beam in which the electrons at the beam waist are flowing parallel
to the axis. In order for the beam to be useable in a linear-​beam tube it must be con-
trolled by magnetic, or electric, fields as described in Chapter 7. Here we shall only
consider magnetic focusing using either a solenoid field or a PPM field. We saw in
Chapter 7 that, in order to ensure smooth flow of the beam, some of the magnetic
flux must be linked to the cathode (see equations (7.54) and (7.96). This condition
is necessary, but not sufficient, to achieve smooth flow. It is also necessary to ensure
that the magnetic field between the cathode and the beam waist is not such as to
cause the beam to scallop. The usual practice is to carry out the electrostatic design
of the gun so that the beam waist radius is equal to the desired beam radius, and
then to design the magnetic field in the beam entry region to ensure that that diam-
eter is maintained without scalloping.

9.3.1 Solenoid Focusing

The magnetic field of the solenoid for a linear-​beam tube is normally terminated
by a soft iron sheet (pole-​piece). The purpose of this is to provide a return path for
the magnetic flux so ensuring that the most efficient use is made of the electro-​or
permanent magnets providing it. As there must be a hole in the pole-​piece, which
is at least big enough for the beam to pass through it, some of the magnetic flux
leaks into the beam entry region. The magnetic field in the beam entry region is
thus controlled primarily by the position of the pole-​piece relative to the cathode
and by the size of the aperture in it. A theoretical expression for the flux leakage
caused by the hole in the pole-​piece is obtained by assuming that the pole-​piece can
be represented by a thin sheet which is a perfect magnetic conductor coinciding
with the inner face of the pole-​piece [30]. The problem is rotationally symmetrical
and described in ellipsoidal coordinates ( ξ, ζ ) which are related to cylindrical polar
coordinates ( r, z ) by the equations

z2 r2
+ 2 =1 (9.35)
aζ
2 2
a (1 + ζ 2 )
34

334 Electron Guns

and
z2 r2
− + 2 = 1. (9.36)
a ξ
2 2
a (1 − ξ 2 )

It follows that

z = aζξ (9.37)

and

r = a (1 + ζ 2 ) (1 − ξ 2 ) (9.38)

where the sheet is in the z = 0 plane and the edge of the hole is at r = a . The solution
of the analogue problem in electrostatics is found from Laplace’s equation in this
system of co-​ordinates subject to the boundary conditions E z → E0 when z → ∞
and E z → 0 when z → −∞. The resulting scalar potential is

 1 
V = aE0 ξ ζ − (ζ cot −1 ξ − 1) . (9.39)
 π 
This is quite a good approximation provided that the material of the pole-​piece is
not close to saturation around the edge of the hole. The parametric equations of
the flux lines are
1
−
 1 ς  2
r = r0 1 − cot −1 ζ −  (9.40)
 π 1 + ζ 2  

and
1
 r2  2
z = ς a 2 −  , (9.41)
 1 + ς2 

where a is the radius of the aperture, za is the axial position referred to the aperture,
and r → r0 as ζ → ∞. Figure  9.13 shows the flux lines calculated using equations
(9.40) and (9.41). For a paraxial ray r  a and ζ = z a so that the flux density on
the axis varies as

 1  z za  
Bz = B0  1 − cot −1   −  , (9.42)
π   a  1 + ( z a )2 
  
as shown in Figure 9.14. It is evident that the flux is very close to its asymptotic
values at z = ± 2a.
The profile of the magnetic field in the beam entry region must be designed in
such a way that the electrons flow without scalloping in a beam having the intended
radius. That means that, at the electrostatic beam waist, the radial component
of velocity must be zero and the angular velocity must have the correct value to
35

Magnetic Field Design for a Pierce Electron Gun 335

Figure 9.13:  Magnetic flux lines in the region of an aperture in a pole-​piece.

Figure 9.14:  Flux density on the axis in the region of an aperture in a pole-​piece.

balance the space-​charge forces. Thus at least two parameters are needed to specify
the magnetic field profile. This can be achieved, in principle, by choosing the diam-
eter of the aperture in the pole-​piece, and the distance of the inner face of the pole-​
piece from the cathode.
If the beam is focused by a uniform magnetic field whose strength is m times the
Brillouin field then the ratio of the flux linked to the cathode to that linked to the
beam is given by (7.57)

Bc rc2 1
= K = 1 − 2 . (9.43)
B0 b2 m

Thus the flux line which passes through the edge of the cathode is asymptotic within
the solenoid to radius r0 given by
1
 1 4
r0 = b 1 − 2  . (9.44)
 m 

Using this value of r0 we can find the value ζ = ζc for which r = rc in (9.40). The
distance between the pole-​piece and the cathode rim is then found, as a function of
a, by substituting r = rc and ζ = ζc in (9.41). Hence the radius of the aperture and
36

336 Electron Guns

its distance from the cathode rim are not independent of each other. In order to
complete the specification of the magnetic field we need some way of deciding how
to choose one of these dimensions so that the flux varies smoothly through the gun.
Now the differential equation governing the edge of the beam is (7.53)
d 2R 1 2 1 1
= β p + β2L K 3 − β2L f ( z ) R,
2
(9.45)
dz 2
2 R R
where β p = ω p u, β L = ω L u , f ( z ) is the term in braces in (9.42), and u0 has been
replaced by u because the electron velocity is not constant. Now let R = Re + ∆R ,
where Re = r b is the normalised electrostatic beam edge, and ∆R the small pertur-
bation caused by the magnetic field. By definition
d 2 Re 1 2 1
= βp . (9.46)
dz 2 2 Re

Therefore, making use of (9.43)

d 2 ( ∆R )  R04 2
≈ β 2
L Re  − f (z)  , (9.47)
dz 2
 Re
4


where it has been assumed that ∆R  Re . At the edge of the cathode the term in
square brackets is zero because from (9.40) R02 Re2 = r02 rc2 = f ( zc ). In order to keep
the perturbation as small as possible we wish to keep this term small by matching
the flux line to the electrostatic beam edge as closely as possible [31]. It is particu-
larly important to do this close to the cathode where u is small and, therefore, β L is
large. Thus we expect that the best choice of the aperture dimensions will be close
to those which make the slope of the flux line at the edge of the cathode equal
to the angle of convergence of the beam [32]. Thus a good estimate of a can be
obtained from

 dr dz 
 d ζ = − tan θ0 . (9.48)
d ζ  c

To find how the choice of aperture dimensions affects the beam, it is necessary to
integrate the equations of motion of the electrons under the combined influence of
the potentials on the electrodes, the space-​charge of the electrons, and the applied
magnetic field. To do this properly requires the use of a computer program such as
EGN2 [20] to find a self-​consistent solution to the problem. However, the effects of
the aperture dimensions on the beam can be illustrated by a simple model which
assumes that the electron flow is conical between the cathode and the theoretical
anode. Then (9.45) is integrated with the initial conditions r = ra and dr dz = − tan θ1
from that point onwards (see Worksheet 9.1). Since it has been assumed that the
electrons follow paraxial paths, it is necessary to determine the effective position of
the pole-​piece by setting z = aζ, rather than by using (9.41), to ensure that descrip-
tion of the magnetic field is self-​consistent. Using the worksheet it can be shown that
37

Magnetic Field Design for a Pierce Electron Gun 337

Figure 9.15:  Magnetic focusing of the beam in the gun illustrated in Figure 9.12.

there is a value of the pole-​piece dimensions for which the electrons form a beam
with minimum scalloping. It is found that the slope of the flux line passing through
the edge of the cathode is then close to − tan θ0. It has also been found that the flux
linked to the beam at the plane of the electrostatic beam waist is close to the cathode
flux, a criterion which was proposed for PPM focusing by Legarra et al. [33].
Figure  9.15 shows the computed position of the pole-​piece aperture, the flux
line through the cathode edge, and the magnetically confined beam profile for the
gun illustrated in Figure 9.12 when m = 2 and a is given by (9.48). The beam shows
a small amount of scalloping which can be reduced, but not quite eliminated, by
adjusting the diameter and axial position of the aperture. The gun has a large angle
of convergence so the use of the paraxial ray approximation is not really valid.
Nevertheless the model does show the way in which the beam is affected by the
choice of the aperture parameters.
Once an initial design has been chosen it can be modelled accurately, and opti-
mised, using a gun simulation program [20–​22, 25, 26]. If necessary, further control
over the distribution of the magnetic field can be provided by placing a soft iron cyl-
inder around the gun, though this has little effect on the field at the cathode unless
it is placed close to it [31]. It is sometimes also useful, at least for developmental
purposes, to place a coil, known as a bucking coil, outside the tube in the plane of
the cathode so that the cathode flux can be adjusted to give optimum beam trans-
mission. When a very small beam is required, as in some millimetre-​wave tubes, a
region of increasing magnetic field may be used to compress the beam [34–​36].
The method described above cannot be applied to a magnetically shielded gun
because no flux passes through the cathode. An alternative approach is based upon
an ideal field which is zero up to the beam waist and equal to the Brillouin field
thereafter [37–​39]. Since this ideal field cannot be realised in practice we assume
that there is a short transitional region where the field is given by
B0
B= (1 − cos α ) , (9.49)
2
38

338 Electron Guns

where 0 ≤ α ≤ π . This expression is a good approximation to that given by (9.42).

A  more accurate approximation is given in [40]. The hole in the pole-​piece is
assumed to be small so that the magnetic field is effectively zero at the plane of
the cathode. The effects of the real, and ideal, magnetic fields can therefore be
computed from the equation for the focal length of a magnetic lens (7.46). Since
the beam radius is already close to its final value in the region where the mag-
netic field is changing we can assume that r is constant and remove it from the
integral. The condition for the equivalence of the real and ideal magnetic fields
is then [39]
2 2
∫1
Breal
2
dz = ∫ Bideal
1
2
dz, (9.50)

where the field is changing between planes 1 and 2. This can be written in terms of α
1 π π
∫ (1 − cos α ) d α = ∫ d α,
2
(9.51)
4 0 αw

where α w is the position of the beam waist. It is straightforward to show that this
condition is satisfied when α w = 5π 8 and that, therefore, that the magnetic field
should be 0.69B0 at the beam waist to achieve smooth Brillouin flow. It is found that
this conclusion is insensitive to the exact form of the variation of magnetic field
[38]. Worksheet 9.1 gives a similar result when m is close to unity.

9.3.2 PPM Focusing

The principles for designing the magnetic field for a beam entering a periodic per-
manent magnet (PPM) stack are the similar to those for solenoid focusing. It was
shown in Chapter 7 that the equilibrium conditions for an electron beam, having
a given mean beam radius and cathode flux, are the same for both solenoid and
PPM focusing when the RMS PPM field is equal to the solenoid field. The maxi-
mum diameter of the PPM focused beam is then greater than that of the equivalent
solenoid focused beam by the amplitude of the ripple. The diameter of the PPM
focused beam is greatest when the magnitude of the axial magnetic flux density
is greatest. Thus, in theory, the beam should be launched with the desired maxi-
mum radius, and with dr dz = 0, at a plane where the amplitude of magnetic field
is greatest.
A method for designing the magnetic field in the entry region was described in
[41] (see Worksheet 9.2). An ideal variation of the magnetic field was defined in
which the electron trajectories followed the magnetic flux lines between the cathode
and the electrostatic beam waist. At that plane the beam entered a cosinusoidally
varying PPM field, as shown by the solid line in Figure  9.16. Thus, in the entry
region, the flux linked to the beam is constant and equal to the cathode flux. The
field profile in Figure  9.16 cannot be realised in practice. However, following an
approach similar to that used in (9.50), it was shown that correct launching of the
39

Magnetic Field Design for a Pierce Electron Gun 339

Figure 9.16:  Magnetic field for the entry of the beam from a Pierce gun into a PPM stack
showing the ideal profile (solid curve) and a real profile (dashed curve)
(copyright 1967, IEEE, reproduced, with permission, from [41]).

beam can be achieved by the use of an equivalent magnetic flux density Be that
satisfies the equation
L 4
L 4 0 ⌠  2πz 
∫ Be2 dz = ∫ B1 ( z ) dz +  B pk
2 2
cos 2  dz, (9.52)
−a −a
⌡0  L 

where B1 is the flux density between the theoretical anode ( z = −a ) and the entrance
plane of the PPM field, B pk is the peak PPM flux density, and the first zero of
the flux density is at z = L 4. It should be noted that this equation assumes that
the electron velocity is constant, and the beam radius is approximately constant,
throughout the transition region. This equation imposes the criterion that the focal
lengths of the magnetic lenses, represented by the real and ideal magnetic field pro-
files, should be the same. A  field profile satisfying this equation is shown by the
dashed curve in Figure 9.16. Rawls et al. showed that nearly identical beam profiles
were produced using the real and ideal field profiles. Beam transmission of 98% was
achieved in a tube designed in this way. The application of this method to the case
where the cathode is shielded is described in [39].
An alternative method for designing the field in the entry region is shown in
Figure 9.17 [33]. In region 1 the field is designed in the same way as for solenoid
focusing with a target magnetic field B0 . In region 3 the peak magnetic field is 2 B0 ,
and the beam has mean radius b with ripple δ . The transition region 2 can be
regarded as a quarter section of a PPM stack in which the peak magnetic field
is B0 . At the end of the section the beam radius must be (1− δ ) b and the electron
trajectories parallel to the axis. If the beam radius is b at the start of region 2 there
is no net inward force and the beam does not converge as required. Thus the beam
340

340 Electron Guns

Figure 9.17:  Magnetic field profile at the entry of a PPM stack

(copyright 1983, IEEE, reproduced, with permission, from [33]).

radius must be greater than b at the end of region 1 and the radius of the beam
waist must be less than b so that there is a net outward force there leading to the
necessary overshoot.

9.4 Other Pierce Guns

The methods used to design Pierce electron guns for cylindrical beams can also
be used to design guns to produce beams having other cross-​sections. Pierce guns
for producing sheet electron beams have become of greater interest in recent years
because of their potential for use in linear-​beam tubes at frequencies of 100 GHz
and above. Hollow beams have occasionally been used in klystrons and TWTs, but
their principal use is in gyrotrons. Most high power tubes with hollow beams have
magnetron injection guns as described in Section 9.5.2.

9.4.1 Guns for Sheet Beams

In principle the design of a Pierce gun to produce a sheet beam proceeds in exactly
the same way as that described for a cylindrical beam, in Section 9.2 [42]. The cur-
rent is assumed to be drawn from a space-​charge limited cathode which is a section
of a circular cylinder given by (5.70). The focal length of the anode slit can be deter-
mined in a manner similar to that for a circular hole by considering the transverse
impulse given to an electron passing through it. The universal beam spreading curve
for a sheet beam can be derived in the same way as that for a cylindrical beam. The
design, construction and testing of sheet-​beam guns has been described in [43–​45].

9.4.2 Guns for Hollow Beams

Pierce guns for the production of hollow cylindrical beams are similar to those for solid
beams, except that the electrons are emitted from a ring on the surface of the cathode.
341

Beam Control Electrodes 341

The anode aperture may be a round hole [42, 46–​48] or a circular slot [49, 50]. In the
former case the perveance is determined from the solution for a spherical space-​charge-​
limited diode in a manner similar to that for a solid beam. In the latter, the solution for
a toroidal space-​charge-​limited diode is used [51]. In each case focusing electrodes are
required both inside, and outside, the beam. The anode lens effect for a circular hole
is given by (9.5) and the equivalent expression for an annular hole is given in [51]. The
spreading of a hollow beam under the influence of space-​charge forces is identical to
that for a solid beam having the same diameter, current, and voltage. However, because
there is no space-​charge in the centre of the beam, the trajectories of electrons on the
inner edge of the beam do not spread. Thus the design of the gun must ensure that the
outer electrons are converging towards the axis whilst the inner electrons are moving
parallel to the axis as they leave the anode lens. Experimental results for toroidal elec-
tron guns having perveances of up to 15 μperv are described in [51].

9.5 Beam Control Electrodes

It is often desirable to be able to switch the electron beam on and off electronically,
or to control the current in the beam independently of the accelerating voltage. This
requires the use of a third electrode so that the gun is a triode rather than a diode.
It has the advantage that the voltage and power required to control the beam may
be much less than if the anode voltage is switched or controlled directly. Three tech-
niques described in the literature are discussed below: Modulating anode; Control
focus electrode; and Control grid [6, 52, 53]. Their typical characteristics are sum-
marised in Table 9.2 [52, 53] where µ is the amplification factor as defined in (6.1)
and µ c is the ratio of the anode voltage to the negative grid voltage at cut-​off.

9.5.1 Modulating Anode and Control Focus Electrode

When the anode of a diode electron gun is isolated from the body of the tube its
potential can be varied independently. It is then known as a modulating anode.

Table 9.2:  Comparison of methods of beam modulation using control electrodes

(© 1994, Springer Nature. Reproduced, with permission, from [53]).

Parameter Modulating Control focus Intercepting Non-​

anode electrode grid intercepting
grid

µ 1 to 3 2 to 10 30 to 75

µc 1 to 3 2 to 10 30 to 150 30 to 250
Capacitance 30 pF 50 to 100 pF 10 to 20 pF 30 to 50 pF
Grid current 0% 0% 10 to 20% 0.1 to 1%
Maximum grid power n/​a n/​a 5W 5W
342

342 Electron Guns

The perveance of the diode gun is fixed by its geometry so that the current depends
upon the potential difference between the cathode and the anode in the usual way.
However, the velocity of the electrons in the interaction region is determined by the
potential difference between the cathode and the body of the tube. Thus the beam
current and the electron velocity can be varied independently of one another. It
is necessary to take care that the beam is focused correctly over the desired range
of voltages and currents. In a second method the focus electrode is isolated from
the cathode so that its voltage can be varied independently. Because the focusing
of the beam is seriously disrupted by changes in this voltage the technique is only
suitable for switching the beam on and off. The negative voltage (with respect to the
cathode) that is required to reduce the current density to zero at the centre of the
cathode is comparable with the positive voltage applied to the anode. The negative
voltage can be reduced somewhat by including a central pin at the same poten-
tial [32, 52]. In both of these methods the control voltage is large but the control
power is small because the electrode can be designed to intercept very little current.
Further information about the design of these two types of beam control electrode
is given in [32].

9.5.2 Intercepting Control Grid

Control of the electron beam can be achieved with voltages much less than the anode
voltage if a control grid is placed close to the cathode and at a uniform distance from
it. The grid is manufactured from a refractory metal, such as molybdenum, or from
pyrolitic graphite, which can withstand the high temperatures close to the cathode
surface whilst maintaining very precise dimensions. The beam current then depends
upon the voltage applied to the control grid in exactly the same way as in a triode
(see Chapter 6). The current is cut off if the grid voltage is sufficiently negative with
respect to the cathode. The negative grid voltage required to make the gun cut-​off is
typically 1–​3% of the anode voltage while a similar positive voltage is needed to turn
the gun fully on [53]. To minimise interception the grid must have transparency (the
ratio of the holes in the grid to the total grid area) in the range 80–​90% [32, 52]. If the
grid voltage is positive then it intercepts 10–​20% of the beam current [53]. Intercepted
electrons may liberate secondary electrons from the grid and there can also be ther-
mionic emission if the temperature of the grid is high enough [54]. For these reasons
guns with intercepting grids are limited to low beam powers and to tubes, such as
the IOT, in which the grid voltage is positive for only a small proportion of the time.
The arrangement of the apertures in the grid can take the form of an array of
circular or hexagonal holes, or of the spaces between radial and concentric ‘wires’.
The section of the cathode opposite each hole in the grid behaves like an individual
electron gun, and it is desirable that these should all have similar characteristics so
that their currents vary together. Where the grid takes the form of a ‘honeycomb’
array of circular holes it can be much more robust, though the transparency may
be as low as 60%, and the cathode surface may be dimpled to improve the electron
optics [32, 52].
34

Beam Control Electrodes 343

The properties of a gridded gun can be calculated as described for a triode in

Chapter 6. Because the grid is very close to the cathode surface it is possible to con-
sider an equivalent planar geometry in which the spacing between the cathode and
the anode is chosen to give the same electric field at the surface of the cathode in the
absence of the grid (c.f. (6.11)) [54]. The dimensions of the actual grid can be con-
verted to those of an equivalent two-​dimensional grid of round wires [19]. Hence the
dependence of the beam current on the grid and anode voltages can be computed,
as described in Section 6.4. In general, the current between the grid and the anode is
not space-​charge limited. It should be noted that the grid dimensions and spacings
used in gridded guns are such that island formation usually occurs close to cut-​off.
The electron trajectories in the region of the grid exhibit strong lens effects caused
by the local fields in a manner similar to that shown in Figure 6.8. Thus, many of
the electrons have appreciable transverse components of velocity at the beam waist,
and the beam is less well-​collimated than the beam from a well-​designed diode gun.
It may be necessary to remove the beam halo by passing the beam through a tunnel
(a ‘beam scraper’) only slightly larger than the intended beam diameter.

9.5.3 Non-​Intercepting Control Grid

The problems of intercepted current make a gun with an intercepting control grid
unsuitable for use in high power tubes (with the exception of IOTs), and the col-
limation of the beam is unsatisfactory for TWTs. These problems can be greatly
reduced by the introduction of a second grid (the shadow grid) very close to the
cathode surface which is maintained at the cathode potential, or very close to it.
The arrangement of this grid is such that it is a projection of the control grid so that
any electrons which pass through the shadow grid also pass through the control
grid without being intercepted by it. The intercepted grid current in such a gun can
be 0.1% or less [6, 53, 55]. Figure 9.18 shows the computed electron trajectories in

Figure 9.18:  Electron trajectories in a shadow-​gridded gun

(copyright 1984, IEEE, reproduced, with permission, from [56]).
34

344 Electron Guns

a shadow-​gridded gun [56]. It can be seen that lens effects in the region of the grid
produce an appreciable beam halo. Although these trajectories carry only a small
fraction of the total beam current they are very difficult to focus. True has shown
that (9.28) applies to gridded guns if the beam radius is taken as that enclosing 95%
of the beam current. He has also shown that the normalised radial component of
the beam velocity can be calculated from the empirical relationship [56]
ur
= 0.0015 × θ0  . (9.53)
u0

Thus the properties of the beam halo are determined to a large extent by the angle
of convergence of the gun. It is desirable to keep this angle as small as possible to
make the beam easier to focus. The procedure for coupling the design of a shadow
gridded gun with PPM focusing is described in detail in [56]. Because the lens effects
in the region of the grid depend upon the grid voltage and hence, the gun perve-
ance, it is found that good collimation of the beam is only achieved over a limited
range of operating conditions.
A number of variants of the basic shadow-​gridded gun have been found to give
improved beam collimation [55, 57]. If the shadow grid is placed in contact with the
cathode the emitting area of the cathode is reduced, but it is found that the beam
collimation is appreciably improved. The shadow grid may also be embedded in
the surface of the cathode, or be formed by a thin patterned layer of non-​emissive
coating deposited on it. Improved beam collimation can also be achieved by isolat-
ing the shadow grid from the cathode and adjusting its potential independently in
a tetrode gun [57]. An ultra-​laminar gun with three grids is described in [55]. More
complex grid structures are used in dual mode guns to enable beams with high and
low powers to be produced by the same gun [6].

9.6 Crossed-​Field Electron Guns

We saw in Section 8.2.2 that a stable sheet electron beam can exist in a region in
which there are uniform static electric and magnetic fields which are perpendicular
to each other and to the direction of electron motion. The variation in the electro-
static potential within the beam is given by (8.22). Now, useful interaction between
an electron beam, and the RF field of an electromagnetic structure, requires that
the spread of electron velocities should be small compared with their mean velocity.
Therefore the thickness of a sheet beam in crossed fields must be small. A simple
method for launching such a beam where the whole structure is immersed in a
uniform magnetic field Bz = − B0 is shown in Figure 9.19. An electron which starts
from rest on the surface of the cathode has velocity ux = 2 E1 B0 when it is moving
parallel to the x axis. In order for it to continue to move in a straight line parallel to
the x axis ux = E0 B0 . Thus a necessary condition for the launching of a sheet beam
is E1 = E0 2 . It is also necessary for the potential at the interface to be continuous,
345

Crossed-Field Electron Guns 345

Figure 9.19:  Arrangement of a simple crossed-​field electron gun.

Figure 9.20:  Arrangement of a Kino gun

(copyright 1960, IEEE, reproduced, with permission, from [59]).

and equal to that required by the conservation of energy [58]. This arrangement can
produce a useful beam, but the width of the emitting strip of the cathode in the x
direction must be small and, therefore, the current is limited by the current density
available.

9.6.1 Kino Gun
A more practical arrangement is the Kino gun shown in Figure 9.20 [59]. The elec-
tron orbits in the gun (to the left of the plane A–​A) are those for a space-​charge
limited magnetron diode given by (8.36) and (8.46) and shown in Figure 8.7. These
trajectories are the correct ones to use because, although the diode is cut off, the
electrons are extracted as a beam and do not return to the cathode. All the trajec-
tories within the gun have the same form apart from a shift in the direction parallel
to the surface of the cathode. Because the top of the trajectory is fairly flat it is pos-
sible for the cathode to be have a finite length, though this must be short compared
with the cyclotron wavelength ( λ c = 2 πux ω c ). Thus, this type of gun is known as
a ‘short gun’. The whole length of the gun from the centre of the cathode to the
plane A–​A is approximately λ c 2. The shapes of the focus electrodes and the gun
anode are used, as in the Pierce gun, to maintain the correct potential distribu-
tion along the edges of the beam. The design and testing of a gun of this type is
described in [60].
The current which can be delivered by a short gun is limited by the maximum
current which can be obtained from the cathode. Thus, if greater beam current
346

346 Electron Guns

is required, it may be necessary to use a cathode which is several cyclotron wave-

lengths long. However, we have seen that there is then uncertainty as to which of
the possible Slater orbits represents reality.
For this type of gun Kino proposed a different solution for the flow of space-​
charge which assumed that the electrons are emitted from the cathode with finite
velocity. The electron trajectories are then found to be parabolic. In certain cir-
cumstances the initial velocity required for this solution to be valid is of the same
order as the thermal velocities of the electrons. This theory has been criticised by
Vaughan [61] who showed that this solution only exists in specific circumstances
which do not correspond to many practical cases. He investigated solutions in
which the initial velocity of the electrons was an additional parameter and showed
that, in a typical case, a well-​collimated beam could be formed in which some
trajectories showed cycloidal motion. The design and testing of an experimental
‘long gun’ based on Kino’s theory is described in [62]. The gun gave the expected
value of the current but the beam thickness was greater than expected. This was
attributed to RF instabilities in the gun. Reviews of the literature on that subject
are given in [63, 64]. It was found that the noise could be considerably reduced by
operating the cathode in the temperature-​limited regime; by tilting the cathode to
a small angle so that there was a small normal component of the magnetic field
[65]; or by placing a grid in front of the cathode [66] which could also be used for
modulating the beam.

9.6.2 Magnetron Injection Gun

Figure 9.21 shows the arrangement of a magnetron injection gun which produces
a hollow cylindrical beam [8]. This gun bears a superficial resemblance to the Kino
gun described in the previous section, rotated about an axis of symmetry. But there
is the important difference that the magnetic field is essentially parallel to the elec-
tron motion rather than perpendicular to it. The cathode takes the form of a trun-
cated cone, and the whole gun is immersed in a strong axial magnetic field. Thus
the region around the cathode is essentially a cut-​off cylindrical magnetron diode
whose space-​charge layer forms a virtual cathode. The tapering of the cathode and
the anode produces an axial electric field which draws the electrons into the beam.
Focus electrodes are used to maintain the correct potential distribution along the
edge of the beam in the gun region. In the transition region the radial electric
field is gradually reduced to zero and the beam tends to its final dimensions. The
area convergence which can be obtained is greater than that of a hollow-​beam
Pierce gun.
The design principles for a magnetron injection gun can be illustrated by refer-
ence to Figure 9.21. As there is no space-​charge inside the hollow beam it follows
that the inner electrons must have no angular velocity. Therefore, by the application
of Busch’s Theorem (7.39), the flux linked to the smaller end of the cathode must
be equal to that linked to the inner electrons of the beam. If it is assumed that
the whole gun is immersed in a uniform axial magnetic field B0 then ri = ra . The
347

Crossed-Field Electron Guns 347

Figure 9.21:  Arrangement of a magnetron injection gun

(copyright 1965, Elsevier, reproduced, with permission, from [8]).

condition for space-​charge balanced flow of the outer electrons is identical to that
for a solid beam given by (7.54)

β2p = 2β2L (1 − K ) , (9.54)

where β p = ω p u0 is defined for a solid beam having the same current, velocity and
outer radius, and β L = eB0 2 m0 u0 . Hence from (7.52) and (9.54)
4
r  β2p
K =  b  = 1− 2 . (9.55)
 ro  2β L

Thus, given the desired parameters of the beam and the strength of the magnetic
field, the larger radius of the cathode can be determined from (9.55). Now the sec-
ond term on the right-​hand side of this equation is equal to unity for Brillouin
focusing of the equivalent solid beam. But, since rb > ri, it follows that K > ( ri ro ) .
4

The ratio of the magnetic field to the Brillouin field given by (7.57) requires that
B0 1
m= = . (9.56)
BB 1− K
Hence
1
m> . (9.57)
1 − ( ri ro )
4

If, for example, ri ro = 0.9 then m > 1.7. It is evident that the minimum magnetic
field required to focus a hollow beam is appreciably greater than for the equivalent
solid beam. From the geometry of the cathode the beam current is

π ( rb2 − ra2 )
I0 = Jc , (9.58)
sin φ
348

348 Electron Guns

Figure 9.22:  Electrode shapes, magnetic field profile and electron trajectories of a magnetron
injection gun for a gyrotron
(copyright 2010, The Electromagnetics Academy, reproduced, with permission from [71]).

where φ is the angle between the slope of the cathode and the axis and Jc is the
cathode current density. This equation determines the angle φ and, hence, the slant
length of the cathode. Methods for calculating the electron trajectories and the
electrode shapes have been described in [67–​70].
Magnetron injection guns are occasionally used in high power linear-​beam tubes
but their main use is in gyrotrons (Chapter 17). Figure 9.22 shows the arrangement
of a typical magnetron injection gun for use in a gyrotron [71]. Comparing this
with Figure 9.21 it can be seen that the slant length of the cathode is small, that
the anode is divided into two sections whose potentials can be adjusted separately,
and that the magnetic field increases with distance so that the beam is compressed.
A detailed description of the design methods for guns for gyrotrons, including two
examples, is given in [72].

References

[1]‌ J. R. Pierce, Theory and Design of Electron Beams. New York: D. van Nostrand, 1949.

[2]‌ M. B. Müller, ‘New points of view in the design of electron guns for cylindrical beams
of high space charge’, Journal of the British Institution of Radio Engineers, vol. 16,
pp. 83–​94, 1956.
[3]‌ U. Tiwary and B. N. Basu, ‘Noniterative method for the synthesis of convergent Pierce
electron guns’, IEEE Transactions on Electron Devices, vol. 34, pp. 1218–​1222, 1987.
[4]‌ B. N. Basu, Electromagnetic Theory and Applications in Beam-​ Wave Electronics.
Singapore: World Scientific, 1996.
[5]‌ J. R. M. Vaughan, ‘Synthesis of the Pierce gun’, IEEE Transactions on Electron Devices,
vol. 28, pp. 37–​41, 1981.
[6]‌ A. S. Gilmour, Jr., Principles of Traveling Wave Tubes. Norwood, MA: Artech House,
1994.
349

References 349

[7]‌ A. H. W. Beck, Thermionic Valves: Their Theory and Design. Cambridge: Cambridge

University Press, 1953.
[8]‌ J. F. Gittins, Power Travelling-​Wave Tubes. London: English Universities Press, 1965.
[9]‌ H. Z. Sar-​El, ‘Revised theory of Pierce-​type electron guns’, Nuclear Instruments and
Methods, vol. 203, pp. 21–​33, 1982.
[10] G. R. Brewer, ‘Formation of high-​density electron beams’, Journal of Applied Physics,
vol. 28, pp. 7–​15, 1957.
[11] R. D. Frost et al., ‘Electron guns for forming solid beams of high perveance and high
convergence’, Proceedings of the IRE, vol. 50, pp. 1800–​1807, 1962.
[12] W. E. Danielson et al., ‘A detailed analysis of beam formation with electron guns of the
Pierce type’, Bell System Technical Journal, vol. 35, pp. 375–​420, 1956.
[13] C. Yang et  al., ‘Improved noniterative method for the synthesis of conver-
gent Pierce  electron guns’, IEEE Transactions on Electron Devices, vol. 53,
pp. 2849–​2852, 2006.
[14] C. K. Birdsall, ‘Aperture lens formula corrected for space charge in the electron stream’,
IRE Transactions on Electron Devices, vol. 4, pp. 132–​134, 1957.
[15] K. Amboss, ‘The effect of the anode aperture in conical flow Pierce guns’, Journal of
Electronics and Control, vol. 13, pp. 545–​572, 1962.
[16] W. Liu and S. Liu, ‘PIC simulation study of electron gun with rotational surface cathode’,
Frontiers of Electrical and Electronic Engineering in China, vol. 6, pp. 556–​562, 2011.
[17] R. True, ‘Gridded Pierce gun design and the PPM focussing of beams from gridded
electron guns’, in International Electron Devices Meeting, pp. 32–​35, 1982.
[18] C. C. Cutler and M. E. Hines, ‘Thermal velocity effects in electron guns’, Proceedings of
the IRE, vol. 43, pp. 307–​315, 1955.
[19] R. True, ‘Calculation and design of grids in Pierce guns’, in 1989 International Electron
Devices Meeting, IEDM ‘89, pp. 215–​218, 1989.
[20] W. B. Herrmannsfeldt, ‘Numerical design of electron guns and space charge limited
transport systems’, Nuclear Instruments and Methods in Physics Research, vol. 187,
pp. 245–​253, 1981.
[21] J. Petillo et  al., ‘The MICHELLE three-​ dimensional electron gun and collector
modeling tool: theory and design’, IEEE Transactions on Plasma Science, vol. 30,
pp. 1238–​1264, 2002.
[22] L. Kumar and E. Kasper, ‘On the numerical design of electron guns’, Optik, vol. 72,
pp. 23–​30, 1985.
[23] J. E. Boers, ‘Digital computer analysis of axially symmetric electron guns’, IEEE
Transactions on Electron Devices, vol. 12, pp. 425–​435, 1965.
[24] V. Hamza, ‘Convergence and accuracy criteria of iteration methods for the analysis
of axially symmetric and sheet beam electrode shapes with an emitting surface’, IEEE
Transactions on Electron Devices, vol. 13, pp. 485–​493, 1966.
[25] B. M. Lewis et  al., ‘Design of an electron gun using computer optimization’, IEEE
Transactions on Plasma Science, vol. 32, pp. 1242–​1250, 2004.
[26] R. True, ‘The deformable relaxation mesh technique for solution of electron optics
problems’, in International Electron Devices Meeting, pp. 257–​260, 1975.
[27] G. Faillon, ‘Technical and industrial overview of RF and microwave tubes for fusion’,
Fusion Engineering and Design, vol. 46, pp. 371–​381, 1999.
[28] T. Lee et al., ‘A fifty megawatt klystron for the Stanford Linear Collider’, in International
Electron Devices Meeting, pp. 144–​147, 1983.
350 Electron Guns

[29] M. N. Pinto et al., ‘An electron gun design for a C-​band TWT’, in SBMO/​IEEE MTT-​S
International Microwave & Optoelectronics Conference, pp. 937–​940, 2011.
[30] W. R. Smythe, Static and Dynamic Electricity, 2nd ed. New York: McGraw-​Hill, 1950.
[31] R. B. Nelson, ‘Shaping the magnetic field for convergent confined-​flow electron guns’,
in 6th International Conference on Microwave and Optical Generation and Amplification,
Cambridge, England, pp. 60–64, 1966.
[32] M. J. Smith and G. Phillips, Power Klystrons Today. Taunton, UK: Research Studies
Press, 1995.
[33] J. R. Legarra et  al., ‘A convergent confined-​flow focusing system for millimeter wave
tubes’, in 1983 International Electron Devices Meeting, pp. 137–​140, 1983.
[34] K. Amboss, ‘Studies of a magnetically compressed electron beam’, IEEE Transactions
on Electron Devices, vol. 16, pp. 897–​904, 1969.
[35] J. A. Seeger, ‘Magnetic compression of axially symmetric Brillouin-​focused electron
beams’, IEEE Transactions on Electron Devices, vol. 16, pp. 1–​5, 1969.
[36] E. A. Ash, ‘Compression and reflection of Brillouin beams’, Journal of Electronics and
Control, vol. 15, pp. 402–​417, 1963.
[37] M. Müller, ‘Magnetische Elektronenoptik in Langstrahl-Verstärker Rohren’, Telefunken
Zeiting, vol. 26, pp. 95–​101, 1953.
[38] V. Bevc et  al., ‘On the design of the transition region of axisymmetric, magnetically
focused beam valves’, Journal of the British Institution of Radio Engineers, vol. 18,
pp. 696–​705, 1958.
[39] J. L. Rawls, ‘An analogue study of periodic permanent magnet focusing’, International
Journal of Electronics, vol. 25, pp. 165–​175, 1968.
[40] A. L. Igritsky, ‘Focusing a beam of electrons in the region of entry into the mag-
netic field of a traveling-​wave tube’, Radio Engineering and Electron Physics, vol. 8,
pp. 117–​123, 1963.
[41] J. L. Rawls et al., ‘PPM focusing of convergent beams emerging from partially shielded
cathodes’, IEEE Transactions on Electron Devices, vol. 14, pp. 301–​305, 1967.
[42] N. B. Kramer and E. G. Todd, ‘Design of hollow and strip beam guns’, IEEE
Transactions on Electron Devices, vol. 10, pp. 394–​404, 1963.
[43] S. K. Wong et al., ‘The design of sheet-​beam electron-​gun for high-​power microwave
sources’, in Particle Accelerator Conference, pp. 1043–​1045, 1999.
[44] K. T. Nguyen et al., ‘High-​perveance W-​band sheet-​beam electron gun design’, in ICOPS
2008: IEEE 35th International Conference on Plasma Science, pp. 179–​180, 2008.
[45] J. Pasour et  al., ‘Demonstration of a 100-​ kW solenoidally focused sheet electron
beam for millimeter-​wave amplifiers’, IEEE Transactions on Electron Devices, vol. 58,
pp. 1792–​1797, 2011.
[46] J. R.  M. Vaughan, ‘Synthesis of a hollow-​beam gun based on the Pierce design’, in
International Electron Devices Meeting, pp. 496–​499, 1981.
[47] J. R.  M. Vaughan, ‘Synthesis of a hollow-​beam Pierce gun’, IEEE Transactions on
Electron Devices, vol. 34, pp. 468–​472, 1987.
[48] J. R.  M. Vaughan, ‘Corrections to “Synthesis of a hollow-​beam Pierce gun”’, IEEE
Transactions on Electron Devices, vol. 34, p. 1885, 1987.
[49] L. A. Harris, ‘Toroidal electron guns for hollow beams’, Journal of Applied Physics, vol.
30, pp. 826–​836, 1959.
[50] H. L. Hartnagel, ‘Electrostatically focused electron gun for hollow beams’, Proceedings
of the Institution of Electrical Engineers, vol. 111, pp. 57–​66, 1964.
 References 351

[51] H. Hartnagel, ‘Toroidal hollow-​beam gun’, Proceedings of the Institution of Electrical

Engineers, vol. 111, pp. 1821–​1829, 1964.
[52] A. Staprans et al., ‘High-​power linear-​beam tubes’, Proceedings of the IEEE, vol. 61,
pp. 299–​330, 1973.
[53] L. Sivan, Microwave Tube Transmitters. London: Chapman and Hall, 1994.
[54] H. E. Gallagher, ‘Gridded electron guns for high average power’, IRE Transactions on
Electron Devices, vol. 9, pp. 234–​241, 1962.
[55] G. Miram et  al., ‘Gridded gun design considerations for pulse and cw operation’, in
Vakuumelektronik und Displays, Garmisch ​Partenkirchen, Germany, pp. 225–​233, 1989.
[56] R. True, ‘A theory for coupling gridded gun design with PPM focusing’, IEEE
Transactions on Electron Devices, vol. 31, pp. 353–​362, 1984.
[57] R. True, ‘An ultra-​laminar tetrode gun for high duty cycle applications’, in International
Electron Devices Meeting, pp. 286–​289, 1979.
[58] O. Doehler, ‘Static characteristics of injection crossed-​field tubes’, in E. Okress, ed.,
Crossed-​Field Microwave Devices, vol. 1. New York: Academic Press, pp. 151–​163, 1961.
[59] G. S. Kino, ‘A design method for crossed-​field electron guns’, IRE Transactions on
Electron Devices, vol. 7, pp. 179–​185, 1960.
[60] T. A. Midford and G. S. Kino, ‘Some experiments with a new type of crossed-​field gun’,
IRE Transactions on Electron Devices, vol. 8, pp. 324–​330, 1961.
[61] J. R. M. Vaughan, ‘Generalization and synthesis of the Kino gun’, IEEE Transactions
on Electron Devices, vol. ED-​28, pp. 41–​48, 1981.
[62] T. A. Midford and G. S. Kino, ‘Experiments with a new type adiabatic crossed-​field
gun’, IRE Transactions on Electron Devices, vol. 9, pp. 431–​439, 1962.
[63] T. van Duzer and J. R. Whinnery, ‘Noise in crossed-​ field electron beams’, in
E.  Okress, ed., Crossed-​Field Microwave Devices, vol. 1. New York: Academic Press,
pp. 327–​357, 1961.
[64] R. M. Gilgenbach et  al., ‘Crossed-​field devices’, in R. J. Barker et  al., eds, Modern
Microwave and Millimetre-​ Wave Power Electronics. Piscataway, NJ:  IEEE Press,
pp. 289–​342, 2005.
[65] M. L. Sisodia and R. P. Wadhwa, ‘Noise reduction in crossed-​field guns by cathode tilt’,
Proceedings of the IEEE, vol. 56, pp. 94–​95, 1968.
[66] R. J. Espinosa and R. R. Moats, ‘Broad-​band injected-​beam crossed-​field amplifiers’,
IEEE Transactions on Electron Devices, vol. 24, pp. 13–​21, 1977.
[67] C. J. Edgcombe, ‘Synthesis of electrodes for axisymmetric systems’, IEE Proceedings –​
A, vol. 135, pp. 125–​134, 1988.
[68] T. Okoshi, ‘An improved design theory of a magnetron injection gun’, IEEE Transactions
on Electron Devices, vol. ED-​11, pp. 349–​356, 1964.
[69] W. E. Waters, ‘A theory of magnetron injection guns’, IEEE Transactions on Electron
Devices, vol. 10, pp. 226–​234, 1963.
[70] G. S. Kino and N. J. Taylor, ‘The design and performance of a magnetron-​injection
gun’, IRE Transactions on Electron Devices, vol. 9, pp. 1–​11, 1962.
[71] U. Singh et al., ‘Numerical simulation of magnetron injection gun for 1 MW 120 GHz
gyrotron’, Progress in Electromagnetics Research Letters, vol. 16, pp. 21–​34, 2010.
[72] J. M. Baird and W. E. S. Lawson, ‘Magnetron injection gun (MIG) design for gyrotron
applications’, International Journal of Electronics, vol. 61, pp. 953–​967, 1986.
10 Electron Collectors and Cooling

10.1 Introduction

Even the best vacuum tubes do not convert all of the DC input power into radio­
frequency power. Therefore the electrons which emerge from the interaction region
still carry appreciable power. This must be dissipated safely, in the form of heat, on
the electrodes on which the spent electrons are collected. The maximum permissible
temperature of the surface is limited by the need to avoid:
• physical damage including re-​crystallisation of the copper (at 700 C [1]), local
melting, and fatigue fractures caused by cyclical heating in pulsed devices;
• liberation of adsorbed gasses by exceeding the bake-​out temperature (350–​550 °C)
used during evacuation of the tube [2–​3]. It has been suggested that the surface
temperature of copper should be limited to 300–​350 °C [4].
The maximum power density which can be tolerated on the surface depends
on the heat transfer from the collector electrodes to some external heat sink, or
heat exchanger, and on the method by which the heat is finally dissipated (see
Section 10.5).
The impact of electrons on metallic surfaces (usually OHFC copper) results in
the liberation of secondary electrons, and the generation of X-​rays (see Chapter 18).
When the primary electron energies are small (a few keV) the X-​rays generated can­
not penetrate through the metal wall of the tube, and there is no external hazard.
At higher voltages X-​rays may escape and suitable shielding must be provided. It
is also necessary to ensure that there is no path by which X-​rays can pass from the
inner surface of the tube through insulators. This is because the attenuation of the
X-​rays by the insulators (usually ceramics) is much less than that by metals. It is
usually necessary to provide X-​ray shielding for tubes with an operating voltage
greater than 15 kV [5–​6] (see Section 20.8).
In gridded tubes and magnetrons the spent electrons are collected directly on the
anode of the tube. The dimensions of the anode are limited by the frequency of the
tube and it is important to consider the power density on the anode surface during
the design process. Because the anode is positive with respect to the rest of the tube
any secondary electrons which may be liberated are re-​collected on the anode and
do not cause any problems. The exception to this is when the potential of the anode
of a tetrode falls below that of the screen grid so that secondary electrons from the
 Linear Beam Tube Collectors 353

anode are then collected on the grid. This is undesirable in power tetrodes and that
region of operation must be avoided (see Chapters 6 and 12).
In klystrons, travelling-​wave tubes (TWTs), inductive output tubes (IOTs), and
gyrotrons the electrons are collected on an electrode, designed for the purpose
known as the collector. Because this electrode does not form part of the RF inter­
action structure of the tube it is possible to allow the beam to expand under the
influence of its space charge so that the local power density is reduced. It is usual
for the collector to be electrically isolated from the body of the tube so that the
body current can be monitored separately to protect the tube. The power density on
the surface of the collector can be reduced still further by operating the collector at
a potential which is negative with respect to the tube body. This technique, known
as collector depression, is discussed in Section 10.3. The advantages of collector
depression are [7]:
• The DC input to the tube is reduced, and the overall efficiency is increased.
• The power which must be dissipated is reduced so that the cooling pumps and
fans can be smaller, and less power is required to run them.
• The intensity and penetrating power of the X-​rays generated by high voltage
tubes is reduced, so that less shielding is required.
However, it is more difficult to cool electrodes which are not at earth potential. In
addition, great care must be taken to ensure that no primary electrons are reflected
by the collector, and that any secondary electrons liberated are trapped within it.
Electrons which flow back into the interaction region form an unwanted feedback
path which may cause the tube to become unstable. The advantages of collector
depression can be increased by the use of multi-​element depressed collectors.
However, the use of many electrodes increases the complexity and cost of a tube
and its power supply, and it is necessary to balance the advantages and disadvan­
tages in determining the design of the collector. The development of multi-​element
depressed collectors was stimulated especially by the need to achieve the highest
possible overall efficiencies for tubes for use in satellites and spacecraft.

10.2 Linear Beam Tube Collectors

At the exit of a linear beam tube the spent beam is freed from the magnetic focusing
field and allowed to spread out, as shown in Figure 7.8. The kinetic energy of the
electrons can then be dissipated on the surface of a collector which typically has the
form shown in Figure 10.1. The emission of secondary electrons in single stage col­
lectors is not a problem, provided that they cannot pass out of the collector and flow
back through the interaction region of the tube. For this reason it is important that
the magnetic focusing field does not penetrate into the collector where it can serve
to guide electrons back towards the interaction region. When the impact energies of
the electrons on the surface of the collector are high (above about 1 keV) then the
dominant effect is the elastic reflection of the primary electrons (see Section 18.3).
354 Electron Collectors and Cooling

Figure 10.1:  Arrangement of a liquid cooled collector.

Figure 10.2:  Schematic spent-​beam distribution curve.

For this reason the collector usually has the form of a long tube with a conical end
so that electrons which strike the walls cannot be reflected directly back towards the
tube body. It is necessary to ensure that the collector does not have any resonant RF
modes which coincide with the operating frequency of the tube, or its harmonics.
If any resonances are excited then the RF fields may perturb the electron paths in
unexpected ways.
The interaction which takes place in any kind of linear beam tube results in a net
transfer of energy from the electron beam to the RF output. Therefore the aver­
age kinetic energy of the electrons in the spent beam is less than that at the input
to the tube. The electrons in the spent beam do not all have the same velocity on
leaving the tube, and some of them usually have velocities greater than the initial
beam velocity. Figures 10.2 shows, schematically, the fraction of the beam current
which would be collected on an electrode at a potential –Vc with respect to the tube
body, normalised to the potential of the cathode ( −Va )1. This curve assumes that
no primary electrons are reflected and that any secondary electrons generated are
re-​collected by the electrode.

1
Note that in this Chapter the potentials of the electrodes are referred to the potential of the tube body rather
than to the potential of the cathode.
 Linear Beam Tube Collectors 355

The detailed shape of the spent-​beam curve depends upon the type of tube, and
the specific operating conditions. However, the spent-​beam curves for all tubes
show the same features:
• As the potential of the collecting electrode is made more negative the whole of
the beam current is collected until a potential –​Vmin is reached. Thus all electrons
have energies greater than Vmin electron volts.
• When Vc > Vmin the fraction of current collected decreases monotonically as the
collection potential is made more negative.
• For potentials beyond −Vmax the collected current falls to zero indicating that no
electrons have energies greater than Vmax electron volts. It is common to find that
Vmax Va > 1, which shows that some of the electrons have been accelerated by the
RF interaction.
It should be noted that a graph such as that shown in Figure 10.2 represents the
time-​average of the velocity distribution. In practice the electron velocity distribu­
tion varies during the RF cycle.
The power carried by electrons which enter the collector in an energy range (in
electron volts) dV at energy V is V dI. It follows that the total power carried by the
spent beam is represented by the area under the curve in Figure 10.2. The dashed
rectangle shows the energy distribution in the unmodulated beam, and the area of
the rectangle represents the DC beam power I 0Va . If losses in the tube are neglected
the difference between the area of the rectangle and the area under the spent-​beam
curve represents the RF power output of the tube. Since the power in the spent
beam is less than that in the initial beam it follows that the thermal dissipation in
the collector is less than the initial beam power. In most cases, however, the collec­
tor is designed to be able to dissipate the full DC beam power. If this is not the case,
it is essential that the DC input power to the tube is cut, whenever the RF input is
removed, to avoid damaging the collector by overheating it.
In order to design a collector we need to know how the power is distributed on
its surface. For simplicity let us consider an unmodulated beam of radius b, hav­
ing uniform current density and electron velocity, emerging from a magnetic field
mBB into a concentric cylindrical collector whose wall radius is rw , as shown in
Figure 10.3. Each electron within the beam will follow the same normalised path
given by

Z = S ( R ) , (10.1)

where Z = β p z , R is the radius of the trajectory normalised to its initial value,

and the function S is defined by the right-​hand side of (7.82). The axial position
at which a trajectory whose initial radius is r strikes the wall of the collector is
given by

r 
Z = S  w  . (10.2)
 r
356 Electron Collectors and Cooling

Figure 10.3:  Expansion of the spent beam into the collector.

Figure 10.4:  Power density on the surface of a collector.

The electrons in an annulus of radial thickness dr strike the collector along a sec­
tion of normalised length dZ such that

dZ d  rw 
= S   . (10.3)
dr dr  r 

The ratio of the power density at the wall of the collector to that in the initial beam
is given by the ratio of the area of the element of the wall to that of the annulus.
Thus, the normalised collector power density is

−1
r dZ r  d  rw  
Pc = =  S    . (10.4)
rw dr rw  dr  r  

Since Z and Pc can be calculated as functions of r from (10.2) and (10.4) it is pos­
sible to plot the variation in the power density, normalised to the power density in
the initial beam, against axial position, as shown by the example in Figure 10.4 (see
Worksheet 10.1). It is evident that the power density is greatest at the point where
 Linear Beam Tube Collectors 357

the beam first strikes the wall of the collector and it diminishes rapidly thereafter
with increasing Z.
The power carried by the spent beam is less than that in the unmodulated beam
by the RF power extracted from it. There are also two other factors which affect the
power density on the collector wall when the beam is modulated:
• The beam is bunched so that the peak current is greater than the DC beam
current I 0 . The extent of the increase depends upon the type of tube and the
drive level. At saturation the peak current in klystrons and TWTs may be as
high as 4 I 0 .
• The average velocity of the electrons in the bunch is less than the DC velocity.
A fuller discussion of these issues is to be found in Chapters 12, 13, and 14. For an
IOT, only the second point applies because the peak current in the bunch is equal to
the peak cathode current (see Chapter 12). As a consequence of both these factors
the beam spreads more rapidly than it would if it were not modulated. In addi­
tion, the electrostatic potential within the beam is depressed by the increased space-​
charge in the bunch, and by the increase in the radius of the conducting boundary
surrounding the beam. The potential depression can be estimated from (7.8) with
r = 0. If we denote the multiplication factors for the current and the velocity by FI
and Fu then (7.8) can be written in the form

V F
= I β2p b2 1 + 2 ln ( rw b ) , (10.5)
Va Fu

where rw is the inner radius of the collector and relativistic effects have been
ignored. So, for example, in a tube where FI = 2, Fu = 0.6, β p b = 0.12 and rw b = 10
the ratio V Va = 0.27. The reduction in the electron velocity caused by the poten­
tial depression can, therefore, be big enough to be important. It increases the
charge density and causes the potential to be still further depressed so that, in
extreme cases the ratio can reach unity. A virtual cathode is then formed at the
entrance to the collector and some of the electrons are reflected by it. Because it
is highly undesirable for any electrons to find their way back into the interaction
region, the formation of a virtual cathode must be avoided. The easiest way to
achieve this is to arrange for the inner wall of the collector to be conical in the
region between its entrance and the point at which the electrons first strike it. It
should be possible to keep the ratio rw b to less than 4 in the critical region. In
the example given above the voltage ratio could be reduced to around 0.18 in
this way.
Particle in cell (PIC) code simulations of collectors show that, if that the space
potential depression is small compared with the beam voltage, the bunches spread
out into the collector without diffusing very much in the axial direction. It is there­
fore possible to estimate the power density distribution on the inside of the collec­
tor wall at different points in the RF cycle by computations using the instantaneous
values of the parameters.
358 Electron Collectors and Cooling

10.3 Collector Depression

Examination of the schematic spent beam distribution curve in Figure 10.2 shows

that the collector can be biased negatively to potential Vmin with respect to the tube
body without causing any of the electrons to be repelled. The electrons then strike
the collector with reduced energies so that energy is recovered. This is achieved
using the power supply arrangement shown in Figure 10.5. The collector is then
said to be depressed and the net DC input power to the tube is

PDC in = I c (Va − Vc ) + I bVa = I 0Va − I cVc . (10.6)

where I c is the collector current, I b is the body current and the potential of the col­
lector with respect to the tube body is −Vc . Substituting from (10.6) into (1.18) the
RF efficiency is

P2
ηrf = . (10.7)
I 0Va − I cVc
where P2 is the useful RF output power at the fundamental frequency, and it has
been assumed that the gain of the tube is great enough for the RF input power to
be negligible. The RF efficiency without collector depression is

P2
ηrf 0 = , (10.8)
I 0Va
so the RF efficiency has been increased by depressing the collector. The depressed
collector efficiency increases, as Vc is increased, up to the point at which electrons
start to be reflected. An improvement in efficiency by a factor in the range 1.3 to
1.4 is typical of what can be achieved by simple techniques. The improvement in

Figure 10.5:  Circuit diagram for a tube with a single-​stage depressed collector.

 Collector Depression 359

Figure 10.6:  Power recovered by a single-​stage depressed collector.

efficiency is a result of the reduction in the heat dissipated, but the price paid is
increased complexity of the power supply and the problem of cooling a high volt­
age electrode. If the potential depression is increased further then some of the
slower electrons are reflected. They can be prevented from entering the tube body
by including an earthed electrode between the magnetic pole piece and the collec­
tor on which any reflected electrons can be collected [8]. The collector then has two
electrodes, but it is known as a single-​stage depressed collector because only one
electrode is depressed. The extent to which the efficiency of a tube can be increased
is limited by the reflection of electrons.
Since the power dissipated in the collector is the integral under the spent beam
curve it follows that the effect of depressing the collector is to reduce the integral by
the shaded area shown in Figure 10.6. The power recovered is given by

Prec = I c Vc . (10.9)

This increases until Vc = Vmin and decreases for greater potential depression because
some electrons are reflected and collected at zero potential. The power carried by
the spent beam entering the collector is

Pent = (1 − ηe ) I 0Va , (10.10)

where ηe is the electronic efficiency defined in (1.20) and RF losses within the tube
are assumed to be negligible. Since the body current is typically of the order of 1%
of the beam current we will assume that I c ≈ I 0 . The efficiency of the collector is
defined as the ratio of the power recovered to the power entering the collector

I 0 Vc Vc
ηcol = = . (10.11)
( e ) 0 a ( ηe )Va
1 − η I V 1 −
Then from (10.7) the RF efficiency with a depressed collector can be written

P2
ηrf = (10.12)
I 0Va (1 − ηcol (1 − ηe ))
360 Electron Collectors and Cooling

Figure 10.7:  Dependence of the RF efficiency of a tube on electronic efficiency and collector

efficiency
(copyright 1970, NASA, reproduced, with permission, from [9]).

so that, making use of (1.22)

ηe ηc
ηrf = , (10.13)
(1 − ηcol (1 − ηe ))
where ηc is the circuit efficiency, and interception of current on the tube body has
been neglected in this simplified derivation. This equation does not depend upon
the details of the collector so it is also valid for the multi-​element collectors dis­
cussed in the next section.
If the effects of RF losses are ignored ηc = 1. Then the effectiveness of collec­
tor depression can be shown by plotting ηrf against ηe with ηcol as a parameter.
Figure 10.7 shows that, for a given electronic efficiency, the RF efficiency increases
as the collector efficiency increases. The increase in the RF efficiency is greatest
when the electronic efficiency is low.

10.3.1 Multi-​Element Depressed Collectors

Consideration of Figure 10.6 shows that still greater improvements in efficiency can
be achieved by using a multi-​element depressed collector, as shown in Figure 10.8.
The electrons are sorted into energy ranges and collected on electrodes at potentials
corresponding to the lowest energy in each range. The equation for the collector
efficiency (10.11) now takes the form
N

∑I V
cn cn

ηcol = n =1
. (10.14)
(1 − ηe ) I cVa
 Collector Depression 361

Figure 10.8:  Circuit diagram for a tube with a three stage depressed collector.

Figure 10.9:  Schematic spent-​beam diagram for a tube with a three-​stage depressed collector.

where Vcn is the potential of the nth collector electrode, I cn is the current collected
by it and

I c = ∑ I cn . (10.15)
n

The corresponding spent-​beam curve is shown in Figure 10.9 and it is clear that

the shaded area, which represents the power recovered, is much greater than in
Figure 10.6. Thus it is possible to increase the efficiency of a collector by increas­
ing the number of electrodes. Note that Figure  10.9 assumes that no current is
collected at earth potential. In practice it is usual for the potential of the first
electrode to be equal to that of the tube body and a small current is collected by it
even with a well-​designed collector. The potentials of the electrodes do not need
to be equally spaced as shown in Figure 10.9 but can be chosen to maximise the
power recovered.
362 Electron Collectors and Cooling

Useful insight into the performance of a tube with a multi-​element depressed col­
lector can be gained by considering a simplified model adapted from that described
by Kosmahl [10]. Let us assume that the spent-​beam curve in Figure 10.9 can be
approximated by a trapezium whose sloping side is defined by voltagesVmin andVmax .
The collector has N depressed electrodes at equally spaced potentials which collect
equal fractions of the beam current. The power entering the collector is

1
Pent = I 0 (Vmax + Vmin ) , (10.16)
2
where it is assumed that the body current is zero. From the geometry of the diagram
the power recovered is

1
Prec = Pent − I 0 (Vmax − Vmin ) . (10.17)
2N
Thus the collector efficiency is

ηcol =
Prec
= 1−
(Vmax −Vmin ) . (10.18)
Pent N (Vmax + Vmin )

Now the electronic efficiency of the tube is

PDC in − Pent Vmax + Vmin

ηe = = 1− (10.19)
PDC in 2Va

so that

Vmax + Vmin = 2Va (1 − ηe ) (10.20)

and, substituting for (Vmax + Vmin ) in (10.18)

1  Vmax 
ηcol = 1 −  − 1 . (10.21)
N  Va (1 − ηe ) 

If it is assumed that there are no accelerated electrons, so thatVmax = Va , then (10.21)

reduces to

ηe
ηcol = 1 − . (10.22)
N (1 − ηe )
In the form of this equation given by Kosmahl the factor N is replaced by ( N − 1)
because of the presence of an additional electrode at potential Va which collects
no current [10]. Equation (10.22) shows that the collector efficiency increases as
N increases, as would be expected, but that it decreases as the electronic efficiency
increases because the fraction of the power entering the collector which can be
recovered decreases. Substituting for the collector efficiency from (10.22) in (10.13),
 Collector Depression 363

and ignoring circuit losses, we find that the efficiency of the tube with a depressed
collector is

N
ηrf = . (10.23)
N +1
Thus, this simple model predicts that the efficiency of the tube depends only on the
number of stages in the collector and is independent of the electronic efficiency.
The overall efficiency increases as the number of stages increases but the improve­
ment obtained by adding extra stages is small if N is greater than 5. Thus, for most
tubes, it is not worth using more than four or five stages because of the additional
cost and complexity of the collector and of the power supply for the tube.
The theoretical efficiency indicated by (10.23) is not achieved in practice because
the spent beam curve is not normally a trapezium, and because no account has
been taken of the effect of electrons which enter the collector with energies
greater than Va. Equation (10.23) also assumes that each electron is collected as
efficiently as possible. This is not always the case, as is shown in the next section.
In particular, if the electronic efficiency changes as a result of changes in the
drive level, then the minimum electron energy Vmin changes. With fixed electrode
potentials the number of stages which are effective changes, and the potentials
are not generally optimal. In practice the electrode potentials are chosen to max­
imise the shaded area below the spent-​beam curve. It is not necessarily the case
that the best overall efficiency is obtained by designing a tube to achieve the best
electronic efficiency and then adding a multi-​element depressed collector. This is
because the shape of the spent beam curve can vary considerably as the design
of the tube is changed. It is therefore necessary to ensure that the shape of the
spent-​beam curve is such that as much of the energy as possible can be recovered
by the collector [11]. Changes in the design of the tube may also result in changes
in the circuit efficiency.

10.3.2 Non-​Ideal Multi-​Element Depressed Collectors

In an ideal multi-​element collector each electron would be collected at the most
negative potential possible for its energy. Thus all electrons for which Vn ≥ Ve > Vn +1
would be collected at potential Vn. It has been shown experimentally that this situa­
tion can be approached quite closely, provided that the space-​charge density is very
low [12]. However, as the space-​charge density is increased some of the electrons
in this energy range continue to be collected at potential Vn−1 so that the collec­
tion efficiency is reduced [12–​13]. To show how this affects the collector efficiency
we will assume that the probability that an electron with energy Ve is collected at
potential Vn is

Ve − Vn
pn (Ve ) = (10.24)
Vn +1 − Vn
364 Electron Collectors and Cooling

and the probability that it is collected at potential Vn−1 is

pn −1(Ve ) = 1 − pn (Ve ) . (10.25)

If Ve < V1 then the electrons are collected at potential V0 = 0 so that p0 = 1. It is also

assumed that pN = 1 when Ve > 2VN − VN −1. Figure 10.10 shows the probability that
electrons will be collected on each electrode of a five-​stage collector as a function of
the normalised electron energy (see Worksheet 10.2). The dashed rectangles show
the current collected by an ideal collector and the triangles show the current col­
lected according to the model defined by (10.24) and (10.25).

Figure 10.10:  Theoretical probabilities that an electron having energy Ve will be collected by

the electrodes of a five-​stage depressed collector.
 Collector Depression 365

Figure 10.11:  Theoretical DC collector efficiency as a function of electron energy.

The DC collector efficiency, that is the collector efficiency for mono-​energetic

electrons, can be computed as a function of the normalised electron energy of the
electrons from

1
ηdc (Ve ) = ∑ p (V ) ⋅V . (10.26)
n e e
Ve n

Figure  10.11 shows the DC collector efficiency for a five-​stage collector with
equally spaced voltages. It is found that the shape of this curve varies very little
if the linear variation of probability is changed to a sine or sine squared curve
(see Worksheet 10.2). The curve is similar to that computed for an experimental
five-​stage collector [13]. The differences can be explained by the collection of some
of the current on electrodes at less favourable potentials. This is caused partly by
imperfections in the design, and partly by secondary electrons.
It has been shown that the RF collector efficiency can be computed to good accu­
racy from the DC collector efficiency and the spent-​beam curve [13]. If the current
collected on a single depressed electrode at a potential Vc is I c then the fraction of
the total current in the small energy range dVe at energy Ve  is

 dI 
dI c =  c  dVe (10.27)
 dVc  Ve

and the power entering the collector associated with this current is

 dI 
dPent = dI cVe =  c  Ve dVe . (10.28)
 dVc  Ve
366 Electron Collectors and Cooling

Thus the total power entering the collector is given by

Vmax
⌠  dI 
Pent =   c  Ve dVe , (10.29)
⌡Vmin  dVc  Ve

where the integral is taken over the full range of electron energies in the spent beam.
Now the fraction of the power entering the collector in the energy range dVe which
is recovered is ηdc (Ve ). Thus, the total power recovered is

Vmax
⌠  dI 
Prec =   c  Ve ηdc (Ve ) dVe . (10.30)
⌡Vmin  dVc  Ve

The collector efficiency with an RF modulated beam is found by dividing (10.30)

by (10.29). Modern multi-​element depressed collectors have collector efficiencies
exceeding 80% and therefore the majority of the electrons are being collected at the
most favourable potential [14–​16] (see Worksheet 10.2).

10.4 Design of Multi-​Element Depressed Collectors

The design of a multi-​element depressed collector begins with knowledge of the

spent-​beam curve for the tube. The shape of the curve depends upon the type of
tube, and upon the level of the RF input power (see Figures  12.19, 13.15, and
14.16). It also depends on the frequency within the working bandwidth of the tube.
Although the energy distribution in the spent beam can be determined experimen­
tally, it is now usual to obtain the information from a large-​signal computer model.
For this purpose a 2.5-​dimensional model is needed since the radial and angular
velocities of the electrons are important. The number of collector stages is chosen
as a compromise between collector efficiency and complexity, and the potentials of
the stages are chosen to maximise the energy recovered.
The task of designing a multi-​element depressed collector is one of arranging
the electrodes in such a way that the electrons are sorted according to their initial
energies, and collected with minimum impact energies. Since these impact energies
may be of the order of a few hundred electron volts, the generation of second­
ary electrons is more important than reflection of primary electrons, though both
processes occur (see Section 18.3). It is important that any secondary electrons,
or reflected primary electrons, are collected on the surface from which they were
emitted to avoid degradation of the collector efficiency. It is also important that
there should be no back-​streaming electrons flowing from the collector into the
interaction region. In a multi-​element depressed collector the presence of a number
of electrodes at different potentials means that, in general, the electrons are flow­
ing through a succession of electrostatic lenses which may be either converging or
diverging. Since the electrons have a range of initial vector velocities, and the effects
 Design of Multi-Element Depressed Collectors 367

Figure 10.12:  Arrangement of a symmetrical multi-​element depressed collector.

of the space-​charge field cannot be ignored, the problem of synthesising the design
of the collector is a difficult one.
Multi-​element depressed collectors can be classified as either symmetric (that is
axi-​symmetric) or asymmetric [17]. Efficient symmetric collectors have the general
arrangement shown in Figure  10.12. The electrons are collected on the backs of
the electrodes so that any secondary electrons are emitted into a retarding field and
tend to be returned to the electrode from which they were emitted. Two essential
features of collectors of this type are the conical shape of the last electrode, and
the spike which protrudes from it along the axis of symmetry. The purpose of the
spike is to provide a radial electric field to aid the dispersion of the beam in a region
where the space-​charge forces are weak. It also opposes the motion of secondary
electrons generated on the cone towards the axis. Ideally the potential of the cone
and the spike should be −1.5Va so that few, if any, electrons would have sufficient
energy to reach them. It is not always convenient to provide this potential from the
power supply to the tube, and it is more usual for the final electrode to be at the
potential of the cathode.
Kosmahl proposed a ‘Dispersive-​Lens collector’ in which the electrodes were
placed along equipotential surfaces determined from a given variation of the
potential along a boundary surrounding the collector [17–​18]. This choice of
the  positions of the electrodes means that the electric fields within the collector
are independent of the sizes of the apertures in the electrodes. The best aperture
sizes are those which are just large enough to allow all the electron trajectories to
pass through them. Since the field variation is smooth within the collector the lens
effects are minimised. In principle it should be possible to design a collector of this
type by finding mutually consistent solutions of Poisson’s equation, and the equa­
tions of motion of the electrons, subject to a prescribed potential distribution on a
cylindrical or spherical boundary surrounding the beam. A Dispersive-​Lens collec­
tor built for a 200W 12 GHz coupled-​cavity TWT had a collector efficiency of 81%
and it was thought that 90% should be achievable [17].
368 Electron Collectors and Cooling

An alternative approach is the ‘Individual-​Lens collector’ [17] which has a similar

form, except that the potentials of the electrodes are not dependent on a prescribed
boundary potential distribution, but may be chosen freely. The potential distribu­
tion around each electrode can then form either a converging or a diverging lens
and it is found that this can be beneficial to the operation of the collector. An
Individual-​Lens collector is normally larger that a Dispersive-​Lens collector for the
same tube but the collector efficiency which can be achieved is greater. The design
of an Individual-​Lens collector for a dual-​mode TWT was described by Dayton
et al. [19].
Most of the early work on efficient multi-​element depressed collectors concen­
trated on symmetric designs because, at that time, no computer programs existed
which could be used to study and optimise asymmetric collectors. Nevertheless, it
was recognised that it was not possible to design a symmetric collector which was
completely free from back-​streaming [18]. Hechtel pointed out that, if space-​charge
effects are ignored, the electric field between a pair of electrodes can be used as
an electron velocity spectrometer [20]. He observed that some field distributions
are inherently focusing whereas others are defocusing. Since electrons, having the
same velocity, enter the collector with a range of angles relative to the axis it is
desirable to employ a focusing field which causes electrons which start from a point
with the same velocity, but at different angles, to arrive at a single point when they
strike an electrode. In particular he showed that the fields in symmetric collectors of
the types described above are inherently defocusing. However, collectors where the
beam is not injected on the axis of symmetry the field can be made to be focusing.
Hechtel showed that, of the types of collector he studied, the hyperbolic asym­
metric collector (Figure  10.13) could be expected to have higher efficiency than

Figure 10.13:  Arrangement of Hechtel’s hyperbolic asymmetric collector

(copyright 1977, IEEE, reproduced, with permission, from [20]).
 Design of Multi-Element Depressed Collectors 369

Figure 10.14:  Arrangement of an asymmetric multi-​element depressed collector.

a parabolic symmetric collector but this conclusion was challenged by Kosmahl

[17]. The use of an asymmetric design also greatly reduces the possibility of back-​
streaming. Hechtel’s hyperbolic collector is potentially rather bulky and the theo­
retical electrode shapes must be approximated by planes and cones for simplicity
of manufacture.
The wide variety of possible configurations for symmetric and asymmetric col­
lectors which can exist was classified by Mihran and Neugebauer [9]. The collec­
tors which have been described so far have zero magnetic field and non-​uniform
electric field. An alternative type of collector which is commonly used because it
is compact is illustrated in Figure 10.14. The collector elements are cylindrically
symmetric but meet at planes which are at an angle to the axis. This produces
a transverse component of the electric field which sweeps secondary electrons
sideways to avoid back-​streaming. An alternative way of introducing asymmetry
into a collector is by use of a transverse magnetic field. This can conveniently be
achieved empirically by placing a small permanent magnet alongside an axisym­
metric collector. Some high-​efficiency collectors employ combinations of tilted
electric fields and magnetic fields, including continuation of the PPM field into the
collector [16].
The design and optimisation of multi-​element depressed collectors requires
the use of three-​dimensional computer models which can take account of space-​
charge, reflected primary electrons, and secondary electrons [21–​ 24]. More
recently, genetic algorithms have been used for efficient optimisation of the shapes
and potentials of the electrodes of both symmetric and asymmetric collectors
[22, 25–​27].

10.4.1 Suppression of Secondary Electrons

Problems associated with secondary electrons can be reduced by ensuring that the
secondary electron emission coefficient of the electrode surfaces is low (see Section
18.3). This is achieved by using materials which have low secondary electron emis­
sion coefficients, such as carbon and graphite, and/​or by creating a textured sur­
face which tends to trap the secondary electrons. This was recognised very early
in the development of depressed collectors, and tests were carried out with elec­
trodes which were carbonised or coated with lamp-​black (carbon black) [8]. These
coatings have very low secondary electron yield but are unsuitable for use in tubes
370 Electron Collectors and Cooling

because of their poor adhesion [28]. Subsequently, several techniques have been
developed [29–​30]:
• Oxygen free high conductivity copper is commonly selected for collector elec­
trodes because of its high thermal and electrical conductivity and ease of fabrica­
tion. However, the secondary electron yield of copper is greater than unity over
a wide range of primary impact energies commonly found in depressed collec­
tors. For this reason surface coatings of materials such as titanium carbide, and
surface texturing by ion bombardment, have been used to reduce the secondary
electron yield. A low-​cost technique for depositing carbon on copper electrodes
is described in [31].
• Pyrolytic graphite is a strong, lightweight, material with low secondary electron
emission yield. It was found in comparative tests that a collector with electrodes
made of pyrolytic graphite had an efficiency which was several percentage points
better than a collector having copper electrodes with the same dimensions.
Pyrolytic graphite with a textured surface gave results which were better then
those with a smooth surface and virtually identical to those obtained using car­
bon black [28].
• High-​purity isotropic graphite has adequate strength for use in collectors and is
readily machined. The secondary electron yield when there is a surface texture is
close to that for carbon black and slightly lower than that for textured pyrolytic
graphite [32].

10.4.2 Reconditioning the Spent Electron Beam

When a linear-​beam tube is operated at, or close to, saturation the spent electron
beam is strongly bunched and the electrons can have appreciable radial veloci­
ties. It is more difficult to collect such a beam efficiently than an unmodulated
beam. That is because electrons, which start from the same point with different
radial velocities, follow different paths through the collector. It has been found
that, if the spent beam does not emerge abruptly from the magnetic focusing
field, but passes through a region where the beam is allowed to expand adi­
abatically to a larger diameter, then the space-​charge and the radial velocities are
reduced and the beam is easier to collect [17, 33–​34]. The magnetic field in this
region typically includes a field reversal. Beam conditioning, in this way, is most
useful for tubes with high electronic efficiencies and/​or high perveances, such as
klystrons and coupled-​cavity TWTs, because the disorder in the spent-​beam is
greatest. However, it has been shown that the use of a refocusing region is not
necessary for most helix TWTs. In that case it is sufficient to allow the beam to
drift through a short length of the PPM field after it has left the helix and before
it enters the collector [35]. As the beam has not been allowed to expand, it is pos­
sible for the collector to be smaller than is the case when a refocusing section is
employed.
 Cooling 371

10.5 Cooling

The greatest part of the heat generated in a tube is in the collector of a linear-​
beam tube or gyrotron, or in the anode of a magnetron or gridded tube. The
body of the tube is heated by interception of electrons, and by RF losses in the
interaction structure. Thus attention must be paid to adequate cooling of all
parts of a tube. The different parts of the tube may be cooled by conduction, by
natural or forced convection of air or liquid, or by boiling liquid. These meth­
ods are reviewed briefly below. All are ways of transferring heat from the very
high power density at the tube surface to some surface where the surface area is
larger, the power density is lower, and the heat can be dissipated more readily. It
is necessary to take care that the efficiency of the cooling system is not reduced
by the accumulation of dust, by contamination of coolants, or by corrosion of
the surfaces. It is also necessary to consider the ambient temperature conditions
in which the equipment will be stored and used. If the ambient temperatures fall
below freezing then it is necessary to ensure that liquid coolants do not freeze,
and have low enough viscosities for the cooling system to work properly when it
is started up. If the ambient temperatures are high then the efficiency of air cool­
ing systems is reduced and there is a risk that liquid coolants may boil, or that
parts of the collector may overheat. General information about cooling systems
is to be found in [5–​6, 36].

10.5.1 Conduction Cooling

Low power tubes, such as helix TWTs, which generate not more than a few hundred
watts of heat can be cooled by conduction from the tube to a suitable heat sink. In
some mobile applications the heat sink can be the external surface of the vehicle
(e.g. an aircraft or a ship). In spacecraft the heat must be radiated into space and
heat pipes may be used to convey the heat from the tube to the radiating surface
[37]. Alternatively, radiating fins can be mounted directly on the outside of the col­
lector. The tube is then mounted so that these fins are on the outside of the space­
craft and can radiate directly into space.

10.5.2 Air Cooling

Natural, or forced, convection of air can be used to cool regions of a tube where
the power density is not too high. The usual method is to use an array of cooling
fins through which air can flow, or can be forced by a fan. This has the advantage
that it is not necessary to provide liquid coolants. It is important that the air intake
is fitted with an effective filter to prevent the accumulation of dust on the cooling
surfaces.
372 Electron Collectors and Cooling

10.5.3 Liquid Cooling

At higher power levels it is common to pump a coolant through suitably placed
narrow channels, as shown in Figure 10.1. The coolant may be de-​ionised water,
but other liquids (e.g. fluorocarbons) have been used. The channels must be small
enough to ensure turbulent flow. There is a trade-​off between the power required to
pump the liquid around the system and the thermal power which can be removed.
It is necessary to avoid cavitation and local boiling since those can cause hot spots.
If, for example, the heat is conducted though a copper wall 1  cm thick, cooled
by water at a maximum temperature of 100 °C, then the maximum power density
is 700 W cm −2 . This figure can be increased to 2 kW cm −2 by using very small
pipes close to the surface to be cooled [4]. In multi-​phase cooling water is pumped
through narrow tubes and permitted to boil locally. Power densities up to 2.5 kW
cm–2 can be handled in this way [36]. The output water temperature is below 100 °C
and the steam generated condenses back into the coolant.

10.5.4 Vapour Phase Cooling

In vapour phase cooling the part of the tube to be cooled is immersed in a liquid
coolant (usually de-​ionised water) which is allowed to boil. The vapour is circu­
lated by natural convection to a heat exchanger where it is condensed. Figure 10.15
shows one possible arrangement. The effective surface area for heat transfer can
be increased by arranging a number of large bore cooling channels around the
collector. The method has the advantage that the temperature of the tube is kept
constant. The power density is up to 150 W cm −2 [36, 38].
In a development of this technique, known as Hypervapotron® cooling, the sur­
face of the part to be cooled is provided with a large number of narrow bore blind
channels into which the liquid flows. At high power densities the heat flux causes
explosive local boiling which expels the heated liquid from the channels. Cooler
liquid is sucked back in and the process is repeated [4, 38]. Power densities of up to
2 kW cm −2 are possible.

Figure 10.15:  Vapour phase cooling of a power gridded tube

(copyright 1976, EEV, reproduced, with permission of Teledyne e2v, from [39]).
 References 373

References

[1]‌ C. Shibata et al., ‘High-​power (500 kW) c.w. magnetron for industrial heating’, Electrical
Engineering in Japan, vol. 111, pp. 94–​100, 1991.
[2]‌ W. P. Bennett, ‘Large-​power-​tube design, processing and testing’, in R. S. Burnap, ed.,
Electron Tube Design. Harrison, NJ: RCA, pp. 763–​791, 1962.
[3]‌ A. S. Gilmour, Jr., Klystrons, Traveling Wave Tubes, Magnetrons, Crossed-​ Field
Amplifiers and Gyrotrons. Norwood, MA: Artech House, 2011.
[4]‌ G. Faillon, ‘Technical and industrial overview of RF and microwave tubes for fusion’,
Fusion Engineering and Design, vol. 46, pp. 371–​381, 1999.
[5]‌ L. Sivan, Microwave Tube Transmitters. London: Chapman and Hall, 1994.
[6]‌ Varian, Technical Manual:  Installation, Operation, Maintenance, Care and Handling
Instructions, General: Microwave Tubes, Magnetron Tubes, Electron Tubes, Publication
no. T.0.00-​25-​251, 1 October 1979.
[7]‌ F. Sterzer, ‘Improvement of traveling-​wave tube efficiency through collector potential
depression’, IRE Transactions on Electron Devices, vol. 5, pp. 300–​305, 1958.
[8]‌ H. J. Wolkstein, ‘Effect of collector potential on the efficiency of traveling-​wave tubes’,
RCA Review, vol. 19, pp. 259–​282, 1958.
[9]‌ T. G. Mihran and W. Neugebauer, ‘Analytical study of a depressed collector for linear
beam microwave amplifiers’, NASA Lewis Research Center, Cleveland, OH, 1970.
[10] H. G. Kosmahl, ‘How to quickly predict the overall TWT and the multistage
depressed  collector efficiency’, IEEE Transactions on Electron Devices, vol. 27,
pp. 526–​529, 1980.
[11] D. S. Komm et al., ‘Advances in space TWT efficiencies’, IEEE Transactions on Electron
Devices, vol. 48, pp. 174–​176, 2001.
[12] T. Okoshi et al., ‘The tilted electric field soft-​landing collector and its application to a
traveling-​wave tube’, IEEE Transactions on Electron Devices, vol. 19, pp. 104–​110, 1972.
[13] W. Neugebauer and T. G. Mihran, ‘A ten-​ stage electrostatic depressed collector
for improving klystron efficiency’, IEEE Transactions on Electron Devices, vol. 19,
pp. 111–​121, 1972.
[14] H. G. Kosmahl and P. Ramins, ‘Small-​ size 81-​to 83.5-​ percent efficient 2-​and
4-​stage depressed collectors for octave-​bandwidth high-​performance TWT’s’, IEEE
Transactions on Electron Devices, vol. 24, pp. 36–​44, 1977.
[15] P. Ramins and T. A. Fox, ‘90-​to 93-​percent efficient collector for operation of a dual-​
mode traveling-​wave tube in the linear region’, IEEE Transactions on Electron Devices,
vol. 26, pp. 1662–​1664, 1979.
[16] G. Faillon et al., ‘Microwave tubes’, in J. A. Eichmeier and M. K. Thumm, eds, Vacuum
Electronics: Components and Devices. Berlin: Springer-​Verlag, pp. 1–​84, 2008.
[17] H. G. Kosmahl, ‘Modern multistage depressed collectors –​a review’, Proceedings of the
IEEE, vol. 70, pp. 1325–​1334, 1982.
[18] H. G. Kosmahl, ‘A novel, axisymmetric, electrostatic collector for linear beam micro­
wave tubes’, NASA Lewis Research Center, Cleveland, OH, 1971.
[19] J. A. Dayton, Jr. et  al., ‘Experimental verification of a computational procedure for
the design of TWT-​refocuser-​MDC systems’, IEEE Transactions on Electron Devices,
vol. 28, pp. 1480–​1489, 1981.
[20] J. R. Hechtel, ‘A novel electrostatic-​focusing depressed collector for linear beam tubes’,
IEEE Transactions on Electron Devices, vol. 24, pp. 45–​52, 1977.
374 Electron Collectors and Cooling

[21] L. Kumar et  al., ‘3-​ dimensional simulation of multistage depressed collectors on
microcomputers’, IEEE Transactions on Electron Devices, vol. 42, pp. 1663–​ 1673,
September 1995.
[22] S. Coco et al., ‘Shape optimization of multistage depressed collectors by parallel evolu­
tionary algorithm’, IEEE Transactions on Magnetics, vol. 48, pp. 435–​438, 2012.
[23] J. Petillo et  al., ‘The MICHELLE three-​ dimensional electron gun and collector
modeling tool:  theory and design’, IEEE Transactions on Plasma Science, vol. 30,
pp. 1238–​1264, 2002.
[24] J. J. Petillo et  al., ‘Recent developments to the MICHELLE 2-​D/​3-​D electron gun
and collector modeling code’, IEEE Transactions on Electron Devices, vol. 52,
pp. 742–​748, 2005.
[25] T. K. Ghosh and R. G. Carter, ‘Optimization of multistage depressed collectors’, IEEE
Transactions on Electron Devices, vol. 54, pp. 2031–​2039, 2007.
[26] Z. Liang et al., ‘Numerical optimization of a multistage depressed collector with sec­
ondary electron emission for an X-​band gyro-​BWO’, IEEE Transactions on Plasma
Science, vol. 37, pp. 2328–​2334, 2009.
[27] T. K. Ghosh, ‘Three-​dimensional modelling and optimization of multistage depressed
collectors’, PhD, Engineering Department, Lancaster University, Lancaster, UK, 2002.
[28] A. N. Curren and T. A. Fox, ‘Traveling-​wave tube efficiency improvement with textured
pyrolytic graphite multistage depressed collector electrodes’, IEEE Electron Device
Letters, vol. 2, pp. 252–​254, 1981.
[29] P. Ramins et  al., ‘Verification of an improved computational design procedure for
TWT-​dynamic refocuser-​MDC systems with secondary electron emission losses’, IEEE
Transactions on Electron Devices, vol. 33, pp. 85–​90, 1986.
[30] A. N. Curren, ‘Carbon and carbon-​coated electrodes for multistage depressed collectors
for electron-​beam devices: a technology review’, IEEE Transactions on Electron Devices,
vol. 33, pp. 1902–​1914, 1986.
[31] B. T. Ebihara et al., ‘TWT efficiency improvement by a low-​cost technique for depos­
ition of carbon on MDC electrodes’, IEEE Transactions on Electron Devices, vol. 34,
pp. 490–​493, 1987.
[32] P. Ramins and B. T. Ebihara, ‘Isotropic graphite multistage depressed collectors –​a pro­
gress report’, IEEE Transactions on Electron Devices, vol. ED-​36, pp. 817–​824, 1989.
[33] N. Stankiewicz, ‘Analysis of spent beam refocusing to achieve optimum collector effi­
ciency’, IEEE Transactions on Electron Devices, vol. 24, pp. 32–​36, 1977.
[34] J. A. Dayton, Jr. et al., ‘Analytical prediction and experimental verification of TWT and
depressed collector performance using multidimensional computer programs’, IEEE
Transactions on Electron Devices, vol. 26, pp. 1589–​1598, 1979.
[35] P. Ramins et al., ‘A reexamination of spent beam refocusing for high-​efficiency helix TWTs
and small MDCs’, IEEE Transactions on Electron Devices, vol. 35, pp. 539–​548, 1988.
[36] EIMAC, Care and Feeding of Power Grid Tubes, 5th ed. San Carlos, CA: CPI Inc. Eimac
Division, 2003.
[37] M. J. Smith and G. Phillips, Power Klystrons Today. Taunton, UK: Research Studies
Press, 1995.
[38] P. Gerlach and C. Kalfon, ‘New technologies in power-​grid tubes and their impact in
high-​power UHF-​TV operation’, Proceedings of the IEEE, vol. 70, pp. 1335–​1345, 1982.
[39] EEV, ‘Preamble –​ Tetrodes’, ed. Chelmsford, UK: Teledyne e2v, 1976.
11 Beam-​Wave Interaction

11.1 Introduction

Microwave tubes depend for their operation on the interaction between an elec-
tron beam and an electromagnetic structure as shown in Figure  11.1. In the
majority of tubes the geometry is rotationally symmetric. The current in the beam
is modulated by the RF electric field of the structure. The RF current in the beam
then induces currents in the structure which result in a transfer of energy from
the beam to the structure. The electromagnetic structure of a klystron is made
up of cavity resonators (see Chapter 3), whilst in a travelling-​wave tube (TWT)
it is a slow-​wave structure (see Chapter 4). This chapter develops the theory of
these two types of interaction as a foundation for the discussion of klystrons
(in Chapter  13), and TWTs (in Chapter  14). The theory of crossed-​field tubes
is developed in Chapters  15 and 16, and the theory of gyrotrons is covered in
Chapter 17.
The equations describing beam-​wave interactions are non-​linear, and it is neces-
sary to use numerical methods to find solutions to them. However, useful insight
can be gained by assuming that the modulation of the beam is small, compared
with its DC properties. In small-​signal theory the equations are replaced by lin-
earised approximations which can be solved more easily. In a further approxima-
tion ballistic theory ignores the space-​charge forces that the electrons exert on each
other. We shall begin by considering the ballistic theory of the interaction between
an electron beam and the RF fields of a cavity resonator.
The effects of space-​charge can be introduced by treating the electron beam as
a compressible fluid. The space-​charge waves which propagate on a linear electron
beam were first studied by Hahn [1] and Ramo [2]. A considerable body of literature
exists on this subject and the reader is referred to [3] and [4] for convenient summar-
ies of much of it. This chapter deals only with space-​charge wave theory as far as it
is needed to understand klystrons and TWTs. Many other possible types of space-​
charge wave devices have been studied in the past including resistive and inductive
wall amplifiers, velocity jump amplifiers, double beam amplifiers, and beam-​plasma
amplifiers, but these are not of practical importance (see [3, 4] and [5] for further
information). It is possible to develop a space-​charge wave theory of crossed-​field
devices [3] but this is of limited value because, as will be seen in Chapters 15 and 16,
the operation of these tubes is always non-​linear.
376 Beam-Wave Interaction

Figure 11.1:  Arrangement of a linear-​beam tube.

Under large-​signal conditions electron overtaking can occur so that the electron
velocities are no longer single-​valued functions of position. Under these condi-
tions space-​charge forces cannot be neglected and the interaction must be modelled
numerically by tracking the motion of sample electrons under the influence of both
circuit and space-​charge fields. The large-​signal theory of beam-​wave interactions
is discussed in Section 11.8.
The analysis of problems in beam-​wave interaction presents a number of pitfalls
which are a result of the negative charge on the electron. Referring to Figure 11.1
we see that, for an electron beam, the convection current in the beam is in the
negative z direction. That is, the conventional current in a linear-​beam tube flows
from the collector to the gun. The reversal of the direction of the current presents
conceptual problems. The author has found that it is much easier to develop the
analysis assuming that the electrons are positively charged, and this approach is
followed in this book. The results for a beam of negatively charged electrons are
easily recovered, if necessary, by reversing the signs of all the currents and voltages.
In many cases this makes no difference, as we shall see. For convenience, the charge
carriers will be referred to as ‘electrons’ even though their charge has been assumed
to be positive.

11.2 Ballistic Theory of Interaction with a Gap

In many microwave tubes the electron beam interacts with the electric field of one
or more gaps in the surrounding drift tube. These are connected to external cavity
resonators or other electromagnetic structures. In the region of the gap the RF
electric field is usually much greater than the space-​charge field. Useful information
about the interaction between the electrons and the gap can therefore be obtained
by ballistic analysis which neglects the space-​charge forces.

11.2.1 Beam Modulation by a Gridded Gap

Figure 11.2 shows an unmodulated electron beam passing through a gridded inter-
action gap having an RF voltage across it. Gridded gaps are not used in tubes because
the grids would rapidly be destroyed by the power of the electrons intercepted by
 Ballistic Theory of Interaction with a Gap 377

Figure 11.2:  Modulation of an electron beam by the RF electric field in the gap between a
pair of grids (electrons positive).

them. Nevertheless it is useful to begin by studying this case because the analysis is
straightforward and provides insight into the interaction. The sign convention used
is that the gap voltage is positive when charges are accelerated. The direction of the
gap current i g is such that power is flowing from the gap to the beam.
The electric field of the gap is given by

Vg
Ez = − exp ( j ωt ) , (11.1)
g
where Vg is the amplitude of the RF voltage across the gap. The non-​relativistic
equation of motion of an electron is

eVg
z=
 exp ( j ωt ). (11.2)
m0 g
Multiplying (11.2) by z, and integrating with respect to time, gives the change in
kinetic energy of an electron (in electron volts) produced by the gap

m0 2 Vg g 2
V1 = z = ∫ exp ( j ωt ) dz. (11.3)
2e g −g 2

In this equation the limits of integration are the times at which an electron enters
and leaves the gap. Now, if the RF component of the beam velocity is small com-
pared with the DC component ( u0 ), and the electron crosses the centre of the gap
when t = t0, then the time at which it reaches position z is given by

z
t= + t0 . (11.4)
u0
Substituting for t from (11.4) into (11.3) gives

Vg exp ( j ωt0 ) g 2
V1 = ∫ exp ( j βe z ) dz, (11.5)
g −g 2
378 Beam-Wave Interaction

Figure 11.3:  Dependence of the coupling factor for a gridded gap on the electron transit angle.

where the electronic propagation constant is defined by

ω
βe = . (11.6)
u0
Evaluating the integral we obtain

sin (θ 2 )
V1 = Vg exp ( j ωt0 ) = MVg exp ( j ωt0 ) , (11.7)
θ2
where θ = βe g is the gap transit angle and M is the gap coupling factor. Figure 11.3
shows how the gap coupling factor varies with the transit angle (see Worksheet
11.1). We note that M is always less than or equal to unity because the field changes
during the finite transit time of the electrons across the gap. Since M tends to unity
as the transit angle tends to zero it follows that the action of a finite gridded gap is
equivalent to that of a very narrow gridded gap with gap voltage MVg . The beam is
said to be velocity modulated.

11.2.2 Ballistic Electron Bunching

The kinetic energy of electrons leaving the gap can be written

1
m0 u 2 = e (V0 + MVg exp ( j ωt0 )) , (11.8)
2
whereV0 is the potential of the cathode relative to the drift region, including the effects
of space-​charge and beam rotation. V0 and Vg are both positive. As the electrons drift
downstream from the gap the variation in velocity causes them to become bunched
together so that the beam current is modulated. If the modulation is small we can
neglect second-​order quantities and the velocity of an electron leaving the gap is

 MVg 
u = u0 1 + exp ( j ωt0 ) . (11.9)
 2V0 
 Ballistic Theory of Interaction with a Gap 379

Figure 11.4:  Ballistic bunching of a velocity modulated electron beam.

The electron reaches a plane z from the gap at time

z  MVg 
t = t0 +
u0 1 − 2V exp ( j ωt0 ) , (11.10)
0

as shown in Figure 11.4. Under small-​signal conditions the electron trajectories do

not cross and therefore the current at z is related to the current at z = 0 by

( I 0 + I1 ) dt = I 0 dt0 . (11.11)
Differentiating (11.10) with respect to t0 we obtain

dt  MVg 
= 1 − j βe z  exp ( j ωt0 ) . (11.12)
dt0  2V0 
where βe is defined in (11.6). Substituting for dt0 dt in (11.11) gives the RF convec-
tion current at the plane z

 MVg βe z 
I1 = j I 0  exp j ( ωt − βe z ) , (11.13)
 2V0 
since ωt0 = ωt − βe z . Thus, in a frame of reference moving with uniform velocity u0 ,
the phase of the charge bunches leads the phase of the gap voltage by 90°. This
conclusion still holds when the charges are negative. However, because I 0 is negative
the conventional RF current I1 is maximum in the negative direction at the centre
of the charge bunches.
It can be shown that, for longer drift lengths, the amplitudes of the harmonic
currents in a bunched electron beam are given by [6–​8]

 nMVg βe z 
I n = 2I 0 J n  n = 1, 2, (11.14)
 2V0 
380 Beam-Wave Interaction

where J n is the Bessel function of the first kind and order n. This equation,
which ignores space-​charge forces, is valid even when electron overtaking occurs.
However, except in the special case of frequency-​multiplier tubes, it is desirable
to avoid electron overtaking as far as possible. Also, in practical tubes the drift
length between interaction gaps is normally much less than the distance to the first
maximum of (11.14) and space-​charge forces are not negligible. The bunching of a
velocity-​modulated beam, including the effects of space-​charge forces, is discussed
in Section 11.4.

11.2.3 Beam Loading of a Gridded Gap

When (11.8) is integrated over one RF cycle it is found that the mean energy of the
modulated beam is equal to the energy of the unmodulated beam. Thus first-​order
analysis suggests that no power is required to modulate the beam. This is a conse-
quence of the assumption that the electrons cross the gap without any change in
velocity. In practice those electrons which are accelerated cross the gap in a shorter
time, and the effective value of M is increased. Thus these electrons gain more
energy than is shown in the first-​order analysis. Similarly, those electrons which are
retarded lose less energy than shown in the first-​order analysis. Hence the power
carried by the beam is increased by velocity modulation and this can be verified by
numerical integration of the equations of motion (see Worksheet 11.3).
To calculate the power delivered to the beam by the gap we consider the gap to be
made up of a series of elementary gaps whose position is z and whose length is dz.
The transit angle of each elementary gap is small so that M = 1 and the gap voltage
is Vg dz g . Then from (11.13) the current at a remote plane z1 due to the modulation
by an elementary gap is

1 Vg
dI1 ( z1 ) = G0 j βe ( z1 − z ) exp  − j βe ( z1 − z ) dz, (11.15)
2 g
where G0 = I 0 V0 and the phase is referred to the phase of the gap voltage at z.
Note that, for negative charges the signs of both I 0 and V0 are reversed so that G0 is
unchanged. Thus the RF current at z1 is

1  Vg  z1
I1 ( z1 ) = G0   ∫ j βe ( z1 − z ) exp  − j βe ( z1 − z ) dz. (11.16)
2  g  −g 2
Making the substitution α = βe ( z1 − z )(11.16) becomes

1  Vg  βe ( z1 + g 2)
I1 ( z1 ) = G0  j α exp ( − j α ) d α. (11.17)
2  βe g  ∫0

Evaluating the integral gives

1  Vg 
I1 ( z1 ) = G0   −βe ( z1 + g 2 ) e − jβe ( z1 + g 2) + j e − jβe ( z1 + g 2) − j  . (11.18)
2  βe g  
 Ballistic Theory of Interaction with a Gap 381

The power flow from the gap to the beam is given by [9]

1 g2 Vg g 2
∫ E z ( z1 ) I1 ( z1 ) dz1 = ∫ I1 ( z1 ) dz1 , (11.19)
* *
P=
2 − g 2 2g − g 2

where the asterisk denotes the complex conjugate. Substituting for I1 from (11.18)
and performing the integration we find that
2
1  Vg   θ  θ θ 
P = G0   1 − cos θ − sin θ + j  cos θ + − sin θ  , (11.20)
2  θ   2   2 2 

where θ = βe g. The real part of this expression is the active power required to modu-
late the beam

1 sin θ 2  sin θ 2  Vg2 Vg2

Pa = G0 − cos θ 2 = Gb , (11.21)
2 θ 2  θ 2  2 2

where Gb is the beam loading conductance. It is straightforward to show that

Gb 1 d 1 dM
= − θ ( M ( θ )) = − M θ
2
. (11.22)
G0 4 dθ 2 dθ
When relativistic effects are important (11.22) becomes [10]

Gb 1 dM
=− Mθ , (11.23)
G0 γ R (1 + γ R ) dθ

where

1
γR = . (11.24)
1 − u02 c 2

Note that the symbol γ R is used in this chapter for the relativistic factor to avoid
confusion with the use of γ for radial propagation constants. The reactive power
transferred to the beam is given by the imaginary part of (11.20)

1 cos (θ 2 )  sin (θ 2 )  Vg2 Vg2

Pr = − G0 − cos ( θ 2 ) = − B , (11.25)
θ 2  θ 2  2 b
2 2

where Bb is the beam loading susceptance. The negative sign in (11.25) arises because
the reactive power flow into a susceptance B is given by

1 1 V2
Pr = VI * = V ( jBV ) = − jB
*
. (11.26)
2 2 2
Thus the presence of the beam can be represented in an equivalent circuit by the
admittance Yb = Gb + jBb . Figure 11.5 shows how the beam loading conductance,
and susceptance, of a gridded gap vary with the transit angle. Because Gb G0 is
382 Beam-Wave Interaction

Figure 11.5:  Beam loading conductance, and susceptance, of a gridded gap.

Figure 11.6:  Arrangement of a gridless interaction gap (electrons positive).

normally less than 0.2 the power required to modulate the beam is only a small
fraction of the DC beam power. The beam loading conductance may be nega-
tive when the transit angle is greater than 2 π . If the sum of the beam loading
conductance and the conductance of the external circuit connected to the gap
is negative then the system is unstable and can break into self-​oscillation (see
Section 13.4.1).

11.2.4 Beam Modulation by a Gridless Gap

Figure 11.6 shows the arrangement of a gridless gap of the type generally used in
microwave tubes. The axial component of the electric field within the gap can be
found using the method described in Section 3.5. It is assumed that E z is known
as a function of z when r = a as discussed in Section 3.5.3. The Fourier transform
of this function is
∞
Γ (β ) = ∫ E (a, z ) exp ( jβz ) dz. (11.27)
z
−∞
 Ballistic Theory of Interaction with a Gap 383

Since the fields must satisfy the wave equation in free space we know that E z varies
with radius as I 0 ( γ r ) where the radial propagation constant γ = β2 − ω 2 c 2 . Thus
the electric field within the gap is given by the inverse Fourier transform
∞
1 ⌠ I0 (γ r)
E z ( r, z ) = Γ (β ) exp ( − j βz ) dβ. (11.28)
2π 
⌡ I0 (γ a)
−∞

We saw, in Section 3.5.3, that the electric field distribution in the region occupied by
the electron beam depends only weakly on the profile of the field at r = a . We shall
therefore assume, for the moment, that E z = −Vg g when r = a and z ≤ g 2, and
zero elsewhere, so that

sin (β g 2 )
Γ (β ) = −Vg . (11.29)
βg 2
It is convenient to remove the dependence on r by taking the average of the first
term in the integral in (11.28) across the beam. We shall also assume that the elec-
tron beam is in confined flow and that the RF current is nearly constant across the
beam. The average electric field acting on the electrons is then
∞
Vg ⌠ 2 I1 ( γ b ) sin (β g 2 )
Ez ( z ) = − ⋅ ⋅ exp ( − j βz ) dβ (11.30)
2π 
⌡ βb I 0 ( γ a ) βg 2
−∞

and the magnitude of the effective gap voltage is

∞
V1 = − ∫ E z ( z ) exp ( j βe z ) dz. (11.31)
−∞

Substituting for the field from (11.30) and exchanging the order of the
integration gives
∞
Vg ⌠ 2 I1 ( γ b ) sin (β g 2 )  ∞ 
V1 =  ⋅ ⋅  ∫ exp ( − j (βe − β ) z ) dz  d β. (11.32)
2 π ⌡ βb I 0 ( γ a ) β g 2  −∞ 
−∞

Following Branch [9] we note that

∞
∫ exp ( − j (β e − β ) z ) dz = 2 π δ(βe − β ) , (11.33)
−∞

where δ( z ) is the Dirac delta function which is defined as a pulse having zero width,
infinite height and unit area. Thus the integral with respect to z is zero except when
β = βe and then

2 I1 ( γ e b ) sin (βe g 2 )
V1 = ⋅ ⋅ ⋅Vg , (11.34)
βe b I 0 ( γ e a ) βe g 2
384 Beam-Wave Interaction

where βe has been used in place of β in the calculation of γ . Then (11.34) can be
written as

V1 = MVg = µ c µ dVg , (11.35)

where the axial coupling factor given by

sin (βe g 2 )
µd = (11.36)
βe g 2

has exactly the same form as the coupling factor for a gridded gap given by (11.7).
If the electric field at r = a is not uniform this expression must be replaced by the
Fourier transform of the gap field for unit gap voltage. When the ends of the drift
tubes have knife edges so that the field in the gap is given by (3.89) then

µ d = J 0 (βe g 2 ) , (11.37)

where J 0 is the Bessel function of the first kind [11].

The average of the radial variation of the field (known as the radial coupling
­factor) is, for confined flow,

2 I1 ( γ e b )
µc = ⋅ . (11.38)
γ e b I0 (γ e a)

An alternative average is based upon an analysis of the exchange of power between

the field and the beam [9]

I 02 ( γ e b ) − I12 ( γ e b )
µb = . (11.39)
I0 (γ e a)

In practical cases, the difference between the numerical values given by (11.38) and
(11.39) is negligible and either expression may be used. For a beam in Brillouin flow
the RF current is a surface current at r = b (see Figure 7.6) so that the radial coup-
ling factor is

I 0 ( γ e b)
µB = . (11.40)
I0 (γ e a)

Figure 11.7 shows curves of the radial coupling factor for Brillouin flow, and
confined flow, plotted against γ e b with b a as a parameter. It should be noted that
(11.40) does not include the effect of the radial component of the electric field on
the electron motion. However Brillouin flow is of limited practical importance
because it is normally necessary to use a magnetic field greater than the Brillouin
field to ensure sufficient beam stiffness. We shall therefore assume that (11.38) is a
satisfactory approximation if m > 1.5. The effects of the radial component of the
electric field may be important if the beam stiffness is small.
 Space Charge Waves on Linear Electron Beams 385

Figure 11.7:  The radial gap coupling factor of an ungridded gap for: (a) Brillouin flow, and
(b) confined flow.

11.3 Space Charge Waves on Linear Electron Beams

The RF properties of electron beams under small-​signal conditions, including

the effects of space charge, can be represented by space-​charge waves. The non-​
relativistic equations which govern the motion of the electrons in a linear electron
beam are the continuity equation

∂ρ
∇⋅J = − . (11.41)
∂t
Newton’s second law of motion

du e
= (Eb + Ec ) , (11.42)
dt m0
where E b is the electric field arising from the space charge of the electrons and Ec is
the field applied to the beam by an external microwave structure. The convection
current in the electron beam is related to the charge density and the velocity by
J = ρu (11.43)

and, finally, the relationship between the charge density and the space-​charge field
is given by Poisson’s equation

ρ
∇ ⋅ Eb = . (11.44)
ε0
These four equations involve four dependent variables so, in principle, they can be
solved. The general solutions are rather complicated but the properties of space-​
charge waves can be understood from a simplified solution.
386 Beam-Wave Interaction

As a first stage of simplification let us assume that the electrons are moving par-
allel to the z axis and that the beam extends to infinity in the transverse direction.
We will also assume that all quantities vary only with z and with time. The four
equations (11.41) to (11.44) then become:

∂J z ∂ρ
= − , (11.45)
∂z ∂t
duz ∂uz ∂u e
= + uz z = ( E zb + E zc ) , (11.46)
dt ∂t ∂z m0

J z = ρuz , (11.47)

and

∂E zb ρ
= . (11.48)
∂z ε0
The left-​hand side of (11.46) is expanded as the sum of: the rate of change of the
velocity with time at constant position, and the rate of change with position as the
motion of the electron is followed. In these equations the vector quantities J, u and
E are assumed to possess only z components. The set of equations is non-​linear
because (11.46) and (11.47) contain terms which are the product of two variables.
The next stage is to linearise the equations by assuming the propagation of waves
so that all the dependent variables can be written in the form

a = a0 + a1 exp j ( ωt − βz ) , (11.49)

where a1  a0 and ω and β are the same for all variables. It is then possible to sep-
arate each equation into constant and time-​varying parts. The constant part of
the solution does affect the velocities of the electrons and it has been common
to assume that it is cancelled out by the static field of a background of positive
ions. Although positive ions do sometimes accumulate in the region of an electron
beam, their effects are undesirable, and such accumulations are to be avoided. In
the theory of space-​charge waves it is better to take account of DC space-​charge
potential depression by including its effects as a correction to the DC electron
velocity, as shown in Section 7.2. The small-​signal equations for the time-​varying
quantities are

βJ1 = ωρ1 (11.50)

e
j ( ω − βu0 ) u1 = ( Eb1 + Ec1 ) (11.51)
m0

J1 = ρ0 u1 + ρ1u0 (11.52)

ρ1
− j βEb1 = . (11.53)
ε0
 Space Charge Waves on Linear Electron Beams 387

For the moment we will assume that the RF field of the external circuit ( Ec1 ) is zero.
Then Eb1 can be eliminated between (11.51) and (11.53) to give

e ρ1
(ω − βu0 ) u1 = ⋅ . (11.54)
ε 0 m0 β
Similarly J1 can be eliminated between (11.50) and (11.52) to give

(ω − βu0 ) ρ1 = βρ0 u1. (11.55)

Equations (11.54) and (11.55) both relate u1 to ρ1 and they can be satisfied simul-
taneously only if

eρ0
(ω − βu0 )2 = . (11.56)
ε0 m
The terms in this equation have the dimensions of angular frequency squared and
the expression on the right-​hand side is recognised as the electron plasma frequency
which was introduced in (7.47) and defined by

eρ0
ω 2p = . (11.57)
ε 0 m0
It should be noted that this remains a positive quantity if the charge is negative
because ρ0 is also negative. When the DC electron velocity is large enough for rela-
tivistic effects to be important the rest mass of the electron is replaced by the lon-
gitudinal mass γ 3 m0 (see (1.7) and [3]). The physical significance of the plasma
frequency can be understood by considering the case when the electrons have no
DC component of velocity so that the solution of (11.56) is

ω = ω p . (11.58)

This is the natural frequency of oscillation of a stationary cloud of electrons which

has been disturbed from equilibrium. The possible solutions to (11.56) can there-
fore be written

ω  ωp
β± = = βe  β p . (11.59)
u0
It is convenient to display these solutions in the form of a dispersion diagram, as
shown in Figure 11.8. The slope of each line is equal to the DC beam velocity u0 .
The lines defined by (11.59) represent two possible wave solutions (β + and β − )
whose phase velocities ( ω β ) are, respectively, greater than and less than u0 . The
waves are compression waves of charge density known as fast (β + ), and slow (β − ),
space-​charge waves. The properties of these waves will be examined in more detail
in Section 11.3.3.
Since it has been assumed that the beam has infinite diameter and the electron
motion is confined to the axial direction it follows that the RF electric field only
has a longitudinal component, as shown in Figure  11.9(a). This field exerts the
388 Beam-Wave Interaction

Figure 11.8:  Dispersion diagram for space-​charge waves neglecting the effects of radial
boundaries.

Figure 11.9:  Electric field due to space charge: (a) without radial boundaries and (b) with
radial boundaries.

forces on the electrons which cause the plasma oscillations. The displacement cur-
rent density within the beam is found, using (11.53), to be

∂Dz ω
= j ωε 0 Eb1 = − ρ1 . (11.60)
∂t β
Comparing (11.60) with (11.50) we see that the displacement current density is
equal and opposite to the convection current density J1 so that the total RF current
density is zero. It follows that there is no RF magnetic field in an infinite beam. We
shall see in the next section that this conclusion is modified when the beam is finite
in the transverse direction.

11.3.1 Effect of Radial Boundaries

Any practical electron beam must have finite dimensions normal to the direction of
electron motion. The RF electric field is then able to spread out into the charge-​free
 Space Charge Waves on Linear Electron Beams 389

region outside the beam, as shown in Figure 11.9(b). The space-​charge forces act-
ing on the electrons are therefore less than those for the infinite beam illustrated in
Figure 11.9(a). Thus the plasma frequency is lower, in finite beam, than it is in an
infinite beam. This effect can be represented by multiplying the right-​hand side of
(11.53) by a factor p2 where 0 ≤ p2 ≤ 1. The right-​hand sides of equations (11.54)
and (11.56) are, likewise, multiplied by p2. Thus the conclusions of the analysis in
the previous section are unchanged, provided that the plasma frequency is replaced
by the effective plasma frequency (ω q ) defined by

ω q = p ω p , (11.61)

where p is known as the plasma frequency reduction factor.

We saw in Section 7.4.3 that the diameter of a magnetically focused electron
beam increases as the beam current is increased, except in the limiting case when
the magnetic field tends to infinity. Thus, in order to determine the value of p in any
particular case it is necessary to solve Maxwell’s equations for a cylindrical electron
beam within a concentric metal cylinder in the presence of a finite axial magnetic
field. Under small-​signal conditions the variations in the beam radius can be repre-
sented by charges on the surface of a beam whose radius is constant. This problem
was studied by Brewer and Mackenzie [12, 13] who showed that the plasma fre-
quency reduction factor is the solution of the equation

 1  1 I1 ( τb ) 1  I1 (βb ) K 0 (βa ) + I 0 (βa ) K1 (βb ) 

1 − p2  τb I ( τb ) = βb  I (βb ) K (βa ) − I (βa ) K (βb )  (11.62)
0  0 0 0 0 
where a and b are the tunnel radius and the beam radius (see Figure 11.9(b)), I 0 ( z ),
I1 ( z ), K 0 ( z ), and K1 ( z ) are modified Bessel functions of the first and second kinds,
and β is the axial propagation constant of the waves. It is assumed that the phase
velocity of the waves is small compared with the velocity of light so that the axial
and radial propagation constants are identical in the region outside the beam.
Inside the beam the radial propagation constant τ is given by

 ω 2p ω 2q − 1
τ 2 = β2  2 2  , (11.63)
 ω p ω 0 − 1
where ω 0 is defined by
2
 e Φk 
ω 20 = ω 2q − 4  (11.64)
 2 π m0 b2 
and Φ k is the magnetic flux linked to the cathode. Making use of (7.52) and (7.58)
we can rewrite (11.64) as

ω 20 =  p2 − 2 ( m2 − 1) ω 2p , (11.65)

where m is the ratio of the focusing field to the Brillouin field which was defined in
(7.56). Equation (11.65) shows that ω 20 is positive and equal to ω 2q when m = 1, that it
390 Beam-Wave Interaction

passes through zero when m  1.02, and is then negative for all higher values of m.
The denominator in (11.63) is positive when m = 1 and has an infinity and changes
sign as ω 0 passes through zero. Thus τ is real for values of m close to unity, and
imaginary for higher values of m. The Bessel functions on the left-​hand side of
(11.62) are computed with real or imaginary arguments as appropriate. Equation
(11.62) can be solved numerically for any desired values of βb, a b and m (see
Worksheet 11.2). Since β is the propagation constant of the waves given by

β ± = βe  βq , (11.66)

where βq = ω q u0 , the value of p should be calculated separately for the fast and slow
space-​charge waves. An initial estimate can be found by setting β = βe and refined by
iteration if necessary. Curves of p versus βb for typical values of b a and m are shown
in Figure  11.10. When the effects of radial boundaries are taken into account the
dispersion diagram for the fast and slow space-​charge waves takes the form shown in
Figure 11.11. Comparison with Figure 11.8 shows that negative values of ω do not
normally occur and that the dispersion curves are asymptotic to those in Figure 11.8
for large values of β. For beam voltages greater than about 20 kV the assumption
that the phase velocity of the waves is small compared with the velocity of light is not
valid. It is reasonable to suppose that β should then be replaced by γ = β 1 − u02 c 2 in
(11.62) and (11.63), and in Figure 11.10. It was found that this correction was neces-
sary to obtain good results for the klystron discussed in Section 13.4.3.
Brewer [12] showed how the division of the RF beam current between beam
radius modulation (surface current) and charge density modulation (body current)
depends on the magnetic field. Figure 2 of that paper shows these currents as being
in anti-​phase with each other. There is apparently a value of the magnetic field for
which there is zero net convection current. This must be wrong, on physical grounds,
and seems to be a consequence of using different sign conventions for the charge
in e m0 and ρ. When this error is corrected it is found, using Brewer’s normalisa-
tion, that the body current should be negative. The way in which these two currents
vary with magnetic field is shown in Figure 11.12 when βb = 0.5 and b a = 0.5. The
curves do not vary very much if different parameter values are chosen. When this
figure is compared with Figure 7.6 it can be seen that the two analyses give identical
results in the limit of small RF currents.

11.3.2 Induced Current

When a space-​charge wave is excited on an electron beam, which is surrounded by
a continuous metal drift tube, the radial RF electric field must be terminated by
induced charges on the wall of the drift tube (see Figure 11.9(b)). The movements
of these charges constitute a longitudinal RF induced current in the wall. It is easy
to see that this must be so by considering Maxwell’s equation

⌠ ⌠⌠  ∂D 
 H × ds =   J +  ⋅ dS. (11.67)
⌡ ⌡⌡ ∂t 
 Space Charge Waves on Linear Electron Beams 391

Figure 11.10:  Plasma frequency reduction factor curves: (a) m = 1, (b) m = 1.5 and (c) m = 3
(Figure 11.4(a) copyright 1956, IEEE, reproduced, with permission, from [12]).

Figure 11.11:  Dispersion diagram for space-​charge waves including the effects of radial
boundaries (nominal scales).
392 Beam-Wave Interaction

Figure 11.12:  Dependence of the RF surface and body currents in an electron beam on the
magnetic field parameter m.

If the system is cylindrically symmetrical and the integral of H is taken in the θ dir-
ection along a line just inside the drift tube wall, then the boundary conditions at
the wall require the current in the wall to be equal and opposite to the integral of
H. The two terms on the right-​hand side of (11.67) are, respectively, the convection
current and the axial component of the displacement current within the drift tube.
Thus (11.67) requires that the induced current in the wall should be

iW = − ( IC + I D ) , (11.68)

where IC and I D are the longitudinal convection and displacement currents within
the drift tube. The convection current can be computed using the results of Brewer
(with the sign of the body current reversed). The displacement current (using
Brewer’s normalisation) given by

I1 ( τb ) 1 − βb ⋅ I1 (βb ) K 0 (βa ) + I 0 (βa ) K1 (βb )  I 0 ( τb )

( I D )norm = + ⋅ . (11.69)
τb  I 0 (βb ) K 0 (βa ) − I 0 (βa ) K 0 (βb )  (βb )2

is in antiphase with the convection current. The wall current can be computed
using (11.69), or by integrating H θ over the surface of the drift tube taking account
of the sign of the convection current. The results obtained by these two methods
agree with each other, confirming the correction to the sign of the body current.
Figure 11.13 shows the ratio ( q ) of the magnitude of the wall current to the con-
vection current as a function of βb, and of the ratio b a, for m = 1, 1.5 and 3. The
reduction in the normalised wall current is explained by the increase in the axial dis-
placement current as the separation between the beam edge and the wall increases,
and as the wavelength on the electron beam decreases. The changes resulting from
changes in the magnetic focusing field, and, therefore, in the proportions of surface
 Space Charge Waves on Linear Electron Beams 393

and body current in the beam, are not so marked. The differences between the
curves for m = 1.5 and m = 3.0 are almost indistinguishable and further increases in
m make no difference. In a microwave linear-​beam tube it is normally desirable that
the coupling between the beam and the surrounding circuits should be as strong as
possible. For this reason βb is normally not greater than 0.5, and b a is not less than
0.5. It can be seen from Figure 11.13 that this choice is necessary to ensure that the
induced wall current is high. It is also evident that, because the curves fall steeply
with increasing βb, the coupling between the beam and the wall at harmonics of the
fundamental frequency falls off sharply with increasing order of harmonic. The
curves in Figure  11.13(c) are identical to those in Figure  11.7(b) except that the
horizontal axis is βb rather than γ b because it has been assumed that the electron
velocity is small compared with the velocity of light. If this restriction is removed
it is seen that the magnitude of the induced current in the wall is the product of the
radial coupling factor and the convection current in the beam. Thus the use of γ b in
place of βb here, and in Figure 11.10, ensures the symmetry of the theory.

Figure 11.13:  Ratio of the RF current in the wall of a drift tube to the RF convection current
in a concentric electron beam: (a) m = 1, (b) m = 1.5 and (c) m = 3.
394 Beam-Wave Interaction

In linear-​beam tubes of all kinds the drift tube enclosing the electron beam
is not continuous, but is broken at intervals by interaction gaps. In the case of
the klystron these gaps only occupy a small portion of the total length of the
tube and it is satisfactory to ignore them when computing the reduced plasma
frequency. In TWTs the gaps occupy a much greater proportion of the length of
the tube so that the assumption of a continuous tunnel is no longer appropriate.
A possible way of compensating for this is to use an effective tunnel radius [14].
It should be understood that this correction to ω q is distinct from the interaction
between the space-​charge waves and the gapped tunnel which is dealt with in
Section 11.6.

11.3.3 Transmission Line Representation of Space-​Charge Waves

It is often useful to represent space-​charge waves by equivalent transmission line
modes. To do this we define an RF voltage, known as the beam kinetic voltage (V1 ),
by invoking the non-​relativistic form of the principle of conservation of energy

1
m0 ( u0 + u1 exp ( j ωt )) = e (V0 + V1 exp ( j ωt )) . (11.70)
2

2
Note that V0 is the potential of the cathode relative to the drift space. It corres-
ponds to the DC beam velocity u0 allowing for the effects of space-​charge potential
depression and rotation in the magnetic field. For small signals the left-​hand side of
(11.70) can be expanded to give

2e
u02 + 2u0 u1 exp ( j ωt ) =
m0
(V0 + V1 exp ( j ωt )) , (11.71)
where the second-​order term u12 has been neglected. This equation is true at all times
and so the RF and DC terms must balance separately to give

u0 u1
V1 = (11.72)
e m0

and

u02
V0 = . (11.73)
2 ( e m0 )

Dividing (11.72) by (11.73) we obtain

V1 2u1
= . (11.74)
V0 u0
Note that this equation remains the same if the sign of the charge is reversed. The
RF current density is obtained from (11.50) and (11.52) as
 Space Charge Waves on Linear Electron Beams 395

ωρ0
J1 = u (11.75)
(ω − βu0 ) 1
so that, making use of (11.66) and (11.74), for the two space-​charge waves

ω J0
J± = ± ⋅ V± . (11.76)
2ω q V0

If it is assumed that the DC and RF current densities are uniform then (11.76) can
be integrated across the cross-​section of the beam to give the RF convection cur-
rents of the space-​charge waves

ω I0
I± = ± ⋅ V± . (11.77)
2ω q V0

These waves can be represented in terms of an equivalent transmission line. The

characteristic admittance, known as the electronic admittance, is

ω I0
Ye = ⋅ , (11.78)
2ω q V0

which is positive regardless of the sign of the electronic charge. The kinetic power
flows in the waves are then

1 2
P± = ± V± Ye , (11.79)
2
where the plus and minus signs refer to the fast and slow space-​charge waves
respectively. Equation (11.79) reveals the surprising fact that the slow space-​charge
wave carries negative power. The direction of the power flow is certainly positive
because, as Figure 11.11 shows, the slow-​wave has a positive group velocity. The
explanation of this paradox is to be found in equation (11.76) which shows that, in
the slow space-​charge wave, the electron density and the velocity are in anti-​phase
with one another. Thus when u > u0 , ρ < ρ0 , and when u < u0 , ρ > ρ0 . It follows that
the average kinetic energy of a beam carrying a slow space-​charge wave is less than
the kinetic energy of the unmodulated beam. In order to set up a slow space-​charge
wave, power must be removed from the beam. This unexpected feature of the slow
space-​charge wave provides the key to understanding all kinds of microwave linear-​
beam tubes. When the beam velocity is relativistic the results are qualitatively the
same but the electronic admittance is given by [3]

ωI 0
Ye = , (11.80)
γ R ( γ R + 1) ω qV0

where the relativistic factor γ R is defined in (11.24) and ω q is calculated using the
relativistic value of u0 but without other relativistic corrections.
396 Beam-Wave Interaction

The analysis above ignores the electromagnetic power associated with space-​
charge waves. It can be shown that the ratio of the electromagnetic power to the
kinetic power in a space-​charge wave is of the order of ω q ω [15]. For tubes having
low space-​charge this ratio is small and the effects of electromagnetic power can
be ignored. In many high power tubes the ratio is around 10% and the electronic
admittance should probably be corrected for this effect.

11.3.4 Space-​Charge Waves on Non-​Ideal Electron Beams

In general, the aim of the tube designer is to create an electron beam in which the cur-
rent density and the axial velocity are uniform throughout and there is no scalloping.
In that way the properties of the beam are close to those of the ideal beam assumed
by theory. In practice all beams are non-​ideal to some extent but, provided the varia-
tions from the ideal are not too great, it is possible to use the theory developed above.
We saw in Chapter 7 that the effects of space-​charge potential depression and
beam rotation should be taken into account when calculating the mean axial vel-
ocity of the electrons. These effects also mean that there is a small variation in the
electron velocities across the beam. This variation is normally ignored, but it could
contribute to the noise carried by the beam through the two-​stream instability (see
[4]) and to the Landau damping of space-​charge waves [16–​18]. These effects are
not normally important in high power microwave tubes.
If the current density is not uniform across the beam but does not vary along its
length then that affects the calculation of the plasma frequency reduction factor.
Some idea of the magnitude of the correction needed can be estimated from the
curves for the plasma frequency reduction factors of hollow beams [19]. If the beam
is scalloping then both the plasma frequency and the plasma frequency reduction
factor vary along the length of the beam. Where the beam radius is a maximum
the local plasma frequency is a minimum and the two effects tend to cancel one
another out. It has been shown that, when the variations are averaged along the
beam, the mean value of ω q does not depart very much from the value for the uni-
form beam [20]. When the beam is focused by a periodic permanent magnet system
there is always a periodic variation of the beam radius. The author has found that
good results are generally obtained if the equilibrium beam radius, calculated as
described in Section 7.6.1, is taken to be the effective radius of the beam.

11.3.5 Higher-Order Modes
The analysis in Section 11.2.1 assumed that the waves which are excited on the
electron beam are in the cylindrical TM 01 mode. However, like circular waveguides,
electron beams can also support higher-order space-​charge modes such as those
illustrated in Figure 11.14. These waves exist as pairs of fast and slow waves like the
fundamental modes discussed above. The higher-​order modes are usually ignored
in the analysis of microwave tubes and only those with azimuthal symmetry could
normally be excited by the fields of axially symmetrical structures. Further infor-
mation can be found in [3, 4].
 Space-Charge Wave Theory of Beam and Gap Interaction 397

Figure 11.14:  Examples of higher-​order space-​charge wave modes: (a) TM11, (b) TM 02

and (c) TM 21.

11.3.6 Cyclotron Waves
We have seen that when an electron beam is confined by an axial magnetic field Bz
the electrons follow helical paths. The space-​charge waves already discussed are
associated with axial displacements of the electrons. Another set of modes exists
for which the displacements are azimuthal. These modes, known as cyclotron
waves, occur in fast and slow wave pairs in the same manner as the space-​charge
waves with propagation constants

ω  ωc
β± = , (11.81)
u0
where the cyclotron frequency ω c = eBz m0 . These waves are normally not excited
in linear beam tubes because the azimuthal component of the electric field of the
electromagnetic structure is usually zero. A  fuller discussion of cyclotron modes
can be found in [4, 21].

11.4 Space-​Charge Wave Theory of the Interaction between a Beam

and a Gap

The interaction between an electron beam and a narrow, gridded, gap (M = 1) can
be expressed in terms of the amplitudes of the transmission line analogue of space-​
charge waves. If the beam is initially unmodulated the boundary conditions are

V+ + V− = Vg (11.82)

and, because the convection current must be continuous at the gap,

I + + I − = Ye (V+ − V− ) = 0, (11.83)

where time variation as exp ( j ωt ) has been assumed and the symbol I indicates a
conventional current. The solution to (11.82) and (11.83) is

1
V+ = V− = Vg . (11.84)
2
398 Beam-Wave Interaction

At a plane which is z from the gap the beam kinetic voltage is

V1 = V+ exp ( − j β + z ) + V− exp ( − j β − z )

= Vg cos (βq z ) exp ( − j βe z ) , (11.85)

where time variation exp ( j ωt ) has been assumed. Similarly the RF current is

I1 = I + exp ( − j β + z ) + I − exp ( − j β − z )

= jYeVg sin (βq z ) exp ( − j βe z ) . (11.86)

The corresponding equation derived using ballistic theory is (11.13). The two equa-
tions give the same result for short drift lengths. It can be seen that bunching of
the beam can be thought of as beating between the fast and slow space-​charge
waves. When βq z = π 2 the velocity modulation is zero and the current modulation
is maximum. Similarly, when βq z = π the velocity modulation is maximum and the
current modulation is zero. This pattern is repeated periodically.
The space-​charge wave theory of the interaction between a beam and a gridless
gap can be developed following the same approach as in Section 11.2.4. A small
element of length dz at position z within the field of the gap functions like a
narrow gridded gap so that the change in the amplitude of the kinetic voltage
produced is

dV1 = − E z ( z ) dz (11.87)

and the RF current leaving the elementary gap is zero. This can be expressed in
terms of the amplitudes of the space-​charge waves using (11.84)

1 1
dV± = dV1 = − E z ( z ) dz. (11.88)
2 2
The amplitudes of the space-​charge waves at a plane z1 are found by superimposing
the amplitudes produced by each element dz. Thus
z
1 1
V± ( z1 ) = − ∫ E z ( z ) exp ( − jβ± ( z1 − z )) dz. (11.89)
2 −∞

Now the amplitudes of the waves do not depend on z1, provided that it lies beyond the
point where the field of the gap has fallen to zero. We may therefore extend the upper
limit of integration to infinity and write the amplitudes of the space-​charge waves as

1 ∞
E z ( z ) exp ( − j β ± z ) dz. (11.90)
2 ∫−∞
V± = −

Equation (11.90) has the same form as (11.31) so we conclude that

1 1
V± = M (β ± )Vg = M ±Vg , (11.91)
2 2
 Space-Charge Wave Theory of Beam and Gap Interaction 399

where β ± has replaced βe in the calculation of the gap coupling factor. Hence the
beam kinetic voltage at a plane outside the field of the gap which is distance z from
the centre of the gap is

1
( )
V1 = Vg M + exp ( j βq z ) + M − exp ( − j βq z ) exp ( − j βe z ) . (11.92)
2
Similarly the RF convection current is

1
( )
I1 = YeVg M + exp ( j βq z ) − M − exp ( − j βq z ) exp ( − j βe z ) . (11.93)
2

11.4.1 Current Induced in a Gap by Space-​Charge Waves

We saw, in Section 11.3.2, that the RF current of a space-​charge wave induces a cur-
rent in a surrounding conducting drift tube. This current is determined by the need
to satisfy the boundary condition for the azimuthal component of the RF magnetic
field ( H θ ) on the surface of the drift tube. Figure 11.15 shows the coupling between
a modulated beam and an external circuit through a gap in the drift tube. The sign
convention used for the gap voltage and current is that they are positive when the
gap field is accelerating, and power is flowing from the gap to the beam.
To first order, H θ is unchanged by the presence of the gap and the boundary
conditions require that the tangential magnetic field is continuous at the gap. Thus
the RF magnetic field of the beam is a source which drives the current in the cir-
cuit external to the gap. In circuit terms this is represented by a current generator
connected across the gap. The current of this generator must be the average of the
induced wall current across the gap. Thus, for the two space-​charge waves
g 2
1
iW ± exp ( − j β ± z ) dz, (11.94)
g − g∫ 2
ig ± = −

Figure 11.15:  Current induced in a gap by a modulated beam (electrons positive).

400 Beam-Wave Interaction

where time variation at frequency ω has been assumed and iW is in phase with RF
convection current I1. Carrying out the integration we find that

sin (β ± g 2 )
ig ± = − iW ± = −µ d (β ± ) iW ± . (11.95)
β± g 2
Thus the gap current is related to the induced current in the wall by the axial gap
coupling factor µ d . Now we saw, in Section 11.3.2, that the wall current is less than
the convection current in the beam by a factor which is equal to the radial coupling
factor defined by (11.38) for a beam in confined flow, and (11.40) for a Brillouin
beam. Thus the current induced in the gap is given by

V+ 
i g = (M + I + + M − I − ) = Ye [M + −M − ]   . (11.96)
V− 

11.4.2 Beam Loading of a Gridless Gap

The power carried by a modulated beam is the sum of the powers of the fast and
slow space-​charge wave (we recall that the slow wave carries negative power).
Equation (11.91) shows that the two space-​charge waves excited by the gap do
not have exactly equal amplitudes. Thus active power supplied to the beam by
the gap is

Pg =
Ye
2
( 2
V+ − V−
2
). (11.97)
Substituting for the voltages from (11.91) we obtain

Ye 1
(M +2 − M −2 ) Vg
2 2
Pg = = Gb Vg , (11.98)
8 2
where Gb is the beam loading conductance. This can be written

Gb 1 ∆M 2
=− θ , (11.99)
G0 4 ∆θ
where G0 = I 0 V0 , θ = βe g, ∆M 2 = M +2 − M −2 and ∆θ = θ + − θ −. Equation (11.99),
which is unchanged by changing the sign of the charge on the electron, may be com-
pared with (11.22) which was derived using ballistic analysis. It is found that the
results given by these two equations are virtually identical if ω q ω < 0.1, which is the
situation in many linear-​beam tubes. Thus it is usually satisfactory to ignore space-​
charge effects when calculating the gap coupling factor and the beam loading [22, 23].
The beam loading susceptance of a gridless gap can be calculated using the equa-
tions in [24]. In general

1
Yb = Ye  I av ′ ( j β − g ) − I av ′ ( j β + g ) , (11.100)
2  
 Space-Charge Wave Theory of Beam and Gap Interaction 401

where Ye is given by (11.40) and the definition of I av ′ is given in the paper. For the
special case where the electric field is uniform in the gap the real part of I av ′ is
given by
2
1  sin (β g 2 ) I 02 ( γ b ) − I12 ( γ b )  1
Re  I av ′ ( j β g ) =   = M (β ) , (11.101)
2
⋅
0( )
  2 β g 2 I γ a 2
 

where the definition of the radial coupling factor is given by (11.73) and the nota-
tion has been changed to be consistent with that used in this book. The real part of
the beam loading admittance given by (11.100) is identical to (11.99). The imagin-
ary part of I av ′ for a uniform field is given by

β g − sin (β g )  I 02 ( γ b ) − I12 ( γ b )

Im  I av ′ ( j β g ) = − ⋅
  β2 g 2 I 02 ( γ a )
∞  
 An Bn
 − 2β g ∑  − 2 2 2 . (11.102)
 (β g + rn ) β g + rn 
2 2 2 2
n =1 

In this equation

2 2
g  βa   u 
rn = λ n 1 −    0  , (11.103)
a  λn   c 

where λ n are the zeroes of the Bessel function J 0 ( λ ),

 xn + yn   rn − 1 + exp ( −rn ) 
An = 2  ⋅
2   , (11.104)
 1 − (βa λ n ) ( u0 c )  
2
rn

 xn + yn   3 − 2rn − (3 + rn ) exp ( −rn ) 

Bn =  ⋅
2 

 1 − (βa λ n ) ( u0 c )  
2
rn3 

2 xn ( rn − 1 + exp ( −rn ))
+ , (11.105)
rn3

g 2 J 02 ( λ n b a )
xn = , (11.106)
a 2 J12 ( λ n )

g 2 J12 ( λ n b a )
and yn = ⋅ . (11.107)
a2 J12 ( λ n )
Then the beam loading susceptance can be calculated. The summation in (11.102)
converges rapidly so that three terms are usually sufficient to give good accuracy
(see Worksheet 11.1). Figure 11.16 shows examples of curves of beam loading con-
ductance and susceptance calculated in this way for a number of gridless gaps [10].
402 Beam-Wave Interaction

Figure 11.16:  Beam-​loading conductance and susceptance of gridless gaps of length

g when b a = 0.7
(copyright 1967, IEEE, reproduced, with permission, from [10]).

11.4.3 Beam Interaction with a Passive Gridless Gap

Bers [22] showed that the effect of a gridless gap can conveniently be represented by
two drift regions separated by an equivalent narrow gridded gap. The interaction
between a beam and a passive gap can be represented by the equivalent circuit
shown in Figure 11.17. The admittance YL represents the properties of the circuit
connected to the gap. The voltage induced in the gap by a modulated beam is

ig Ye V1+ 
Vg = − =− [M + −M − ]   , (11.108)
YT YT V1− 
where the initial modulation of the beam at the plane of the gap is represented by
the amplitudes of the space-​charge waves (V1± ) and YT = Yb + YL . The negative sign
in (11.108) is present because power must be transferred from the beam to a passive
circuit. The modulation added to the beam by the gap voltage is given by (11.91).
Thus the modulation of the beam leaving the gap is

V2 +  V1+  Ye M +  V1+ 

V  = V  − 2Y M  [M + M − ] V 
 2 −   1−  T  −  1− 

V1+  Ye  M +2 M + M −  V1+ 
=  − M M . (11.109)
V1−  2YT  + − M −2  V1− 
 Space-Charge Wave Theory of Beam and Gap Interaction 403

Figure 11.17:  Equivalent circuit for the interaction between an electron beam and a passive
gap (electrons positive).

The matrices for the drift regions take the form

V2 +  exp ( − j β + z ) 0  V1+ 

V  =  0

exp ( − j β − z ) V1− 
(11.110)
 2−  
and it is assumed that the drift regions extend outside the space occupied by
the field of the gap. Thus the interaction between an electron beam and a ser-
ies of passive gaps can be represented by cascaded matrices (see Section 13.2).
The interaction with a series of gaps that are connected together is discussed in
Section 11.6.
The relationship between the transmission line and space-​charge wave param-
eters describing the modulation of the beam can be written

V1   1 1  V+ 
 I  = Y −Ye  V− 
. (11.111)
 1  e
Inverting this equation

V+  1 1 Ze  V1 
V  = 2 1 −Z   I  . (11.112)
 −  e 1

From (11.95) the modulation of a beam by a gap is

V+  1 M + 
V−  = 2 M Vg . (11.113)
 −
After the gap, from (11.111) and (11.113),

V2  1  1 1  M +  1  M+ + M−   M 
 I  = 2 Y   
−Ye  M − 
Vg =  Vg = G M Vg , (11.114)
2 Ye (M + − M − )
 2  e  b 
where

1
M= (M + + M − ) (11.115)
2
404 Beam-Wave Interaction

and, from (11.98),

Ye
Gb = (M +2 − M −2 ) . (11.116)
4
Similarly from (11.96) and (11.112) the induced current is

Ye 1 Ze  V1   Gb  V1 
ig = [M + −M − ]    =  M    , (11.117)
2 1 −Ze   I1   M   I1 
where V1 and I1 describe the modulation of the beam entering the gap. The effect of
the gap on a modulated beam is then given by

V2  V1  1  M 
 I  =  I  − Y G M  i g . (11.118)
 2   1 T  b 
Substituting for i g from (11.117) gives, after some rearrangement [25]

 G M2 
 1− b −
 2 
V Y YT  V1 
I  =  G2
T
 . (11.119)
 2 Gb   I1 
− 2 1− 
b

 M YT YT 
This equation is unchanged if the sign of the charge on the electron is reversed.

11.5 Continuous Interaction with a Slow-​Wave Structure

Figure 11.18 shows, schematically, the continuous interaction between an electron

beam and a uniform slow-​wave structure (circuit). At any given frequency the cir-
cuit can be regarded as a transmission line having characteristic impedance Zc on
which the propagation constant is β0. The potential difference between the surface
of the structure and a surrounding earthed conductor is

Vc = V1 exp j ( ωt − β0 z ) . (11.120)

Figure 11.18:  Interaction between an electron beam and a uniform slow-​wave structure

(electrons negative).
 Continuous Interaction with a Slow-Wave Structure 405

The amplitude of the axial component of electric field at the surface of the circuit
is given by (4.11). When the phase velocity is much less than the velocity of light
this becomes

∂Vc
E z (a ) = − = j β0Vc . (11.121)
∂z
If it is assumed that the electron beam is in confined flow (which has been seen to
be a good approximation if m > 1.5) then the effective electric field acting on the
electrons is the average over the cross-​section of the beam so that

E z = j β0 µ c ( γ 0 )Vc , (11.122)

where the radial coupling factor µ c is given by (11.38) and γ 0 = β20 − ω 2 c 2 . The
relationship between the electric field acting on the beam and the power flow along
the circuit can be written as
2
Ez
= µ c2 Zc = ZC , (11.123)
2β02 P
where ZC is the Coupling Impedance of the circuit averaged over the cross-​section
of the beam. In the limiting case when the radius of the beam is vanishingly small
then (11.123) becomes

E z (0)
2
Zc
= = ZP , (11.124)
2β20 P I0 (γ 0a)
2

where ZP is known as the Pierce Impedance of the circuit [26].

Now consider the interaction taking place in a short length dz, as shown in
Figure 11.18. The changes in the amplitudes of the space-​charge waves caused by
the field of the circuit are given by (11.88). These must be added to those caused by
the propagation of the waves already existing on the beam so that

∂V± 1 ∂Vc
= − j β ±V± + µ c . (11.125)
∂z 2 dz
The signs in this equation assume that the charge on the electron is negative. This is
essential for consistency with the conventional voltages and currents on the circuit.
The amplitude of the wave on the circuit (Vc ) can be expressed as the sum of the
amplitudes of the forward and backward waves so that (11.125) becomes

∂V± 1 ∂
= − j β ±V± + µ c (V f + Vb ) . (11.126)
∂z 2 dz
Note that we have assumed here that the beam does not load the circuit appreciably
so that the modulation of the beam can be expressed in terms of the circuit voltage.
This is justified because the beam impedance is much greater than that of the circuit.
406 Beam-Wave Interaction

The RF displacement current flowing from the beam into the element of the cir-
cuit can be found by applying the equation

⌠⌠  ∂D 
  J +  ⋅ dS = 0 (11.127)
⌡⌡ ∂t 
S

to the closed surface bounded by the planes z and z + dz and the surface of the cir-
cuit at r = a . Then the radial displacement current flowing into the element of the
circuit is

∂ ∂I
diR = − ( IC + I D ) dz = −µc C dz, (11.128)
∂z ∂z
where IC and I D are the convection current and the axial displacement current, as
defined in (11.68). The radial displacement current is divided equally between the
forward and backward directions so that the increments in the forward and back-
ward waves on the circuit are

1 ∂I
dV = − µ c Zc C dz. (11.129)
2 ∂z
This increment is added to amplitudes of the waves which already exist on the cir-
cuit. The convection current can be expressed as the sum of the currents in the fast
and slow space-​charge waves so that the forward and backward waves on the circuit
satisfy

∂V f 1 ∂
= − j β0V f + µ c ZcYe (V+ − V− ) (11.130)
∂z 2 ∂z
and

∂Vb 1 ∂
= j β0Vb − µ c ZcYe (V+ − V− ) . (11.131)
∂z 2 ∂z
Note that we have assumed that the coupling is weak so that the characteristic
impedances of the modes are not significantly altered by it. We have also assumed
that the coupling takes the form of induction of current on the structure. This
is justified because the impedance of the beam is much higher than that of the
structure.
Equations (11.126), (11.130) and (11.131) can be rearranged as coupled-​mode
equations (see Section 1.3.8). The result is

V f   β0 0 − Aβ + − Aβ −  V f 
   0 −β0 Aβ + Aβ −  Vb 
∂ Vb
j  = , (11.132)
∂z V+   Bβ0 − Bβ0 β+ 0  V+ 
    
V−   Bβ0 − Bβ0 0 β −  V− 
 Continuous Interaction with a Slow-Wave Structure 407

where

1
A= µ c ZcYe (11.133)
2
and

1
B= µ c . (11.134)
2
If the voltages are normalised, following Louisell [21], by dividing them by the
square root of the appropriate characteristic impedance, then the magnitudes of
the coupling terms are typically of the order of 1% of the diagonal terms. This
confirms the validity of the assumption that the coupling as weak, and that the
propagation constants and characteristic impedances of the modes are not changed
by it. The propagation constants of the four coupled modes are the eigenvalues of
the matrix so that

[C ] − β [ I ] = 0, (11.135)

where [C ] is the matrix in (11.132) and [ I ] is the unit matrix. Expanding the deter-
minant gives the dispersion equation of the coupled system

[β20 − β2 ] βq2 − (βe − β)2  = −µc2 ZcYeβqβ0β2 . (11.136)

Substitution for Ye using (11.78) gives

1
[β20 − β2 ] βq2 − (βe − β)2  = − 2 (µc2 ZcG0 ) βeβ0β2 . (11.137)
where G0 = I 0 V0 . This is a quartic equation in β which therefore has four roots at
any given frequency. The bracketed term on the right-​hand side of this equation is
the ratio of the coupling impedance defined by equation (11.123) to the DC imped-
ance of the beam. It may also be noted that the right-​hand side of the equation is
independent of ω q.
To examine the solutions of (11.137) let us consider first the case when the cou-
pling between the beam and the circuit is removed, by setting µ c = 0. The solutions
are then

β = ±β0 (11.138)
for the forward and backward waves modes on the circuit in the absence of the
beam, and

β = βe  βq (11.139)

for the fast and slow space-​charge waves on the beam. Figure  11.19(a) shows
the uncoupled modes defined by these equations. The axes are normalised to the
synchronous point ( ω s , β s ) which is defined here as the point of intersection of
408 Beam-Wave Interaction

Figure 11.19:  Continuous interaction between an electron beam and a uniform slow-​wave

structure: (a) uncoupled modes and (b) coupled modes. The real parts are shown by
solid and dashed lines. The imaginary parts multiplied by 50 by the dotted lines. At the
synchronous point A = 0.0014, B = 0.25 , and ω q ω = 0.1.

the forward circuit wave, and the slow space-​charge wave, curves in the absence
of coupling.1 The parameter values are typical of those found in helix TWTs but
for simplicity it has been assumed that the reduced plasma frequency is constant.
Coupled-​mode theory (see Section 1.3.8) leads us to expect type C coupling between
the forward wave on the circuit, and the slow space-​charge wave, in the region where
they are close to synchronism. Figure 11.19(b) shows the coupled-​mode diagram
computed from (11.132). The coupling produces a region where a pair of complex
conjugate solutions exists. These are known as the growing wave and the decaying
wave. It is evident that the backward wave on the circuit is hardly affected by the
coupling. Similarly, the fast space-​charge wave is not much affected by the cou-
pling, especially if the space-​charge density is high.
The solutions of (11.137) were first derived by Pierce, who assumed that the
backward wave was unchanged, and solved the resulting the cubic equation [26, 27].
He employed normalised parameters which have remained part of the common

This definition differs from that commonly used (β0 = βe ) but, in the author’s opinion the definition used
1

here is to be preferred.
 Continuous Interaction with a Slow-Wave Structure 409

language of TWT theory although they are not an ideal choice. The Pierce param-
eters describing a loss-​less tube are:
• The gain parameter
1 1
I Z 3
G Z 3
C= 0 C =  0 C  , (11.140)
4V0  4 

which resembles the constant term on the right hand side of (11.137).
• The velocity parameter

1  u0 
b= − 1 , (11.141)
C  v p 

where v p is the phase velocity of the waves on the circuit.

• The space-​charge parameter

2
1  ωq ω 
QC =  1 + ω ω  . (11.142)
4C 2  q 
Curves giving the solutions to the quartic equation (11.136) in terms of the Pierce
parameters were published by [28, 29]. Today it is better to program a computer to
find the eigenvalues of (11.132) directly (see Worksheet 11.4).
Useful insight can be obtained from the two-​wave approximation which assumes
that only coupling between the slow space-​charge wave and the forward wave on the
circuit is important. The approximate equation is obtained from (11.132) by drop-
ping the terms in Vb and V+ to give

∂ V f   β0 − Aβ −  V f 
j = . (11.143)
∂z V−   Bβ0 β −  V− 
The eigenvalues of this matrix are the solutions of the quadratic dispersion equation

 1 
β2 − (β0 + β − ) β + 1 + µ c2 ZcYe  β0β − = 0, (11.144)
 4 
which can be solved using the usual formula to give

1 1
β= (β 0 + β − ) ± (β0 − β − )2 − β0β − µc2 ZcYe . (11.145)
2 2
This solution reveals a number of important things about the travelling-wave inter-
action. If the term under the square root is negative then β has a pair of complex
conjugate roots. These are the growing and decaying waves whose amplitudes vary
with z as exp ( ± α z ) where

1
µ c2 ZcYe β0 β − − (β0 − β − ) . (11.146)
2
α=
2
410 Beam-Wave Interaction

The real part of each of these roots is given by the first term of (11.145) which is
the mean of the two uncoupled roots. The validity of the two-​wave approximation
increases as the plasma frequency increases because the coupling between the slow
space-​charge wave and the forward wave on the circuit is reduced. Table 11.1 shows
a comparison between the values of the four propagation constants computed from
(11.132), and the approximate values obtained from (11.145) (see Worksheet 11.4).
The same parameter values are used as in Figure 11.18. The figures in brackets are
those for the uncoupled modes. It can be seen that the difference between the two
sets of values is small.
At the synchronous point (11.145) becomes

 µ 
β = β0 1 ± j c ZcYe  , (11.147)
 2 
where the positive sign corresponds to the growing wave. This is very close to the
condition that the imaginary part of β is greatest. Thus the maximum gain per cir-
cuit wavelength ( λ 0 = 2 π β0 ) is given by

 (  )
Gain = 20 log 0 exp πµ c ZcYe  = 27.3µ c ZcYe dB λ 0−1 . (11.148)

The parameter µ c2 ZcYe is typically in the range 0.01 to 0.02 corresponding to gain
of around 3 to 4 dB per wavelength. It is not surprising that the gain increases
with coupling impedance (corresponding to increasing interaction field for a given
power flow), and with the electronic admittance of the beam (corresponding to
increasing current and decreasing voltage).
If we define x = β − β0 , y = 2α β0 and C ′ = µ c2 ZcYe then (11.146) can be written as

y = C ′x − ( x − 1) . (11.149)
2

Figure  11.20 shows curves of y against x for a range of typical values of C ′.
Careful examination of these curves shows that they are not quite symmetrical.
The peak of the curve lies very close to x = 1 but moves gradually to the right as C ′

Table 11.1:  Comparison between 2-​wave and 4-​wave models of the continuous interaction between an
electron beam and a slow-​wave structure

ω ωs Model β β0 β β0 β β0 β β0

0.5 2 waves 1.010 1.076 (0.744) (–1.000)

4 waves 1.012 1.075 0.742 –1.000
1.0 2 waves 1.001+0.019i 1.001–​0.019i (0.829) (–1.000)
4 waves 1.001+0.019i 1.001–​0.019i 0.828 –1.000
1.5 2 waves 0.986+0.005i 0.986–​0.005i (0.858) (–1.000)
4 waves 0.988 0.985 0.856 –1.000
2.0 2 waves 0.996 0.962 (0.872) (–1.000)
4 waves 0.997 0.961 0.871 –1.000
 Discrete Interaction with a Slow-Wave Structure 411

Figure 11.20:  Illustration of the dependence of the bandwidth of a TWT calculated from

(11.149) on C ′ = µ 2 ZcYe.

increases. The curves show that the gain per unit length is very nearly proportional
to C ′ . Figure  11.20 also shows that the bandwidth of the interaction increases
as C ′ increases. Since the bandwidth has been defined in terms of the ratio β − β0
the bandwidth in terms of frequency depends also upon the angle of intersection
between the dispersion curves for these two modes. The angle can be reduced by
reducing the effective plasma frequency of the beam, and also by adjusting the
shape of the dispersion curve of the slow-​wave circuit. Although the curves in
Figure  11.20 show the general properties of the travelling-wave interaction, it is
important to remember that the gain curves for real TWTs differ from them in
detail because of dependence of ω q, µ c ,Zc , and Ye on frequency.
The analysis presented so far has ignored the effect of losses in the slow-​wave
structure. In practice all real slow-​wave structures have some transmission loss in
the absence of the electron beam. Also it is common to introduce additional loss
to ensure stability against oscillations. The effects of the cold loss of the circuit can
be included in the theory by replacing β0 for the forward wave by (β0 − j α 0 ). The
two-​wave approximation to the solution of the dispersion equation (11.145) shows
that the growing wave then grows as exp ( α − α 0 2 ). The effect of this is to reduce
the overall gain of a section of slow-​wave structure by half the cold loss in decibels.
A more exact figure can be arrived at in particular cases by allowing for the cold loss
when solving (11.132).

11.6 Discrete Interaction with a Slow-​Wave Structure

In high-​power TWTs the slow-​wave structure is not continuous and the field act-
ing on the beam is the superposition of the fields of a series of gaps [30]. It is
self-​evident that this is true for coupled-​cavity, ring-​bar, and other similar periodic
slow-​wave structures (see Chapter 4). In fact the field of a helix should really also
be considered in the same way, particularly for the space harmonic fields. Whilst it
412 Beam-Wave Interaction

Figure 11.21:  Interaction between an electron beam and a periodic slow-​wave structure

(electrons negative).

is possible to model the forward-​wave interaction in a helix TWT as continuous, it

does not provide an adequate description for backward wave interactions.
In order to model the interaction between an electron beam and a periodic slow-​
wave structure let us consider a single cell of the circuit whose pitch (length) is p,
as shown in Figure 11.21. For consistency the polarity of the gap voltage has been
reversed so that the electron charge is negative. Within this cell we will use the drift-​
gap-​drift model described in Section 11.4.3. This model is satisfactory for the circuits
used in space-​harmonic TWTs because the gaps are small compared with the pitch
of the circuit, and the fields of adjacent gaps do not overlap to any appreciable
extent. There is a problem, however, if the model is applied to forward-​fundamental
coupled-​cavity TWTs because the length of the gap is then comparable with the
pitch of the circuit and the fields of adjacent gaps do overlap. Any attempt to model
the simultaneous interaction between an electron and the fields of two, or more, gaps
results in great difficulties. The best solution seems to be to regard the total electric
field within one cell of the circuit as ‘belonging’ to one gap. This approach results in
discontinuities in the field at the junctions between the cells, and all the induced cur-
rent is regarded as ‘belonging’ to that cell rather than being divided between adjacent
cells. Nevertheless, the results obtained show that the approximation is acceptable.
These issues will not be pursued any further here and we will assume that overlap-
ping between the fields of adjacent gaps can be neglected.
The matrix describing the interaction between the beam and a localised gap is
found, by reference to Figure 11.17, and using (11.91) and (11.96) to be

V2 +   1 0 M+ 2  V1+ 
V  =  0  
 2−   1 M− 2  V1−  . (11.150)
 iL  M +Ye −M −Ye Yb  Vg 
This matrix is unchanged when the signs of the currents and voltages are reversed.
The term Yb is typically of the order of 4 × 10 −6 which is very small compared with
the other terms and can, therefore, be neglected. The physical explanation for this
is that the beam impedance of several hundred kilohms is much greater than the
total impedance of the circuit (a few hundred ohms) so that the beam does not load
the circuit to an appreciable extent. This must be distinguished from the situation
 Discrete Interaction with a Slow-Wave Structure 413

which exists when the beam interacts with a resonant circuit where the beam and
circuit impedances are comparable with each other. Equation (11.150) can be writ-
ten in partitioned form as

v2   I B2  v1 
iL  =  B Y  Vg  (11.151)
 3 b 

V  1 M 
where [v ] =  + , [ I ] is the 2×2 unit matrix, [ B2 ] =  +   and [ B3 ] = Ye [M + −M − ].
V−  2 M − 
The generalised matrix for localised coupling to a transmission line, when there
is no localised energy storage is, from (4.22),

V2 f   1 0 − NZc 2  V1 f 

  
V2 b  =  0 1 NZc 2  V1b  , (11.152)
 Vg   N N 0   iL 

where Zc is the characteristic impedance of the line and N = ZT Zc from (4.28).

Equation (11.152) can be written in partitioned form as

V2   I S2  V1 
Vg  = S 0  iL 
(11.153)
   3

V  NZc  − 1
where [V ] =  f  , [S2 ] = and [S3 ] = [ N N ]. Eliminating Vg and iL
Vb  2 1 
between (11.151) and (11.153) yields

V2   I −Yb [S2 ][S3 ] [S2 ][ B3 ] V1  =

[G ] v 1  . (11.154)
V
v  =   
 2   [ B2 ][S3 ] I   v1   1

We will assume that the gap is at the centre of the cell so that the drift matrix [ D ]
for half the pitch is

V2 f  exp ( − j β0 p 2 ) 0 0 0  V1 f 

V   0 exp ( 0 )
j β p 2 0 0
 
 2b  =   V1b  .
 V1+ 
V
 2+   0 0 exp ( − j β + p 2 ) 0
V2 −   
 0 0 0 exp ( − j β − p 2 ) V1− 

(11.155)
Thus the complete representation of the cell is

[V2 ] = [D ][G ][D ][V1 ] = [ P ][V1 ]. (11.156)

The normal modes of the coupled system are the eigenvalues of the matrix [ P ].
These are found in the usual way by requiring that [V2 ] = λ [V1 ] so that the values of
λ are the solutions of the dispersion equation

[P ] − λ [I ] = 0. (11.157)
414 Beam-Wave Interaction

Figure 11.22:  Dispersion diagram for coupling between an electron beam and a periodic
slow-​wave structure. The imaginary part ( × 50 ) is shown by the dotted line.

Since [ P ] is a 4 × 4 matrix, equation (11.157) has four roots which may be expressed
in terms of the propagation constants of the coupled system using

λ n = exp ( − j β n p ) . (11.158)

Thus (11.157) represents the properties of a discretely coupled system in the same
way that (11.135) represents the properties of a continuously coupled system. For
any particular combination of beam and circuit it is possible to find a numerical
solution to (11.157) (see Worksheet 11.5). Figure 11.22 shows the coupled modes
obtained for a typical case where ω and β are normalised to the synchronous values.
The predicted region of gain is very similar to that seen in Figure 11.18(b). There
is, however, one important difference because the periodic nature of the interaction
also permits interactions with space-​harmonic waves.
The interaction with the m = −1 space harmonic of the backward wave on the
circuit is shown in more detail in Figure 11.23. This diagram shows Type B cou-
pling between the backward circuit wave and the fast space-​charge wave, and Type
D coupling between the backward circuit wave and the slow-​space-​charge wave
(see Section 1.3.8). These interactions are predicted by coupled-​mode theory. To
understand their implications we must apply the correct boundary conditions. For
coupling between the backward wave and the fast space-​charge wave the only per-
mitted solutions are those in which the power in the forward wave is progressively
reflected into the backward wave. Thus, for a sufficiently long region of coupling,
the power in the forward wave is completely reflected leading, to a stop band in the
dispersion diagram. For coupling between the backward wave and the slow space-​
charge wave, however, the transfer of power from the slow space-​charge wave to the
backward wave causes the amplitude of the space-​charge wave to grow. It is there-
fore possible for a backward wave with finite amplitude at the input to be created
by the injection of a very low noise power in the space-​charge wave. The result is
the possibility of backward-​wave oscillation, which is discussed in the next section.
 Backward-Wave Interactions 415

Figure 11.23:  Detail of the dispersion diagram for coupling between an electron beam and a
periodic slow-​wave structure showing the interaction with the backward wave.

11.7 Backward-​Wave Interactions

It was shown in the previous section that the space-​charge waves can interact with
the m = −1 space harmonic of the backward wave on the structure. This leads to the
possibility of amplification of the backward wave. Since energy must be conserved
for the system comprising the beam and the slow-​wave structure, it is not possi-
ble for any amplification to result from the interaction with the fast space-​charge
wave. The possibility of an interaction between the backward circuit wave and the
slow space-​charge wave has generally been studied using the theory of a continu-
ous interaction [3, 31–​33]. The equations are developed in exactly the same man-
ner as in Section 11.5 except that the sign of the characteristic impedance of the
slow-​wave structure is reversed because the power flow is in the negative direction.
In order to understand the implications of this it is sufficient to use the two-​wave
approximation so that (11.144) becomes

 1 
β2 − (β −1 + β − ) β + 1 − µ c2 ZcYe  β −1β − = 0, (11.159)
 4 
where β −1 = −β0 + 2 π p and circuit impedance Zc is that for the appropriate space
harmonic. The solutions to this equation are

1 1
β= (β −1 + β − ) ± (β −1 − β − )2 + β −1β − µc2 ZcYe (11.160)
2 2
and we see at once that they are always real. At the synchronous point β −1 = β − and

 µ 
β = β −1 1 ± c ZcYe  . (11.161)
 2 
Thus the amplitude of the backward wave must vary as

( ) ( )
Vb =  A exp j β −1 zµ c ZcYe 2 + B exp − j β −1 zµ c ZcYe 2  exp ( − j β −1 z ) ,
 
(11.162)
416 Beam-Wave Interaction

where A and B are constants. If a backward wave interaction takes place in a

system of length L where the output end of the circuit is matched, then Vb must
be finite when z = 0 and zero when z = L . The conditions for this are that A = B
(
so that the amplitude of the backward wave varies as cos β −1 zµ c ZcYe 2 and )
β −1Lµ c ZcYe 2 = π 2 . Thus the oscillations are just possible when
2
 π 
Ye =  Yc . (11.163)
 β −1Lµ c 
Substituting forYe from (11.78) gives the least current for which oscillation can occur
2
2ω qV0  π 
I0 = Yc . (11.164)
ω  β −1Lµ c 
For helix slow-​wave structures µ c is calculated for the TM11 mode. More exact results
for the start-​oscillation current can be obtained by solving the full dispersion equa-
tion and applying the appropriate boundary conditions. Good agreement has been
shown between the theoretical and experimental values of start-​oscillation current
[34]. If the current is insufficient to permit oscillations to occur, and a backward
wave is launched at the output end of the structure, then the boundary conditions
are modified and an amplified wave appears at the input end. However, because the
interaction produces a standing wave, the growth is sinusoidal rather than exponen-
tial, as in the case of the forward-​wave interaction.
Equation (11.159) was derived assuming that all quantities varied sinusoidally
with time. To examine the rate of growth of the oscillations with time it is necessary
to allow the frequency to be complex [31]. We therefore replace ω by ( ω − j δ ) so that

jδ
β −1 → β −1 − , (11.165)
vp

where v p is the phase velocity of the wave on the circuit. The sign of the imaginary
part is positive because the slope of the dispersion curve of the backward wave is
negative. Similarly

jδ
β− → β− + . (11.166)
u0
It can be shown that, when there are no reflections at the output of the section,
oscillations can only occur at the synchronous point [31]. Then (11.160) becomes

2
jδ  1 1  1  1 1
β = β −1 +  −  ± µ c2 ZcYe β2−1 − δ 2  +  , (11.167)
2  v p u0  2  v p u0 

where it is assumed that δ  ω so that second-​order terms may be neglected. It

can be seen that β is now complex. When the boundary conditions are applied it is
found that δ increases as the beam current is increased beyond the start-​oscillation
 Large-Signal Modelling of Beam-Wave Interactions 417

current. Thus, in general, both β and ω are complex when a backward wave inter-
action occurs, confirming the existence of an absolute instability. Their values are
found by solving the dispersion equation subject to the appropriate boundary con-
ditions [33]. Once oscillation has commenced the amplitude increases with time
until it is limited by non-​linear effects.

11.8 Large-​Signal Modelling of Beam-​Wave Interactions

The fundamental equations for the interaction between an electron beam and an
RF electric field in one dimension are (11.45) to (11.48). The theory of space-​charge
waves was developed by linearising these equations using the small-​signal assump-
tion in (11.49). The result is an Eulerian analysis which treats the electron beam as
a compressible fluid. It is possible to develop this analysis so that it applies under
large-​signal conditions [35–​37]. The method breaks down whenever electrons over-
take one another so that the properties of the beam at a given plane are no longer
a single-​valued function of position.
In general, large-​signal theory must use Lagrangian analysis in which the elec-
tron beam is described by a set of sample electrons. The motions of these electrons
are found by self-​consistent numerical integration of the equations of motion, tak-
ing account of the space-​charge field and the circuit field. It is possible to use com-
mercially available particle in cell (PIC) codes for this purpose [38]. These codes
model the whole electromagnetic system of the tube, including the electron dynam-
ics, using either a finite difference or finite element method. The model can include
the motion of large numbers of sample electrons in three dimensions so that it is
possible to model a complete tube. This approach is very time consuming even with
the fastest modern computers.
For many purposes it is better to use specially constructed models in which the
properties of the electromagnetic structure outside the interaction region are mod-
elled by equivalent circuits. This has the advantage that their properties are known
and do not have to be re-​computed at each time step. It is also possible to achieve
good results using quite small numbers of sample electrons each of which is asso-
ciated with a disc, or ring, of charge. The disc models divide the unmodulated
beam into identical transverse slices and track the electron motion only in the axial
direction using radially averaged values of the electric field. The motion of a disc
is defined by two coordinates ( z, z ) which are functions of time and the model is
therefore one-​dimensional. Ring models subdivide the discs into concentric rings
of charge which are able to move independently of one another. In these models
azimuthal symmetry is assumed, the motion of a disc is defined by five coordinates
( )
r, z, r, θ , z , and they are therefore known as 2.5-​dimensional models. It is usual to
track the motion of the set of sample electrons which enter the interaction space
in one RF cycle. In disc models 24 identical discs are usually sufficient.
The fundamental equations describing the problem ((11.45) to (11.48)) show that
the variables depend on both distance and time. In order to carry out the numerical
418 Beam-Wave Interaction

integration of the equations it is necessary to choose one of these to be the inde-

pendent variable. When the independent variable is z (a distance stepping model)
the coordinates of the electrons in a 1D model are (t, z ). Conversely in a time step-
ping model their coordinates are ( z, z ). As the solution proceeds it is necessary to
compute the current induced in the electromagnetic structure at intervals, and the
electric field of the space-​charge. Both of these are continuous functions which
have to be reconstructed from the information available from the sample electrons.
The convection current at a plane, which is strictly periodic in time, is determined
from the times at which the sample electrons cross that plane. The electric field of
the space-​charge is found from the positions of the sample electrons at a given time.
Thus the convection current can be computed directly in a distance stepping model,
but the space-​charge field must be calculated by interpolation. Conversely, in a time
stepping model the space-​charge field can be computed directly, but interpolation is
necessary to find the induced current. Examples of computer programs using both
methods are to be found in the literature (see, for example: time step [39, 40], and
distance step [41, 42]).
The variation of the beam current with time at any plane is computed by asso-
ciating a periodically repeated pulse of current with each sample electron. This
current can be expressed as a set of harmonic currents by Fourier analysis, and the
results summed over all the electrons, to give the amplitude and phase of each har-
monic of the beam current. It might be thought that this method would be subject
to numerical noise when the modulation of the beam is small, but that has not been
found to be a problem.
The calculation of the space-​charge field is more difficult because the distribution
of charge density in space is not periodic. Thus, if the time-​step method is used, it is
necessary either to assume an approximate periodic distribution, or to track more
than a single wavelength of electrons so that the forces on the electrons at the ends
of the group are calculated correctly. It is important to be able to model electrons
which overtake one another, and also those which overtake those in an adjacent
wavelength. With the distance-​step method the positions of the electrons are found
by interpolation on data from previous steps, or by extrapolation to future steps.
This does not raise as many problems as might appear at first because the space-​
charge forces have short range, and those arising from discs at a distance greater
than half a wavelength are usually negligible. Also, as we shall see, it is necessary to
make multiple passes through the interaction space to find a self-​consistent solution
so it is possible to store the arrival times of the electrons at each plane. Thus data
from a previous iteration can be used when calculating the space-​charge field of
electrons whose positions are beyond the current plane. The actual calculation of
the space-​charge field from the positions of the electrons can be achieved by assign-
ing charges to the nearest nodes of a spatial grid and then solving Poisson’s equation
numerically [40, 43]. However, the space-​charge calculation forms the innermost of
a number of nested loops in the computational algorithm and it therefore has a
very important effect on the time taken by the computation. For this reason it is
better to compute the field of a single disc, or ring, and store the results in a look-​up
 Large-Signal Modelling of Beam-Wave Interactions 419

table [41, 44] (see Worksheet 11.6). The infinitely thin discs used in early models led
to problems with singularities in the field as electrons passed each other. That prob-
lem is removed by using discs having finite thickness. In 1D models it is possible to
use compressible discs defined by the charge between each pair of sample electrons
[45] but the additional complexity is not justified by the results.
In order to find the motion of the electrons it is also necessary to know the self-​
consistent field of the electromagnetic structure surrounding the beam. If the struc-
ture is excited from an external source then the initial field is known. However,
the interaction between the beam and the structure changes that field, and it is
necessary to make multiple passes through the structure in order to achieve a self-​
consistent solution. The self-​consistency can be checked by computing the net flow
of energy through a closed boundary surrounding the interaction space because
that should be zero in the steady state. When the electromagnetic structure has no
external source, but is driven by a modulated beam, iteration is again necessary to
achieve a self-​consistent solution. Many computer codes have been written for the
simulation of klystrons and TWTs but, almost without exception, they are propri-
etary and not generally available outside the organisations which produced them.
An exception to this is AJ-​DISK which is a 1D large-​signal klystron model [11].
Simple Mathcad models for klystrons and helix TWTs can be found in Worksheets
13.3, 13.5, and 14.3.

11.8.1 Large-​Signal Model of the Beam–​Gap Interaction

The large-​signal interaction between an electron beam and the RF field of an inter-
action gap can be investigated by numerical integration of the equations of motion
of the electrons. To obtain accurate results it is necessary to use a three-​dimensional
model [46, 47]. However, the main features can be studied qualitatively using a
one-​dimensional disc model of the electron beam [48–​51]. In Worksheet 11.6 the
motion of a single wavelength of electrons is tracked in time, and the space-​charge
forces are calculated by assuming that the positions of the electrons are periodic
in space. This assumption is not strictly valid, but it is adequate for purposes of
illustration. The results of such calculations are commonly displayed in the form
of an Applegate diagram as shown in Figure 11.24 (c.f. Figure 11.4). The slopes
of the electron trajectories are proportional to their velocities. It is convenient to
normalise the time and position by plotting θ = βe z against φ = ωt because then
electrons which have velocity u0 have unit slope. The figure shows the trajectories of
12 electrons distributed through one RF period when gap centre is at z = 0, the nor-
malised effective RF voltage MVg V0 = 0.2, and space-​charge forces are neglected.
The maximum accelerating field is at ωt = π 2 and the formation of the bunch is
clearly visible.
For many purposes it is convenient to use a modified Applegate diagram in which
the axes are interchanged. The phase is plotted relative to that of an electron travel-
ling at constant velocity (i.e. φ ′ = φ − θ). In this diagram the trajectory of an electron
whose velocity is u0 is horizontal, while increased (reduced) velocity is indicated by
420 Beam-Wave Interaction

( )
Figure 11.24:  Applegate diagram for ballistic bunching of an electron beam  MVg V0 = 0.2 .

a negative (positive) slope. Figure 11.25(a) shows a modified Applegate diagram,

including space-​charge effects, when ω q ω = 0.1 and the beam is modulated by a
gap at z = 0 with MVg V0 = 0.1. The relative phase of the gap voltage is π 2, so
that it is accelerating when 0 < φ′ < π and retarding when − π < φ ′ < 0. Under these
conditions the bunching is just in the small-​signal regime and the diagram shows
repeated plasma oscillations whose wavelength is approximately 10 λ e. These oscilla-
tions show that there is a periodic exchange of energy between the kinetic energy of
the electrons and the potential energy stored in the bunch. At the plane of where the
bunching is greatest the trajectories are almost parallel to the horizontal axis, indi-
cating that the electron velocities are approximately u0 , as predicted by small-​signal
theory. The relative phase of the bunch is close to zero and it therefore leads the
phase of the gap voltage by approximately 90°, as expected from (11.13). However
the diagram is not quite symmetrical about φ ′ = 0 because a finite input gap has been
modelled and second-​order effects are not negligible. In Figure 11.25(b) MVg V0
has been increased to 0.2, the bunching is more pronounced, and the plane of the
bunch has moved closer to the input gap. There is now a visible spread in the elec-
tron velocities at the plane of the bunch, and the subsequent trajectories are only
approximately periodic. When the gap voltage is increased further (Figure 11.25(c),
MVg V0 = 0.4) the electron trajectories cross one another so that the bunch is not
well-​defined and there is a considerable spread of velocities.
The harmonic components of the beam current, as a function of position, can be
calculated from the arrival times of the sample electrons at successive planes in z,
as described above. Figure 11.26 shows how the harmonic currents vary with axial
position for three different values of the effective gap voltage. In Figure 11.26(a)
(MVg V0 = 0.02) the behaviour is very close to that predicted by small-​signal the-
ory, and the amplitude of the first current harmonic is proportional to sin (βq z ).
It should be noted, however, that even in this case the amplitudes of the higher har-
monic components are not zero. The remaining figures show that, as the effective
 Large-Signal Modelling of Beam-Wave Interactions 421

Figure 11.25:  Modified Applegate diagrams for bunching of an electron beam with space-​
charge (ω q ω = 0.1): (a) MVg V0 = 0.1, (b) MVg V0 = 0.2 and, (c) MVg V0 = 0.4.

Figure 11.26:  Normalised harmonic currents as a function of axial position for ω q ω = 0.1

and different values of the effective gap voltage: (a) MVg V0 = 0.02, (b) MVg V0 = 0.1
and (c) MVg V0 = 0.2.
422 Beam-Wave Interaction

gap voltage is increased, the maximum of the first harmonic increases, and moves
closer to the bunching gap, whilst the amplitudes of the higher harmonics increase
more rapidly, so that the behaviour is increasingly non-​linear.

11.8.2 Modulation of an Electron Beam by a Gap

When an electron beam passes through an interaction gap the field of the gap is
generally much greater than the space-​charge field. It is therefore possible to neglect
the space-​charge field to a first approximation when investigating the interaction.
Figure 11.27(a) shows the profile of the radially averaged axial electric field acting
on the electrons for a typical case ( b = 0.6a, g = a ). The dashed vertical lines show
the positions of the edges of the gap in the beam tunnel. The gap used for this
illustration has a small-​signal gap coupling factor M = 0.79 and the ratio of the gap
voltage to the beam voltage is 1.52 corresponding to MVg V0 = 1.2. The modified
Applegate diagram in Figure 11.27(b) shows that at this gap voltage the electrons
which lose energy (moving upwards in the figure) remain in the gap much longer
than those which gain energy. They therefore experience a much greater change in

Figure 11.27:  Modulation of an electron beam by a gap under large-​signal conditions

when space-​charge is neglected: (a) variation of the axial electric field with
position, (b) the modified Applegate diagram, (c) the variation of the electron
velocities with position and (d) the dependence of the final electron energy on
phase (βe a = βe g = 1; βe b = 0.6; MVg V0 = 1.2 ) .
 Large-Signal Modelling of Beam-Wave Interactions 423

the phase of the field. In extreme cases the phase of the field is reversed during the
transit time of an electron. This behaviour is also illustrated in Figure  11.27(c),
which shows the normalised velocities of 12 electrons as they pass through the gap.
Only those electrons with the greatest final velocities are accelerated continuously,
all the others stay in the gap long enough to experience a reversal of the field. The
normalised velocities of the slowest and fastest electrons are approximately 0.45
and 1.45 as they leave the gap. Figure 11.27(d) shows the ratio of the final electron
energy to the initial electron energy as a function of the phase at which each electron
would reach the centre of the gap if the field of the gap were zero. It is evident that
under these conditions there is asymmetry between those electrons which are accel-
erated and those which are retarded. The dotted line in the figure shows the electron
energies computed using the small-​signal approximation (11.8). That model gives
poor results especially for the electrons which are retarded because it assumes that
they pass through the gap with constant velocity. Examination of Figure 11.26(c)
suggests that better results could be obtained assuming that an electron has its ini-
tial velocity ( ui ) up to the centre of the gap, and its final velocity (u f ( φ )) thereafter.
Then the final energy is given by

V f ( φ ) = V0 +
1
2
( )
M ( ui ) + M (u f ( φ )) Vg cos ( φ ) , (11.168)

which can be solved by iteration to find V f . The results of this second approxi-
mation are shown by the solid line in Figure 11.27(d). This equation predicts the
velocities of the fastest and slowest electrons quite well but fails to reproduce
the phase shift calculated by the disc model. The accuracy of (11.168) increases as
the gap voltage is reduced.
The beam loading conductance can be calculated by computing the change in the
kinetic energy of an unmodulated beam produced by the field of the gap [52] (see
Worksheet 11.3). Using this method it is found that the beam loading conductance
is given by (11.23) up to MVg V0 = 0.6. The results are only slightly affected by the
inclusion of space-​charge forces, as has been shown by the similarity between the
results of ballistic analysis (11.22) and space-​charge wave theory (11.99).

11.8.3 Current Induced in a Passive Gap by a Modulated Beam

We have seen that, when a modulated beam passes through an interaction gap, a
current is induced in the gap. Under small-​signal conditions the induced current
can be calculated by assuming that the magnitude of the RF current in the beam
is constant within the gap. The gap voltage is the product of the induced current
and the impedance presented to the gap by beam loading and by the external cir-
cuit. Under large-​signal conditions this assumption is no longer valid and the cur-
rent must be calculated by finding self-​consistent values of the induced current and
the gap voltage [53]. It is normally the case that the external circuit is passive and
therefore the time average of the power in the beam can only be reduced by the
interaction.
424 Beam-Wave Interaction

Figure 11.28:  Field and circuit representations of the interaction between a modulated

electron beam and a gap in the wall of the surrounding drift tube.

Figure 11.28 shows the interaction between the convection current ( J (t )) and the
electric field of a gap in the wall of the surrounding drift tube ( E (t )). The sign con-
vention used for the gap voltage (Vg ) and the induced current (i g ) is the same as in
Figure 11.15. The induced current is in antiphase with the conventional current of
the electrons and, therefore, in phase with the convection current. The instantane-
ous power transferred to the beam from the gap is given by [54]

i g (t )Vg (t ) = − 
∫∫∫ J (t ) ⋅ E (t ) dv. (11.169)
In a one-​dimensional model this becomes
z2

i g (t )Vg (t ) = − ∫ I ( z, t ) E z ( z, t ) dz, (11.170)

z1

where I is the convection current in the beam, E z is the axial component of the
electric field averaged over the cross-​section of the beam, and the integral is taken
over the region in which E z ≠ 0. Now the electric field is proportional to the gap
voltage so we may write

Vg (t )
E z ( z, t ) = − f ( z ) . (11.171)
g
Substituting in (11.170) gives the instantaneous induced current
z
1 2
i g (t ) = I ( z, t ) f ( z ) dz. (11.172)
g z∫1

But the currents are harmonic in time so that they can be expressed as a Fourier
series. Then, equating the harmonics on the two sides of the equation gives the
complex amplitude of the nth harmonic of the induced current in terms of the com-
plex amplitude of the nth harmonic of the convection current

1 z2
f ( z )I n ( z ) dz. (11.173)
g ∫z1
i gn =
 Large-Signal Modelling of Beam-Wave Interactions 425

Figure 11.29:  Phasor diagrams for the interaction between a modulated electron beam and a
passive cavity resonator: (a) current phasors, and (b) gap voltage phasors.

In small-​signal theory it is assumed that the induced current depends only on the
modulation of the beam entering the gap. The gap current associated with the re-​
modulation of the beam by the field of the gap is represented by the beam loading
admittance. In large-​signal theory the gap current computed by numerical evalua-
tion of (11.173) includes both effects. The cavity resonator outside the gap presents
impedance Z g to the current. That impedance is negligible except close to the signal
harmonic to which it has been tuned. Thus the gap voltage is

Vg (t ) = −Z g ( nω ) i gn exp ( jnωt ) . (11.174)

Mutually consistent values of the complex induced current and the complex gap
voltage can be found by iteration.
The impedance of the gap is real when the signal frequency harmonic is equal
to the resonant frequency of the cavity ( ω 0 ), it has positive (inductive) reactance
when ω < ω 0 , and negative (capacitive) reactance when ω > ω 0 (see Section 3.2.1).
The phase angle of the impedance varies from π 2 to − π 2 as the signal frequency
increases (see Figure  3.4). The interaction between the modulated beam and the
cavity can be understood by considering the phasor diagrams in Figure 11.29. The
phases are referred to the instant at which the centre of the bunch lies at the centre
of the gap. The phase of the RF convection current ( I1 ) is − π , and the phase of the
induced current in the gap is zero, as shown in Figure  11.29(a). Figure  11.29(b)
shows the positions of the phasor representing the gap voltage at three values of
the modulation frequency ( ω ). The phase of the current modulation added to the
beam by the gap ( ∆I1 ) lags the phase of the gap voltage by approximately π 2.
The phases of the additional modulation for the three frequencies are shown in
Figure 11.29(a). In many cases the additional modulation is the dominant modula-
tion downstream from the gap. The special case of the output gap of an IOT or a
klystron is considered in more detail in the next section. The application to electron
426 Beam-Wave Interaction

bunching in klystrons, by cavities whose resonant frequency differs from the modu-
lation frequency, is considered in section 13.3.3.

11.8.4 Power Transfer in an Output Gap

The resonant frequency of the output cavity of an IOT or a klystron is normally
very close to the frequency of modulation. The impedance presented to the induced
current in the gap is therefore largely resistive. It is dominated by the external load
in order to minimise the loss of RF power in the cavity.
To investigate how the power transferred to the output gap depends upon the
current waveform, and upon the properties of the gap, we will consider the set of
idealised current waveforms. These have been constructed to give the greatest pos-
sible value of I1 I 0 for a given number of harmonics, subject to the condition that
the total current can never be negative. They are defined by

I ( φ, n ) = 2 π (1 + cos φ ) , (11.175)
n

where n is an integer. These functions are maximally flat at the zeroes of current.
The DC current is given by
π
1
I ( φ, n ) d φ (11.176)
2 π −∫π
I0 =

and the harmonic currents are

π
1
I ( φ, n ) cos ( mφ ) d φ. (11.177)
2 π −∫π
Im =

Table 11.2 shows the peak currents, and the amplitudes of the harmonic currents,
of these idealised waveforms normalised to the DC current. Figure 11.30 shows the
first four waveforms of the set. As the number of harmonics is increased the ampli-
tude of the first harmonic increases and the width of the bunch decreases. For n ≥ 4
very nearly all the electrons are grouped within a phase range of 180°.

Table 11.2:  Peak currents and harmonic current components of idealised bunching waveforms

Number of harmonics 1 2 3 4 5 6

I pk I 0 2.00 2.67 3.20 3.66 4.06 4.43

I1 I 0 1.000 1.333 1.500 1.600 1.667 1.714
I2 I0 0.333 0.600 0.800 0.952 1.071
I3 I 0 0.100 0.229 0.357 0.476
I4 I0 0.029 0.079 0.143
I5 I 0 0.008 0.026
I6 I0 0.002
 Large-Signal Modelling of Beam-Wave Interactions 427

Figure 11.30:  Idealised current waveforms having the maximum possible amplitude of the
first harmonic for different numbers of harmonic components.

The range of usefulness of these equations can be investigated by comparing

the results with those obtained directly using the disc model, without space-​
charge, so that the effects of potential energy are not considered. It is assumed
that the electrons all enter the gap with velocity u0 . This problem was studied
by Hechtel [51] who showed that the optimum conversion efficiency decreased
with increases of the normalised tunnel radius ( γ e a ) and the normalised gap
length (βe g ). Detailed results were only obtained for I1 I 0 = 1.44 and the wave-
form assumed was not given. The same problem can be studied using Worksheet
11.6 with the waveforms given by (11.175) and very similar results are obtained.
Under small-​ signal conditions the maximum power delivered to the gap is
given by

1
Pgap = MI1Vg . (11.178)
2
If the RF beam current and the gap voltage are fixed then the power is propor-
tional to the gap coupling factor (M ) so that it decreases as M decreases. That is the
case when the normalised tunnel radius ( γ e a ) and the normalised gap length (βe g )
increase, exactly as found by Hechtel.
The physical origin of the beam loading conductance of an unmodulated beam is
the imbalance between the powers transferred to electrons which are accelerated or
retarded. When the input beam is modulated, the proportions of electrons in these
two classes change. Then the time-​average transfer of power between the beam and
the gap depends upon the magnitude of the modulation, and upon the phase differ-
ence between the bunch centre and the field of the gap. The power delivered to the
gap is the time-​average of the product of the current and the effective gap voltage.
Making use of (11.168) this can be written approximately as
π
Pgap = −
1
4 π −∫π
( )
I ( φ ) M ( u0 ) + M (u f ( φ )) Vg cos ( φ + Φ ) d φ, (11.179)
428 Beam-Wave Interaction

Figure 11.31:  Efficiency of the extraction of energy from an ideally bunched electron beam
comparing results from the disc model and equations (11.179) and (11.182): (a) with
Φ = 180° , and (b) with MVg V0 = 1.0  (βe a = βe g = 1; βe b = 0.6 ).

where φ = ωt, I ( φ ) is the beam current at the centre of the gap, and Φ is the phase of the
gap voltage when the bunch is at the centre of the gap. This equation is found to give
good agreement with the power computed using the disc model when the bunch length
is 180° or less. Figure 11.31(a) shows a comparison between the electronic efficiency
calculated using the disk model, and that computed using (11.179), for extraction of
energy from an idealised bunch with n = 6. The disk model shows that the transfer of
energy is greatest when MVg V0 = 1.2. The electronic efficiency decreases for greater
gap voltages because some electrons remain in the gap for a long time, so that they
experience a large change of phase and are re-​accelerated, taking energy from the gap.
For the ideally bunched beam, and the one-​dimensional model without space-​charge,
considered here, this effect that sets a limit on the electronic efficiency of the interac-
tion. The energy of the slowest electron leaving the gap is around Vmin = 0.15 V0 when
g = a . This figure decreases slowly as the length of the gap is reduced.
Figure 11.31(b) compares the results of the disc model with those from (11.179)
for energy transfer from the bunch to the field when MVg V0 = 1.0 and the phase of
the bunch relative to the field of the gap is varied. It is found that these results are
almost independent of the normalised length of the gap.
By analogy with small-​signal theory we can define an effective large-​signal gap
coupling factor by

2 Pgap
M eff = , (11.180)
I1Vg

where I1 is the first harmonic of the beam current under large-​signal conditions.
Then the efficiency of the transfer of power from the beam to the gap is
 Large-Signal Modelling of Beam-Wave Interactions 429

M eff I1Vg 1 M eff I1 MVg

ηe = = ⋅ ⋅ ⋅ . (11.181)
2 I 0V0 2 M I 0 V0
The value of M eff can be calculated using the disk model or, approximately, using
(11.179). When the beam is tightly bunched the majority of the electrons cross the
gap centre close to the phase of the maximum retarding field. Then the exit velocities
of the electrons are approximately umin = u f ( π ) and a useful approximation to M eff  is

1
M eff ≈
2
(M (u0 ) + M (umin )). (11.182)
The electronic efficiency calculated using (11.182) is close to that given by the disc
model up to MVg V0 = 1.0, as shown in Figure 11.30(a) when V0 = 25 kV so that
relativistic effects are not important.
The effective coupling factor that gives maximum efficiency was calculated using
the disc model, without space-​charge, for I1 I 0 in the range 1.33 to 1.67; γ e a in
the range 0.2 to 3.0; and βe g in the range 0.1 to 5.0. When the results were plot-
ted against the small-​signal coupling factor (M ) the result was the universal curve,
shown in Figure 11.32. The symbols represent different combinations of the param-
eters. The curve is closely fitted by the empirical approximation

1
M eff = (M + M 6 ) . (11.183)
2
The corresponding values of Vg V0 were about 1.5 when M = 0.8. They increased
at varying rates, depending upon the parameters chosen, as M decreased. It is
important to remember that this result was obtained using a one-​dimensional disc
model in which space-​charge was ignored and a uniform electric field in the gap
was assumed. The value of the effective gap coupling factor for understanding the
large-​signal operation of IOTs and klystrons, and the effects when the bunches are

Figure 11.32:  Dependence of the effective gap coupling factor giving maximum efficiency
upon the small-​signal coupling factor.
430 Beam-Wave Interaction

not ideal, are explored in Chapters 12 and 13. The effects of space-​charge on the
maximum efficiency of energy transfer to the gap are discussed in Section 13.3.6.

References

[1]‌ W. Hahn, ‘Small signal theory of velocity modulated electron beams’, Gen. Elec. Rev,
vol. 42, pp. 258–​270, 1939.
[2]‌ S. Ramo, ‘The electronic-​wave theory of velocity-​modulation tubes’, Proceedings of the
IRE, vol. 27, pp. 757–​763, 1939.
[3]‌ A. H. W. Beck, Space-​Charge Waves and Slow Electromagnetic Waves. London: Pergamon
Press, 1958.
[4]‌ R. E. Trotman, Longitudinal Space-​Charge Waves. London: Chapman & Hall, 1966.
[5]‌ M. T. Vlaardingerbroek and K. R. U. Weimer, ‘Beam-​plasma amplifier tubes’, Phillips
Research Review, vol. 27, pp. 275–​284, 1966.
[6]‌ K. Spangenburg, Vacuum Tubes. New York: McGraw-​Hill, 1948.
[7]‌ D. L. Webster, ‘Cathode-​ray bunching’, Journal of Applied Physics, vol. 10, pp. 501–​508,
1939.
[8]‌ B. B. van Iperen and H. J. C. A. Nunnink, ‘Harmonics in velocity-​modulated cylindrical
electron beams’, Philips Research Reports, vol. 20, pp. 432–​461, 1965.
[9]‌ G. M. Branch, Jr., ‘Electron beam coupling in interaction gaps of cylindrical sym-
metry’, IRE Transactions on Electron Devices, vol. 8, pp. 193–​207, 1961.
[10] E. J. Craig, ‘The beam-​loading admittance of gridless klystron gaps’, IEEE Transactions
on Electron Devices, vol. 14, pp. 273–​278, 1967.
[11] G. Caryotakis, ‘Klystrons’, in R. J. Barker et al., eds, Modern Microwave and Millimeter-​
Wave Power Electronics. Piscataway, NJ: IEEE Press, pp. 107–​170, 2005.
[12] G. R. Brewer, ‘Some effects of magnetic field strength on space-​charge-​wave propaga-
tion’, Proceedings of the IRE, vol. 44, pp. 896–​903, 1956.
[13] L. A. MacKenzie, ‘Space charge waves in finite magnetic fields’, in The Fourth International
Conference on Microwave Tubes, The Hague, The Netherlands, pp. 663–668, 1962.
[14] V. Srivastava and R. G. Carter, ‘Effect of boundaries on the space charge potential
in coupled cavity travelling wave tubes’, IEE Proceedings I:  Solid-​State and Electron
Devices, vol. 133, pp. 185–​188, 1986.
[15] W. H. Louisell and J. R. Pierce, ‘Power flow in electron beam devices’, Proceedings of the
IRE, vol. 43, pp. 425–​427, 1955.
[16] R. L. Gunshor, ‘Decay of space-​charge waves on electron beams’, Journal of Applied
Physics, vol. 37, pp. 1904–​1911, 1966.
[17] M. Caulton, ‘Damping of waves in electron beams’, Journal of Applied Physics, vol. 38,
pp. 1839–​1855, 1967.
[18] M. Caulton et  al., ‘Experimental evidence of Landau damping in electron beams’,
Journal of Applied Physics, vol. 33, pp. 800–​803, 1962.
[19] G. M. Branch and T. G. Mihran, ‘Plasma frequency reduction factors in electron
beams’, IRE Transactions on Electron Devices, vol. 2, pp. 3–​11, 1955.
[20] G. M. Branch, Jr. et al., ‘Space-​charge wavelengths in electron beams’, IEEE Transactions
on Electron Devices, vol. 14, pp. 350–​357, 1967.
[21] W. H. Louisell, Coupled Mode and Parametric Electronics. New York: Wiley, 1960.
[22] A. Bers, ‘Linear space-​charge theory of gap interaction between an electron beam
and electromagnetic fields’, in Microwave and Optical Generation and Amplification,
Munich, pp. 53–60, 1960.
 References 431

[23] C.-​L. Huang et al., ‘AC-​space-​charge effects on gap coupling coefficient of a klystron
cavity’, IEEE Transactions on Plasma Science, vol. 40, pp. 828–​834, 2012.
[24] E. J. Craig, ‘Relativistic beam-​loading admittance’, IEEE Transactions on Electron
Devices, vol. 16, pp. 139–​139, 1969.
[25] G. Faillon, ‘Klystrons de puissance à large bande’, Revue Technique Thomson-​CSF, vol.
8, p. 139, June 1976.
[26] J. R. Pierce, Traveling-​Wave Tubes. Princeton, NJ: D. van Nostrand, 1950.
[27] J. R. Pierce, ‘Theory of the beam-​type traveling-​wave tube’, Proceedings of the IRE, vol.
35, pp. 111–​123, 1947.
[28] C. K. Birdsall and G. R. Brewer, ‘Traveling wave tube characteristics for finite val-
ues of C’, Transactions of the IRE Professional Group on Electron Devices, vol. 1, pp.
1–​11, 1954.
[29] G. R. Brewer and C. K. Birdsall, ‘Traveling-​wave tube propagation constants’, IRE
Transactions on Electron Devices, vol. 4, pp. 140–​144, 1957.
[30] G. S. Kino et al., ‘Small-​signal and large-​signal theories for the coupled-​cavity TWT’,
in 6th International Conference on Microwave and Optical Generation and Amplification,
Cambridge, UK, pp. 49–​53, 1966.
[31] D. L. Bobroff, ‘The buildup of oscillations in an electron beam backward-​wave oscilla-
tor’, IEEE Transactions on Electron Devices, vol. 12, pp. 307–​312, 1965.
[32] B. O. Nilsson and C. E. Hagstrom, ‘A two wave theory of traveling-​ wave tubes
and backward-​ wave oscillations’, IEEE Transactions on Electron Devices, vol. 22,
pp. 869–​880, 1975.
[33] K. Minami et al., ‘Analysis of starting currents in a backward wave oscillator with finite
structure length’, Journal of the Physics Society Japan, vol. 61, pp. 3566–​3575, 1992.
[34] H. R. Johnson, ‘Backward-​ wave oscillators’, Proceedings of the IRE, vol. 43,
pp. 684–​697, 1955.
[35] S. K. Datta et al., ‘Nonlinear Eulerian hydrodynamical analysis of helix traveling-​wave
tubes’, IEEE Transactions on Electron Devices, vol. 45, pp. 2055–​2062, 1998.
[36] J. G. Wöhlbier et  al., ‘The multifrequency spectral Eulerian (MUSE) model of a
traveling wave tube’, IEEE Transactions on Plasma Science, vol. 30, pp. 1063–​1075,
2002.
[37] C. C. Motta, ‘A large-​signal analysis of a ring-​bar TWT’, in IEEE 34th International
Conference on Plasma Science, pp. 875–​875, 2007.
[38] T. M. Antonsen, Jr. et al., ‘Advances in modeling and simulation of vacuum electronic
devices’, Proceedings of the IEEE, vol. 87, pp. 804–​839, 1999.
[39] B. E. Carlsten and P. J. Tallerico, ‘Self-​consistent klystron simulations’, Nuclear Science,
IEEE Transactions on Nuclear Science, vol. 32, pp. 2837–​2839, 1985.
[40] J. R. M. Vaughan, ‘Calculation of coupled-​cavity TWT performance’, IEEE Transactions
on Electron Devices, vol. 22, pp. 880–​890, 1975.
[41] V. Srivastava and R. G. Carter, ‘A fast large-​signal model for coupled-​cavity TWTs’,
IEEE Transactions on Electron Devices, vol. 35, pp. 2068–​2076, November 1988.
[42] D. Chernin et al., ‘A three-​dimensional multifrequency large signal model for helix trav-
eling wave tubes’, IEEE Transactions on Electron Devices, vol. 48, pp. 3–​11, 2001.
[43] T. Shintake, ‘FCI  –​field charge interaction program for high power klystron simu-
lations’, in Proceedings of the 1989 IEEE Particle Accelerator Conference, vol. 1,
pp. 94–​96, 1989.
[44] J. R. Hechtel, ‘The effect of potential beam energy on the performance of linear beam
devices’, IEEE Transactions on Electron Devices, vol. 17, pp. 999–​1009, 1970.
432 Beam-Wave Interaction

[45] C. J. Edgcombe, ‘Increased efficiency for klystron amplifiers’, IEE Journal on Solid-​
State and Electron Devices, vol. 1, pp. 62–​68, 1977.
[46] T. G. Mihran, ‘The effect of space charge on bunching in a two-​cavity klystron’, IRE
Transactions on Electron Devices, vol. 6, pp. 54–​64, 1959.
[47] H. G. Kosmahl and L. U. Albers, ‘Three-​dimensional evaluation of energy extraction in
output cavities of klystron amplifiers’, IEEE Transactions on Electron Devices, vol. 20,
pp. 883–​890, 1973.
[48] S. E. Webber, ‘Ballistic analysis of a two-​cavity finite beam klystron’, IRE Transactions
on Electron Devices, vol. 5, pp. 98–​108, 1958.
[49] S. E. Webber, ‘Large signal bunching of electron beams by standing-​wave and traveling-​
wave systems’, IRE Transactions on Electron Devices, vol. 6, pp. 365–​372, 1959.
[50] S. E. Webber, ‘Some calculations on the large signal energy exchange mechanisms in
linear beam tubes’, IRE Transactions on Electron Devices, vol. 7, pp. 154–​162, 1960.
[51] J. R. Hechtel, ‘DC-​ to-​
RF energy conversion in ungridded klystron gaps’, IEEE
Transactions on Electron Devices, vol. 16, pp. 212–​217, 1969.
[52] J. R. M. Vaughan, ‘A model for the klystron cavity gap’, IEEE Transactions on Electron
Devices, vol. 32, pp. 2482–​2484, 1985.
[53] T. Kageyama, ‘A large-​signal analysis of broad-​band klystrons with design applica-
tions’, IEEE Transactions on Electron Devices, vol. ED-​24, pp. 3–​12, 1977.
[54] B. E. Carlsten et  al., ‘Accuracy of the equivalent circuit model using a fixed beam
impedance for klystron gain cavities’, IEEE Transactions on Plasma Science, vol. 26,
pp. 1745–​1749, 1998.
12 Gridded Tubes

12.1 Introduction

Gridded tubes include triodes, tetrodes, and inductive output tubes (IOTs). In these
devices the current emitted by a thermionic cathode is modulated by the voltage
applied to a control grid as discussed in Chapter 6. The time varying current deliv-
ers RF power to an output circuit. The technology of triodes and tetrodes is mature
and most of the literature dates from 1960 or earlier. Up to that time small valves
were still the dominant type of active electronic device, and much of the litera-
ture concentrates on them. The subject was fully discussed in a number of books
which contain much greater detail than is possible here [1–​5]. The state of the art
of high power triodes and tetrodes was reviewed in [6]. The principle of the IOT
was known for many years, but this type of tube did not become commercially
­important until the development of pyrolytic graphite grids [7, 8]. Triodes and
­tetrodes are now mainly used in high power oscillators and amplifiers up to around
200 MHz. Inductive output tubes are high power amplifiers in the range 200 MHz
to 1.3 GHz [9, 10].

12.1.1 Gridded Tube Amplifiers

Before we proceed to examine the properties of gridded tubes in detail it is neces-
sary to understand how they are used in power amplifiers. Figure 12.1 shows the
circuit of a common cathode tetrode amplifier with a tuned anode (tank) circuit.
In UHF amplifiers this circuit takes the form of a cavity resonator, but at lower
frequencies lumped components are used. At very low frequencies a resistive anode
load may be used, but this is unsatisfactory in the VHF band and above because of
the effects of parasitic capacitance. At the resonant frequency the load in the anode
circuit comprises the shunt resistance of the resonator in parallel with the load
resistance. If the load impedance has a reactive component then it detunes the res-
onator and can be compensated for. The DC electrode potentials are maintained by
the power supplies shown and the capacitors provide RF coupling and decoupling.
An alternative circuit which is frequently used is the common grid circuit shown
in Figure 12.2. This circuit is easier to construct, using tubes with coaxial electrodes,
and there is better isolation between the input and the output circuits because the
RF voltage on both grids is zero. The other important difference is that in this
434 Gridded Tubes

Figure 12.1:  Common cathode tetrode amplifier.

(copyright 1992, CERN, reproduced, with permission, from [11]).

Figure 12.2:  Common grid tetrode amplifier

(copyright 1992, CERN, reproduced, with permission, from [11]).

circuit the full anode current flows in the input circuit with the result that the input
impedance and the gain are both lower than for the grounded cathode circuit.

12.1.2 Classes of Amplification

Figure 12.3 shows typical characteristic curves of a tetrode. In the absence of an
RF input the currents flowing to the electrodes are determined solely by the DC
voltage applied to them. The tube is then said to be at its quiescent (Q) point. Tuned
amplifiers are normally operated in one of three modes known as classes A, B, and
C as described below. The difference between these lies in the choice of the Q point.
The Q point marked in Figure 12.3 lies on the curve of zero anode current at an
anode voltage of 10 kV. This is achieved by setting the control grid voltage to –240
V. The screen grid is at RF ground and a DC potential of 900V.
 Introduction 435

Figure 12.3:  RS 2058 tetrode characteristic curves with load line

(copyright 1986, Siemens, reproduced, with permission of Thales Electron Devices,
from [12]).
436 Gridded Tubes

When an RF input voltage is applied between the cathode and the control grid,
the RF cathode current is in phase with the RF grid voltage. If the anode load is
tuned to resonance then it presents a pure resistance to the fundamental compo-
nent of the RF current. The tank circuit is not resonant at the higher harmonics of
the anode current. Thus, the RF voltage on the anode is sinusoidal and in antiphase
with the control grid voltage. The tube operates on a straight load line passing
through the Q point whose slope is determined by the anode load resistance. A typi-
cal load line is shown in Figure 12.3. When the instantaneous control grid voltage is
100 V the anode current is a little over 40 A and the anode voltage is 1.5 kV.
The operation of a gridded-​tube power amplifier can be explored using a simple
theory in which it is assumed that the tube obeys the 3 2 power law, and the grid
currents are negligible. The analysis is presented here in terms of a triode, but tet-
rode amplifiers show the same behaviour. Let the control grid voltage be

Vg (t ) = Vgo + Vg1 cos ( ωt ) , (12.1)

where ω is the frequency of operation. The anode current is given by (6.1),
3
 V 2
I a (θ ) = K Vgo + Vg1 cos (θ ) + a  , (12.2)
 µ
when the term in brackets is positive. The conduction angleα is defined as the range
of θ for which the anode current is non-​zero. Thus from (12.2)

Va  α
Vgo + = −Vg1 cos   . (12.3)
µ  2
Substituting into (12.2) gives
3 3

I a (θ ) = KVg1 2 ( cos (θ ) − cos ( α 2 )) 2 . (12.4)

The peak anode current is

3 3
I pk = KVg1 2 (1 − cos ( α 2 )) 2 , (12.5)
so that the anode current may be written
3
 cos (θ ) − cos ( α 2 )  2
I a (θ ) = I pk   . (12.6)
 1 − cos ( α 2 ) 
Figure  12.4 shows normalised anode current waveforms for some representative
conduction angles. The classes of amplification are defined in terms of the conduc-
tion angle as shown in Table 12.1.
Using Fourier analysis, the anode current can be expressed as

I a (t ) = I a 0 + I a1 cos ( ωt ) + . (12.7)
 Introduction 437

Table 12.1:  Power amplifier classes

Class Conduction angle (degrees)

A 360
AB 180 to 360
B 180
C <180

Figure 12.4:  Normalised anode current waveforms for representative conduction angles.

If it is assumed that the output circuit is resonant at the signal frequency, the anode
voltage can be written

Va = Va 0 − Va1 cos ( ωt ) , (12.8)

because Va1 is in antiphase with I a1. The DC input power is

P0 = I a 0Va 0 (12.9)

and the RF output power is

1
P2 = I a1Va1 . (12.10)
2
Then the electronic efficiency is

P2 I V
ηe = = a1 a1 . (12.11)
P0 2 I a 0Va 0
438 Gridded Tubes

Figure 12.5:  (a) Variation of electronic efficiency with conduction angle, and (b) variation of
DC and RF current with conduction angle.

Because the amplifier is a power amplifier the amplitudes of the RF voltage and cur-
rent should be as great as possible. They are limited only by the need to ensure that
the anode voltage is always greater than the grid voltage in a triode, or the screen
grid voltage in a tetrode, and that the tube ratings are not exceeded. Figure 12.5(a)
shows the efficiency as a function of the conduction angle and of Va1 Va 0 . It can be
seen that the efficiency increases as Va1 Va 0 increases and as the conduction angle
decreases. The reason for this can be seen in Figure 12.5(b), which shows the varia-
tion of the DC and RF currents with conduction angle. The peak current is limited
by the ratings of the tube. Therefore the DC current, the RF current, and the power
output of the amplifier, decrease as the conduction angle decreases. However,
the DC current decreases more rapidly than the RF current so that the efficiency
increases for a given ratio Va1 Va 0 (see Worksheet 12.1).
The input power of a common cathode amplifier is

1
P1 = Vg1I g1 , (12.12)
2
where I g1 is the RF current flowing to the control grid. For a common grid
amplifier it is

1 1
P1 = Vg1 ( I g1 + I a1 ) ≈ Vg1I a1 . (12.13)
2 2
The grid current is zero except when the grid is positive so I g1  I a1 and the grid cur-
rent can be ignored in the first approximation. The gain of the amplifier in decibels

P 
GdB = 10 log  2  dB. (12.14)
 P1 
Thus the gain of a common cathode amplifier is always greater than that of a
common grid amplifier having the same output power. As the conduction angle
 Triodes 439

decreases, the amplitude of the RF grid voltage must be increased to maintain

the same peak anode current. Therefore the input power increases and the gain
decreases.
The power dissipated on the anode is the difference between the DC input power
and the RF output power. Therefore it decreases, as the conduction angle is reduced,
at constant RF output power. In many cases the performance of a tube is limited by
grid temperature raised by radiation from the cathode, electron bombardment, and
RF power dissipation [13]. The RF power dissipation is proportional to the square
of the RF anode current. It is therefore possible to increase the peak anode current,
while maintaining constant RF anode current, as the conduction angle decreases.
However, grid dissipation caused by electron bombardment increases as the peak
anode current increases, and this sets a limit to the peak anode current. For many
purposes class B operation offers a good compromise between high efficiency and
high gain.
Two tubes driven in anti-​phase can be connected to the same output circuit so
that each conducts in turn to form a push-​pull amplifier. The tubes are operated
approximately in class B but the conduction angle is increased slightly to make
the output waveform more nearly sinusoidal (reduction of ‘cross-​over distortion’).
Further information on this and other types of power amplifier can be found in
[13–​16].

12.2 Triodes

Figure 12.6 shows a schematic view of a high power triode. The cathode, grid, and
anode are usually in the form of concentric cylinders with the anode on the outside
for ease of cooling. Triodes for use at microwave frequencies have planar geometry.
The cathode may be an indirectly heated oxide cathode, or be made of thoriated

Figure 12.6:  Arrangement of a high power triode.

440 Gridded Tubes

Figure 12.7:  Characteristic curves of the ML-​5681 high power triode

(copyright 1952, IEEE, reproduced, with permission, from [19]).

tungsten wire which is heated directly by the passage of current through it. The
grid is formed from metallic wire or sheet, or from pyrolytic graphite. The anode is
made of oxygen-​free high conductivity copper and cooled by forced convection of
air or water or by vapour phase cooling (see Chapter 10). The insulators are usu-
ally alumina ceramics, though glass was used in older tubes. Further details of the
construction of these tubes can be found in [6, 13, 17, 18].
Figure  12.7 shows the constant current characteristic curves of a typical high-​
power triode [19]. As long as the grid potential is negative with respect to the cathode
no electrons are collected by the grid. However, when the grid potential is positive
some of the electrons strike the grid and may liberate secondary electrons. The grid
current shown is therefore a balance between the primary and secondary electron
currents. The anode voltage must always exceed the grid voltage to avoid collection
on the grid of secondary electrons emitted from the anode. The chief disadvantage of
the triode is that the capacitance between the grid and the anode provides unwanted
feedback between the input and output circuits. This limits the maximum operating
frequency to around 200 MHz for coaxial triodes and 4 GHz for planar triodes.
 Triodes 441

12.2.1 Case Study: The ML-​5681 Triode

The theory of the triode developed in Chapter 6 can be illustrated by computing
the characteristic curves of a triode for which data is available [19, 20]. This tube,
designed to give 50 kW output power at up to 110 MHz, had coaxial geometry with
a helical grid supported on 16 equally spaced longitudinal rods. The electron source
was a thoriated tungsten filament comprising 16 parallel wires equally spaced in a
circle and centred mid-​way between the grid support rods. The characteristic curves
of this tube are shown in Figure 12.7. Examination of the data in this figure reveals
the following information:
• The anode current curves are approximately parallel throughout the region in
which the grid current is zero. The slopes of these curves correspond to values
of μ in the range 22–​25 with a stated design value of 23. The variation in μ is
presumably the effect of space-​charge between the grid and the anode, which has
been ignored in the theory.
• The curves of constant cathode current, computed by adding the grid current to
the anode current, are extensions of the straight sections of the anode current
curves. From this it may be concluded that, if there is emission from the grid, few
electrons are collected by the cathode.
• A plot of total current to the power 2/​3 against the grid voltage is close to being
a straight line. Thus the tube is free from island formation (see Section 6.4.2).
The departures from the straight line at higher currents are a consequence of the
variations in μ.
• Figure 12.8 shows experimental data for the ratio of the grid current to the anode
current for anode currents in the range 10A to 80A and grid currents in the range
1A to 20A. The points are a good fit to the empirical line given by

Ig Vg
= 0.67 − 0.2, (12.15)
Ia Va

showing that space-​charge effects are not important. Figure 12.8 also shows
lines derived from (6.61) and (6.66), neither of which is a good fit to the
experimental data (see Worksheet 12.2). The greater slope of the line given
by (12.15) may be the result of the collection of electrons on the grid support
wires, and the offset may be caused by focusing effects of the fields in three
dimensions.
The approximate internal dimensions of the tube are given in the paper or may
be found by measurements on the photographs in it. The results are summarised in
Table 12.2. The amplification factor computed from the data in Table 12.2 using (6.36)
is 23.9, which lies in the range found from the characteristic curves. The use of equations
derived for planar triodes is justified because the distance between the filament and the
anode is small compared with the inner radius of the anode. It has been found from
experience that a filament of the kind used in this tube is equivalent to a continuous
442 Gridded Tubes

Table 12.2:  Key dimensions of the ML-​5681 triode

Parameter Dimensions Normalised dimensions

Grid diameter 88 mm

Grid spacing (a) 3.5 mm 1.0
Grid wire radius (r) 0.3 mm 0.086
Cathode-​grid spacing (d1) 3.18 mm 0.91
Grid-​anode spacing (d2) 9.5 mm 2.72
Effective cathode length (Lc) 165 mm
Number of grid sections (Ns) 16

Figure 12.8:  ML-​5681 triode: comparison between measured and calculated grid currents.

cathode whose active sections have a width equal to twice the distance between the
filament and the grid ([5] p.191). Thus the area of the equivalent cathode is

Aeff = 2 N s d1Lc . (12.16)

The effective area corresponds to only about one-​third of the perimeter of the grid
so it is safe to assume that the screening factor of the grid is not affected by the
grid support rods. The cathode current can be computed for any values of the grid
and anode voltages using (6.56) and (12.16). The normalised cathode-​grid spacing
is close to unity so there is virtually no island formation and this is confirmed by
direct calculation. The characteristic curves computed using (12.15) for the grid
current shown in Figure 12.9 are in good agreement with those in Figure 12.7.
The validity of these characteristic curves as a description of the tube at its maxi-
mum operating frequency can be established by examining the time taken for elec-
trons to move from the cathode to the grid since it is in that region where they are
moving most slowly. For typical class B operation with a minimum anode voltage
of 1.5 kV, and a maximum grid voltage of 515 V, the transit time given by (5.20)
 Tetrodes 443

Figure 12.9:  ML-​5681 triode: theoretical characteristic curves.

using the effective diode voltage is 8% of an RF period. The increase in the mag-
nitude of the RF current computed using (5.95) is 5%. The conclusion that the
use of the static characteristics is valid under these conditions is confirmed by the
figures given for a class B amplifier in the tube data sheet. At the same frequency
the effective length of the filament is 1 16 of the free space wavelength. The overall
conclusion is that the methods described above provide a good initial estimate of
the properties of a tube. They are therefore useful for conceptual design calcula-
tions, and for gaining insight into the relationship between the dimensions of a tube
and its electrical properties.

12.3 Tetrodes

Figure  12.10 shows the arrangement of a high-​power tetrode in which a second

grid (the screen grid) has been placed between the control grid and the anode. The
characteristic curves of a typical high power tetrode are shown in Figure 12.3. They
resemble the curves of a triode shown in Figure 12.7, but the dependence of the
anode current on the anode voltage is much less because the screen grid keeps the
444 Gridded Tubes

Figure 12.10:  Arrangement of a high-​power tetrode.

effective anode voltage almost constant. It is important that the anode potential
should not fall below that of the screen grid because then secondary electrons liber-
ated from the anode are collected on the screen grid and the thermal dissipation on
that grid is increased. The collection of secondary electrons also distorts the char-
acteristic curves by reducing the anode current. The current collected by the control
grid, when it is positive with respect to the cathode, is seen to be largely independent
of the anode voltage. It is usual in tetrodes to arrange the position of the wires of
the screen grid so that they are shielded, to some extent, by the wires of the control
grid so that fewer electrons are intercepted by the screen grid.

12.3.1 Case Study: The RS 2058 Tetrode

The theory of the tetrode, described in Chapter 6, can be validated by comparison
with experimental results in the same way as for a triode. If the detailed dimensions
are not available it is still possible to fit the model to experimental data and infer the
dimensions of the tube. As an example we will consider the RS 2058 tetrode whose
characteristic curves for one value of screen grid voltage are shown in Figure 12.3
[12]. This tube is a coaxial tetrode with a directly heated thoriated tungsten cath-
ode. It is designed for use as an RF amplifier at frequencies up to 220 MHz with
an output power of 64 kW CW and up to 150 kW pulsed. The ratings of this
tube for operation as a class B, grounded grid, RF power amplifier are shown in
Table 12.3 [12].
The first step is to analyse the characteristic curves to determine whether there
is island formation. This can be done by plotting the 2/​3 power of the total current
against the control grid voltage when the anode and screen grid voltages are con-
stant. Figure 12.11 shows the plot for an anode voltage of 6 kV and various screen
grid voltages. The lines are effectively straight for control grid voltages greater than
 Tetrodes 445

Table 12.3:  RS 2058 tetrode: Ratings for operation as a class B,

grounded grid, power amplifier

Frequency 220 MHz

Anode voltage 12 kV
Screen grid voltage 1000 V
Control grid voltage –350 V
Peak RF control grid voltage 320 V
DC cathode current 35 A
Peak cathode current 100 A
Anode dissipation 90 kW

Figure 12.11:  RS 2058 tetrode: graph of cathode current to the power 2/​3 against grid voltage
determined from the characteristic curves for 6 kV anode voltage and three screen grid
voltages.

–100 V and have appreciable curvature for lower voltages, showing that island for-
mation is occurring. We saw in Section 6.4.2 that the straight portions of these
curves correspond to an analysis in which island formation is ignored. Thus, setting
E x = 0 in (6.87), the intercept of the projection of the straight line on the horizontal
axis is

Vg = − Dg′ (Vs + DsVa ) . (12.17)

If values of Dg′ and Ds are assumed then Vg can be calculated for pairs of values
of Vs and Va and compared with values calculated directly from the characteristic
curves. Using iteration, values of Dg′ and Ds can be found which minimise the RMS
difference between the data and the theoretical model.
Next we note, from (6.46) and (6.87) that, at cut-​off

Vg = − Dg′ ( a 2 ) ⋅ (Vs + DsVa ) (12.18)

446 Gridded Tubes

Table 12.4:  Constants of the RS 5058 tetrode determined from the characteristic curves

Ds Dg′ Dg′ (a /2)

0.019 0.15 0.23

where, from (6.88),

Dg ( y )
Dg′ ( y ) = , (12.19)
1 + Ds (1 + d3 d 2 )
where Dg ( y ) is given by (6.45). Then Dg′ ( a 2 ) can be calculated from the voltages
at cut-​off. These calculations were carried out for Vs = 700 V, 900 V and 1100 V and
Va = 6 kV and 12 kV with the results shown in Table 12.4.
The same constants can be calculated from the dimensions of the tetrode using
the theory described above. In the absence of information to the contrary we will
assume that the spacings and wire diameters of the two grids are the same. We can
take a = 1mm without loss of generality since all the equations involve ratios of
the dimensions. The tube is then defined in terms of four constants: r, d1 , d 2 and d3
which must be chosen so that the calculated values of the penetration factors are
equal to the values given in Table  12.4. Since there are four unknowns, but only
three equations, the solution obtained is not unique. However, it is found that, by
using the results of the calculation as trial values, the process converges onto a sta-
ble solution after a few iterations. This solution has the property that the values are
almost constant if one of the trial values is altered slightly. The results were used
to calculate the set of curves shown in Figure 12.12 for comparison with those in
Figure  12.11. The area of the cathode was chosen so that the current calculated
agreed with that in the characteristic curves. It can be seen that the two sets of
curves are in good agreement with one another.
The characteristic curves can also be computed for comparison with Figure 12.3
by first finding the current intercepted by the control grid, as described in Section
6.4.1. The primary current division factor is calculated from (6.62) with D = Dg,
and the ratio of the grid current to the anode current is found using (6.61) with Va
replaced by Ve . This turns out to be quite a good approximation to the grid current
in this case. Figure 12.13 shows the curves of constant anode current for compari-
son with Figure 12.3. The curves of constant control grid current (not shown) are
very similar to those in Figure 12.3 (see Worksheet 12.3).
The agreement between the calculated and measured characteristic curves is
good, apart from the curvature at low anode voltages caused by the collection
of current on the screen grid. This current is a small fraction of the total current
except at large positive control grid voltages and low anode voltages, when the
space-​charge between the screen grid and the anode becomes important in deter-
mining the trajectories of the electrons, as discussed in Section 6.7.2. Under other
operating conditions the trajectories are focused between the wires of the screen
 Tetrodes 447

Figure 12.12:  RS 2058 tetrode: graph of cathode current to the power 2/​3 against grid voltage
computed for 6 kV anode voltage and three screen grid voltages.

Figure 12.13:  Calculated characteristic curves of the RS 2058 tetrode (Vs = 900 V ).

448 Gridded Tubes

Table 12.5:  Computed dimensions for the RS 2058 tetrode

Parameter Possible dimensions Normalised dimensions

Grid spacing (a) 2.4 mm 1.0

Grid wire radius (r) 0.145 mm 0.06
Cathode-​control grid spacing (d1) 0.8 mm 0.33
Control grid –​screen grid spacing (d2) 2.2 mm 0.92
Screen grid-​anode spacing (d3) 20.7 mm 8.58
Cathode diameter 37 mm
Cathode length (Lc) 85 mm
Anode inner diamter 84 mm

grid by the electrostatic fields, and the intercepted current is small [21]. There is no
simple way of estimating the screen grid current.
The final values of the normalised parameters for the tube are shown in Table 12.5.
It must be emphasised that these have been obtained by fitting the model to the
measured data and are not the actual parameters of the tube. The scaling of the
parameters can be estimated from the transit angle for typical class B operation. If
we assume that βe d1 = 0.1, and that the amplitude of the RF control grid voltage is
350 V, then d1 = 0.8 mm and all the other dimensions can be found by scaling from
the normalised values.
If the length of the active region is taken to be λ 16 = 85 mm then the diameter
of the cathode is 37 mm, which is plausible for a thoriated tungsten mesh cathode
that emits over its whole surface [13]. The internal diameter of the anode is then
84 mm which can be compared with the external diameter of the anode water cool-
ing jacket (125 mm). The maximum cathode current density is 1 A cm–2 which is
well within practical limits for this kind of cathode [13]. It may be concluded that
the methods adopted have yielded a plausible set of dimensions for this tube.

12.4 Design of Triodes and Tetrodes

Information about design procedures for high power triodes and tetrodes is avail-
able in a number of sources [6, 13, 19, 22]. Triodes and tetrodes are used in power
oscillators for industrial heating, FM and AM broadcast transmitters, SSB commu-
nications transmitters, pulsed radar transmitters and RF power sources for particle
accelerators. They have also been used as high power switches. Because of limita-
tions of space we will only consider the design of tetrodes for use as CW power
amplifiers operating in conditions close to class B.  The reader is referred to the
sources cited for further information. To show how a tetrode might be designed we
will consider the design of a tube similar to the RS 2074 tetrode which delivers up
to 600 kW of CW output power at frequencies up to 110 MHz [12]. The steps in the
design calculations are as follows:
 Design of Triodes and Tetrodes 449

i) From Figure  12.5(a) it is expected that the conversion efficiency of the tube
operating in class B will be around 70%. For an output power of 600 kW the
DC input power required is 860 kW. Tubes of this kind generally have thori-
ated tungsten cathodes and operate at anode voltages up to around 20 kV [13].
But high anode voltages lead to greater risk of voltage breakdown both inside
and outside the tube and so we will choose a figure of 15 kV. (Note that, for
purposes of illustration, the use of the exact figures in the data sheet has been
avoided.) Then the DC current is 60 A and, from Figure 12.5(b), the peak cur-
rent is 204 A. It is desirable for the control grid voltage to be positive for only
a small part of the cycle to minimise the control grid current, as shown by the
load line in Figure 12.3. From that figure we see that the sum of the two grid
currents is typically about 20% of the maximum anode current. Thus the cath-
ode must be able to deliver a peak current of 255 A.
ii) Let us assume that the screen grid voltage is 10% of the DC anode voltage and
that the penetration factor of the control grid (Dg ) is 0.2. Then the control grid
voltage at cut-​off is approximately

Vg 0 = DgVs = −300 V. (12.20)

Now the maximum positive control grid voltage in a class B amplifier must not
be too great to avoid excessive interception of current on the control grid. We
will assume that the maximum control grid voltage is 100 V so that the amp-
litude of the RF control grid voltage is Vg1 = 400 V. The spacing between the
cathode and the control grid must be small enough to avoid transit time effects
at the maximum operating frequency. Let us assume that βe d1 = 0.05 and that
( )
βe = ω 2eVg1 m0 so that d1 = 0.86 mm.
iii) To avoid island formation let us choose a = d1 and assume that d 2 = d1. Then the
radius of the control grid wires can be found by making the value of Dg calcu-
lated from (6.36) equal to the figure assumed above. If the grid is to be a tung-
sten or pyrolitic graphite mesh then the penetration factor can be calculated by
using the geometrical screening factor in the calculations [23, 24].
iv) Let us assume that the penetration factor of the screen grid ( Ds ) is 0.02. The
screen grid wire radius, and the distance between the screen grid and the anode,
are adjusted to achieve this figure. It may be noted that the wires of the screen
grid are often thicker than those of the control grid to make it more robust and
that, typically, d3 ∼ 10 d 2 .
v) The current density can now be calculated as a function of the electrode
voltages using (6.93). The area of the cathode is computed from the peak
current calculated above and, hence, the current density at the maximum
control grid voltage with the chosen screen grid voltage and minimum anode
voltage.
vi) If it is assumed that the active length of the tube is λ 16 then the diameter of
the cathode and of the other electrodes can be calculated.
vii) The control grid current can be calculated as described in Section 6.4.1 and
then the characteristic curves can be plotted.
450 Gridded Tubes

Table 12.6:  Leading dimensions of a 600 kW, 110 MHz tetrode

Parameter Dimensions

Grid spacing (a) 0.86 mm

Control grid wire radius (rg) 0.04 mm
Screen grid wire radius (rs) 0.052 mm
Cathode-​control grid spacing (d1) 0.86 mm
Control grid–​screen grid spacing (d2) 0.86 mm
Screen grid-​anode spacing (d3 ) 6.9 mm
Cathode diameter 38 mm
Cathode length (Lc) 170 mm
Anode inner diameter 55 mm

The complete set of dimensions calculated in this way is shown in Table 12.6. The
figures are comparable with those of the RCA 4648 tetrode [6] apart from the diam-
eters of the grid wires which may be too thin. The peak cathode current density is
1.3 A cm−​2, which is within the practical range [13]. The perimeter of the anode is
well below the value which would permit the excitation of the TE11 mode. The set of
calculations described above can readily be incorporated in a worksheet which ena-
bles the effects of different design choices to be examined quickly (see Worksheet
12.4). In practice the results of this calculation would be informed by experience
and would then be confirmed by more detailed calculations [22].
The anode power density in class B operation is just under 1 kW cm−​2 which is
comfortably within the limit for power dissipation with turbulent forced convection
cooling [25] (see Section 10.5.3). The electrode spacings must be great enough to
avoid voltage breakdown. Whilst it is difficult to calculate figures for the geometry
of such a tube, it may be noted that the voltages and spacings are similar to those
of the RCA 4648 tetrode [6]. The heater power required for the cathode can be esti-
mated using the method described in [18].
This section has discussed the principles of design of coaxial tetrodes and it could
equally be applied to triodes and to planar tubes. Information about the technology
used can be found in the literature [6, 13, 18]. A number of variants of the standard
coaxial design exist including [6, 13, 26]:
• Tubes comprising multiple unit triodes or tetrodes connected in parallel within
a single vacuum envelope. This system has the advantage that it can readily be
scaled to higher power levels, provided that the diameter does not become so big
that the TE11 mode can be excited.
• Tubes in which the cathode is outside the anode.
• Beam power tubes in which additional electrodes are used to focus the electrons,
so that the interception of electrons by the grids is reduced.
• Tubes in which there is a second anode insulator, so that the coaxial space within
the anode can be located in the centre of the external resonator, rather than at
 Design of Triode and Tetrode Amplifiers 451

Figure 12.14:  State of the art of commercially available high power tetrodes (CW Class B
operation).

one end. It is then possible to arrange that the standing wave in the resonator has
an anti-​node of RF voltage at the centre of the anode, and a node of RF current.
Thus the heat dissipation caused by RF currents in the screen grid is much less
than in a conventional tetrode. This technique enables the RF power output to be
doubled at a given frequency [26].
Figure  12.14 shows the state of the art of commercially available high power
tetrodes.

12.5 Design of Triode and Tetrode Amplifiers

The process by which a tetrode amplifier can be designed is best explained by means
of an example [27]. This is based upon a 62 kW, 200 MHz, amplifier used in the
CERN SPS [28]. The example was chosen because sufficient information is avail-
able about the amplifier to verify the results of the calculations. The amplifier uses
a single RS2058CJ tetrode [12] operating with a DC anode voltage of 10 kV, and
900 V screen grid bias. The design procedure described below is based upon that
given in [29, 30]. The actual amplifier is operated in class AB but quite close to class
B. For simplicity class B operation is assumed in the calculations which follow.
i) The first step is to estimate the probable efficiency of the amplifier. We will
assume that the minimum anode voltage is 1.5 kV. Using the method described
in Section 12.1.2 we expect that the efficiency of the amplifier will be around
72% (see Figure 12.5(a)). This estimate can be adjusted later, if necessary, when
the actual efficiency has been calculated. Then the DC power input required to
obtain the desired output power is

P2 = 62 / 0.72 = 86 kW. (12.21)

452 Gridded Tubes

The DC anode voltage was chosen to be Va0 = 10 kV so the mean anode

current is

I a0 = 86 / 10 = 8.6 A. (12.22)
ii) The theoretical value of Ipk from the Fourier analysis is 3.6 I a 0 but when other
factors including island formation, space-​charge effects and grid current are
taken in to account it is found that this figure lies in the range 3.6 to 4.4. If we
take the factor to be 4.0 then

I pk = 4 × 8.6 = 34 A. (12.23)
iii) Next we construct the load line on the characteristic curves for the tube, as
shown in Figure 12.3, by joining the point (1.5 kV, 34 A) to the quiescent point
(10 kV, 0 A). We note that this requires the control grid voltage to swing slightly
positive with a maximum of +70 V.
iv) The DC and RF currents are found by numerical Fourier analysis of the anode
current waveform, using values read from Figure 12.3 at intervals of 15°. The
anode voltages are given by

Va = 10 − 8.5 cos θ, (12.24)

where θ is the phase angle. The results are shown in Table 12.7. Then using the
formulae given in [30]

1
I a0 = ( 0.5I 0 + I15 + I 30 + I 45 + I 60 + I 75 ) (12.25)
12
and

1
I a1 = ( I 0 + 1.93I15 + 1.73I 30 + 1.41I 45 + I 60 + 0.52 I 75 ), (12.26)
12
where the subscripts refer to the phase angles in degrees. Using the data from
Table 12.7 it is found that:

I a0 = 8.9 A, (12.27)

I a1 = 15.0 A. (12.28)

v) The DC input power is

P0 = I a 0 Va 0 = 89 kW. (12.29)
The amplitude of the RF voltage is

Va1 = 10.0 − 1.5 = 8.5 kV (12.30)

and the RF output power is

1
P2 = Va1 I a1 = 64 kW, (12.31)
2
 Design of Triode and Tetrode Amplifiers 453

Table 12.7:  Anode currents at 15o phase intervals taken from Figure 12.3

Phase (degrees) 0 15 30 45 60 75 90
Va (kV) 1.5 1.8 2.6 4.0 5.7 7.8 10.0
Ia (A) 34 32.5 28 18 8 3 0

which is very close to the desired value and gives an efficiency of 72%, as origi-
nally assumed.
vi) The effective load resistance is

RL = Va1 / I a1 = 570 Ω. (12.32)

The source impedance of the output of the amplifier can be found by noting
that, if the RF load resistance is zero and the anode voltage is constant, the
peak anode current is 46 A for the same RF voltage on the control grid. Thus
the short circuit RF current is 20 A and the anode source resistance ( Ra ) is
1.7 kΩ.
vii) To find the input impedance of the amplifier we note that the amplitude of the
RF control grid voltage is

Vg1 = 245 + 70 = 315 V (12.33)

and that, for grounded grid operation, the amplitude of the RF input current
is the sum of the RF anode and control grid currents

I c1 = I a1 + I g1 ≈ I a1 . (12.34)

viii) The amplitude of the RF control grid current ( I g1 ) may be obtained by read-
ing the control grid currents off Figure 12.3 at 15o intervals and employing Eq.
(12.26). The result is 0.67 A, which is small compared with the RF anode cur-
rent, and can be neglected in the first approximation. The RF input resistance
is then

R1 = Vg1 / I a1 = 20 Ω. (12.35)

ix) Finally, we note that the input power is

1
P1 = Vg1I a1 = 2.5 kW (12.36)
2
and that the power gain of the amplifier is

GdB = 10 log (64 / 2.5 ) = 14 dB. (12.37)

Table 12.8 shows a comparison between the figures calculated above and those
reported in [31]. The differences between the two columns of Table 12.8 are
attributable to the difference between the actual class AB operation and the
class B operation assumed in the calculations.
454 Gridded Tubes

Table 12.8:  Comparison between actual, and calculated, parameters of the

amplifier described in [31]

Parameter Actual Calculated

Anode voltage (kV) 10 10

Anode current (A) 9.4 8.9
Screen grid voltage (V) 900 900
Control grid voltage (V) –200 –245
RF output power (kW) 62 64
RF input power (kW) 1.8 2.5
Gain (dB) 15.4 14
Efficiency (%) 64 72

Figure 12.15:  Arrangement of a tetrode amplifier

(copyright 2011, CERN, reproduced, with permission, from [10]).

12.5.1 Practical Details

Figure 12.15 shows a simplified diagram of a tetrode amplifier. The tube is oper-
ated in the grounded grid configuration with coaxial input and output circuits. The
outer conductors of the coaxial lines are at ground potential and they are sepa-
rated from the grids by DC blocking capacitors. The anode resonator is a re-​entrant
coaxial cavity which is separated from the anode by a DC blocking capacitor. The
output power is coupled through an impedance matching device to a coaxial line.
The anode HT connection and cooling water pipes are brought in through the cen-
tre of the resonator.
The electrodes of the tube form coaxial lines with characteristic impedances of
a few ohms. We have seen above that the input impedance of the amplifier is typ-
ically a few tens of ohms, and the output impedance a few hundred ohms. Thus
the terminations of both the input and output lines are close to open-​circuits. The
 Inductive Output Tubes (IOTs) 455

anode resonator therefore has one end open circuited and the other short circuited,
and it must be an odd number of quarter wavelengths long at resonance. Typically
the resonator is 3/​4 of a wavelength long. In that case the point at which the output
coaxial line is coupled into the resonator can be used to transform the impedance
to provide a match. There is also a voltage node towards the lower end of the outer
part of the resonator and this can be used to bring connections through to the
anode. The higher-​order modes of the cavity can be troublesome, and it is usually
necessary to damp them by the selective placing of lossy material within the cavity,
or by using coupling loops connected to external loads. The cathode heater con-
nections must incorporate some means of decoupling from the RF circuit. Further
information about the design of gridded tube amplifiers can be found in [17, 28, 32].

12.6 Inductive Output Tubes (IOTs)

Tetrode amplifiers suffer from the disadvantage that the same electrode, the anode,
is part of both the DC and the RF circuits. The output power is limited by screen
grid and anode dissipation. In addition, the electron velocity is least when the cur-
rent is greatest because of the voltage drop across the output resonator. This means
that the gap between the screen grid and the anode must be small to avoid transit
time effects. To get high power at high frequencies it is necessary to employ high-​
velocity electrons, and to have a large collection area for them. This can be achieved
by separating the electron collector from the RF output circuit. The possibility
that this might be done was originally recognised by Haeff in 1939, but it was not
until 1982 that the concept became practical, following the development of laser-​
machined pyrolytic graphite grids [7, 8]. Haeff called his invention the ‘Inductive
Output Tube’ (IOT) but it is also known by the proprietary name Klystrode ®.
Figure 12.16 shows a schematic diagram of an IOT. The electron beam is formed
by a gridded Pierce electron gun and confined by an axial magnetic field (not
shown). The RF input circuit is formed by the cathode and the grid in a manner sim-
ilar to that in a tetrode amplifier. The gun is biased so that the conduction angle is

Figure 12.16:  Arrangement of an inductive output tube (IOT)

(copyright 2010, IEEE, reproduced, with permission, from [33]).
456 Gridded Tubes

Figure 12.17:  Electron trajectories in an IOT.

approximately 180° and the electron bunches formed by the cathode and the grid are
accelerated through the constant potential difference between the cathode and the
anode. A detailed diagram of the input cavity resonator for an IOT is given in [34].
The bunched electron beam passes through a cavity resonator where the azimuthal
magnetic field induces a current in the cavity, hence the name of the tube (see Section
11.8.4). Because the cavity is tuned to the repetition frequency of the bunches the
RF electric field in the interaction gap is maximum in the retarding sense when the
centre of a bunch is at the centre of the gap. This resembles the operation of a class
B amplifier with the important difference that the input electron velocity is constant
throughout the RF cycle. Figure 12.17 shows a typical Applegate diagram for an
IOT (see Worksheet 12.5). Because the electron velocity is high it is possible to use
a much longer output gap than in a tetrode. Other advantages of the IOT over the
tetrode are that it does not need a DC blocking capacitor in the RF output circuit,
because the cavity is at ground potential, and that it has higher isolation between
input and output and a longer life than an equivalent tetrode amplifier. These advan-
tages are offset to some extent by the need for a magnetic focusing field. The typical
gain is 20–​23 dB which is appreciably higher than that of a tetrode amplifier and
high enough for the input to a 60 kW tube to be supplied by a solid state driver stage.
IOTs have been designed for use in UHF TV transmitters, and for powering par-
ticle accelerators [35–​38]. For both these applications the linearity of the power
transfer curve is a valuable asset (see Figure 12.22(a)). Where increased bandwidth
is required (e.g. for digital TV transmission) a double-​tuned output cavity is used
[34, 37] (see Section 3.2.4). A general review of the principles and technology of
IOTs and related devices is given in [39].

12.6.1 Bunch Formation

The beam current is given as a function of time by (12.6). To avoid transit time
effects, and to minimise the RF voltage required to modulate the beam, the grid
 Inductive Output Tubes (IOTs) 457

must be placed very close to the cathode so that βe d1  0.7 at the maximum effec-
tive diode voltage (see Section 5.7). Calculations based on the 1.3 GHz tube dis-
cussed in Section 12.6.5, for example, suggest that the separation between the grid
and the cathode should be less than 0.3 mm when the amplitude of the RF grid
voltage is 350 V. At lower forward voltages the transit angle is increased and tran-
sit time effects become important. The electron velocity at the grid is smaller at
the start and the finish of each bunch than it is at the bunch centre. Hence the
bunch waveform is distorted from that shown in Figure 12.4 because the currents
at the start and the finish of the bunch are delayed with respect to the bunch centre
[40]. The very small spacing between the cathode and the grid must be maintained
accurately at the working temperature of the gun. It sets a limit on the maximum
frequency of operation and means that island formation is likely to occur. The cath-
ode and the grid must be manufactured with high precision and have high thermal
stability to ensure that the intended separation is maintained, and to avoid the risk
of a short circuit between them.
The penetration factor of the grid may vary over its surface for two reasons.
In the first place the electric field of the anode is not normally constant over the
surface of the grid [41]. It may be necessary to compensate for that by varying the
spacing of the grid wires and/​or the distance between the grid and the cathode. In
the second place the diameter of the grid may be an appreciable fraction of the
free-​space wavelength. If we use the same guidelines as in the design of triodes and
tetrodes then the cathode diameter should be less than λ 8. For the example used
above where the frequency is 1.3 GHz the cathode diameter should be less than
30 mm. If the penetration factor is not constant for any reason, or if island forma-
tion exists, then the effective power law of the gun is greater than 3/​2 (see Section
6.4.2). The input circuit of an IOT is effectively that of a common grid amplifier so
that the RF input power is given by (12.13).
Other ways of modulating the beam, which could permit operation at higher
frequencies, have been investigated theoretically including: photo-​cathodes [42–​44],
gated field emission arrays [45], and deflection modulation [46].

12.6.2 Space-​Charge Debunching

After the electrons have left the grid they are accelerated by the field between the
grid and the anode, and space-​charge forces are normally negligible. However,
when the electron bunches have passed into the drift tube, then space-​charge
forces dominate. The radial space-​charge forces are counteracted by the force
of the magnetic focusing field, but the axial space-​charge forces tend to cause
debunching so that the conduction angle increases. The effect of space-​charge
forces can be calculated by numerical integration of the equations of motion of a
set of electrons distributed within the bunch, as described in Section 11.8. Since
any increase in the bunch length (conduction angle) causes a reduction in conver-
sion efficiency (see Figure12.5(a)) it is desirable to keep the drift length as short
as possible.
458 Gridded Tubes

12.6.3 Power Transfer in the Output Gap

The extraction of power from the electron bunches in the output gap can be com-
puted using a disk model or by using the simplified theory in Section 11.8.4 (see
Worksheet 12.5). If the effective output load resistance, including the effects of
losses in the output cavity, is RL then the minimum energy of electrons leaving
the gap is

Vmin = Va − M eff
2
I1RL , (12.38)

where Va is the voltage of the anode relative to the cathode, and the effective gap
coupling factor is given by (11.182). Self-​consistent values of Vmin and umin can be
found by iteration using (11.182) and (12.38). Then the electronic efficiency is

M eff I1Vg
ηe = . (12.39)
2 I 0Va
The usefulness of this approach for modelling the performance of an IOT is dem-
onstrated in the case study in Section 12.6.5. For typical dimensions, and class B
operation, I1 I 0 = 1.65, M = 0.8, M eff = 0.53 and MVg Va = 1.2 when the gap volt-
age is adjusted to give the optimum efficiency. Inserting these figures in (12.39) gives
an electronic efficiency of 66%. This is shown by the dashed line in Figure 12.18(a)
for comparison with the efficiency of CW IOTs given in manufacturers’ data sheets.
Figure 12.18(b) shows data from the same sources for CW output power.
The maximum output power of an IOT is limited by the DC current which
can be drawn from the cathode, while avoiding excessive interception by the grid,
and by the maximum voltages which can be applied without voltage breakdown.
Information about the design of a 1.3 GHz, 100 kW CW, IOT is given in [47].
A 500 kW pulsed IOT at 425 MHz is described in [48]. The use of PIC codes to
model IOTs has been described in [40, 49, 50]. The possibility of developing higher
power IOTs with multiple beams, annular beams, or radial beams has been studied
in [51–​56].

Figure 12.18:  State of the art of commercial CW IOTs: (a) efficiency, and (b) output power.

 Inductive Output Tubes (IOTs) 459

12.6.4 IOT Collectors

The conversion efficiency of a IOT at low drive levels can be improved considerably
by the use of a multi-​element depressed collector. The energy of the electrons in the
spent beam can be expressed as a function of the phase angle as

Vs (θ ) = Va − M eff Vgap cos θ. (12.40)

The slowest electrons are those at the centre of the bunch where θ = 0. In the simple
model used here the bunch profile is symmetrical, and the part of the beam current
which has energy greater than Vs is given by
π
1
I s (θ ) = I (θ ) d θ, (12.41)
π ∫θ

where θ lies in the range ( 0 : π ). The spent beam curves can be plotted, using
these equations, for comparison with the results from the disc model, shown in
Figure 12.19(a). These curves were obtained by adjusting the RF grid voltage to
give an output power of 12 kW in each case. Figure 12.19(b) shows how the spent-​
beam curves computed using the simple model vary with output power. Although
these are superficially similar to those for a klystron (see Section 13.3.7) there are
some important differences. The IOT does not exhibit saturation because the maxi-
mum power is limited by the electron gun. The tube is typically operated in class
AB so that the few electrons crossing the output gap in the accelerating phase of
the gap field are only mildly accelerated. Thus there are fewer fast electrons than is
normally the case for a klystron. In addition, the mean beam current drops as the
output power is reduced so that the electronic efficiency remains high at low output
power levels.
Figure  12.20 shows typical dependence of the RF efficiency of an IOT with
a five-​stage depressed collector on the normalised output power (see Worksheet
12.6). Because the electronic efficiency is higher at low output power, than is the

Figure 12.19:  Spent-​beam curves for an IOT: (a) comparison between different models, and
(b) spent-​beam curves for different output powers.
460 Gridded Tubes

Figure 12.20:  Dependence of the efficiency of an IOT with a five-​stage depressed collector on

normalised output power.

case with a klystron, the overall efficiency is also high at low output powers. It has
been shown that, with careful design of the collector, almost constant efficiency of
around 65% can be achieved over the range from 25% to 100% of full output power
[34, 57, 58]. This performance is very valuable for applications such as digital televi-
sion broadcasting where the signal has a high peak to average power ratio.

12.6.5 Case Study: The 116LS IOT

The 116LS IOT was designed to generate up to 16 kW CW at 1.3 GHz for power-
ing particle accelerators [59, 60]. This tube has been modelled, using the methods
described above, for comparison with the published experimental data [33]. Where
information about the tube is not available appropriate assumptions can be made
as described below. The description that follows differs slightly from that previously
published.
It is assumed that the properties of the electron gun can be described by (12.2) in
which K, μ and the power law are treated as adjustable constants. The anode volt-
age is 25 kV and the DC grid bias voltage (Vg 0 ) is –105 V, as stated in [59]. Then
the DC and fundamental RF currents ( I 0 , I1 ) can be calculated as functions of the
amplitude of the RF grid voltage (Vg1 ). Since IOTs are operated as common grid
amplifiers the RF input impedance is given by

Vg1
Rin = . (12.42)
I1 (Vg1 )

If the RF source impedance is Rs then the RF input power is given by

2
1  R + Rs  Vg21
Pin =  in . (12.43)
8  Rin  Rs
 Inductive Output Tubes (IOTs) 461

Figure 12.21:  (a) Input impedance, and (b) DC beam current, of an IOT as functions of the
RF input power
(copyright 2010 IEEE, reproduced, with permission, from [33]).

It should be noted that I 0 Rin and Pin are functions of Vg1 with Rs as a parameter.
The input impedance and the DC beam current can therefore be plotted against the
RF input power, as shown in Figure 12.21(a). If the source impedance is assumed
to be 220 Ω then a good fit to the experimental data can be obtained, as shown in
Figure  12.21(b), by setting K = 0.42 mA.V −1.5, µ = 150 and the power law to 1.5.
The principles governing the design of the type of gun used in this tube were dis-
cussed in Section 12.6.1.
Since dimensions of the output gap are not given in [59] it is assumed
that: βe a = 1.0, βe b = 0.6, and βe g = 1.0 where βe = ω u0 , a is the drift tube radius, b
is the beam radius and g is the length of the output gap. These parameters are typ-
ical of those used in IOTs and klystrons. The gap coupling factor can then be calcu-
lated as a function of the electron velocity. Starting from an initial guessed value of
the velocity ( umin ) of the slowest electrons leaving the gap self-​consistent values of
umin and Vgap are found for each drive power level using (12.38), as described above.
The load resistance RL is used as an adjustable parameter. Note that this procedure
differs slightly from that described in [33]. The results using the modified procedure
show better agreement with experiment. Graphs of the output power and efficiency
against the input power were plotted (see Figure  12.22) and it was found that a
good fit to experimental data could be achieved by setting RL = 28 kΩ. The 1 dB
bandwidth of this tube was 3 MHz. If it is assumed that this is determined by the
properties of the output cavity then QL = 220 and R Q = 120.
The output power and efficiency were also calculated using the disc model with-
out space-​charge. The values of Vgap were determined by iteration to give the correct
load resistance at each input power level. The results of these calculations are shown
in Figure 12.22 for comparison with the results of the simple model. It should be
noted that the curve of output power against input power is very nearly a straight
462 Gridded Tubes

Figure 12.22:  (a) RF output power, and (b) electronic efficiency of an IOT as functions of

RF input power.

Figure 12.23:  (a) Gain, and (b) normalised minimum electron velocity of an IOT as functions
of RF input power.

line. The shape of this curve is found to depend upon the quiescent point selected,
and it is most nearly linear when there is a small quiescent current. Thus an IOT
has performance which is close to that of an ideal linear amplifier. Figure  12.23
shows the gain, and the normalised minimum electron velocities, plotted against
RF input power. The gain of this tube was a little lower than that typically achieved
in IOTs (around 20 dB). It was increased to around 21 dB in an improved ver-
sion of the tube that showed evidence of island formation [60]. The minimum elec-
tron velocity in the spent beam, normalised to the DC beam velocity, is seen to be
around 0.5 so that no electrons are reflected by the output gap. The comparison
between the results from the simple model and the disc model shows that the per-
formance of an IOT can be modelled with good accuracy using the simple model,
as described above.
 References 463

The effects of space-​charge can be included in the disc model if the drift length
is known. For example, the assumption of a drift length five times the length of the
gap, gave a reduction in efficiency of five percentage points, and a reduction in gain
of 1 dB at an input power of 300 W. Agreement with the measured data could be
achieved by changing the load resistance assumed.

References

[1]‌ W. G. Dow, Fundamentals of Engineering Electronics. New York: John Wiley, 1937.

[2]‌ A. H. W. Beck, Thermionic Valves. Cambridge: Cambridge University Press, 1953.
[3]‌ R. S. Burnap, Electron Tube Design. Harrison, NJ: Radio Corporation of America, 1962.
[4]‌ J. W. Gewartowski and H. A. Watson, Principles of Electron Tubes. Princeton, NJ: D. van
Nostrand Co. Inc., 1965.
[5]‌ K. R. Spangenberg, Vacuum Tubes. New York: McGraw-​Hill, 1948.
[6]‌ T. E. Yingst et al., ‘High-​power gridded tubes –​1972’, Proceedings of the IEEE, vol. 61,
pp. 357–​381, 1973.
[7]‌ A. V. Haeff and L. S. Nergaard, ‘A wide-​band inductive-​output amplifier’, Proceedings
of the IRE, vol. 28, pp. 126–​130, 1940.
[8]‌ D. Preist and M. Shrader, ‘The klystrode –​an unusual transmitting tube with potential
for UHF-​TV’, Proceedings of the IEEE, vol. 70, pp. 1318–​1325, 1982.
[9]‌ R. G. Carter, ‘Acceleration technologies for charged particles:  an introduction’,
Contemporary Physics, vol. 52, pp. 15–​41, 2011.
[10] R. G. Carter, ‘R.F.  power generation’, in R. Bailey, ed., Proceedings of the CERN
Accelerator School ‘RF for Accelerators’, 8–​17 June 2010, Ebeltoft, Denmark, Geneva:
CERN, pp. 173–​207, 2011.
[11] R. G. Carter, ‘Review of RF power sources for particle accelerators’, in S. Turner,
ed., CERN Accelerator School:  RF Engineering for Particle Accelerators, vol.
1. Geneva: CERN, pp. 269–​300, 1992.
[12] Siemens, Transmitting Tubes Data Book 1986/​87. Siemens AG, 1986.
[13] W. P. Bennett, ‘Large-​power-​tube design, processing and testing’, in R. S. Burnap, ed.,
Electron Tube Design. Harrison, NJ: RCA, pp. 763–​791, 1962.
[14] N. O. Sokal, ‘RF power amplifiers, classes A through S-​how they operate, and when to
use each’, in Electronics Industries Forum of New England, 1997. Professional Program
Proceedings, pp. 179–​252, 1997.
[15] F. H. Raab et  al., ‘Power amplifiers and transmitters for RF and microwave’, IEEE
Transactions on Microwave Theory and Techniques, vol. 50, pp. 814–​826, 2002.
[16] S. Kostic et al., ‘Increasing efficiency and output power of HFHPTA by injection of two
harmonics’, IEEE Transactions on Broadcasting, vol. 47, pp. 32–​37, 2001.
[17] EIMAC, Care and Feeding of Power Grid Tubes, 5th ed. San Carlos, CA: CPI Inc.
Eimac Division, 2003.
[18] P. Gerlach and C. Kalfon, ‘New technologies in power-​grid tubes and their impact in
high-​power UHF-​TV operation’, Proceedings of the IEEE, vol. 70, pp. 1335–​1345, 1982.
[19] R. H. Rheaume, ‘A coaxial power triode for 50-​kW output up to 110 Mc’, Proceedings
of the IRE, vol. 40, pp. 1033–​1037, 1952.
[20] Machlett Laboratories Inc., ‘ML-​5681 Data Sheet’, 1962.
464 Gridded Tubes

[21] C. S. Bull, ‘The alignment of grids in thermionic valves’, Journal of the Institution
of Electrical Engineers  –​Part III:  Radio and Communication Engineering, vol. 92,
pp. 86–​92, 1945.
[22] S. G. McNees, ‘Two megawatts RF power tetrode’, IEEE Transactions on Nuclear
Science, vol. NS-​20, pp. 422–​423, 1973.
[23] Y. Kusunose, ‘Calculation of characteristics and the design of triodes’, Proceedings of
the Institute of Radio Engineers, vol. 17, pp. 1706–​1749, 1929.
[24] L. C. Scholz, ‘Calculation of fields and currents’, in R.S. Burnap, ed., Electron Tube
Design. Harrison, NJ: Radio Corporation of America, pp. 159–​201, 1962.
[25] G. Faillon, ‘Technical and industrial overview of RF and microwave tubes for fusion’,
Fusion Engineering and Design, vol. 46, pp. 371–​381, 1999.
[26] G. Clerc et al., ‘A new generation of gridded tubes for higher power and higher frequen-
cies’, in Particle Accelerator Conference, Vancouver, BC, pp. 2899–​2901, 1997.
[27] R. G. Carter, ‘R.F. Generation’, in R. Bailey, ed., High Power Hadron Machines: CERN-​
2013-​001. Geneva: CERN, pp. 45–​69, 2013.
[28] W. Herdrich and H. P. Kindermann, ‘RF power amplifier for the CERN SPS operating
as LEP injector’, IEEE Transactions on Nuclear Science, vol. NS-​32, pp. 2794–​2796,
October 1985.
[29] EIMAC, ‘Tube performance computer for RF amplifiers’, CPI Inc., 1969.
[30] EEV, ‘Preamble –​Tetrodes’. Chelmsford, UK: Teledyne e2v, 1976.
[31] H. P. Kindermann et  al., ‘The RF power plant of the SPS’, IEEE Transactions on
Nuclear Science, vol. NS-​30, pp. 3414–​3416, August 1983.
[32] T. Fujisawa et al., ‘The radiofrequency system of the RIKEN ring cyclotron’, Nuclear
Instruments and Methods in Physics Research Section A:  Accelerators, Spectrometers,
Detectors and Associated Equipment, vol. 292, pp. 1–​11, 1990.
[33] R. G. Carter, ‘Simple model of an inductive output tube’, IEEE Transactions on Electron
Devices, vol. ED-​57, pp. 720–​725, 2010.
[34] R. S. Symons, ‘The constant efficiency amplifier’, in Annual Broadcasting Engineering
Conference, pp. 523–​530, 1997.
[35] H. Bohlen et al., ‘Inductive output tubes for particle accelerators’, in European Particle
Accelerator Conference, Lucerne, Switzerland, pp. 1111–1113, 2004.
[36] H. Bohlen et  al., ‘IOT RF power sources for pulsed and cw linacs’, in LINAC 2004,
Luebeck, Germany, pp. 574–​577, 2004.
[37] R. Heppinstall and G. Clayworth, ‘The inductive output tube’, EBU Technical Review,
pp. 43–​52, 1997.
[38] R. Heppinstall and G. Clayworth, ‘The inductive output tube –​a modern UHF ampli-
fier for the terrestrial television transmitter’, GEC Review, vol. 13, pp. 76–​85, 1998.
[39] E. G. Zaidman and M. A. Kodis, ‘Emission gated device issues’, IEEE Transactions on
Electron Devices, vol. 38, pp. 2221–​2228, 1991.
[40] K. Nguyen et al., ‘Analysis of the 425-​MHz klystrode’, IEEE Transactions on Electron
Devices, vol. 38, pp. 2212–​2220, 1991.
[41] R. True, ‘Calculation and design of grids in Pierce guns’, in 1989 International Electron
Devices Meeting, IEDM ‘89, pp. 215–​218, 1989.
[42] J. Welch, ‘Results from the SLAC Lasertron’, NASA STI/​Recon Technical Report N, vol.
88, p. 26647, 1988.
[43] M. Shrader et al., ‘Pre-​bunched beam devices: efficient sources of UHF and microwave
power’, in International Electron Devices Meeting, pp. 342–​345, 1985.
 References 465

[44] E. Garwin et al., ‘An experimental program to build a multimegawatt lasertron for super
linear colliders’, IEEE Transactions on Nuclear Science, vol. 32, pp. 2906–​2908, 1985.
[45] M. A. Kodis et al., ‘Operation and optimization of gated field emission arrays in induct-
ive output amplifiers’, IEEE Transactions on Plasma Science, vol. 24, pp. 970–​981, 1996.
[46] C. Wilsen et al., ‘Gridless IOT for accelerator applications’, in IEEE Particle Accelerator
Conference, pp. 2556–​2558, 2007.
[47] C. Yates et  al., ‘Performance results of a 1.3 GHz 100 kW CW IOT’, in IEEE
International Vacuum Electronics Conference, Monterey, CA, pp. 107–​108, 2014.
[48] D. H. Preist and M. B. Shrader, ‘A high-​power Klystrode with potential for space appli-
cation’, IEEE Transactions on Electron Devices, vol. 38, pp. 2205–​2211, 1991.
[49] P. Schütt et al., ‘Computer simulations of inductive output tubes’, in European Particle
Accelerator Conference, pp. 1233–​1235, 1998.
[50] H. P. Freund et  al., ‘Time-​ domain simulation of inductive output tubes’, IEEE
Transactions on Plasma Science, vol. 35, pp. 1081–​1088, 2007.
[51] D. Preist, ‘Prospects for very high power high efficiency RF generators’, in Microwave
and Optical Generation and Amplification, Cambridge, UK, pp. 146–​150, 1966.
[52] H. Bohlen et  al., ‘Ways to decrease high-​voltage requirements in high-​power micro-
wave tubes’, in Displays and Vacuum Electronics, Garmisch-​Partenkirchen, Germany,
pp. 269–​273, 1998.
[53] H. Bohlen, ‘Inductive output tubes: history and present status’, in Displays and Vacuum
Electronics, Garmisch-​Partenkirchen, Germany, pp. 93–​98, 2001.
[54] H. S. Kim and H. S. Uhm, ‘Analytical calculations and comparison with numerical data
for annular klystrode’, IEEE Transactions on Plasma Science, vol. 29, pp. 875–​880, 2001.
[55] L. Ives et al., ‘12.2: A 350 MHz, 200 kW CW, multiple beam IOT’, in IEEE International
Vacuum Electronics Conference (IVEC), Monterey, CA, pp. 215–​216, 2010.
[56] E. Wright et al., ‘High-​power multiple-​beam IOT design’, in IEEE Thirteenth International
Vacuum Electronics Conference (IVEC), Monterey, CA, pp. 143–​144, 2012.
[57] C. Yates et  al., ‘Performance characteristics of an MSDC IOT amplifier’, IEEE
Transactions on Electron Devices, vol. 48, pp. 116–​121, 2001.
[58] R. Symons et  al., ‘Prototype constant-​efficiency amplifiers’, IEEE Transactions on
Broadcasting, vol. 47, pp. 147–​152, 2001.
[59] J. Orrett et al., ‘IOT testing at the ERLP’, in European Particle Accelerator Conference,
Edinburgh, Scotland, pp. 1382–​1384, 2006.
[60] A. Wheelhouse, ‘Improved 1.3 GHz inductive output tube for particle accelerators’, in
European Particle Accelerator Conference, Edinburgh, Scotland, pp. 1373–​1375, 2006.
13 Klystrons

13.1 Introduction

We saw in the previous chapter that the performance of an IOT is limited in both
power and frequency by the use of a control grid to modulate the electron beam.
At higher frequencies and powers it is necessary to modulate the beam in some
other way. In the klystron this is achieved by velocity modulation of an electron
beam by a cavity resonator excited by an external RF source, as described in
Section 11.2.1. When an output cavity, tuned to the signal frequency, is placed
in the region where the beam is bunched the result is the simple two-​cavity klystron
illustrated in Figure 13.1. Radio frequency current is induced in the second cavity
in exactly the same way as in an IOT. At resonance the cavity presents a resistive
impedance to the induced current, and the phase of the field across the gap is in
anti-​phase with the RF beam current. Electrons which cross the gap within ±90°
of the bunch centre are retarded and give up energy to the field of the cavity. Since
more electrons cross the second gap during the retarding phase, than the acceler-
ating phase, there is a net transfer of energy to the RF field of the cavity. Thus the
klystron operates as an amplifier by converting some of the DC power input into
RF power in the output cavity.
In practice the gain and efficiency of a two-​cavity klystron are too low to be
of practical value. It is therefore usual to add further cavity resonators in order
to increase the gain, efficiency and bandwidth of the tube. Figure 13.2 shows
the arrangement of a multi-​cavity klystron. The electron beam is formed by a
diode electron gun whose perveance is typically in the range 0.5 − 2.0 µA V −1.5 .
The function of all the cavities, except the last, is to form electron bunches from
which RF power can be extracted by the output cavity. The first and last cav-
ities are tuned close to the centre frequency and have low Q factors determined
largely by the coupling to the external waveguides. The intermediate (idler)
cavities normally have high Q and are tuned to optimise the performance of
the tube. The long electron beam is confined by an axial magnetic field to avoid
interception of electrons on the walls of the drift tube. Radio frequency power
passes into and out of the vacuum envelope through ceramic windows, and
the spent electrons are collected by a collector in exactly the same way as in
an IOT.
 Introduction 467

Figure 13.1:  Arrangement of a two-​cavity klystron.

Figure 13.2:  Arrangement of a multi-​cavity klystron.

The tube illustrated in Figure 13.2, whose cavities are wholly within the vacuum
envelope, is known as an integral cavity klystron. An alternative arrangement is to
insert a cylindrical ceramic window in the cavity as shown in Figure 13.3 so that the
outer portion is at atmospheric pressure. Tubes with this kind of cavity are known
as external cavity klystrons [1–​3]. The external part of the cavity is made in two
halves which are bolted around the tube body. A good electrical connection at the
joint is ensured by the use of metallic spring fingers. The cavity can be tuned by
sliding walls as shown. The tuning and coupling arrangements are made simpler by
the use of external cavities because they are outside the vacuum envelope. A tun-
ing range of 50–​60% is possible with this arrangement. The advantages of external
cavity klystrons are: easier tuning; easier coupling; and cheaper replacement tubes.
However, they are limited to powers under 100 kW and frequencies less than 1 GHz
by voltage breakdown in the part of the cavity at atmospheric pressure, and by the
need to ensure good electrical contact at the junction between the external cavity
and the tube body. The window is in line of sight with the electron beam and may be
damaged by direct electron bombardment or a multipactor discharge (see Section
468 Klystrons

Figure 13.3:  Arrangement of an external klystron cavity.

18.8). The task of replacing a failed tube is made more difficult because of the need
to transfer the external cavities to the new tube, but they should then need little or
no retuning.
This chapter considers the theory and design of klystrons. Small signal models
are based on the space-​charge wave theory in Section 11.4, while large signal mod-
els represent the electron beam by discs, or rings, of charge as described in Section
11.8. Klystrons fall into two main classes:
• Broadband tubes have bandwidths up to 10% at frequencies from UHF to around
18 GHz. Their applications include radar, UHF television transmitters, satellite
ground stations and tropospheric scatter systems. [1, 4].
• Narrow band, high efficiency, tubes are used chiefly in particle accelerators and
plasma heating and in nuclear fusion experiments at frequencies below 4 GHz
[5, 6].
The design of tubes in these classes is discussed in Section 13.4. There is a brief
review of the principles of multiple beam and sheet beam klystrons. Further infor-
mation can be found in [3, 7–​10]

13.2 Small-​Signal Klystron Theory

Under small-​signal conditions a klystron can be modelled using the space-​charge

wave theory introduced in Chapter 11. The cavities are represented by equivalent
circuits, as shown in Figure 13.4. The coupling factors, and beam loading admit-
tances, of the interaction gaps can be calculated using (11.36), (11.38), and (11.100).
In practical cases they can be determined by cold and hot test measurements, or by
numerical modelling [11, 12].
 Small-Signal Klystron Theory 469

Figure 13.4:  Klystron equivalent circuits: (a) input cavity, and (b) idler and output cavities.

The beam loading conductance (Gb ) is typically around 0.15 times the DC beam
conductance (G0 ) when g = a , and the gap transit angle is 1 radian (see Figure 11.16).
Thus, for a non-​relativistic beam, we can write

Gb ~ 0.15K Va , (13.1)

where K is the perveance and Va is the beam voltage. If K = 0.5 µA V −1.5 and
Va = 10 kV, both of which are at the lower end of the range of typical values, then
Gb ~ 7.5 µS . The R Q of a typical re-​entrant cavity is of the order of 100 Ω. The
practical unloaded Q is around 6000 at 1 GHz, and decreases by a factor of 4 for
every decade increase in frequency, if the surfaces of the cavity are smooth (see
Chapter 3). Thus the shunt conductances of the cavities range from around 1 μS
at 1 GHz, to 10 μS, or more, at 30 GHz. This is comparable with the beam loading
conductance. The reactive component of the beam loading is normally negligible
[13] but, if necessary, it can be compensated for by slight adjustments to the tuning
of the cavities.

13.2.1 Input Cavity
The equivalent circuit of the input cavity is shown in Figure 13.4(a). The source
current and conductance are transformed from those for the external source by the
input coupler (see Section 3.6). Then the input gap voltage is

is
Vg = , (13.2)
Gs + Yc + Yb
where the cavity admittance is obtained from (3.5) as

  ω ω0  
Yc = Gc 1 + jQU  − , (13.3)
  ω 0 ω  

where Gc is the shunt conductance. The forward power in the input waveguide is

is2
Pin = . (13.4)
8Gs
470 Klystrons

Then, from (13.2) and (13.4) the input gap voltage is

2 8Gs
Vg1 = 2
Pin . (13.5)
Gs + Yc + Yb

If the input is matched at resonance then Gs  Gb + Gc, where Gb is the real part of
Yb, and the power transferred to the beam is

1 2 4GsGb
P1 = Vg Gb = 2
Pin , (13.6)
2 2 ∆ω
2Gs + jQU Gc
ω0
where ∆ω = ω − ω 0 . But QE Gs = QU Gc so that (13.6) becomes

1 Gb
P1 = ⋅ Pin (13.7)
(1 + Q ( ∆ω ω ))
E
2
Gs

This equation shows that the gain of the input cavity is maximum at resonance and
that it is reduced by losses in the cavity. Away from resonance the input is no longer
matched and the gain is reduced by reflected power. The modulation of the beam
by the input gap can be expressed in terms of the space-​charge wave amplitudes
using (11.91)

V1+  1 M + 
V  = 2 M Vg1 . (13.8)
 1−   −
The transformation of the RF beam voltage and current by a drift length d is given
by (11.110)

V2 +  exp ( − j β + d ) 0  V1+  V1+ 

V  =     = [ D ]   . (13.9)
 2−   0 exp ( − j β − d ) V1−  V1− 

13.2.2 Idler Cavities

Figure  13.4(b) shows the equivalent circuit for an idler cavity including the pos-
sibility of an external load (though this is unusual). The beam is represented by
a current source having magnitude i g and admittance equal to the beam admit-
tance (Yb ). In small-​signal analysis it is usual to use the beam-​loading admittance
of an unmodulated beam in all cavities, though this is not strictly correct [14, 15].
If the imaginary part of Yb is small compared with the real part (Gb ) then Yb can be
replaced by Gb without serious error. The cavity extracts power from the modulated
beam and uses it to add fresh modulation to the beam. Because the cavity imped-
ance is high, a large gap voltage is produced by a small induced current. If the cavity
losses are neglected, the kinetic power in the beam leaving the cavity is equal to that
entering. Analysis of the interaction between the beam and the cavity is represented
by the transfer matrix in (11.109)
 Small-Signal Klystron Theory 471

 Y Ye 
 1 − e M +2 − M+M− 
 2+ 
V 2YT 2YT V  V 
V  =  Y   1+  = [C ]  1+  , (13.10)
 Y  V V1− 
 2− 
 − 2Y M + M −
e
1 − e M −2   1− 
 T 2YT 
where

YT = Yc + Yb + Ye (13.11)

is the admittance of the cavity, the beam, and the external circuit, in parallel with
one another. If the klystron has N cavities separated by ( N − 1) drift spaces the volt-
age and current modulation at the output gap is given by

VN +  1 M + 
V  = 2 [ DN −1 ][CN −1 ][C2 ][ D1 ] M Vg1 , (13.12)
 N−   −
where the properties of the gaps and of the drift spaces may differ from one another.

13.2.3 Output Cavity
The voltage induced in the output gap is given by (11.108) as

Ye VN + 
VgN = − [M + M− ]   . (13.13)
YT VN − 
The output power is

1
Re (YE ) VgN (13.14)
2
Pout =
2
so that, finally, the power gain in decibels is

P 
GdB = 10 log  out  . (13.15)
 Pin 
This method of calculation is incorporated in Worksheet 13.1.
An alternative formulation using the transmission line representation of the
beam employs the cavity transfer matrix in (11.119) [13]

 G M2 
 1− b −
V2   YT YT  V1 
I  =  G2  . (13.16)
 2 Gb   I1 
 − M 2Y 1− 
b

 T YT 
The drift matrix is

V2   cos βq z jZe sin βq z  V1 

 I  = exp ( − j βe z )  jY sin β z cos βq z   I1 
, (13.17)
 2  e q
472 Klystrons

where time variation as exp ( j ωt ) has been assumed. At the input, from (11.114)

V1   M 
 I  = G M Vg1 (13.18)
 1  b 
and, at the output, from (11.117)

1  Gb  VN 
VgN =  M M    . (13.19)
YTN  IN 
The small-​signal can be computed by cascading the transfer matrices as before.
The voltage at the output combines contributions from the individual cavities so
that there is a peak of voltage at the resonant frequency of each cavity. In addition
the phasor addition of the contributions of the cavities produces maxima and min-
ima as the voltages add and subtract in phase. It can be shown that the complex
gain can be written
N −1

∏ (s − z )
n
G (s) = A n=2
N
, (13.20)
∏ ( s − pn )
n =1

where s is the complex frequency and A is a constant [3, 7]. This function has poles
in the complex frequency plane when s = pn and zeroes when s = zn .
The way in which the poles and zeroes arise can be illustrated by a simple approxi-
mate analysis based on cascaded matrices, as described above. The tube is considered
to be made up of series of stages, each of which comprises one cavity with the follow-
ing drift length. As the modulation grows along the length of the tube the RF current
in the beam increases while the electron velocity remains approximately constant. The
RF current in the beam entering a stage drives the gap voltage of the cavity. The vel-
ocity modulation imposed by the gap is transformed into additional current modula-
tion by the drift region. Thus, to a first approximation, the current gain in a stage can
be computed by ignoring the velocity modulation at its start. This analysis assumes
that the contribution to the gain from cavities which are not adjacent is negligible.
The overall gain of the tube is found by summing the gains of the stages in decibels.
The interaction with a gap derived from (13.16) is

 M2 
−
V2   YT 
 I   G  I1 , (13.21)
= 
 2  1− b
 Y 
 T 

where I1 is the RF beam current before the gap. The modulation of the beam after
a drift length d is found by pre-​multiplying (13.21) by the drift matrix in (13.17) so
that the RF current at the end of a drift region of length d is given by
 Small-Signal Klystron Theory 473

 Y  G  
I 3 = exp ( − j βe d )  − j e M 2 sin (βq d ) + 1 − b  cos (βq d ) I1 . (13.22)
 YT  YT  
Then the current gain in decibels is

 G 2 Y
2

GdB = 10 log  1 − b cos 2 (βq d ) + e M 2 sin2 (βq d ) . (13.23)
 YT YT 
If there is no added loss, or external loading, thenYT ≈ Gb at the resonant frequency
of the cavity. Then the second term is dominant and the gain contributed by the
stage is approximately

 M 2Ye 
GdB  20 log  sin (βq d ) . (13.24)
 YT 
For typical values of the parameters this has a value in the range 20 to 40 dB. When
the signal frequency is far from resonanceYT is dominated by the cavity susceptance
so that

QU 2 ∆ω
YT  j . (13.25)
Rc ω 0
If we now assume that YT  Gb then (13.22) becomes

I3 Y R ω
= cos (βq d ) − e c 0 M 2 sin (βq d ) . (13.26)
I1 2 QU ∆ω
This expression is zero when

∆ω Ye Rc
= ⋅ M 2 tan (βq d ) , (13.27)
ω0 2 QU
which shows that, when βq d < 90° the frequency of the zero lies above the resonant
frequency of the cavity, and that it increases as βq d increases. The frequency of
the zero lies below the resonant frequency of the cavity when βq d > 90° . These
general conclusions are confirmed by more accurate calculations as illustrated, for
typical parameters, in Figure 13.5. The maxima and minima of the curves corre-
spond to the poles and zeroes of the transfer function of the stage in the complex
frequency plane. The transfer functions of the input and output cavities, which do
not have zeroes, only contribute poles to the overall transfer function. Hence, for
a klystron with N cavities there are N poles and N − 2 zeroes. The zeroes produce
dips in the gain, and an important part of the design of a klystron is the selection
of the cavity and drift-​tube parameters in such a way that the variation in gain
within the band is minimised. The design of broad-​band klystrons is discussed in
Section 13.4.1.
474 Klystrons

Figure 13.5:  Gain of a klystron stage (cavity plus drift length) as a function of frequency and
drift length  (R Q = 100, Q = 4000, M = 0.79, Gb = 10 µS, Ye = 2 mS) .

13.2.4 Simplified Small-​Signal Model

A simplified small-​signal model of a klystron can be constructed by assuming that
the contribution to the RF beam current made by a gap is negligible so that (13.18)
becomes

V1 = MVg1 . (13.28)

The RF beam current is given as a function of position by (11.86)

I 2 ( z ) = jYe M1Vg1 sin (βq z ) exp ( − j βe z ) . (13.29)

and the voltage induced in a cavity by this current is

M2 I2 Y
Vg 2 = − = − j e M1M 2Vg1 sin (βq d ) exp ( − j βe d ) , (13.30)
YT 2 YT 2
where d is the drift length between the cavities. The voltage induced in each cavity
excites waves which, in turn, induce voltages in each of the subsequent cavities,
and so on. Because the analysis is linear, the final voltage induced in the output
cavity is the phasor sum of the voltages induced by all the waves which have been
excited. This can be thought of in terms of a number of signal paths in parallel,
each of which has its own gain and phase shift as a function of frequency. These
paths represent all the possible connections between the input and the output.
Thus for a five-​cavity tube, for example, we find:  one path which connects the
input and the output directly; three which take in one intermediate cavity; three
which take in different pairs of cavities; and one which takes in all three inter-
mediate cavities –​a total of eight paths [7, 16, 17]. This model is implemented in
Worksheet 13.2.
 Large-Signal Behaviour of Klystrons 475

13.2.5 Overall Performance

Figure 13.6 shows a comparison between the calculated and measured small-​signal
gain for a 3kW klystron for use in satellite ground stations. The cavities of this tube
are tuneable and their exact frequencies in the experimental tube are not known. The
frequencies of the experimental results were scaled by 1.25% to match the calculated
results. The figure shows results from Worksheets 13.1 and 13.2. The gain was found
to be quite sensitive to the beam radius. In the example shown, the best results were
obtained when the radius was reduced by about 10% relative to the theoretical value.
It is often difficult to obtain a reliable value for the beam radius, especially if the
current density is not uniform, though good results are usually given by using the
equilibrium beam radius in the calculations. This tube has four stages (five cavities)
so that the average small signal gain is about 11 dB per stage. The cavities are tuned
for broad-​band operation (see Section 13.4.1). If the cavities in the model are all
tuned to the same frequency the gain per stage at the band centre is increased to 19
dB. However some tubes suffer from feedback oscillations if they are synchronously
tuned so it does not follow that this tube could be operated with synchronous tuning.

13.3 Large-​Signal Behaviour of Klystrons

Unlike an IOT, the DC power input to a klystron is constant, regardless of the level
of the RF output power. Therefore it is usually desirable to operate a tube with
high RF power output to obtain the best conversion efficiency. Under large-​signal
conditions the behaviour of a tube becomes non-​linear and the linearised theory in
the previous section is not valid. It is therefore necessary to use a large-​signal model
as described in Section 11.8. Details of time-​step and distance-​step models, and
of the application of PIC codes, can be found in the literature [18–​20]. The main
features of the large-​signal behaviour of a klystron can be illustrated using a simple

Figure 13.6:  Comparison between calculated and measured small-​signal gain of a klystron.

476 Klystrons

time-​step disc model (see Worksheet 13.3). The power transferred to the output gap
tends to be an over-​estimate when computed by a one-​dimensional model, because
the radial variations of the space-​charge field and the field of the gap, and the radial
motion of electrons, have been neglected [21, 22].
We saw in Section 11.8.4 that the RF power extracted from a bunched beam by the
output cavity increases with the first harmonic of the beam current ( I1 ). In the small-​
signal regime, the amplitude of I1 at a fixed plane varies linearly with the normalised
input gap voltage (see (13.29)). When the RF power in the beam is small, the output
gap voltage is small compared with Va so that the electrons cross the gap with aver-
age velocities close to u0 . The gap coupling factor is then independent of the level
of modulation, and the output power is proportional to the input power. As the RF
power in the beam increases the behaviour of the tube becomes non-​linear because:
• The amplitude of the fundamental RF current is not linearly dependent on the
input gap voltage.
• The position at which the amplitude of the current is greatest varies with the drive
level (see Figure 11.26).
• The effective coupling factor of the output gap decreases because the electrons at
the centre of the bunch slow down as they pass through the gap.
The plot of Pout against Pin is then not linear at high drive levels but saturates as
shown in Figure 13.7. The difference between the gain at saturation and the small-​
signal gain, known as the gain compression of the tube, is typically in the region of
5 to 7 dB. The output power of a klystron therefore depends on both the effective-
ness with which the beam is bunched, and on the power conversion efficiency in the
output gap. When the frequency of the input signal is varied within the working
band of the tube it is found that the saturated output power is constant to first
order. However, the saturated gain and hence the input power required to saturate
the tube, vary with frequency [13].

Figure 13.7:  Typical non-​linear power transfer characteristics of a klystron.

 Large-Signal Behaviour of Klystrons 477

13.3.1 Klystron Sections

It is convenient to consider the design of a multi-​cavity klystron, such as that
illustrated in Figure  13.2, in terms of three sections, as shown in Figure  13.8
[23]. Each section is made up of one, or more, stages each comprising a cavity
with the following drift length. The cavities can be represented by the equiva-
lent circuits shown in Figure  13.4 if the gap coupling factors are adjusted for
large-​signal effects where necessary. The purpose of the first section is to provide
sufficient gain to produce a strongly bunched beam over the working bandwidth
of the tube. The second section increases the fundamental RF current by com-
pressing the bunches to enable the maximum power to be extracted from them
by the output section. The factors influencing the design of each section are
discussed below.
The objective of the bunching process is that the final value of I1 I 0 should be
as great as possible, with minimum velocity spread and few crossing trajectories. It
has been suggested [24] that a useful figure of merit for bunching is

1 I1 umin
F1 = , (13.31)
2 I 0 u0
where umin is the velocity of the slowest electron within the bunch. The maximum
possible value of this figure of merit is F1 = 1, and it decreases sharply when electron
crossovers occur. It has been found to be a good predictor of efficiency at saturation
and to be useful for the optimisation of the bunching [13, 25]. A definition based
on the minimum electron energy in the input bunch could also be justified. However,
the velocity of the slowest electron is not a good measure of the velocity spread in
the bunch as a whole [26]. An improved figure of merit is

1 I1  ∆u 
F2 = 1 −  , (13.32)
2 I 0  u0 

where ∆u is the root mean square of the velocity spread in the bunch. The appli-
cation of this figure of merit to the design of klystrons is discussed in Section
13.3.3.

Figure 13.8:  The sections of a klystron.

478 Klystrons

13.3.2 Initial Bunching Section

The input cavity is normally working in the small-​signal regime so that the analysis
in Section 13.2.1 is valid. The resonant frequency of the cavity is often close to
the centre frequency of the tube but it may be elsewhere in the working frequency
band in a broadband klystron (see Section 13.4.1). The normalised voltage in the
input gap can be estimated by recalling that the RF power transferred to the beam
is given by (11.98)

1 2
Pg = Gb Vg , (13.33)
2
and that the output power of the tube is

Pout = ηe I 0V0 . (13.34)

Thus the gain is

2
2 ηe I 0V0 2 ηeG0 V0
G= 2
= . (13.35)
Gb Vg Gb Vg

Now, typically for a broadband klystron, ηe ~ 0.5, Gb G0 ~ 0.15 and the power gain
is 40 dB so that G = 10 4. Substituting these figures into (13.35) we obtain the nor-
malised input gap voltage

MVg
~ 0.02, (13.36)
V0
where it has been assumed that, typically, M ~ 0.8.
Klystron input cavities may be affected by multipactor discharge (see Section
18.8), especially between the tips of the drift tubes. These discharges are usually
of low power but they add loading to the cavity and introduce noise [27–​29]. The
problem may be reduced or eliminated by coating the surface with a material having
a low secondary electron emission coefficient, by castellation of the drift-​tube tips,
or by changing the surface topology in some other way [3].
The remaining cavities in the initial bunching section make a major contribution
to the gain of the tube. Their parameters are chosen to equalise the gain over the
bandwidth of the tube, and to create well-​formed bunches for compression by the
following section. Although small-​signal theory is sometimes used to model bunch
formation we shall see that the role of harmonic modulation of the beam can be
important. The resonant frequencies of the cavities generally lie within the band-
width of the tube. Their Q factors may be adjusted by the addition of lossy coatings
or by coupling to external loads [30]. In this section the RF power in the beam is
small compared with the DC beam power. Thus, the power dissipated in the cavity
losses is small and has a negligible effect on the overall efficiency of the tube. The
gap voltage induced in a cavity is determined by the RF current in the input beam
 Large-Signal Behaviour of Klystrons 479

so that, using small-​signal theory, the amplitude of the velocity modulation added
to the beam is

M 2 I1
V1 = , (13.37)
QU 2 ∆ω
Gb + Gc + j ⋅
Rc ω 0

where ω 0 is the resonant frequency of the cavity. Taking Rc Q0 = 100 Ω as a typical

figure, it can be seen that the imaginary term in the denominator becomes compar-
able with the beam conductance estimated from (13.1) when the frequency devi-
ation is of the order of 1%. For greater frequency deviations the imaginary term is
dominant and (13.37) can be written approximately as

V1  R  ω  I
 G0  M 2 c   0  1 . (13.38)
V0  QU   2 ∆ω  I 0
Thus, if large bandwidth is required, it is desirable to maximise the beam conduct-
ance by choosing a high beam perveance, and to maximise M 2 ( Rc Q0 ) [13, 31].
This parameter depends essentially on the choice of the normalised radius of the
drift tube and on the gap transit angle. It has been found that the optimum transit
angle is 1.4 radians [13].

13.3.3 Final Bunching Section

The cavities in this section are tuned to frequencies above the working band of
the tube so that the phase of the gap voltage leads the phase of the RF beam
current by just under 90°. The centre of the bunch crosses the gap when the RF
field is zero and the bunch is compressed. The extent to which the bunch can be
compressed in this way depends upon both the form of the input bunch and the
space-​charge density in the beam. The influence of these factors on bunch com-
pression by a single cavity can be explored using the disk model in Worksheet
13.4. Figure 13.9 shows, as an example, the compression of bunches in a 25 kV, 1
μPerv, beam when the current waveform in the input beam is defined by (11.175)
with n = 1.5 so that I1 I 0 = 1.2. The depression of the potential by the space charge
of the bunches causes some of the kinetic energy in the beam to be converted to
potential energy. The initial electron velocity has been reduced to account for this.
The figure of merit F2 is greatest, at a distance z λ e = 0.845 from the centre of the
gap, when the normalised gap voltage MVg Va = 0.33. The normalised RF cur-
rent at this plane is I1 I 0 = 1.63. Figure 13.9(b) shows the computed current wave-
form with the ideal current waveform for n = 4 for comparison. The bunching
can also be represented by plotting the relative phase at which sample electrons
reach a plane against their initial phase, as shown in Figure 13.9(c) [32]. These
results were obtained using the parameters βe a = 0.5, βe b = 0.3, βe g = 0.5 . When
the case in Figure  13.9 was recomputed with all the parameter values doubled
480 Klystrons

Figure 13.9:  Compression of a bunch by a cavity: (a) electron trajectories (b) current

waveforms, and (c) phases of sample electrons.

the maximum normalised RF current was 1.78 at z λ e = 1.005. Thus the detailed
results depend upon the choice of parameters but it is to be expected that the
general trends will be the same.
The effects of perveance and of the initial bunching of a beam having the same
parameters are shown in Figure  13.10. Figure  13.10(a) shows that the normal-
ised RF current in the final bunches increases, tending to a limit which depends
upon the perveance, as the initial bunches become tighter. However, for perveance
greater than 1.0, it is found that a point is reached where the initial bunches are
already optimal and cannot be compressed further if the maximisation of F2 is
used as the criterion. The figure also shows that, beyond a certain point, further
tightening of the input bunches yields only a marginal improvement in the final
bunches. Figure 13.10(b) shows that the distance from the cavity to the plane of
optimal bunching decreases as the output bunches get tighter, and also as the per-
veance increases.
 Large-Signal Behaviour of Klystrons 481

Figure 13.10:  (a) Variation of normalised RF current in compressed bunches, and

(b) variation of the optimal drift length with the normalised RF current in the input beam
and with beam perveance.

13.3.4 Output Section

The output cavity is normally tuned close to resonance at the band centre so that
the gap field is in anti-​phase with the bunches to remove as much energy from them
as possible. From the discussion in Section 11.8.4 we know that the electronic effi-
ciency is given by

1 I1 Vg
ηe = M eff , (13.39)
2 I0 V0
where Vg is the output gap voltage and M eff is the effective gap coupling factor
defined by (11.180). The ratio I1 I 0 is determined by the bunching in the previous
sections. It has a theoretical maximum value of 2 for a delta-​function bunch. The
electronic efficiency which can be achieved is determined by a number of factors:
i) The tendency for some electrons to be re-​accelerated leads to an optimum value
for energy transfer to the gap, as shown in Figure 11.31. In extreme cases some
electrons may be reflected by the gap. If these pass back down the drift tube
towards the electron gun they may provide positive feedback causing the tube
to break into oscillation. We saw in Section 11.8.4 that if M = 0.8 then the opti-
mum gap voltage is 1.5V0 and M eff ~ 0.55 so that M eff Vg V0 ~ 0.82 . This analysis
neglected the effects of space-​charge which are discussed in Section 13.3.6.
ii) The electrons leaving the gap must have sufficient residual velocity to drift out of
the field of the gap, without the formation of a virtual cathode at the entrance
to the collector. For example, if M eff Vg V0 = 0.9 then the electrons leave the gap
with about 10% of their initial energy and velocities around 30% of their ini-
tial velocity. This is probably close to the limit set by this requirement. We note
that this limit may be greater than that set in (i) above but it is also influenced by
482 Klystrons

potential depression caused by the high space-​charge density just after the gap (see
Section 13.3.6).
iii) The velocity spread in the bunches as they enter the output gap tends to reduce
the exit velocities of the slowest electrons and to increase the risk that they may
be reflected. For this reason it is desirable for the velocity spread to be small.
iv) Storage of potential energy in the bunches reduces the kinetic energy available
for conversion to RF power. The increase in the space-​charge density in the
bunch entering the output gap leads to increased potential depression which can
be estimated from (7.8) by replacing the mean beam current by the peak cur-
rent. Thus the velocity of the bunch is less than u0 and the effective gap voltage
must be reduced accordingly. When the velocity of the bunch is reduced by the
gap the space-​charge density and the stored potential energy are increased. The
power balance equation is

PDC = PRF + PKE + PPE , (13.40)

where PRF is the power transferred to the gap, PKE and PPE are the kinetic and
potential power in the spent beam, and the RF power used to modulate the
beam is neglected. Thus the electronic efficiency is

PKE PPE
ηe = 1 − − . (13.41)
PDC PDC
The importance of the potential energy term increases as the average charge
density in the beam increases. It can lead to a reduction in efficiency computed
from the kinetic power by several percentage points in beams with high perve-
ance [33]. This suggests a possible alternative figure of merit for bunching

Vb
F3 = ⋅ F2 , (13.42)
Va
where Vb is the energy of the bunched beam in eV calculated from the kinetic
energies of the electrons.
v) It is essential to avoid voltage breakdown in the output gap (see Section 18.7.1).
The calculations which have been presented so far assume that the gap voltage
can be as large as necessary to obtain the optimum extraction of energy from
the bunches. The breakdown condition for a radio-​frequency gap is often taken
to be given by Kilpatrick’s formula [34]

f = 1.6 E max
2
exp ( −8.5 E max ) , (13.43)

where f is the frequency in MHz and E max is the maximum electric field on the sur-
faces of the gap in MV m−​1. The maximum electric field is much greater than the
average field in the gap, defined by E0 = Vg g , because of the concentration of
the field on the tips of the drift tubes. These are commonly rounded to reduce the
maximum field as far as possible. An example of the output gap design for a high
 Large-Signal Behaviour of Klystrons 483

power klystron is given in [35]. In this 50 MW tube operating at 2856 MHz the out-
put gap voltage at saturation was 418 kV, corresponding toVg Va = 1.32. The maxi-
mum electric field was 36 MV m–​1 and the average gap field was 26 MV m–​1 giving a
field enhancement factor of 1.4. Substitution of these figures into (13.43) gives
the maximum safe frequency of operation of the gap as 1.6 GHz. However, it
is known that fields up to twice this limit can be sustained if the surfaces are
clean and well-​prepared [34]. Thus the criterion in (13.43) may be regarded as
conservative.
Performance beyond that of a single output cavity can be achieved by using two
or more coupled output cavities in which the gap field is less than it would be in
an equivalent single cavity [36, 37]. This arrangement, known as an extended inter-
action output cavity, is usually operated in the π-mode with the fields in adjacent
cavities in antiphase. The cavity resembles a short section of coupled-​cavity slow-​
wave structure with short-​circuited ends. A description of a five-​section, π-mode,
extended interaction cavity is given in [38]. At very high powers it has been found
better to operate the cavities in the 2 π-mode so that adjacent cavities are in phase
with one another [39].
The efficiency of klystrons decreases with increasing frequency, as shown by the
data for CW klystrons in Figure 13.11(a). The decrease is the result of design com-
promises and the effects of increasing losses in the output cavity. The figure shows
the trend line

ηe = 0.65 − 0.0225 f , (13.44)

where f is the frequency in GHz. Figure 13.11(b) shows the data for pulsed klys-
trons with the same trend line. The data follows the same trend apart from two
outlying points.
Large-​signal modelling of klystrons shows that, at saturation, the harmonic cur-
rents in the beam have large amplitudes and, in particular for the second har­
monic, I 2 I 0 ~ 1. However, because the output cavity is resonant, a low impedance
is presented to the harmonic current, and the output power at harmonic frequencies
is small. Measurements on a SLAC high power klystron at a fundamental power of
31 MW showed second and third harmonic outputs of –​24 dBc and –​39 dBc respec­
tively [40]. Clearly this would not be the case if any of the higher-order modes of
the output cavity were to present appreciable impedance to the beam at harmonic
frequencies. It is therefore important to check the frequencies of the higher-order
modes as part of the design of the cavity.

13.3.5 Output Coupling

The discussion to this point has assumed that the coupling of the output cavity to
the external load has been arranged so that it is matched. In practice it is not always
possible to ensure that the load is perfectly matched to the output of the klystron
even when a circulator is placed between them. It is therefore necessary to consider
the effect on the performance of the tube if the load is mismatched. For simplicity
484 Klystrons

Figure 13.11:  Efficiencies of klystrons from published data sheets plotted against

frequency: (a) CW tubes, and (b) pulsed tubes.

we will assume that the current induced in the cavity is not affected by the change in
the match and that the cavity is resonant. From Figure 13.4(b) we see that the load
is driven by a current source i g whose source admittance is Gc + Gb. From (11.108)
the gap voltage is

ig
Vg = , (13.45)
Gb + Gc + YL
where YL = GL + jBL is the load admittance referred to the gap. When the load is
correctly matched to the tube BL = 0, GL = GL 0 = Gc + Gb and

ig
Vg 0 = (13.46)
2GL 0
 Large-Signal Behaviour of Klystrons 485

as expected. Dividing (13.45) by (13.46) gives

Vg 2GL 0 2
= = , (13.47)
Vg 0 GL 0 + GL + jBL 1 + gL + jbL

where the lower case symbols represent admittances which have been normalised to
the source conductance. This can be written
Vg
= 1 + ρ exp ( j φ ) , (13.48)
Vg 0

where ρ is the magnitude of the voltage reflection coefficient of the load and φ is
its phase. If the tube is designed to give maximum efficiency with a matched load
then the output power is reduced when the amplitude and phase of the load are
changed (see Figure 11.31). Contours of constant load power can be plotted on a
Smith chart of the normalised load admittance referred to the output flange of the
tube. The result is known as a Rieke diagram (see Figure 13.12). The contours of
constant power are not circles because the effective gap coupling factor depends on
the gap voltage. This effect increases as the mismatch increases. The shaded area of
the diagram represents a region in which the tube must not be operated because the
gap voltage would be too great. The use of the output flange, rather than the gap,
as the reference plane has caused the contours to be rotated about the centre of the
chart relative to their theoretical positions.

13.3.6 Theoretical Limits to Efficiency

It has been observed by a number of authors that the maximum efficiency of a
klystron decreases with increasing perveance [41–​44]. Taking the average of the
empirical formulae suggested gives

Figure 13.12:  Example of a Rieke diagram for a klystron

(courtesy of Thales Electron Devices and Georges Faillon).
486 Klystrons

Figure 13.13:  Efficiencies of klystrons from published data sheets plotted against perveance
showing the trend line given by (13.43): (a) CW tubes (the triangles show points computed
using Worksheet 13.4), and (b) pulsed tubes.

I0
ηe = 0.82 − (0.2 × 10 −6 ) . (13.49)
Va1.5
Figure 13.13(a) shows points obtained from published data sheets for CW klystrons
with the line given by (13.49) for comparison. It should be noted that this line does
not take account of the decrease in efficiency with increasing frequency shown in
Figure 13.11. The two effects can be combined to produce a general expression for
the maximum possible efficiency

ηe = 0.85 − 0.2 × µPerv. − 0.0225 × f (GHz ) . (13.50)

The data for pulsed klystrons is shown with the same trend line in Figure 13.13(b). It
is clear that the efficiencies of pulsed tubes can be greater than predicted by (13.49).
 Large-Signal Behaviour of Klystrons 487

The reason for this is that many pulsed tubes operate at higher beam voltages than
CW tubes (see Tables 13.2 and 13.3). At these voltages relativistic effects become
important and it is possible to remove more kinetic energy from the electrons.
The variation of efficiency with perveance can be explained by the effects of
space-​charge density. Thus we expect that the maximum attainable efficiency will be
related to ω 2q ω 2 . If relativistic effects are ignored then
2
 ωq  p2 1 I
 ω  = ⋅ ⋅ 0 . (13.51)
2 πε 0 2 e m0 (βe b ) Va1.5
2

Now, typically, βe b < 1 and b a ~ 0.6. In this parameter range p is approximately

proportional to βe b for a magnetically focused electron beam, as can be seen from
Figures 11.10(b) and (c). Thus ω 2q ω 2 is approximately proportional to the perve-
ance of the beam for all tubes.
The disc model in Worksheet 13.4 was used to explore the reasons for the
dependence of efficiency on perveance for non-​relativistic electrons. The nor-
malised beam and gap parameters were taken to be βe a = 0.5; βe b = 0.3; βe g = 0.5
giving a small signal gap coupling factor M = 0.945 . Thus the energy given up
by the electrons was generally not limited by re-​acceleration in the output gap.
These parameters are similar to those used in tubes designed for high efficiency.
The effective coupling factor taken from Figure 11.32 was 0.81. The beam enter-
ing the penultimate cavity was assumed to be ideally bunched, as described in
Section 13.4.2. For each value of the perveance the case giving the highest value
of I1 I 0 in Figure 13.10(a) was chosen. The voltage across the penultimate gap
was adjusted to maximise F2 and the output gap was placed at the plane where
the figure of merit was maximum. The amplitude of the output gap voltage was
then adjusted to minimise the final kinetic energy of the electrons. The relative
phase of the output gap voltage was fixed at −​180° which is close to the phase
at which the bunches experience the maximum retarding field. This procedure
was found to give the best conversion efficiency. The results, shown as triangles
on Figure 13.13(a), are similar to those given by (13.49). They are an indication
of the maximum efficiency which can be achieved when transverse motion of
the electrons is not important, other design constraints do not apply, and the
power loss in the output cavity is negligible. In all the cases examined, apart
from 0.2 μPerv, the efficiency was found to be limited by the reflection of slow
electrons by a potential minimum after the output gap (see Figure 13.14). From
these results it appears that the maximum efficiency of non-​relativistic tubes
tends to 85% in the limit of small perveance and low frequencies, as suggested
in (13.50).
The space-​charge density in the beam (which increases with increasing perve-
ance) limits the maximum RF current in the compressed bunches entering the
output cavity. It also limits the energy which can be extracted from the bunches
because of the need to avoid reflection of electrons by potential depression in
the spent beam. In order to achieve the highest possible efficiency at a given
perveance:
488 Klystrons

Figure 13.14:  Normalised electron velocities for maximum efficiency with a 1 μPerv beam.

• The beam entering the output cavity must be tightly bunched, with minimum
velocity spread, and the fraction of electrons in the inter-​bunch region must be
as small as possible. The techniques used to reduce this fraction in high efficiency
tubes are discussed in Section 13.4.2.
• The small-​signal gap coupling factor of the output cavity must be increased to
the point where reflection of electrons by potential depression in the spent beam
is the limiting factor.
The theoretical limit may not be achieved in practice because of:
• The need to avoid voltage breakdown in the output gap.
• The effects of radial motion of electrons.
• Design compromises, especially in broad-​band tubes.
• Losses in the output cavity, which increase as the frequency increases.

13.3.7 Electron Collection

In some cases the overall efficiency of a klystron can be increased by collector
depression. The technique is not suited to high-​power tubes because of the diffi-
culty of cooling multiple high-​voltage electrodes (see Chapter 10). The distribution
of the energies of the electrons in the spent beam can be computed using a disc
model. However, it has been found that useful results can be obtained for a broad-​
band tube using an approximate method (see Worksheet 13.5). This can be illus-
trated by applying it to the tube described in [45] which had a saturated efficiency
( ηsat ) of 51%. For such a tube the beam is not strongly bunched at the output cavity
so it is reasonable to model the current waveform at saturation using (11.175) with
n = 2. The waveform at other drive levels is assumed to be

I (θ, X )
= 1 + 1.33X cos θ + 0.33X 2 cos 2θ (13.52)
I0
 Large-Signal Behaviour of Klystrons 489

where θ = ωt and X is the ratio of the voltage at the input gap to the value at satur-
ation. Then the amplitude of the first harmonic current is

I1
= 1.33X . (13.53)
I0
The normalised dimensions of the output gap are taken to be   βe a = 1, βe b = 0.6, βe g = 1
and the effective voltage of the output gap at saturation is given by

M eff Vg I0
= 2 ηsat = 0.77. (13.54)
Va I1
The normalised energy of the slowest electrons is Vmin Va = 0.23 and the effective
gap coupling factor at saturation is found, from (11.182), to be M eff = 0.6. Then the
normalised resistance of the output load is

I 0 Vg I0
RL = ⋅ , (13.55)
Va Va M eff I1

where losses in the output cavity have been neglected. For output levels below satu-
ration the effective gap voltage Veff ( X ) can be computed as a function of X by itera-
tion. Then the energy of electrons in the spent beam as a function of the phase of
the output gap voltage is given by

Vs (θ, X ) = Va − Veff ( X ) cos (θ ) . (13.56)

Since the current waveform is symmetrical the current collected by an electrode

whose potential is Vs (θ, X ) is

1 π
IC (θ, X ) = I (θ, X ) d θ. (13.57)
π ∫θ

The output backoff, in decibels, is given by

 M eff ( X ) 
OBO = 20 log  X  . (13.58)
 M eff (1) 

Figure 13.15 shows the spent-​beam curves computed at saturation and three dif-
ferent levels of output backoff. It should be noted that an appreciable fraction of
the electrons have been accelerated so that Vs > Va . This is typical of the proper-
ties of the spent beam in a klystron and it limits the improvement in efficiency
which can be achieved by collector depression. Despite the simplifications and
approximations used to calculate them, these curves are very similar to those
calculated using a disc model [45]. Note that their shapes are very different from
the idealised curve in Figure  10.9, and from the typical curve for a TWT in
Figure 14.16.
490 Klystrons

Figure 13.15:  Typical spent-​beam curves for a broad-​band klystron at saturation, and three
levels of output backoff.

The tube used in this example was designed for UHF TV transmission [45]. In
analogue TV service it was operated close to saturation for brief periods during
the synchronising pulses. For the greater part of the time the average power was
less than half of that at saturation [46, 47]. Figure 13.15 shows that the range of
energies in the spent beam is then relatively small so that an efficient multi-​stage
collector can be designed. In the two cases cited five-​stage depressed collectors were
used. The power required was then found to be less than half that without col-
lector depression leading to a substantial reduction in the operating costs of the
transmitter.
The other main use of klystrons with depressed collectors is in space where
high overall efficiency is essential. An experimental pulsed 5.3 GHz klystron
for a satellite borne synthetic aperture radar, having an RF efficiency of 40%,
achieved an overall efficiency of nearly 50% with a four-​stage collector (includ-
ing a spike at cathode potential) [48]. The spent-​beam curve of this tube is shown
in Figure  7.4 in [3]. Experiments with a klystron having an RF efficiency of
40–​50% gave an overall efficiency at saturation of around 70% with a ten-​stage
collector [49].

13.3.8 Terminal Characteristics

The performance of a klystron is appreciably affected by variations in the beam
voltage, signal frequency, and output match. Figure 13.7 shows that, when a klys-
tron is operated at or close to saturation, the output power is insensitive to varia-
tions of input power and, by extension, to variations of beam voltage. The effects
on the phase of the output signal are more serious because of the distance from the
input to the output. If the distance from the centre of the input gap to the centre
of the output gap is L then the phase difference between the input and the output
is approximately
 Klystron Design 491

ωL
φ≈ . (13.59)
u0
Thus, if the normal beam voltage is 90 kV, the tube length is 1.17 m, and the fre-
quency is 500 MHz, the sensitivity of phase to changes in the beam voltage is –​5.8
degrees per kV.
The small-​signal transfer function of a klystron contains both poles and zeroes
as shown by (13.20). If the zeroes are outside the working band of the tube then the
transfer function is essentially that of a multi-​resonator filter. The performance of the
klystron as a system element is governed by this. It is found that the filling times of
the cavities are much greater than the electron transit time along the tube. Thus the
transient behaviour is determined by the Q factors of the cavities. It can be determined
to good accuracy from the characteristics of the tube in the frequency domain [50,
51]. Under large-​signal conditions the non-​linearity of the tube leads to signal distor-
tion and the generation of inter-​modulation products [52, 53] (see Section 1.6.4).

13.4 Klystron Design

The design of any klystron is likely to start from a statement of requirements which
includes the frequency, output power, gain, and bandwidth. Although many tubes
are designed by scaling from existing devices it is instructive to consider the con-
ceptual design process. Many of the design options are common to all tubes and
the choices of parameters are usually constrained by physical factors. A  typical
sequence of design decisions is:
i) The DC beam power required is obtained from the RF output power using an
estimate of the tube efficiency based upon previous experience (see Figures 13.11
and 13.13). This estimate depends upon whether the primary design goal is high
efficiency or broad bandwidth and it can be revised in the light of subsequent
calculations. Thus

PDC = PRF ηrf . (13.60)

ii) The beam current and voltage are selected to give acceptable perveance, conver-
gence and cathode loading in the gun. The choice is influenced by whether high
gain or high efficiency is most important. If the perveance is K then
0.4
P 
Va =  DC  (13.61)
 K 
and

I 0 = PDC Va . (13.62)

The beam voltage must be low enough to avoid voltage breakdown in the gun
[41, 54].
492 Klystrons

iii) The normalised beam and tunnel radii ( γ e b and γ e a ) are chosen to achieve a
compromise between strong coupling to the gaps and the strength of magnetic
focusing field required for adequate beam stiffness. The beam filling factor ( b a )
is typically in the range 0.5 to 0.6 to achieve low beam interception and γ e a ≤ 1.0.
The tunnel radius must be small enough to ensure that the beam tunnel is
strongly cut-​off to avoid unwanted coupling between the cavities.
iv) The cathode radius is chosen to achieve a compromise between low cathode
loading (to achieve a long lifetime) and small area convergence (to get good
beam optics).
v) The choice of the number of cavities, their dimensions, and the drift lengths
depends upon the desired gain, bandwidth and efficiency. This is discussed fur-
ther below. It may be necessary to make a compromise between the drift lengths
and the cavity heights (which increase with increasing R Q).
The conceptual design process is best carried out using a spreadsheet, or other
computer aid, which enables the results of design options to be investigated rapidly.
The computer aids include simple small-​signal and large-​signal models of the kind
illustrated in Worksheets 13.1, 13.2, and 13.3. Figure 13.16 shows the technological
limits to the power of klystrons [55]. Further information about factors limiting the
design of klystrons is given in [41].
Once a promising conceptual design has been found it is possible to proceed to
detailed design in which every aspect is checked using computer models. Particular
aspects include the design of the cavities and couplers; the electron optics of the
gun, the focusing system and the collector; and the cooling system. The large-​signal
performance must be predicted more accurately by using a three-​ dimensional
large-​signal model.

Figure 13.16:  Approximate technology limitations on klystron performance: (1) size (CW

and pulsed); (2) Output window (CW); (3) output cavity power density and cyclical fatigue
(CW); and (4) radio-​frequency breakdown (pulsed)
(courtesy of H.P. Bohlen).
 Klystron Design 493

13.4.1 Broad-​band Klystrons

The design of broad-​band klystrons for telecommunications, television broadcast-
ing, and radar has been discussed by a number of authors [2, 13, 30, 31, 56]. The
general principles are to use stagger tuning of the cavities and a high perveance
beam to reduce the Q of the cavities by high beam loading. The cavities may also
be loaded internally, or externally, to reduce their Q factors. A small-​signal model
may be used to find the best tuning of the cavities for broad bandwidth, but a
large-​signal model must be used to find the effects of cavity tuning on efficiency.
The zeroes in the transfer function of the tube are arranged to give minimum rip-
ple in the gain of the tube, within its operating band, by suitable choice of the drift
lengths (see Section 13.2.3). This can be achieved by placing cavity poles close to
zeroes which fall within the band [3, 7, 17]. Good results have been obtained when
the frequencies of the cavities increase progressively down the tube. The drift length
preceding each cavity is adjusted so that the pole of one cavity is close to the zero
of the previous one [30, 31]. It is usual to tune one or more final bunching cavities
to frequencies lying above the operating band to increase the efficiency of the tube.
Because the zeroes attributable to the individual sections of the tube generally lie
above the cavity resonant frequencies, the gain usually falls more rapidly above the
band than below it (see Figure 13.6). Some examples of tuning patterns are shown
in Table 13.1.
The bandwidth of the output cavity can be increased by coupling it to the out-
put waveguide by a filter, such as a second resonant cavity (see Section 3.2.4), or
by using an extended interaction cavity [13, 47]. Extended interaction cavities may
also be used elsewhere in place of conventional cavities to produce an extended
interaction klystron (EIK) [36]. Under some circumstances the current modulation

Table 13.1:  Examples of cavity tuning patterns for

broad-​band klystrons (relative frequencies)

Cavity [13] [31] [30]

number
1 0.975 0.956 0.96
2 0.955 0.960 0.95
3 1.015 0.985 1.01
4 1.045 0.985 1.04
5 1.050 1.02 1.06
6 1.092 1.02 1.08
7 1.000 1.047 1.10
8 –​ 1.05 1.00
9 –​ 1.05 –​
10 –​ 1.067 –​
11 –​ 1.00 –​
494 Klystrons

Figure 13.17:  Trajectories in a high-​efficiency klystron with long drift lengths.

excited on the beam by one cavity can deliver power to the same cavity so that
the beam loading conductance becomes negative and the cavity self-​oscillates [57].
These are known as monotron oscillations and are the basis of monotron and
extended interaction oscillators [58–​61]. In the ultra-​wide bandwidth Clustered-​
CavityTM klystron the individual cavities are replaced by closely-​spaced clusters of
cavities [17, 62].

13.4.2 High-​Efficiency Klystrons

High-​efficiency klystrons have been developed for powering particle accelerators
and for plasma heating in nuclear fusion experiments [7, 35, 38, 39, 43, 63–​65]. To
achieve high efficiency the fundamental RF beam current entering the output cavity
and the effective gap coupling factor should both be as high as possible (see Section
13.3.6). The initial bunching process should gather as many electrons as possible
into a phase range of 180° while avoiding crossing trajectories. The bunches can
then be compressed using several stages, if necessary (see Figure 13.9).
Two techniques for achieving high efficiency were discussed in [23]. The paper
does not give full details of the parameters of the tubes considered but the calcula-
tions can be made using Worksheet 13.4 if some assumptions are made about the
missing information. Figure 13.17 shows three stage bunching of a 1.0 μPerv beam.
The first stage creates bunches with I1 I 0 = 0.86 at z λ e = 2.28. It can be seen that
the inner trajectories start to diverge at that plane whilst the outer ones continue
to converge. The phase focusing effect of the second cavity is greatest at ±90° with
respect to the bunch centre. The best bunch compression is achieved by placing the
second cavity after the plane for optimum bunching so that more of the electrons
are captured into the bunch. The second stage of compression gives I1 I 0 = 1.3 at
z λ e = 5.4. The third cavity, which is also placed a little beyond the plane of optimal
bunching, gives I1 I 0 = 1.8 at z λ e = 6.6. The growth of the first and second harmon-
ics of the RF beam current along the tube is shown in Figure 13.18. Figures 13.17
and 13.18 are similar to Figure 4 in [23]. The chief disadvantage of this method
of achieving high efficiency is that the drift lengths are long and increase as the
 Klystron Design 495

Figure 13.18:  Growth in the first and second harmonic currents on the beam in a high-​
efficiency klystron with long drift lengths.

perveance decreases. The bandwidth of high-​efficiency tubes designed in this way is

comparable with that of conventional klystrons. An example of a tube of this type
is discussed in Section 13.4.3. Further information can be found in [25, 32, 65, 66].
At an early date in the development of klystrons it was observed that much bet-
ter initial bunching could be achieved if the gap field had a saw tooth waveform
[67]. Although this cannot be achieved in practice, a useful approximation can be
achieved by using one, or more, cavities tuned to the second harmonic of the signal
frequency [23]. Such a cavity can be thought of as compensating for the astigma-
tism in the preceding fundamental mode cavity. Figure 13.19 shows the effect of a
second harmonic cavity in a 0.5 μPerv tube, based on Lien’s data. The first cavity
acting alone produces a bunch with I1 I 0 = 0.9 at z λ e = 2.9. A cavity tuned close
to the second harmonic frequency is placed after the first cavity at z λ e = 1.03. The
tuning of the cavity is adjusted so that the outlying electrons are gathered into the
bunch. Then I1 I 0 = 0.95 at z λ e = 3.2 while the amplitude of I 2 I 0 decreases. At
this plane about 90% of the electrons are contained in a phase range of ±90° with
respect to the bunch centre, as shown in Figure 13.20. The final bunching stage then
comprises two cavities. The first, placed at z λ e = 1.8, increases the bunching to
give I1 I 0 = 1.4 at z λ e = 3.6. The second at z λ e = 3.2 produces a final bunch with
I1 I 0 = 1.8 at z λ e = 3.9. The growth in the first and second harmonics of the beam
current is shown in Figure 13.21. Figures 13.19 and 13.21 show the same general
features as ­figures 2 and 3 in [23]. Comparison with Figure 13.17 shows that the use
of a harmonic cavity greatly reduces the total drift length required to produce tight
electron bunches. This technique is usually used in high efficiency tubes. Its chief
disadvantage is that it is inherently narrow band.
Klystrons which have been developed specifically for use in accelerators are com-
monly known as super-​power klystrons. Tables 13.2 and 13.3 summarise the state
of the art for these tubes. The beam voltage is limited by the need to avoid voltage
breakdown in the electron gun. It can be seen from the tables that the typical beam
voltages are higher for pulsed tubes than for continuous wave tubes because the
breakdown voltage is higher for short pulses than for steady voltages. The beam
496 Klystrons

Figure 13.19:  Trajectories in a high efficiency klystron with a second harmonic cavity.

Figure 13.20:  Bunching at z λ e = 3 in a high efficiency klystron with a second harmonic

cavity: (a) arrival phase plotted against initial phase, and (b) current waveform.

Figure 13.21:  Growth in the first and second harmonic currents on the beam in a high
efficiency klystron with a second harmonic cavity.
 Klystron Design 497

Table 13.2:  Characteristics of typical continuous wave super-​power klystrons

Tube TH 2089 VKP-​7952 TH 2103C*

Manufacturer Thales CPI Thales

Frequency 352 700 3700 MHz
Beam voltage 100 95 73 kV
Beam current 20 21 22 A
RF output power 1.1 1.0 0.7 MW
Gain 40 40 50 dB
Efficiency 65 65 44 %

*
  This tube was developed for heating plasmas in nuclear fusion experiments

Table 13.3:  Characteristics of typical pulsed super-​power klystrons

Tube [39] [68] [69]

Frequency 2.87 3.0 11.4 GHz

Pulse length 1.0 1.0 1.6 μsec
Beam voltage 475 610 506 kV
Beam current 620 780 296 A
RF output power 150 213 75 MW
Gain 59 58 60 dB
Efficiency 51 44 50 %

current is limited by the current density which is available at the cathode, and by
the area of the cathode which decreases with frequency. The saturation current
density of thermionic cathodes is greater for short (microsecond) pulses than for
DC operation.

13.4.3 Case Study: The SLAC 5045 Klystron

The design of the SLAC 5045 50 MW klystron has been described in detail and
provides a convenient case study [35, 70]. The requirement was for a pulsed
tube at 2856 MHz with 6 μs pulses at a pulse repetition frequency of 180 Hz.
From Figure 13.13(b) it can be seen that an efficiency of at least 45% should be
attainable at a perveance of 2.0 µA V −1.5 . Although a higher efficiency could be
obtained by reducing the perveance, a higher beam voltage would be required,
leading to increased problems with voltage breakdown. The beam voltage could
be reduced by choosing a higher perveance but the efficiency would be reduced
and it would be more difficult to produce a well-​collimated beam. The DC beam
power required is 100 MW giving a voltage of 315 kV and current of 350 A for
the perveance chosen (see (13.60) to (13.62)). If the beam filling factor b a = 2 3
498 Klystrons

then the effective beam voltage is reduced to 290 kV by space-​charge potential

depression (see (7.8)).
The normalised beam and tunnel radii chosen were γ e b = 0.5 and γ e a = 0.75
so that b = 10 mm and a = 15 mm. Using these beam parameters with Worksheet
13.4, as described in Section 13.3.6, the theoretical maximum electronic efficiency
was calculated to be 52%. This confirms that the limiting efficiency increases as
the beam voltage increases as a result of relativistic effects. The plasma frequency
including the effects of relativity is 790 MHz and the Brillouin field is 0.63 T. From
Figure 13.10(a) we see that the maximum RF current at the output cavity is 1.5I 0.
From Table 11.2 and Figure 11.30 it is estimated that the peak current in the bunch
is 3.2 I 0. Figure 7.5 shows that the stiffness of the beam should be adequate if the
solenoid field is greater than 1.75 times the Brillouin field. The power consumed by
the solenoid produces a significant reduction in the overall efficiency of a pulsed
tube. We therefore choose the minimum field ratio giving a solenoid field of 0.11
T.  The power required by the solenoid has been eliminated in some more recent
tubes by using PPM focusing [43]. The tube has a Pierce electron gun of conven-
tional design with a dispenser cathode and an average loading of 5.6 A cm −2 .
The product of the potential difference between the electrodes of the gun and the
maximum electric field is 7.1 × 103 kV 2 mm −1 which is close to the maximum for
this pulse length given in [54]. Careful selection of the electrode materials together
with extensive polishing and careful conditioning resulted in satisfactory rates of
voltage breakdown. Because the beam is strongly relativistic it is important to use
γ e = βe2 − ω 2 c 2 rather than βe when calculating the plasma frequency reduction
factor. The reduced plasma frequency is 210 MHz and the reduced plasma wave-
length is 1.08 m.
The details of the interaction structure of the tube are shown in Table 13.4. Most
of the data is taken from [35] but the cavity frequencies are those given in [70].
The tube had been redesigned after the first paper to reduce its gain and increase
its stability, but it is not clear whether the changes were restricted to retuning
of the cavities. The coupling factors of the gaps were computed as described in
Section 11.2.4 using the field profile given in (3.90) with kg = 4.
This tube was modelled using Worksheet 13.3 with beam voltage 315 kV, beam
current 354 A, for comparison with the data given in [70]. The cavity voltages in
the first four cavities were obtained using the small-​signal model in Worksheet 13.1.
The results obtained are shown in Table 13.5.
Figures 13.22 and 13.23 show the electron trajectories computed and the growth
in the RF current along the tube. The first three drift angles are approximately
βq z ≈ 18° so that the first four cavities and the drift lengths which follow them form
the initial bunching section of the tube as described in Section 13.3.2. The long drift
length after the fourth cavity serves to increase the bunching as described in Section
13.4.2. The effective gap coupling factor calculated using (11.180) is 0.4. This is
smaller than the optimum figure given in Figure 11.32 and suggests that the exter-
nal Q of the output cavity may not be optimal. The differences between the results
of measurement and calculation may be because radial motion of the electrons has
 Other Klystrons 499

Table 13.4:  Data for the SLAC 5045 klystron

Cavity No. Cavity frequency R /​Q Q0 Qext Gap M Drift

(MHz) (Ohms) βe g z λq

1 2860 80 2000 200 0.53 0.879 0.052

2 2870 75 2000 0.56 0.878 0.051
3 2890 87 2000 0.64 0.873 0.051
4 2910 96 2000 0.86 0.856 0.257
5 2970 96 2000 0.90 0.852 0.102
6 2853 85 2000 21 1.26 0.814

Table 13.5:  SLAC 5045 klystron: comparison between measured

and calculated and results at saturation

Measured [70] Calculation

Power input (W) 500 50

Power output (MW) 50 47
Efficiency (%) 45 42
Gain (dB) 50 60

been ignored or because the current density in the beam is not uniform. Overall the
results confirm the validity of Worksheet 13.3 as a means of studying the design of
klystrons.
A tube of this kind can be designed by working forwards through the tube adjust-
ing the position and properties of each cavity in turn to optimise the bunching fig-
ure of merit for the next stage. It is difficult to achieve an optimum design by hand
because of the large number of parameters which may be adjusted. If a sufficiently
reliable large signal computer model is available the overall performance can be
optimised using a search algorithm [71].

13.5 Other Klystrons

A number of variants of the klystron are of importance for high-​power, high-​

efficiency, high-​frequency and broad-​band operation. The multiple-​beam klystron
and the sheet beam klystron are discussed briefly below and further information
can be found in [7]. The reflex klystron oscillator in which the electron beam passes
twice through a single cavity, having been reflected by a negative electrode, is of
renewed interest as a possible source of power at sub-​millimetre wavelengths [72,
73]. The electro-​statically focused klystron has the attraction that it does not require
a heavy focusing magnet, but the need to combine electrostatic focusing with the
500 Klystrons

Figure 13.22:  Computed electron trajectories in the SLAC 5045 klystron.

Figure 13.23:  Computed harmonic current growth in the SLAC 5045 klystron.

interaction structure limits the value of this concept [3, 74]. The relativistic klys-
tron, in which powers in the gigawatt region are generated by a very intense, high
energy, electron beam, lies outside the scope of this book [75].

13.5.1 Multiple-​Beam Klystrons

We have seen that the efficiency of a klystron is determined by the perveance of the
electron beam so that, to get high efficiency, it is necessary to use a high-​voltage,
low-​current beam. The use of high voltages produces problems with voltage break-
down and it is therefore difficult to obtain very high power with high efficiency. One
solution to this problem is to use several electron beams within the same vacuum
envelope. A  klystron designed in this way is known as a multiple-​beam klystron
(MBK). The individual beams have low perveance to give high efficiency whilst the
output power is determined by the total power in all the beams.
The principle of the MBK has been known for many years [76] but, until recently,
the only such tubes constructed were for military applications in the former Soviet
Union [77–​79]. The first MBK designed specifically for use in particle accelerators
was the Thales type TH1801 having seven beams whose performance is shown in
 Other Klystrons 501

Table 13.6:  Characteristics of a multiple beam klystron

Type TH 1801

Frequency 1300 MHz

Beam voltage 115 kV
Beam current 133 A
Number of beams 7
Power 9.8 MW
Pulse length 1.5 msec
Efficiency 64 %
Gain 47 dB

Figure 13.24:  Arrangement of a cavity for a multiple beam klystron.

Table 13.4 [80]. Figure 13.24 shows an example of a cavity similar to those used

in that tube. Note that the need to accommodate the beams constrains the shape
of the cavity. The gap coupling factor of the central gap differs from those of the
other gaps because the fields are not identical. The drift tubes may be provided
with noses to increase the fields in the interaction gaps though this can make the
cavities more difficult to construct. Coaxial cavities may be used in place of simple
502 Klystrons

re-​entrant cavities [81, 82]. One of the main design challenges for an MBK is in the
electron optics, since only a slight non-​uniformity in the magnetic field can cause
the beams to be deflected sideways. Further information about multiple-​beam klys-
trons is given in [7].

13.5.2 Sheet Beam Klystrons

Multiple-​beam klystrons are not well-​suited to high frequencies because of the dif-
ficult electron optics. An alternative way of combining high power with low perve-
ance is to use a sheet electron beam [7, 83, 84]. This approach is particularly of
interest for devices operating at millimetre and sub-​millimetre wavelengths.

References

[1]‌ R. S. Symons, ‘Klystrons for UHF television’, Proceedings of the IEEE, vol. 70, pp.
1304–​1312, 1982.
[2]‌ G. O. Chalk and C. N. O’Loughlin, ‘Klystron amplifiers for television’, Electronic
Equipment News, February 1968.
[3]‌ M. J. Smith and G. Phillips, Power Klystrons Today. Taunton, England:  Research
Studies Press Ltd., 1995.
[4]‌ L. Sivan, Microwave Tube Transmitters. London: Chapman and Hall, 1994.
[5]‌ R. G. Carter, ‘R.F.  power generation’, in R. Bailey, ed., Proceedings of the CERN
Accelerator School ‘RF for Accelerators’, 8–​17 June 2010, Ebeltoft, Denmark. Geneva:
CERN, pp. 173–​207, 2011.
[6]‌ R. G. Carter, ‘Acceleration technologies for charged particles:  an introduction’,
Contemporary Physics, vol. 52, pp. 15–​41, 2011.
[7]‌ G. Caryotakis, ‘Klystrons’, in R. J. Barker et al., eds, Modern Microwave and Millimeter-​
Wave Power Electronics. Piscataway, NJ: IEEE Press, pp. 107–​170, 2005.
[8]‌ A. Staprans et al., ‘High-​power linear-​beam tubes’, Proceedings of the IEEE, vol. 61, pp.
299–​330, 1973.
[9]‌ G. Faillon et al., ‘Microwave tubes’, in J. A. Eichmeier and M. K. Thumm, eds, Vacuum
Electronics: Components and Devices. Berlin: Springer-​Verlag, pp. 1–​84, 2008.
[10] A. S. Gilmour, Jr., Klystrons, Traveling Wave Tubes, Magnetrons, Crossed-​ Field
Amplifiers and Gyrotrons. Norwood, MA: Artech House, 2011.
[11] J. R. M. Vaughan, ‘The input gap voltage of a klystron’, IEEE Transactions on Electron
Devices, vol. 32, pp. 2510–​2511, 1985.
[12] B. E. Carlsten and P. Ferguson, ‘Numerical determination of the matching conditions
and drive characteristics for a klystron input cavity with beam’, IEEE Transactions on
Electron Devices, vol. 44, pp. 894–​900, 1997.
[13] G. Faillon, ‘Klystrons de puissance à large bande’, Revue Technique Thomson-​CSF, vol.
8, pp. 289–​331, June 1976.
[14] B. E. Carlsten et  al., ‘Accuracy of the equivalent circuit model using a fixed beam
impedance for klystron gain cavities’, IEEE Transactions on Plasma Science, vol. 26, pp.
1745–​1749, 1998.
 References 503

[15] R. S. Symons and J. R. M. Vaughan, ‘Modification of klystron beam loading by ini-
tial velocity modulation of the beam’, in International Electron Devices Meeting, pp.
885–​888, 1990.
[16] G. R. White, ‘Small-​ signal theory of multicavity klystrons’, IRE Transactions on
Electron Devices, vol. 6, pp. 449–​457, 1959.
[17] R. S. Symons and R. M. Vaughan, ‘The linear theory of the Clustered-​CavityTM
Klystron’, IEEE Transactions on Plasma Science, vol. 22, pp. 713–​718, 1994.
[18] T. M. Antonsen, Jr. et al., ‘Advances in modeling and simulation of vacuum electronic
devices’, Proceedings of the IEEE, vol. 87, pp. 804–​839, 1999.
[19] R. G. Carter, ‘Computer modelling of microwave tubes  –​a review’, in 2nd IEEE
International Vacuum Electronics Conference 2001, Noordwijk, Netherlands, pp.
393–​396, 2001.
[20] L. D. Ludeking et  al., ‘Computational modeling’, in R. J. Barker et  al., eds, Modern
Microwave and Millimetre-​Wave Power Electronics. Piscataway NJ:  IEEE Press, pp.
507–​585, 2005.
[21] T. G. Mihran, ‘The effect of space charge on bunching in a two-​cavity klystron’, IRE
Transactions on Electron Devices, vol. 6, pp. 54–​64, 1959.
[22] H. G. Kosmahl and L. U. Albers, ‘Three-​dimensional evaluation of energy extraction in
output cavities of klystron amplifiers’, IEEE Transactions on Electron Devices, vol. 20,
pp. 883–​890, 1973.
[23] E. L. Lien, ‘High-​efficiency klystron amplifiers’, in Eighth International Conference on
Microwaves and Optical Generation and Amplification, Amsterdam, The Netherlands,
pp. 11–​21 to 11–​27, 1970.
[24] T. G. Mihran, ‘The effect of drift length, beam radius, and perveance on klystron
power conversion efficiency’, IEEE Transactions on Electron Devices, vol. 14, pp.
201–​206, 1967.
[25] P. J. Tallerico, ‘Design considerations for the high-​power multicavity klystron’, IEEE
Transactions on Electron Devices, vol. 18, pp. 374–​382, 1971.
[26] T. G. Mihran et al., ‘Electron bunching and output gap interaction in broad-​band klys-
trons’, IEEE Transactions on Electron Devices, vol. 19, pp. 1011–​1017, 1972.
[27] R. F. Koontz et al., ‘Anomalous electron loading in SLAC 5045 klystron and relativistic
klystron input cavities’, in IEEE Particle Accelerator Conference, pp. 159–​161, 1989.
[28] J. R. M. Vaughan, ‘Multipactor’, IEEE Transactions on Electron Devices, vol. 35, pp.
1172–​1180, 1988.
[29] C. Hill and R. G. Carter, ‘Investigation of possible multipactor discharge in a klystron
input cavity’, in 2006 IEEE International Vacuum Electronics Conference held jointly
with 2006 IEEE International Vacuum Electron Sources, pp. 81–82,  2006.
[30] R. L. Metivier, ‘Broadband klystrons for multimegawatt radars’, Microwave Journal,
vol. 14, pp. 29–​32, April 1971.
[31] G. Faillon, ‘A 200 kilowatts S band klystron with TWT bandwidth capability’, in
International Electron Devices Meeting, pp. 283–​286, 1973.
[32] A. Y. Baikov et al., ‘Toward high-​power klystrons with RF power conversion efficiency on
the order of 90%’, IEEE Transactions on Electron Devices, vol. 62, pp. 3406–​3412, 2015.
[33] J. R. Hechtel, ‘The effect of potential beam energy on the performance of linear beam
devices’, IEEE Transactions on Electron Devices, vol. 17, pp. 999–​1009, 1970.
[34] W. Peter et al., ‘Criteria for vacuum breakdown in RF cavities’, IEEE Transactions on
Nuclear Science, vol. 30, pp. 3454–​3456, 1983.
504 Klystrons

[35] T. Lee et al., ‘A fifty megawatt klystron for the Stanford Linear Collider’, in International
Electron Devices Meeting, pp. 144–​147, 1983.
[36] M. Chodorow and T. Wessel-Berg, ‘A high-​efficiency klystron with distributed interac-
tion’, IRE Transactions on Electron Devices, vol. 8, pp. 44–​55, 1961.
[37] T. G. Lee, ‘Multiple extraction cavities for high-​power klystrons’, IEEE Transactions on
Electron Devices, vol. 40, pp. 1329–​1334, 1993.
[38] W. Luebke and G. Caryotakis, ‘Development of a one megawatt CW Klystron’,
Microwave Journal, vol. 9, pp. 43–​47, 1966.
[39] T. Lee et  al., ‘The design and performance of a 150-​MW klystron at S band’, IEEE
Transactions on Plasma Science, vol. 13, pp. 545–​552, 1985.
[40] W. Fowkes and E. Wu, ‘Multimode harmonic power output measurement of SLAC
high power klystrons’, SLAC-​PUB-​3009, November 1982.
[41] R. S. Symons, ‘Scaling laws and power limits for klystrons’, in International Electron
Devices Meeting, pp. 156–​159, 1986.
[42] A. Beunas et al., ‘A high power long pulse high efficiency multi beam klystron’, in 5th
Modulator-​Klystron Workshop for Future Linear Colliders, 2001.
[43] R. M. Phillips and D. W. Sprehn, ‘High-​power klystrons for the Next Linear Collider’,
Proceedings of the IEEE, vol. 87, pp. 738–​751, 1999.
[44] E. Jensen and I. Syratchev, ‘CLIC 50 MW L-​ band multi-​beam klystron’, in AIP
Conference Proceedings, p. 90, 2006.
[45] E. W. McCune, ‘A UHF-​TV klystron using multistage depressed collector technology’,
in IEEE International Electron Devices Meeting, pp. 160–​163, 1986.
[46] E. W. McCune, ‘Klystron performance using a multistage depressed collector’, in 1987
International Electron Devices Meeting, pp. 157–​159, 1987.
[47] W. Schmidt, ‘Multi-​stage-​depressed-​collector klystron for high-​efficiency UHF trans-
mitter in high and medium power range’, in International Broadcasting Convention, pp.
44–​45, 1990.
[48] D. Perring, ‘A design for a pulsed, PPM focused depressed collector klystron for space
(SAR) applications’, in Vakuumelektronik und Displays, Garmisch-​Partenkirchen,
Germany, 1989.
[49] W. Neugebauer and T. G. Mihran, ‘A ten-​stage electrostatic depressed collector for
improving klystron efficiency’, IEEE Transactions on Electron Devices, vol. 19, pp.
111–​121, 1972.
[50] R. G. Carter and R. O. Jenkins, ‘Studies of the transient response of a klystron’, in
IEEE International Vacuum Electronics Conference, Monterey, USA, pp. 312–​313,
2008.
[51] T. L. Lavine et  al., ‘Transient analysis of multicavity klystrons’, in IEEE Particle
Accelerator Conference, pp. 126–​128, 1989.
[52] K. Yamamoto, ‘Nonlinearities of multicavity klystron amplifiers’, IEEE Transactions
on Electron Devices, vol. 24, pp. 648–​654, 1977.
[53] J. G. Wohlbier and J. H. Booske, ‘Nonlinear space charge wave theory of distortion in a
klystron’, IEEE Transactions on Electron Devices, vol. 52, pp. 734–​741, May 2005.
[54] G. Faillon, ‘Technical and industrial overview of RF and microwave tubes for fusion’,
Fusion Engineering and Design, vol. 46, pp. 371–​381, 1999.
[55] H. Bohlen, ‘Vacuum electronic device limitations for high-​power RF sources’, in CWRF
2008, CERN, Geneva, 2008.
 References 505

[56] T. Kageyama, ‘A large-​signal analysis of broad-​band klystrons with design applica-

tions’, IEEE Transactions on Electron Devices, vol. ED-​24, pp. 3–​12, 1977.
[57] G. M. Branch, Jr., ‘Electron beam coupling in interaction gaps of cylindrical sym-
metry’, IRE Transactions on Electron Devices, vol. 8, pp. 193–​207, 1961.
[58] J. J. Barroso, ‘Design facts in the axial monotron’, IEEE Transactions on Plasma Science,
vol. 28, pp. 652–​656, 2000.
[59] J. J. Barroso, ‘Electron bunching in split-​cavity monotrons’, IEEE Transactions on
Electron Devices, vol. 56, pp. 2150–​2154, 2009.
[60] B. E. Carlsten and W. B. Haynes, ‘Discrete monotron oscillator’, IEEE Transactions on
Plasma Science, vol. 24, pp. 1249–​1258, 1996.
[61] W. R. Day and J. A. Noland, ‘The millimeter-​wave extended interaction oscillator’,
Proceedings of the IEEE, vol. 54, pp. 539–​543, 1966.
[62] R. S. Symons et  al., ‘An experimental Clustered-​CavityTM klystron’, in International
Electron Devices Meeting, pp. 153–​156, 1987.
[63] E. G. Schweppe et  al., ‘Design and results of a 1.3 MW CW klystron for LEP’, in
Particle Accelerator Conference, vol. 2, pp. 1178–​1180, 1993.
[64] C. O’Loughlin et  al., ‘Progress in high power klystron manufacturing at EEV’, in
European Particle Accelerator Conference, pp. 1906–​1908, 1994.
[65] T. G. Mihran et al., ‘Design and demonstration of a klystron with 62 percent efficiency’,
IEEE Transactions on Electron Devices, vol. 18, pp. 124–​133, 1971.
[66] V. A. Kochetova et al., ‘Criteria for optimum bunching and the shape of the optimum
electron bunch in a drift klystron’, Radio Engineering and Electron Physics, vol. 26, pp.
85–​91, 1981.
[67] A. H. W. Beck, Space-​Charge Waves and Slow Electromagnetic Waves . London: Pergamon
Press, 1958.
[68] S. Choroba et al., ‘Performance of an S-​band klystron at an output power of 200MW’,
in XIX International Linac Conference, Chicago, IL, pp. 917–​919, 1998.
[69] D. Sprehn et al., ‘Current status of the next linear collider X-​band klystron develop-
ment program’, in EPAC 2004, Lucerne, Switzerland, pp. 1090–​1092, 2004.
[70] G. T. Konrad, ‘High power RF klystrons for linear accelerators’, SLAC-PUB-3324, 1984.
[71] C. J. Lingwood et  al., ‘Automatic optimization of a klystron interaction structure’,
IEEE Transactions on Electron Devices, vol. 60, pp. 2671–​2676, 2013.
[72] J. W. Gewartowski and H. A. Watson, Principles of Electron Tubes. Princeton, NJ: D. van
Nostrand, 1965.
[73] J. R. Fletcher et  al., ‘Design considerations for submillimeter-​wave reflex klystrons’,
IEEE Transactions on Microwave Theory and Techniques, vol. 52, pp. 2344–​2351, 2004.
[74] B. Vancil et  al., ‘A medium power electrostatically focused multiple-​beam klystron’,
IEEE Transactions on Electron Devices, vol. 54, pp. 2582–​2588, 2007.
[75] M. V. Fazio et al., ‘A 500 MW, 1 μs pulse length, high current relativistic klystron’, IEEE
Transactions on Plasma Science, vol. 22, pp. 740–​749, 1994.
[76] M. R. Boyd et al., ‘The multiple-​beam klystron’, IRE Transactions on Electron Devices,
vol. 9, pp. 247–​252, 1962.
[77] A. N. Korolyov et al., ‘Multiple-​beam klystron amplifiers: performance parameters and
development trends’, IEEE Transactions on Plasma Science, vol. 32, pp. 1109–​1118, 2004.
[78] E. A. Gelvich et al., ‘The new generation of high-​power multiple-​beam klystrons’, IEEE
Transactions on Microwave Theory and Techniques, vol. 41, pp. 15–​19, 1993.
506 Klystrons

[79] A. S. Pobedonostev et  al., ‘Multiple-​ beam microwave tubes’, in IEEE MTT-​S
International Microwave Symposium Digest, vol. 2, pp. 1131–​1134, 1993.
[80] A. Beunas and G. Faillon, ‘10 MW/​1.5ms, L-​band multi-​beam klystron’, in Displays and
Vacuum Electronics, Garmisch-​Partenkirchen, Germany, pp. 257–​262, 1998.
[81] T. Habermann et al., ‘High-​power high-​efficiency L-​band multiple-​beam klystron devel-
opment at CPI’, IEEE Transactions on Plasma Science, vol. 38, pp. 1264–​1269, 2010.
[82] R. Zhang and Y. Wang, ‘Design of the RF circuit for a coaxial cavity high-​power multiple-​
beam klystron’, IEEE Transactions on Electron Devices, vol. 61, pp. 909–​914, 2014.
[83] G. Caryotakis et al., ‘Design of a 11.4 GHz, 150-​MW, sheet beam, PPM-​focused klys-
tron’, High Energy Density and High Power RF, vol. 691, pp. 22–​33, 2003.
[84] J. Pasour et al., ‘Sheet beam extended interaction klystron (EIK) in W band’, in IEEE
14th International Vacuum Electronics Conference, Paris, France, pp. 1–​2, 2013.
14 Travelling-​Wave Tubes

14.1 Introduction

The output of a klystron or an IOT incorporates a resonant circuit so that it is

effectively operating as a class B or C amplifier. The instantaneous bandwidth is
therefore 10% or less. In order to get greater bandwidth the electron beam must
interact with a travelling wave supported by a non-​resonant structure. We saw in
Section 11.5 that gain occurs when the slow space-​charge wave on a beam interacts
with the forward wave of a slow-​wave structure. A  device which operates in this
way is known as a travelling-​wave tube (TWT). TWTs can be divided into two main
classes: those in which the interaction between the beam and the slow-​wave struc-
ture can be regarded as continuous (e.g. helix TWTs, see Section 14.2.1); and those
in which it is discontinuous (e.g. coupled-​cavity TWTs, see Section 14.2.2). However,
TWTs of all kinds have much in common so that it is convenient to discuss them
together, while drawing attention to the differences where those are important. The
next section discusses the small-​signal theory of both types of TWT. This is fol-
lowed by a review of the properties of TWTs under large signal conditions and of
various aspects of their design. The final section considers the design of TWTs for
different applications requiring, respectively, wide bandwidth, high efficiency, high
frequency, or high power.
The slow space-​charge wave carries negative kinetic power and its amplitude
increases as power is transferred to the slow-​wave structure along the length of the
tube. If the match at the output of the tube is not perfect then the reflected signal
passes back down the slow-​wave structure, attenuated by the cold loss of the struc-
ture. A further reflection occurs if the match at the input of the structure is not per-
fect, leading to the possibility of feedback oscillations. The maximum permissible
gain in a section of a TWT can be found by considering the equivalent network for
a single section shown in Figure 14.1. The RF terminations are matched ‘cold’ (in
the absence of the beam) to achieve around 20 dB return loss. The matches in a ‘hot’
tube are affected by the beam loading [1]. If the forward gain of the section is A,
the voltage reflection coefficients of the terminations are ρ, and the cold loss of the
structure is L then output voltage is

Vout = A (Vin ± Vout ρ2 L ) (14.1)

508 Travelling-Wave Tubes

Figure 14.1:  Equivalent network for a single section of a TWT.

in which the two signs represent the extreme cases where the reflected signal is in
phase, and in anti-​phase, with the input signal. Equation (14.1) can be re-​written as

A
Vout = Vin . (14.2)
1 ± Aρ2 L
If the reflection is in anti-​phase then (14.2) shows that it is possible for the loop gain
to be greater than unity so that the tube will self-​oscillate. For smaller loop gain the
variation in the electrical length of the section with frequency means that a ripple
is superimposed upon the gain curve of the tube as the phase of the reflected signal
varies [1]. This discussion can be illustrated by some typical figures. Suppose that
ρ = 0.1 ( −20 dB), L = 0.5 ( −6 dB) and the gain ripple should not exceed 1 dB peak
to peak. Then Aρ2 L ≤ 0.058 so that the maximum voltage gain of the section is
11.5 ( 21 dB). Thus the gain of a single section of a TWT must normally be restricted
to 20–​23 dB to achieve acceptable gain ripple. To achieve greater overall gain the
tube must comprise two or more sections which are separated by breaks (known as
severs) in the slow-​wave structure, or by attenuators which block the passage of the
reflected wave.

14.1.1 Helix and Helix-​Derived TWTs

The properties of slow-​wave structures were discussed in Chapter 4 where we saw
that the helix slow-​wave structure provides the best approximation to a wide-​band
dispersion-​ free structure. TWTs incorporating helix slow-​ wave structures can
achieve bandwidths up to three octaves or more. The output power is limited to a
few hundred watts (depending on the frequency). Helix TWTs find applications in
communications systems (especially satellite systems), electronic countermeasures
and instrumentation [2–​5].
Figure 14.2 shows the general arrangement of a helix travelling-​wave tube body.
The electron beam is generated by a Pierce electron gun, it is PPM focused, and
the spent beam is commonly collected by a multi-​element depressed collector. The
helix is usually formed from tungsten or molybdenum tape, having a rectangular
cross-​section to ensure good thermal contact with the support rods. The rods them-
selves are made of a ceramic such as anisotropic boron nitride (ABN) or synthetic
diamond [6]. Beryllia is now usually avoided because of its toxicity. The helix may
be brazed to the support rods to reduce the thermal resistance [7]. The input and
 Introduction 509

Figure 14.2:  Arrangement of the body of a helix TWT. The Pierce electron gun, PPM
focusing system, and the collector are not shown.

output connections may be made using coaxial lines. The tube is normally divided
into two sections by an attenuator created by spraying lossy material onto the sup-
port rods [5, 8, 9]. Meander-​line structures (see Section 4.2.2) have been consid-
ered for use in low-​cost planar TWTs and for micro-​fabrication for millimetre wave
TWTs [10].
Because the helix is supported by dielectric rods, through which all heat dissi-
pated on it must be conducted, the mean power of a helix TWT is limited by heat
dissipation in the helix. In addition, the reduction in the coupling impedance as
the pitch of the helix is increased makes them unsuitable for use at voltages above
25 kV. Their performance can vary with time because of changes in the properties
of the attenuators and support rods [11, 12]. They are also susceptible to ‘Power
Holes’ which cause unexpected dips in output [13]. Higher powers can be obtained
at the expense of bandwidth by using the ring-​bar and ring-​loop structures (see
Section 4.4). Compared with the helix, ring-​bar and ring-​loop structures are disper-
sive, and therefore narrow band, but they are capable of higher voltage operation.
The hot bandwidth is typically around 10%. Because these structures are supported
by dielectric rods in the same way as a helix their mean power capability is similar
and they are therefore used chiefly for medium power pulsed applications. Examples
of such tubes are: a 200 kW pulsed ring-​bar TWT with bandwidth 1.2 to 1.4 GHz
and a 15kW pulsed ring-​loop TWT with bandwidth 3.1 to 3.5 GHz.

14.1.2 Coupled-​Cavity TWTs
TWTs able to generate high peak and mean powers employ all-​metal structures
such as the folded waveguide, and coupled-​cavity, slow-​wave structures described in
Sections 4.5 and 4.6. Figure 14.3 shows the arrangement of the body of a coupled-​
cavity TWT. The interaction is discontinuous, like that in a klystron, rather than
continuous as in a helix TWT. At each interaction gap the gap voltage adds to
the beam modulation, and the beam modulation induces a current in the cavity.
The cavity voltages are also linked via the forward and backward waves propagat-
ing on the structure. The structure is usually divided into two or more sections by
severs. Each section may be terminated internally by sever loads made from lossy
510 Travelling-Wave Tubes

Figure 14.3:  Arrangement of the body of a coupled-​cavity TWT.

Figure 14.4:  Power flows in one cavity of a coupled-​cavity TWT.

ceramic material as shown. Alternatively the sever power may be coupled out into
waveguides and dissipated in microwave loads either inside, or outside, the vacuum
envelope [14]. Coupled-​cavity TWTs are used for high power communication links
and frequency agile and pulse compression radar [2].
Figure 14.4 shows the power flows in a cavity which combine to determine the
gap voltage, namely the forward and backward wave powers P+ and P− , and the
power transferred from the beam Pb. The power transferred from the beam divides
equally in the forward and backward directions, and the associated signal voltages
are added as phasors to those of the forward and backward waves. The backward
wave assumes a more important role than in helix tubes because the backward wave
components do not cancel each other to the same extent. This effect is particularly
important at small values of the phase shift per cavity.

14.1.3 Energy Conversion in a TWT

The process by which energy is transferred from the electron beam to the slow-​
wave structure of a TWT is essentially the same for TWTs of all kinds. For sim-
plicity we will consider a helix TWT where the dispersion diagram for uncoupled
forward waves is shown schematically in Figure 14.5. We recall that the gain per
unit length is greatest at the synchronous point where the slow-​space-​charge wave
 Introduction 511

Figure 14.5:  Schematic dispersion diagram of a helix TWT.

is synchronous with the forward wave on the helix (see Figure 11.20). This means
that the DC beam velocity ( u0 ) represented by the slope of the dashed line must be
greater than the phase velocity of the wave on the helix (v p = ω 0 β0 ).
The behaviour of the electrons as they move through the wave on the helix can
be understood by considering a simple model in which space-​charge forces and the
spatial growth of the wave are ignored [15]. In the laboratory frame of reference the
electrostatic potential at the synchronous point is

V = µ cV0 cos ( ω 0 t − β0 z ) , (14.3)

where V0 is the amplitude of the wave on the helix. Now let us define a coordin-
ate system which is moving with the wave so that z ′ = z − v p t. In this system the
potential is

V ′ = µ cV0 cos (β0 z ′ ) . (14.4)

If an electron has velocity u in the laboratory frame then its velocity in the moving
frame is

u ′ = u − v p . (14.5)

Because the beam voltage in helix TWTs is less than 10 kV we can ignore relativistic
effects so that the kinetic energy of the electron in the moving frame is

1 1 1 1
m0 u ′ 2 = m0 (u − v p ) = m0 u 2 + m0 v 2p − m0 uv p . (14.6)
2
T′ =
2 2 2 2
The total energy of the electron in the moving frame is the sum of the kinetic and
potential energies

1 1
W′ = m0 u 2 + m0 v 2p − m0 uv p − eµ cV0 cos (β0 z ′ ) . (14.7)
2 2
512 Travelling-Wave Tubes

This function does not vary with time and it must therefore be a constant for the
electron. The constant can be determined by observing that u = u0 when V0 = 0
so that

u2 u u  u0 
− 2 − A cos (θ ) − 0  v − 2 = 0, (14.8)
v 2p vp vp  p 
where A = 2µ c eV0 m0 v 2p is the normalised amplitude of the wave and θ = β0 z ′ . The
parameter A can be expressed in terms of the beam voltage as

µ cV0 u02
A= ⋅ . (14.9)
Va v 2p

The solution of (14.8) is

2
u u 
= 1 ± A cos (θ ) +  0 − 1 . (14.10)
vp  vp 

The right-​hand side of (14.10) has real values for all θ when A ≤ A0 = (u0 v p − 1) .
2

For larger values of A, real values of u v p only exist for a limited range of values
of θ, indicating that the electron has been trapped by the wave. Figure 14.6 shows
how the normalised velocity of the electron varies with θ for a range of values of
A when u0 v p = 1.05 . This figure, in which the trajectory of an electron is defined
by the coordinates ( z ′, u ), is known as a phase space diagram. It is a useful tool for
understanding the operation of a TWT. From (14.9) we find that the normalised
effective helix voltage at the point where trapping commences is given by
2
µ cV0  v p 
= 1 −  . (14.11)
Va  u0 
This has the numerical value 0.0023 for the case shown in Figure 14.6. Thus, up
to the point at which the electrons begin to be trapped, the change in the mean
electron velocity is small, and the transfer of energy to the wave is a small correc-
tion. Further information, including the relativistic form of (14.7) which applies to
coupled-​cavity TWTs, is given in [15].
In a travelling-wave tube the electrons enter the field of the wave with uniform
velocity when the amplitude of the wave is small. They then follow a curve such as
Orbit 2 in Figure 14.6 and move through the wave from left to right, being acceler-
ated on the left-​hand side of the diagram and retarded on the right-​hand side. The
electrons tend to bunch together in the retarding phase of the wave. The phase of
the bunch relative to the wave remains constant while individual electrons enter and
leave the bunch, and the average kinetic energy of the electrons decreases. A helpful
analogy is given by the behaviour of a steady stream of cars moving along a hilly
road [5]. The energy lost by the electrons increases the power in the wave which
therefore increases in amplitude. This process continues until the amplitude of the
 Small-Signal Theory 513

Figure 14.6:  Phase space diagram for an electron for which u0 v p = 1.05 for various values of
the normalised potential A showing trapping when A ≥ A0 .

wave is great enough to reduce the relative velocity of some electrons to zero so
that they are trapped by the wave, as shown by Orbit 3. Any further increase in the
amplitude of the wave (A) causes the trapped electrons to move in a closed orbit,
such as that shown by Orbit 4, around which they move in a clockwise direction.
Figure 14.7 shows examples of phase space diagrams, for the same parameters,
which have been computed by integrating the trajectories of sample electrons through
the field of a helix when space-​charge effects are ignored (see Worksheet 14.2).
Figures 14.7(a) and (b) show the bunching of the electrons up to the point at which
trapping commences. Although many of the electrons are in the retarding phase
of the field ( 0 ≤ φ π ≤ 1) a substantial proportion remain in the accelerating phase
( −1 ≤ φ π ≤ 0). The mean velocity of the electrons remains close to the initial value.
Figure 14.7(c) shows the majority of electrons following an orbit such as Orbit 4 in
Figure 14.6, but a few lie outside it. The electrons which are trapped in the retard-
ing field have lost appreciable energy, and the amplitude of the wave increases sub-
stantially. The slowest electrons are just about to move back into the accelerating
phase of the field. Figure 14.7(d) corresponds to saturation of the output power.
The electrons in the accelerating phase have begun to gain energy from the field
but this is compensated by those in the retarding phase which have lost energy. The
total kinetic energy is a minimum but there are still a few electrons whose velocity
is greater than u0 .

14.2 Small-​Signal Theory

The basic theory of interaction between an electron beam and a slow-​wave struc-
ture was considered in Chapter 11 for both continuous and discrete interactions. It
was shown that the coupling between the normal modes of the slow-​wave structure
514 Travelling-Wave Tubes

Figure 14.7:  Phase space diagrams computed from the electron trajectories for the same
parameters as Figure 14.6 (u0 v p = 1.05).

and of the electron beam results in four waves, three in the forward direction and
one in the opposite direction. In order to model a complete TWT the boundary
conditions at the ends of each section of the tube are used to find the amplitudes
of the four waves.

14.2.1 Small-​Signal Theory of Helix TWTs

Figure 14.8 shows a section of a slow-​wave structure of length L which is continu-
ously coupled to an electron beam. The two ends of the circuit are connected to
external transmission lines having characteristic impedance Zc which is assumed to
be the same as the characteristic impedance of the helix in the absence of the beam.
The signal in the uncoupled regions can be expressed in terms of the voltages of
the forward and backward waves on the transmission line (V f ,Vb ), and the fast and
 Small-Signal Theory 515

Figure 14.8:  External connections to a section of a TWT.

slow space-​charge waves on the electron beam (V+ ,V− ). In the coupled region the
signal is represented by the voltages on the helix of the four coupled modes (Wn ).
At the interfaces between the coupled and uncoupled regions the helix voltage v,
the helix current i, the beam kinetic voltage V, and the RF beam current I. must be
continuous. To construct a small-​signal model of a section of a helix TWT we must
find the relationships between the mode voltages and the voltages and currents at
the interfaces.
Consider, first, the relationships between the voltage and current on the transmis-
sion line and the amplitudes of the uncoupled forward and backward waves. These
can conveniently be written as a matrix equation

v   1 1  V f 
i  = Y −Yc  Vb 
. (14.12)
 c
Similarly for the electron beam

V   1 1  V+ 
 I  = Y −Ye  V− 
. (14.13)
 e
Where Ye is the electronic admittance of the beam. These two equations can be
combined in the form

v   1 1 0 0  V f 
i  Yc −Yc 0 0  Vb 
 =   = [C1 ][U ] , (14.14)
V   0 0 1 1  V+ 
 
 I   0 0 Ye −Ye  V− 

where the column vector [U ] contains the amplitudes of the uncoupled

normal waves.
To find the relationships for the coupled modes we express each variable in terms
of the amplitudes of the normal mode voltages on the helix. Setting ∂ ∂z = − j β in
(11.130) gives

1
βV f = β0V f + βµ c Zc I (14.15)
2
516 Travelling-Wave Tubes

since I = (V+ − V− )Ye. Similarly, from (11.131) we obtain

1
βVb = −β0Vb − βµ c Zc I . (14.16)
2
Adding (14.15) to (14.16)

βv = β0 Zc i (14.17)
or

β
i= v. (14.18)
β0 Zc
Subtracting (14.16) from (14.15) gives

βZc i = β0 v + βµ c Zc I . (14.19)
Then, substituting for i from (14.18) and re-​arranging we obtain

β2 − β20
I= v. (14.20)
β0 βµ c Zc
In the same way (11.126) yields

1
βV+ = β +V+ + βµ c v (14.21)
2
and

1
βV− = β −V− + βµ c v. (14.22)
2
Subtracting (14.22) from (14.21) gives

βZe I = β +V+ − β −V−

= (βe − βq )V+ − (βe + βq )V− (14.23)
= βe Ze I − βqV

or

 β − β
V = e  Ze I . (14.24)
 βq 
Substituting for I from (14.20) gives

 β − β   β2 − β20  Ze
V = e   v. (14.25)
 βq   β0 β  µ c Zc
The magnitude of each physical variable can then be expressed as the sum of the
voltages of the coupled normal modes:
 Small-Signal Theory 517

4
v = ∑Wn , (14.26)
n =1

4
βn
i=∑ Wn , (14.27)
n =1 β 0 Z0

4
β2n − β20 Wn
I =∑ ⋅ , (14.28)
n =1 β n β 0 µ c Zc
and

4 
β − β n   β2n − β20  Ze
V = ∑ e Wn , (14.29)
n =1  βq   β n β0  µ c Zc

where Wn are the amplitudes of the helix voltages of the normal modes in a uni-
form coupled section, and β n are the solutions of the dispersion equation (11.136).
Equations (14.26) to (14.29) can be written in matrix form as

v  W1 
i  W 
  = [C2 ]   . (14.30)
2

V
  W3 
 
 I  W4 
Finally, equations (14.14) and (14.30) can be combined to give the connec-
tion matrix which relates the amplitudes of the coupled modes to those of the
uncoupled waves

[W ] = [C2 ]−1 [C1 ][U ] = [C ][U ]. (14.31)

where [W ] is the column vector on the right hand side of (14.30) and [C ] is
the coupling matrix. Since the four coupled waves are normal modes, their
propagation through the circuit can be expressed as a diagonal transfer matrix
[S ] where
S ( z )nn = exp ( − j β n z ) . (14.32)

Then the vector at the output of the section is expressed in terms of the vector at
the input by

V2 f  V1 f  V1 f 

V  V  V 
 2 b  = [C ] S ( L ) [C ]  1b  = T ( L )  1b  , (14.33)
−1

V2 +  V1+  V1+ 
V2 −  V1−  V1− 

where L is the length of the section. At the input of the first section of a TWT the
forward wave is generally known, and the beam is unmodulated. The backward
518 Travelling-Wave Tubes

wave in the input transmission line is not known but, if the output end of the cir-
cuit is well-​matched, it can be assumed that there is no backward wave there. Thus
the boundary conditions are mixed, with three specified at the input, and one at the
output. The amplitude of the input wave can be set to 1 without loss of general-
ity since the analysis is linear and all other parameters will then be scaled to unity.
Inserting the boundary conditions into (14.33) yields

V2 f  1 
0  V 
  = T ( L1 )  1b  , (14.34)
V2 +  0 
V2 −  0 

where L1 is the length of the section and V1b is the backward wave in the input wave-
guide. The presence of the beam changes the match at the input of the section.
Thus the backward wave is the sum of the wave reflected by that mismatch and
the backward wave emerging from the tube. From the second row of the matrix we
know that

T ( L1 )2,1 + T ( L1 )2,2 V1b = 0 (14.35)

so that V1b can be calculated and substituted into the right-​hand side of (14.33).
The output vector can then be calculated directly with the check that V2 b = 0 . For a
single-​section tube the gain is

GdB 1,2 = 20 log V2 f . (14.36)

At any point within the section the wave amplitudes are given by

1 
V 
W ( z ) = S ( z ) [C ]  1b  (14.37)
0 
0 

and the amplitude of the forward wave, in decibels, as a function of position is

A ( z ) = 20 log ∑W ( z )
n , (14.38)
n

where the summation excludes the backward wave. Figure 14.9 shows a typical plot
of the forward wave amplitude in decibels against distance along the helix. After
some distance the amplitude of the growing wave is much greater than that of the
other two forward waves. The graph is then almost a straight line corresponding to
exponential growth of the growing wave. The boundary conditions at z = 0 mean
that some of the input wave is reflected and that all three forward waves have non-​
zero amplitudes. Thus, the projection of the graph back to z = 0 intercepts the verti-
cal axis at a negative value known as the launching loss whose value is typically in
the range 3 to 6 dB.
 Small-Signal Theory 519

Figure 14.9:  Signal growth along a two-​section TWT.

When a TWT is divided into two or more sections by severs, the signal is car-
ried between the sections by the space-​charge waves on the beam. Since these are
normal modes of the beam they propagate without change of amplitude, but their
relative phase at the start of the next section depends upon the length of the sever.
The uncoupled space-​charge wave amplitudes at the end of a sever of length Ls  are

V3+  exp − j β + Ls 0  V2 + 

V3 −  =  0 exp − j β − Ls  V2 − 
. (14.39)

If the terminations of the second section are matched then the amplitude of the
forward wave at the input of the section, and the backward wave at its output, are
both zero. Inserting these boundary conditions into (14.33), and denoting the input
and output of the section by subscripts 3 and 4 respectively we obtain

V4 f  0 
0  V 
  = T ( L2 )  3b  , (14.40)
V4 +  V3+ 
V4 −  V3 − 

where L2 is the length of the section. The value of the unknown parameter V3b in the
vector on the right-​hand side of the equation can be determined from the second
row of the matrix using

T ( L2 )2,2 V3b + T ( L2 )2,3 V3+ + T ( L2 )2,4 V3 − = 0 (14.41)

and inserted into the input vector to enable the output vector to be determined.
The small-​signal gain of the whole tube is then found by assuming that V1 f = 1
so that

GdB 1,4 = 20 log V4 f . (14.42)

520 Travelling-Wave Tubes

The wave amplitudes at the start of the second section can be determined using
(14.31). Because the boundary conditions must be satisfied at the beginning of the
section, the amplitude of the growing wave at the start of the section is less than
that at the end of the previous section. This difference is the sever loss, shown in
Figure 14.9 which is typically about 6 dB [16]. Like the launching loss this figure
depends upon the parameters of the tube and the frequency of operation. This
model is implemented in Worksheet 14.1.
The method can readily be extended to tubes with more than two sections, or
those in which the parameters of the sections differ from one another. It can also
be applied to tubes where the parameters of the helix are not constant throughout
a section, if each uniform sub-​section is modelled by a separate transfer matrix and
the results cascaded to represent the whole section. The same approach can be used
if a distributed attenuator is used in place of a physical sever, provided that the
propagation constants of the sub-​sections are adjusted to represent the cold loss
profile of the attenuator.

14.2.2 Small-​Signal Theory of Coupled-​Cavity TWTs

In Chapter 11 it was shown that the transfer matrix for the interaction between the
electron beam and a single cavity within a coupled-​cavity slow-​wave structure is
given by (11.156)

V2 f  V1 f 
V   
 2 b  = [ P ] V1b  . (14.43)
V2 +  V1+ 
   
V
 2−  V1− 
Thus the transfer matrix for a section of a TWT comprising N cavities is

V2 f  V1 f 
V  V 
 2 b  = [ PN ][ PN −1 ][ P2 ][ P1 ]  1b  , (14.44)
V2 +  V1+ 
   
V2 −  V1− 
which allows for the possibility that the cavities may not be identical to one another.
Equation (14.44) takes the place of (14.33) for continuous coupling, and the
boundary conditions can be applied in the same way as before. The chief diffe-
rence between continuous and discrete coupling is in the effect of the backward
wave generated by the interaction. With continuous coupling the backward wave
components induced in each element of the slow-​wave structure tend to cancel one
another out and the net backward wave is small. With discrete coupling, however,
this does not happen if the phase shift per cavity is small, that is, at frequencies
below the synchronous point. If the continuous interaction theory is used with
the parameters of a coupled-​cavity tube it is found that, in this region, the forward
wave gain is overestimated and the backward wave amplitude is under-​estimated
 Large Signal Effects 521

[17]. Thus, in a coupled-​cavity TWT, appreciable power may appear at the input
port of a section as a result of the interaction within it.

14.3 Large Signal Effects

At large signal levels a TWT becomes non-​linear and exhibits power saturation
like a klystron. The large-​signal behaviour of a TWT can be modelled with a
Lagrangian model using discs or rings of charge, as described in Section 11.8.
Early work in the field used normalised variables based on the Pierce parameters
[18–​20] but, on the whole, this makes the results difficult to interpret. More recent
authors have used physical parameters [21–​25]. Eulerian models, which are exten-
sions of small-​signal methods, require less computational effort than Lagrangian
models but have decreasing accuracy as saturation is approached [26, 27]. Some
modelling has been carried out using three-​dimensional PIC codes but very large
computer resources are needed for this purpose [28–​30]. For most purposes one-​
dimensional disk models provide the best compromise between accuracy and
computational efficiency. However, one-​dimensional models become less accur-
ate when γ e b > 1 because radial variations in the electric field then mean that the
motion of an electron varies appreciably with its radial position. In addition, these
models cannot predict radial motion of the electrons and the current intercepted
by the slow-​wave structure.
The results presented below were generated using a one-​dimensional model
of a helix TWT similar to that in [31]. In a model of this kind it is necessary to
track the electrons through a section of the tube and then use iteration to achieve
a self-​consistent solution. The process is then repeated for subsequent sections.
However, the author has found that results which closely resemble those given
by such a model can be obtained using a single pass if the voltage on the helix
is first calculated using a small-​signal model. As saturation is approached this
method over-​estimates the forward power on the helix which must therefore be
calculated, instead, from the loss of kinetic energy of the sample electrons. The
difference between the forward wave power assumed using small-​signal theory
and that calculated from the change in the kinetic energies of the electrons means
that the model does not represent a self-​consistent solution. However, it is found
that the results are generally very close to those obtained from a more accurate
model. The only exception is that the harmonic currents in the beam are too
great. This is because the excessive helix voltage close to the output confines the
trapped bunches more tightly than if the correct voltage profile were used. Better
results are obtained if the amplitude of the wave on the helix is adjusted to show
saturation, as described in Section 14.3.6. This model has been implemented in
Worksheet 14.2. It has not been validated for use as a design tool, but it provides
a simple way of exploring the large-​signal behaviour of a TWT.
Figure  14.10 shows the growth in the forward wave on a helix TWT which is
operated at the synchronous point. The tube has a single uniform section and
522 Travelling-Wave Tubes

Table 14.1:  Parameters of the TWT used in the illustrations

Beam voltage 6.0 kV

Beam current 135 mA
Beam radius 0.34 mm
Helix radius 0.68 mm
Frequency 11.7 GHz
Propagation constant (β0 ) 1700 m−​1
Pierce impedance 37 Ω

Figure 14.10:  Non-​linear signal growth in a single-​section TWT.

parameters, similar to those used in X band space TWTs, as given in Table 14.1.

They are identical to those in Figure  14.7, with the difference that space-​charge
effects have been included. The inclusion of space-​charge forces reduces the satu-
rated efficiency computed from 21% to 17%.
It is convenient to think of the operation of a TWT in terms of four regions as
shown by the vertical dividing lines in Figure 14.10:
i) An initial region in which the interaction between the wave on the helix and the
electron beam is established.
ii) A region in which the tube is linear and obeys small-​signal theory. This region
may be taken to end at the point where the velocity of the slowest electrons is
equal to the phase velocity of the growing wave on the helix (vw ) so that they
are on the point of being trapped by the wave.
iii) The non-​linear region in which the electrons are increasingly trapped and the
output power saturates.
iv) The region beyond saturation in which the tube is overdriven.
The details of the interaction can be understood by considering further results from
the model which were obtained under the same conditions.
 Large Signal Effects 523

Figure 14.11:  Amplitudes of the first and second harmonics of the beam current in a TWT
as a function of axial position.

Figure 14.12:  Phase of the first harmonic of the beam current in a TWT relative the wave on
the helix as a function of axial position.

Figures 14.11 and 14.12 show the amplitudes of the first and second harmonics
of the beam current, and the phase of the first harmonic relative to the phase of
the wave on the helix, as functions of axial position. In the second region the amp-
litude of the current grows exponentially. The phase is almost constant, and close
to the maximum retarding phase ( π 2 ). It should be noted that the amplitude of
the second harmonic current is appreciable even in this region where the behaviour
of the tube is apparently linear. The small-​signal relative phase, calculated from
(14.20) by assuming that the wave on the helix is dominated by the growing wave, is
shown by the broken line in Figure 14.12. In the third region energy is taken from
the bunches as more and more of the electrons are trapped, the phase of the first
harmonic current moves away from the maximum retarding phase, and the magni-
tude of the current grows more slowly and eventually passes through a maximum.
524 Travelling-Wave Tubes

Figure 14.13:  Typical modified Applegate diagram for a single-​section TWT.

As the output power saturates, the phase of the first harmonic current passes from
the retarding to the accelerating phase of the wave on the helix.
Figure 14.13 shows the modified Applegate diagram for the section. In this dia-
gram the phase velocity of the growing wave on the helix (vw ) is used as the ref-
erence, and the trajectories of electrons whose velocities are greater than vw have
positive slopes. In the first region, and most of the second region, the perturbation
of the trajectories is imperceptible and the electrons move through the wave with
almost constant velocity. By the end of the second region the bunching is visible
and the electrons are just beginning to be trapped. More and more of the electrons
are trapped by the wave as they move into the third region. It can be seen that the
trapped electrons oscillate within the potential well, as described in Section 14.1.3.
A detailed experimental study of electron trapping was reported in [32].
Figure 14.14 shows phase space diagrams at different points along the tube for
comparison with Figure 14.7. At small-​signal levels (Figures 14.7(a) and 14.14(a))
the figures are very similar, except that the effect of the space-​charge forces is to
move the bunch centre closer to the maximum retarding phase. It can be seen that
the phase of the maximum charge density is close to that for the minimum veloc-
ity, as is the case for a slow space-​charge wave. At the point at which trapping
commences (Figures 14.7(b) and 14.14(b)) the normalised velocity of the slowest
electron u vw = 1. The phase of the bunch has started to move towards the neutral
phase, and the space-​charge forces produce a distortion of the diagram. The plane
at which trapping commences is very close to that for the zero space-​charge case.
When the slowest electron is just beginning to be re-​accelerated (Figure 14.14(c)),
the space-​charge forces prevent the electrons at the front of the bunch from enter-
ing the trapped orbit, but they remain in the retarding phase of the helix wave. At
saturation (Figure 14.14(d)), the effects of space-​charge produce a much more dis-
ordered diagram in which the velocities of the slowest electrons are greater than in
the space-​charge free case. This explains the reduction in saturated efficiency caused
by the space-​charge forces.
 Large Signal Effects 525

Figure 14.14:  Phase space diagrams computed from the electron trajectories for the same
parameters as in Figure 14.7 with the inclusion of space-​charge forces.

Finally, Figure 14.15 shows the variation in the mean velocity of the electrons,
normalised to the phase velocity of the growing wave on the helix. In the first
region, and for most of the second region, the mean electron velocity is constant
and greater than the phase velocity of the wave. In the third region the electrons
lose energy and, at saturation, the mean electron velocity is a minimum. Beyond
saturation the electrons start to regain energy from the wave and the mean veloc-
ity increases. Figure 14.16 shows the spent-​beam distribution curve at saturation.
Comparing this to the idealised curve in Figure 10.9, we can see that the overall
efficiency of this tube could be increased substantially by using a multi-​element
depressed collector.
The results which have been presented in Figures 14.10 to 14.16 all relate to oper-
ation at the synchronous point. The first three regions which have been described
correspond approximately to the initial bunching, final bunching, and output
regions in a klystron. The results show the typical behaviour of a helix TWT under
526 Travelling-Wave Tubes

Figure 14.15:  Mean electron velocity, normalised to the helix phase velocity, as a function of
axial position in a TWT.

Figure 14.16:  A typical spent-​beam distribution curve at saturation.

large-​signal conditions but variations in detail are found when the operating point
is changed (see Section 14.3.3).

14.3.1 Dimensionless Parameters

Although TWTs show complicated non-​linear behaviour, the principles of dimen-
sional analysis can be applied to them (see Section 1.7.1) [33]. Thus efficiency, which
is a dependent dimensionless parameter, is a function of the set of independent
dimensionless parameters describing the tube. If we restrict our attention to uni-
form tubes having a solid beam and cylindrical symmetry then the requirement
for geometrical similarity is satisfied by fixing the value of b a. Since this is typi-
cally in the range 0.5 to 0.6 for many tubes we shall assume that it is constant. It is
also necessary to assume that the length of the output section of the tube is large
enough to ensure that the efficiency is not reduced and that the effects of harmon-
ics are negligible (see Sections 14.3.3 and 14.3.4). Then the scale of the tube can
be defined by the single dimensionless parameter γ e a because the phase velocity,
coupling impedance, and plasma frequency reduction factor are all functions of
it. We will assume that the magnetic focusing field is strong enough to suppress
radial motion of the electrons and that relativistic corrections can be ignored. The
 Large Signal Effects 527

remaining independent dimensionless parameters can be found by rewriting the

small-​signal determinantal equation (11.137) as

1
[Γ 20 − Γ 2 ] Γ 2q − (1 − Γ 2 ) = − 2 Γ1Γ 0 Γ 2 , (14.45)
where

β
Γ= (14.46)
βe
is a complex dependent parameter. The independent parameters (which can also be
expressed in terms of the Pierce parameters) are:

β0 − j α 0
Γ0 = ( = 1 + Cb − jCd ) , (14.47)
βe
where α 0 defines the loss per unit length of the circuit. For a loss-​less circuit this
becomes the ratio of the initial electron velocity to the phase velocity of the slow-​
wave structure

β0 u0
Γ0 = = ( = 1 + Cb) ; (14.48)
βe v p

the ratio of the coupling impedance to the impedance of the beam

I 0 µ c2 Z0
Γ1 = (= 4C 3 ) ; (14.49)
V0
and the ratio of the reduced plasma frequency to the signal frequency

βq ω q  2C QC 
Γq = = =  . (14.50)
βe ω  1 − 2C QC 

We saw in Section 13.3.6 that this parameter can also be expressed in terms of the
dimensionless beam radius and the perveance of the electron beam. Thus an alter-
native dimensionless parameter to Γ q  is
2
1 I  βe b Γ q 
Γ =
2
K ⋅ 10.5 =   , (14.51)
2 πε 0 2 e m0 Va  ωq ω p 
which is the perveance divided by a constant (33.0 µA V −1.5 ) . The presence of the
parameter C in (14.47) to (14.50) means that the Pierce parameters are only inde-
pendent of one another when C is constant. The parameters defined here have
the advantage that their physical significance is clearer and that it is easier to vary
them independently. Two tubes which are described by the same set of independent
dimensionless parameters are expected to have the same saturated efficiency and
gain (apart from the effects of cold loss).
528 Travelling-Wave Tubes

Where full data is available for an existing tube it is possible to develop a new tube
by scaling, if the two tubes are described by the same set of independent dimension-
less parameters. Schindler [34] showed that if γ e a and C are kept constant then QC
is constant when both b a and the beam perveance are constant, and the velocity
parameter is chosen to give either maximum gain or maximum efficiency. Thus it
is possible to scale a tube in such a way that all the dimensionless parameters are
simultaneously constant. There is limited scope for scaling a tube to higher power
levels, at the same frequency, because of the need to keep the temperature of the
slow-​wave structure below an acceptable level.

14.3.2 Dependence of Efficiency on Normalised Parameters

An approximate theory of the dependence of the efficiency of a TWT, with no cold
loss, on its operating point was given by Slater [15]. It was assumed that, at satura-
tion, all the electrons were trapped having a velocity ( usat ) as much below the phase
velocity of the growing wave and the initial velocity was above it. Thus

vw − usat ≈ u0 − vw , (14.52)

where vw is the phase velocity of the growing wave. Strictly speaking this should be
the phase velocity under large signal conditions but the error involved in using the
small-​signal phase velocity is small [32]. Then the efficiency is approximately

2
usat
2  2v 
ηe ≈ 1 − = 1 −  w − 1 . (14.53)
u02
 u0 

When space-​charge effects are included the small-​signal propagation constant of

the growing wave is given, in the two-​wave approximation, by (11.145) so that

u0 β0 + β − 1  u0 ωq 
= =  +1+ . (14.54)
vw 2βe 2  vp ω 

At the synchronous point ( ω = ω s )

u0 ω q u0
= 1+ = . (14.55)
vw ωs vp

Substituting from (14.55) into (14.53) gives

2
 1 − ωq ωs 
ηe ≈ 1 −   . (14.56)
 1 + ωq ωs 

Thus to a first approximation the efficiency should depend only on the normalised
reduced plasma frequency. If we assume that ω q ω s  1 then the term in brackets
can be expanded to give
 Large Signal Effects 529

ωq
ηe ≈ 4 (14.57)
ωs
so that the efficiency increases with increasing space-​charge density. We note that
this is the opposite to the case for klystrons.
Figure 14.14(d) shows the distribution of the velocities of sample electrons
at saturation when the tube defined in Table 14.1 is operated at the synchron-
ous point and u0 vw = 1.063. It can be seen that, in this case at least, the assump-
tion made in (14.52) is fairly good. The electronic efficiency calculated from
(14.56) is 22% which may be compared with the efficiency of 17.2% calcu-
lated using the disc model. The chief error arises from the fact that only 20
out of 24 sample electrons have been trapped, so that it would be reasonable
to reduce the efficiency estimated from that equation by a capture ratio of
20 24 = 0.83. The result is 18.5%. The remaining difference between the results
can be explained by noting that the mean velocity of the electrons which are
not trapped is a little greater than the initial velocity. This reduces the effective
capture ratio to 0.78.
In terms of the Pierce parameters (14.56) becomes

ηe
≈ 8 QC . (14.58)
C
An experimental study of the factors affecting the efficiency of a TWT [35]
assumed that

ηe
≈ 4θH , (14.59)
C
where θ is a function of the Pierce loss parameter d (defined in (14.47)) such that
θ = 1 when the slow-​wave structure is loss-​less. The function H was obtained from
an unpublished memorandum by Pierce and plotted as a graph of H against QC.
Comparing (14.59) with (14.58) we see that H = 2 QC and this function is a found
to be a very good fit to the curve in [35] when QC ≥ 0.25. The discrepancy for small
values of QC is not surprising because the two-​wave approximation assumed in
(14.54) is not valid when the space-​charge is small. When experimental results were
plotted against L C = 54.5 d (where L is the cold loss in dB per circuit wavelength, C
is the Pierce gain parameter, and d is the Pierce loss parameter) it was found that θ
was given to a useful approximation by [35, 36]

L
log10 θ = − = −2.18 d . (14.60)
25C
Thus the efficiency of a TWT operating at the synchronous point can be estimated
using (14.59), or (14.56) if cold loss is neglected. If the effects of space-​charge
potential depression are ignored (13.51) can be written at the synchronous point as
530 Travelling-Wave Tubes

ωq p µPerv
= (14.61)
ω s βe b 33

so that the efficiency can be estimated from the perveance. Typically, the beam filling
factor b a ~ 0.67 and βe b ~ 0.6, so that p βe b ~ 0.55 for many tubes. When the effi-
ciencies of over fifty TWTs of different types were plotted against 0.55 µPerv 33 it
was found that nearly all the points lay between theoretical curves calculated using
(14.56) with multiplying factors 0.3 and 0.7. The multiplying factor takes account
of the capture ratio, and the effects of space-​charge potential depression and the
cold loss of the helix. Other factors affecting the efficiency include design choices
such as the choice of synchronous point, and tapering of the slow-​wave structure
(see Section14.3.7). There was good correlation between points derived from the
data in [37] and a theoretical curve with a multiplying factor of 0.7.
The space defined by the Pierce parameters was explored in a classic series of
experiments by Cutler [38]. The results were summarised in two graphs showing
contours of ηe C plotted on the (QC , γ e b ) plane for maximum gain per unit length
(synchronism) and maximum output power. These graphs showed that:
i) at synchronism ηe C > 3 when γ e b ~ 0.6 ± 0.05 and QC ~ 0.25 ± 0.05;
ii) at maximum efficiency ηe C > 3.5 when γ e b ~ 0.5 ± 0.1 and QC ~ 0.2 ± 0.05.
It was assumed that ηe C was constant when all the other parameters were
constant. The Pierce parameters of the tube, defined in Table  14.1, are
γ e b = 0.57, C = 0.062 and QC = 0.22. Calculations using the model in Worksheet
14.2 gave the results ηe C = 2.7 at synchronism and ηe C = 3.4 at maximum effi-
ciency, which are close to Cutler’s figures.
Cutler’s experimental results included phase space diagrams which closely
resemble those in Figures  14.7 and 14.14. The graphs also included compari-
sons with the results of large-​signal calculations which showed poor agreement
because the radial variation of the field of the slow-​wave structure had been
neglected in them. Cutler concluded that there is a broad optimum in the region of
γ e b = 0.5, C ≤ 0.14 and QC = 0.2 where it should be possible to achieve efficiencies
greater than 30%. The results in this paper have been widely used to guide the design
of TWTs. However, the use of QC as a space-​charge parameter, and the fact that
H = 1 when QC = 0.25, have led to the widespread assumption that the efficiency
depends on C rather than on ω q ω [8, 14, 39, 40]. This misunderstanding arose
because Slater’s original work, which ignored the effects of space-​charge, suggested
that ηe ≈ 2C . Pierce and Cutler were aware that the correct expression should be
ηe ≈ −4Cy1 where y1, which is the growing wave parameter, ranges between −1 2
and –​2 depending upon the operating point and the value of the space-​charge par-
ameter. With this correction the assumptions are exactly the same as have been
made in the derivation of (14.56). Thus the assumption that the efficiency is pro-
portional to C assumes, wrongly, that y1 is constant.
Other experiments described in a less well-​known paper by Caldwell and Hoch
[41] covered similar ground though with a more restricted range of parameters. It
 Large Signal Effects 531

was concluded that there was no simple relationship between the efficiency and the
Pierce parameters. The saturated efficiency at the synchronous point was found to
be independent of the beam current but related to the normalised reduced plasma
frequency. These results, therefore, tend to support the validity of (14.57) as an
approximation.

14.3.3 Dependence of Efficiency on the Operating Point

Equation (14.53) shows that the efficiency of a TWT with no circuit loss should
be determined by the ratio u0 vw at the operating point. It is useful to note that
u0 vw is always greater than unity when the tube has gain, whereas u0 v p can
be less than unity. Thus, the efficiency given by (14.53) is always positive. The
propagation constant of the growing wave in the two-​wave approximation is
given by (11.145) as

1
β= (β0 + β − ) (14.62)
2
so that

u0 β0 + β − 1  β0 + βq 
= =  + 1 . (14.63)
vw 2βe 2  βe 
Thus, the ratio u0 vw can be increased by one, or more of:
i) reducing βe by increasing the beam voltage, keeping the frequency, the reduced
plasma frequency and the pitch of the helix constant [42];
ii) increasing β0 by increasing the frequency, keeping the beam voltage and helix
pitch constant;
iii) increasing β0 by reducing the pitch of the helix (and thus reducing v p ), keeping
the beam voltage and the frequency constant;
iv) increasing βq by increasing the beam current, keeping the beam voltage and the
frequency constant.
It should be noted that the effect of changing u0 vw by different means does not
yield exactly the same results, because other parameters may be changed also. Thus,
the effects of increasing the beam voltage at constant DC power, constant perve-
ance, and constant current differ from one another. Similarly, when the frequency
is changed, the impedance of the helix also changes. A simple helix has positive dis-
persion (i.e. v p decreases with frequency) and the slow space-​charge wave has nega-
tive dispersion so that u0 vw increases with increasing frequency. When the helix is
heavily loaded it may have negative dispersion and then the ratio u0 vw may be kept
almost constant if the dispersion of the helix is chosen correctly [43].
If the beam voltage of a TWT is varied at constant perveance then all the nor-
malised parameters, except u0 v p , are almost constant. Curve (a) in Figure 14.17
shows how the small-​signal growth rate Im (β ) β0 varies with u0 vw as the beam
532 Travelling-Wave Tubes

Figure 14.17:  Effect of the velocity parameter u0 vw under small-signal conditions on (a) the

normalised growth rate (Im (β ) β0 ) and, (b) the phase of the RF beam current relative to
the helix voltage.

voltage is varied at constant perveance, using the same parameters as before. The
normalised reduced plasma frequency is almost constant and equal to 0.06. Curve
(b) in the figure shows the phase of the RF beam current relative to the wave on the
slow-​wave structure under small-​signal conditions when the growing wave is fully
established. The bunch is in the retarding phase of the field for all values of u0 vw
for which the gain per unit length is non-​zero. When the gain is maximum the phase
of the bunch is close to the maximum retarding field. As the ratio u0 vw decreases,
the bunch moves towards the neutral phase where the forces acting on the bunch
tend to compress it.
The dependence of the saturated efficiency, and of the maximum normalised
RF beam current, on the velocity parameter u0 vw are shown in Figure 14.18. The
graphs were computed for a uniform section of a TWT by changing the beam volt-
age at constant perveance, using Worksheet 14.2 and the parameters in Table 14.1.
The Pierce parameters (C = 0.06 and QC = 0.22 ) were approximately constant.
Curve (a) shows that the saturated efficiency increases with u0 vw , as expected, and
that it reaches a maximum close to the top of the range for which the tube has
gain. Curve (b), obtained by multiplying the result from (14.53) by a capture ratio
of 0.78, is a good fit to curve (a). Figure 14.11 shows that the RF beam current
typically reaches a maximum somewhere between the plane at which the electrons
begin to be trapped, and the plane of saturation. Curve (c) in Figure 14.18 shows
the magnitude of the maximum value of the normalised RF beam current ( I1 I 0 ) as
a function of u0 vw . The maximum RF current increases as u0 vw decreases and the
phase of the bunch moves closer to the neutral phase. The curve has a maximum
towards the lower end of the range for which the tube has gain. We shall see that the
conditions for maximum efficiency, and maximum RF current, have an important
role in TWT design.
 Large Signal Effects 533

Figure 14.18:  Variation of the large-​signal properties of a TWT as the beam voltage is

changed at constant perveance (u0 vw = 1.06 at the synchronous point): (a) saturated
efficiency; (b) saturated efficiency estimated from (14.53) multiplied by 0.78 and;
(c) maximum magnitude of the normalised RF beam current.

The most comprehensive theoretical investigation of the effect of changes in the

operating point of a TWT was carried out by Rowe using the Pierce parameters
[19]. The difficulty with using these parameters is that changes in the operating
point generally result in changes in two or more of the parameters. Thus it is not
straightforward to use Rowe’s results to predict the behaviour of a tube when the
operating point is changed. However it is found that the Pierce parameters C and
QC are approximately constant when the beam voltage is changed at constant
perveance. Rowe’s results show that the efficiency then increases linearly with the
Pierce velocity parameter b and reaches a maximum close to b = 2 which lies just
below the point where the gain falls to zero. Figure 14.19 shows curves of efficiency
against u0 vw , obtained by re-​plotting Rowe’s data [19]. The lowest point on each
curve is the synchronous point. The figure also shows theoretical curves calculated
using (14.53) for capture ratios of 0.8 and 0.6. The computed data shows the same
trend as the theoretical curves. An increased value of C is needed to maintain the
capture ratio as the space-​charge is increased. Other results given by Rowe showed
that the output power is reduced by 10 dB when the cold loss per circuit wavelength
divided by the Pierce parameter C is increased by about 25 dB, in agreement with
(14.60). These results, which were obtained using a ring model of the beam, also
showed that the best efficiency was obtained with a beam filling factor ( b a ) in the
range 0.5 to 0.75 corresponding to an optimum value of γ e b in the range 0.75 to 1.0.
Further insight into the effects of variations in the operating point is given by
Figures 14.20 and 14.21, which relate respectively to the conditions for maximum
RF current and maximum efficiency (see also [44]). In the former, the RF current is
greatest at saturation, only two of the sample electrons have not been captured by
the wave and the velocities of the trapped electrons do not differ very much from vw .
534 Travelling-Wave Tubes

Figure 14.19:  Computed efficiency of a TWT as a function of u0 vw derived from data given

in [19] with C and ω q ω as parameters.

In the latter the maximum RF current occurs just after the plane at which trapping
commences, and fewer of the electrons have been trapped. The trapped electrons
have a much greater variation of velocity and at least two of the slowest electrons
have drifted into the following bunch. Thus the bunc hes are much better formed
in the first case but the small change in energy when they are trapped by the wave
means that the efficiency is low.
An experimental study by Gerchberg and Niclas [45] showed that, under condi-
tions for maximum gain, the phase of the circuit field is almost constant. Thus, the
change in the relative phase seen in Figure 14.12 is almost entirely due to the chan-
ging phase of the RF beam current as the electrons lose energy. However, under
conditions for maximum power output, the phases of both the current and the
circuit wave changed with power level until the input power was a few dB below
that required to saturate the tube. For greater drive levels the phase of the cur-
rent remained almost constant while the phase of the circuit voltage continued to
change until the two were in quadrature at saturation. This behaviour is reflected in
kink in the phase curve in Figure 14.21(d).

14.3.4 Effect of a Sever

The tube modelled in the preceding sections is impracticable because the gain is
too high for a single-​section tube to be stable. Thus, in practice, it would be divided
into two, or more, sections by severs. The efficiency of a tube is not affected by a
sever provided that the small-​signal gain of the output section exceeds a critical
 Large Signal Effects 535

Figure 14.20:  TWT behaviour at the working point at which the RF current is greatest
(u0 vw = 1.031): (a) modified Applegate diagram, (b) phase space plot at saturation,
(c) normalised first and second harmonic currents, and (d) phase of the RF beam current
relative to the wave.

value [46–​48]. When the gain falls below this value the efficiency is rapidly reduced.
Thus Scott found that the efficiency of a particular tube was halved when the small-​
signal gain of the output section was 5 dB less than the critical value. The most
important physical reason for the reduction in efficiency is the dispersion of the
electron bunches in the sever, and in the initial part of the output section where the
electric field of the circuit is weak. To avoid this it is necessary for the sever to be
placed at a position where small-​signal conditions apply [42]. If the space-​charge
density in the beam is increased, then the length of the output section required to
avoid a reduction in efficiency also increases [47]. From Figure 14.9 it can be seen
that, for the tube modelled, the length of the output section required to re-​establish
the interaction after the sever corresponds to a small-​signal gain of about 10 dB.
Similarly, from Figure  14.10, we see that the small-​signal gain from the plane at
which trapping commences, to the plane of saturation, is about 15 dB. It seems
reasonable to suppose that the efficiency will not be affected by the sever if the
interaction is fully re-​established by the plane at which trapping commences. Thus,
in this case, the small-​signal gain should be at least 25 dB in the output section. This
536 Travelling-Wave Tubes

Figure 14.21:  TWT behaviour at the working point at which the efficiency is greatest
(u0 vw = 1.095) : (a) modified Applegate diagram, (b) phase space plot at saturation,
(c) normalised first and second harmonic currents, and (d) phase of the RF beam current
relative to the wave.

is comparable with the results in [47, 48]. If the interaction is linear at the plane of
the sever, then the reduction in efficiency caused by dissipation of power in the sever
load, or in the attenuator, is negligible.
When the tube is operated away from the synchronous point, the small-​signal
gain of the output section decreases, and the efficiency may be reduced compared
with that of a single-​section tube. This effect is greater when u0 vw is greater than
the value at the synchronous point. Thus, in broad-​band tubes the position of the
sever should be chosen to achieve satisfactory performance at the highest frequency
[48, 49]. It has also been shown that the efficiency of a severed TWT can be greater
than that of an unsevered tube if the length of the section after the sever is slightly
greater than the critical length [47, 49]. This is caused by a change in the phase of
the bunches relative to the phase of the wave on the slow-​wave structure after the
sever. Overall, the length of the final section of a TWT is a compromise between
the requirements to maximise the efficiency, and to minimise the gain ripples. This
problem becomes more acute when the space-​charge density is high.
 Large Signal Effects 537

14.3.5 Harmonics
At high drive levels the output of a TWT contains appreciable harmonic power.
Figure 14.22 shows typical beam current waveforms at saturation for three differ-
ent values of the velocity parameter u0 vw . The current waveforms were produced
by superimposing the current pulses defined by the phases of successive pairs of
electrons given by

 ωQ 
 if φ j ≤ φ < φ j +1 
I ( φ ) =  φ j +1 − φ j  . (14.64)
 0 otherwise 

where φ = ωt and Q is the charge carried by one disc. It can be seen that all the
bunches have a high harmonic content. As u0 vw increases the width of the bunch

Figure 14.22:  Typical beam current waveforms at saturation for (a) u0 vw = 1.031,

(b) u0 vw = 1.063, and (c) u0 vw = 1.095.
538 Travelling-Wave Tubes

increases and it has a tendency to split into two parts, one in the accelerating, and
one in the decelerating, phase of the field. The helix can propagate waves at harmon-
ics of the signal frequency and, although the coupling impedance decreases rapidly
with increasing order of harmonic, some power is coupled to the helix. Thus there
may be appreciable power output at the second, and higher, harmonic frequencies
depending upon the dispersion and impedance characteristics of the helix.
There are special problems in broad-​band helix tubes when the harmonics of
signals at the low frequency end of the band may lie within the operating range
of the tube. It is not unknown for the harmonic power to exceed the fundamental
power [20, 50]. The total RF power at saturation is then approximately equal to
the saturated output power of an equivalent tube in which harmonic generation
is negligible. The level of harmonic generation is found to depend only slightly
on the gain per unit length of the tube. However, it is strongly affected by the
ratio of the phase velocities at the first and second harmonic frequencies (v p 2 v p1 ).
There is strong second harmonic generation, with a corresponding reduction in
the efficiency at the first harmonic frequency, when this parameter is slightly less
than unity. However, when the ratio is reduced still further, the relative amplitude
of the second harmonic is much smaller and the efficiency at the first harmonic
is only a little less than that calculated when the effects of the second harmonic
are ignored. Typically the increased level of the second harmonic occurs in the
range 0.94 < v p 2 v p1 < 1.0 [20]. The second harmonic power increases as the coup-
ling impedance at the second harmonic frequency increases. This behaviour can be
explained in qualitative terms by the effect on the second harmonic beam current
of the interaction at both the first, and second, harmonic frequencies. If these
two processes are in anti-​phase with one another, some degree of cancellation will
occur, and the amplitude of the second harmonic wave on the helix is reduced.
The output of a tube is also affected by internal, and external, reflection of sig-
nals at the second harmonic frequency. It has been observed that the perform-
ance of a TWT can be altered by the injection of a second harmonic signal [51].
Thus, in general, the effects of the second harmonic (at least) must be included in
large-​signal simulations of helix TWTs. However, the dispersion of the helix in the
tube modelled in the preceding sections is great enough for harmonic effects to be
negligible.

14.3.6 Transfer Characteristics

In the previous sections the large-​signal behaviour of a TWT has been considered
as a function of distance along the tube at constant drive level. This is related to
the terminal properties of the tube, represented by the amplitude and phase of the
signal at the output relative to that at the input. Figure 14.23(a) shows a typical
plot of output power against input power for a TWT (both in decibels relative to
saturation) which is very similar to that for a klystron in Figure 13.7. At low drive
levels the graph is a straight line having unit slope. The non-​linearity in amplitude
 Large Signal Effects 539

Figure 14.23:  Typical characteristics of a TWT: (a) amplitude transfer, and (b) gain

compression.

Figure 14.24:  Typical phase transfer characteristic of a TWT.

is shown by the gain compression curve in Figure 14.23(b) which shows the gain
relative to small-​signal gain as a function of input power. The linear region of tube
is often taken to extend up to the 1 dB compression point shown in Figure 14.23(a).
Because the mean electron velocity decreases as the power output is increased,
the phase of the signal at the output of a TWT is a function of the input power, as
illustrated in Figure 14.24. Under small-​signal conditions the output phase must be
independent of the input power. It can be seen from Figures 14.23 and 14.24 that
the phase is a more sensitive indicator of non-​linearity than the amplitude. This is
confirmed by Figure 14.12 which shows that the phase of the current departs appre-
ciably from the small-​signal value within the region where the amplitude variation is
linear. It is common to specify the linearity of a TWT in terms of the derivatives of
the curves in Figures 14.23(a) and 14.24, as shown in Figure 14.25(a) and (b). These
are the amplitude modulation to amplitude modulation (AM/​ AM) conversion
540 Travelling-Wave Tubes

Figure 14.25:  Typical derivative transfer characteristics of a TWT: (a) AM/​AM conversion,

and (b) AM/​PM conversion.

expressed in dB/​dB, and the amplitude modulation to phase modulation (AM/​PM)

conversion expressed in degrees/​dB. For telecommunications applications it is desir-
able that both these parameters should be as small as possible (see Section 1.6.4).
A good approximation to the graph of output power against position can be
found by assuming that the tube is linear up to the plane at which electron trapping
commences, and that the gain then follows a cubic curve until the slope is zero at the
plane of saturation. In this region the gain in decibels is given by


GdB = 20 α log ( e ) ( z − zt ) −
( z − zt )3  , (14.65)
2
 3 ( zs − zt ) 

where α is the imaginary part of the small-​signal propagation constant of the grow-
ing wave, zt is the plane at which trapping commences, and zs is the plane of satu-
ration. The positions of these two planes can be established, using a large-​signal
disc model, with the assumption that the wave amplitude on the helix is given by
small-​signal theory. Then (14.65) can be used to correct for the gain compression in
the large-​signal region. The planes zt and zs are hardly changed by the revision of
the profile of the wave on the helix and it is found that the model then has greater
self-​consistency (see Worksheet 14.2). Thus the gain compression at saturation is

20
∆GdB = − α log ( e ) . (14.66)
3
When the input power to a tube is increased by 1 dB the effect is equivalent to increas-
ing the linear region of the tube by a section whose length would provide 1 dB of
small-​signal gain. This remains true provided that small-​signal conditions exist over
some part of the length of the tube. The small-​signal gain per unit length is

dG0,dB
= 20 α log ( e ) . (14.67)
dz
 Large Signal Effects 541

Then, differentiating (14.65) and dividing by (14.67) we find an expression for the
AM/​AM conversion in dB/​dB

dG
⋅
dz
= 1−
( z − zt ) (14.68)
2

dz dG0,dB ( zs − zt )2
This gives a parabolic variation from unity in the linear region to zero at saturation,
which may be compared with Figure  14.25(a). The AM/​AM conversion is least
when the distance between zt and zs is least, which generally occurs close to the
synchronous point.
A simplified model of AM/​PM conversion can be derived from small-​signal the-
ory by neglecting the backward wave [52]. Then, from (11.130), the forward wave
is given by

∂V f 1 ∂I
= − j β0V f + µ c Zc 1 . (14.69)
∂z 2 ∂z
Now let

V f = V f 0 ( z ) exp j ( ωt − β0 z ) (14.70)

so that the phase of the voltage is referred to the phase of a wave travelling with the
cold phase velocity of the helix. Substituting in (14.69) the wave voltage at z + dz
is given by

1 ∂I
V f 0 ( z + dz ) = V f 0 ( z ) + µ c Zc 1 dz. (14.71)
2 ∂z
But we may write

∂ I1 ∂I dz
dz = 1 dt = j ωI1 (14.72)
∂z ∂t u0
so that

1 dz
V f 0 ( z + dz ) = V f 0 ( z ) + j ωµ c Zc I1 ( z ) exp ( j φ ( z )) , (14.73)
2 u0
where φ ( z ) is the phase of the RF beam current with respect to the wave. Then, if
change of the amplitude of the wave in the element dz is small, the rate of change
of the phase of V f 0 with z is

dΦ 1 I1 ( z )
= βe µ c Zc cos ( φ ( z )) , (14.74)
dz 2 Vf 0 (z)

which depends upon the ratio of the amplitude of the RF beam current to the amp-
litude of the wave voltage, and also upon the phase difference between them. These
are determined by the non-​linear interaction between the electron beam and the
wave on the helix that also leads to the generation of harmonic currents. If the ratio
of the phase velocities of the helix at the first and second harmonic frequencies lies
542 Travelling-Wave Tubes

in the range for which appreciable harmonic interaction takes place then the AM/​
PM conversion is affected [53]. The AM/​PM conversion tends to be small when the
variation in φ ( z ) with z is small, and this occurs at low values of the ratio u0 vw  [43].

14.3.7 Tapers
We have seen that, at saturation, the phase of the first harmonic of the beam cur-
rent relative to the voltage wave is zero, and no further energy can be extracted from
the electrons. However, if the slow-​wave structure is modified, so that the phase
velocity of the wave is reduced, then it is possible to extract more energy from the
electrons, provided that they remain trapped by the wave. Such a change in the
structure is referred to as a taper [16, 42, 54]. The change in the phase velocity in a
taper can be continuous, or in a series of steps. It is found that the saturated power
output of a tube can be increased in this way by about 2–​3 dB. In general tapering
the slow-​wave structure reduces its characteristic impedance so that the amplitude
of the wave corresponding to a given power is reduced. Thus the wave does not
trap the electrons as effectively. A point is reached, eventually, where further taper-
ing yields no improvement because the space-​charge forces reduce the trapping of
the electrons by the wave. Also, since some of the electrons are not trapped by the
wave at saturation, the effect of the taper may be to move them into the accelerat-
ing phase of the field. Hence it is easiest to increase the efficiency of a TWT using
an output taper if the beam perveance is low, and the bunches are well-​formed
[44]. Conversely, little improvement in efficiency can be obtained by reducing the
phase velocity when an untapered tube is operated at the condition for maximum
efficiency. However, it is sometimes found that the relative phase tends to π as satu-
ration is approached in an untapered tube, operated at maximum efficiency, and
that there is a big variation in the velocities of the electrons within the bunch. It has
been shown that the efficiency can then be increased by the use of a positive taper,
in which the phase velocity increases. This taper recaptures the bunch so that the
relative phase decreases towards saturation in the usual manner [45, 55].
Early papers on tapering concluded that the best efficiency is obtained when the
taper starts at the saturation plane of an untapered tube, and that the detailed
profile of the taper does not have much effect on the efficiency achieved [16, 42].
However, this strategy tends to produce an AM/​AM transfer curve which is concave
just before saturation so that the tube is non-​linear over a greater range of values
of the input power. More recent studies have shown that the taper may start before
saturation; typically at the plane of maximum RF current [44, 56]. Since it is usu-
ally desirable to optimise the performance of a tube over a band of frequencies, the
taper should be designed to give the greatest improvement in efficiency at the end of
the band where the untapered efficiency is low. In helix TWTs the variation in phase
velocity is usually achieved by changing the pitch of the helix which means that
the gaps between the turns and, hence, the coupling impedance, are reduced. Since
only a single parameter is varied there is necessarily a compromise between the
properties of the helix at different frequencies within the working band. However,
 Large Signal Effects 543

in coupled-​cavity TWTs the dispersion curve can be altered in a number of ways so

that it is possible to design tapers whose properties are closer to optimum through-
out the band [44, 54, 57]. It is also possible to increase the efficiency of a tube by
dividing the slow-​wave structure into two or more sections which have uniform
pitch. The sections are insulated from one another and have DC voltages applied to
them so that the electron bunches are accelerated, moving them into the retarding
phase of the wave [42]. This adds considerably to the complexity of the tube and
does not appear to have been used except in experimental devices.
Tapering the slow-​wave structure can also be used to enhance the performance of
TWTs in other ways including [42, 58, 59]:
• reducing harmonic output power;
• improving linearity and reducing the AM/​PM distortion;
• equalising the performance across the frequency band;
• improving tolerance to changes in the beam voltage.
It has already been observed that the AM/​AM conversion of a tube depends upon
the distance between the plane at which trapping commences, and the plane of sat-
uration. In an untapered tube this distance is least close to the synchronous point,
but it can be further reduced by the use of a tapered structure. We have also seen
that the AM/​PM conversion is strongly influenced by the change in the phase of the
RF beam current with respect to the wave voltage. Thus, improved linearity can be
achieved by adjusting the phase velocity of the slow-​wave structure to keep the ratio
u vw constant as the electrons lose energy to the wave [60]. Such a taper, known as
a dynamic velocity taper, can achieve improved linearity together with an increase
of 1 to 2 dB in the efficiency.

14.3.8 Stability
A particular problem with helix TWTs, especially those designed to work at high
power levels, or broad bandwidths, is a tendency to backward wave oscillations
(BWO) when the beam interacts with the n = −1 space-​harmonic of the backward-​
wave on the helix. The starting conditions for oscillation can be found using small-​
signal theory (see Section 11.7). Backward-​wave oscillations are characterised by a
threshold beam current above which the oscillation occurs, and a sharply defined
frequency which is dependent on the beam voltage. Equation (11.164) shows that
the start-​oscillation current is inversely proportional to the square of the length
of the section. Thus, the risk of backward-​wave oscillations increases sharply as
the length of a section is increased. The stability of a tube with given small-​signal
gain can be increased by increasing the number of severs, with the penalty of addi-
tional tube length to compensate for the increase in sever loss. It has been shown,
however, that the stability of a section can also be increased by introducing a step
change in the helix pitch at some point in its length. If the average pitch of the sec-
tion remains constant, then the effect on the forward-​wave gain is small, whilst the
stability against backward-​wave oscillations is determined by the stability of each
544 Travelling-Wave Tubes

uniform section [61–​63]. Other techniques for increasing stability include the use of
resonant loss using patterns printed on the support rods [64], and the use of exter-
nal filter structures to damp the unwanted mode [65, 66]. This analysis assumes that
only a single mode can propagate on the helix. To avoid the possibility of feedback
oscillations involving higher-order modes it is important that they are strongly cut-​
off. The lowest mode of concern is the TE11 waveguide mode in the space between
the helix and the shield and it is important that the cut-​off frequency of this mode
is above the working band of the tube.
In coupled-​cavity TWTs oscillations can occur at the band edges if the beam
voltage is reduced. At these frequencies the group velocity tends to zero so that
the slow-​wave structure supports a standing wave. The coupling impedance of a
normal coupled-​cavity structure tends to infinity at the upper cut-​off frequency
(see Figure 4.28(b)) [67]. Band-​edge oscillations are caused by the interaction of
the slow space-​charge wave with a backward-​wave space-​harmonic close to the
band edge. The frequency of oscillation varies only slightly with beam voltage.
These oscillations present particular problems in cathode pulsed TWTs because
the beam voltage passes through the critical range for oscillation on the rise and
fall of the pulse. The result is a characteristic RF pulse shape with the oscillations
appearing as ‘Rabbits’ Ears’ on the rise and fall of the pulse. If this pulse were
viewed using a spectrum analyser it would be seen that the ‘Rabbits’ Ears’ were at
a different frequency from the main part of the pulse. They can be suppressed by
careful design, and by ensuring that the pulse rises and falls too quickly for the
oscillation to build up. The start-​oscillation conditions can be calculated using
the methods described in [68] or by using a small-​signal gain calculation program.
Frequency selective loss is sometimes introduced to prevent these oscillations
[69]. It has also been shown that they can be suppressed by tapering the slow-​wave
structure [70]. A third approach is to use a slow-​wave structure designed to work
in the inverted mode (see Section 4.5) so that the coupling impedance is zero at
the upper cut-​off frequency [71]. There is some reduction in performance because
the coupling impedance at the centre of the pass-​band is reduced. It is necessary
to ensure that the beam cannot interact with the first higher-​order mode, which
then has infinite coupling impedance at its lower cut-​off frequency. Backward-​
wave oscillations may occur in the higher order modes of a coupled-​cavity struc-
ture, and it may also be possible for feedback oscillations to occur involving a
higher-​order mode [72]. Not all higher-​order modes have appreciable interaction
fields but it is necessary to check for possible problems of this kind when a tube
is being designed.

14.4 TWT Design

The design procedure for TWTs is similar to that for klystrons described in Section
13.4. However, the great variety of types of TWT, and of the slow-​wave structures
incorporated in them, makes it much more difficult to make generalisations about
 TWT Design 545

the process. The sections which follow discuss some of the chief issues involved in
the design of TWTs of different types.

14.4.1 Case Study: An Octave Bandwidth Helix TWT

Helix TWTs with a bandwidth of about one octave have been used in many com-
munications and radar systems. The conceptual design procedure can be illustrated
by considering a tube with the specification in Table 14.2 [39].
The design calculations can conveniently be carried out using a spreadsheet so
that effects of different design choices can be investigated quickly and easily (see
Worksheet 14.3). The principal steps in the calculations for the tube in Table 14.2 are:
i) Since the power output at the band edges is to be at least 1 kW, the RF power
at the band centre must be at least 1.26 kW. The bandwidth required means
that only a helix slow-​wave structure is suitable. However, the mean output
power of 10 W is well within the thermal capabilities of such a structure in this
frequency band. The DC beam power required is found by dividing the RF
power by an estimate of the efficiency of the tube at the band centre based on
previous experience.
ii) The perveance of the electron gun is chosen in the range 0.1 to 2.0 µA V −1.5 . In
this case a high perveance is desirable to reduce the risk of DC voltage break-
down, to ensure that the coupling impedance of the helix slow-​wave structure
is as high as possible, and that the electronic efficiency is high. The anode volt-
age (Va ) and the beam current ( I 0 ) can then be calculated.
iii) The beam filling factor ( b a ) is chosen in the range 0.5 to 0.7. Then the space-​
charge potential depression as the beam passes through the helix can be cal-
culated using (7.8). The potential on the axis (Vh ) is used to calculate the mean
velocity of the electrons ( u0 ) (see Section 7.2).
iv) The normalised beam radius at the band centre ( γ e b ) is chosen in the range
0.5 to 1.0. A large value of this parameter leads to an appreciable radial varia-
tion in the RF field of the helix acting on the electrons, and reduced coupling
impedance through the radial coupling factor. Since γ e b increases with fre-
quency these effects are particularly important at the top of the frequency band
of the tube. The beam radius ( b ) and the helix radius ( a ) can be calculated.
v) It is now possible to calculate the plasma frequency of the beam and the Brillouin
field. The magnetic field factor m is chosen to give the best compromise between

Table 14.2:  Outline statement of requirements for a pulsed TWT

Frequency 2.0 to 4.0 GHz

Power output (1 dB band edges) 1.0 kW
Duty cycle 1%
Gain (at 1 kW) 33 dB
546 Travelling-Wave Tubes

beam stiffness and the size and weight of the focusing magnet. This tube is suit-
able for PPM focusing and it is necessary to check that the peak field strength
can be achieved without saturating the pole-​pieces. Typically m is in the range
1.5 to 3.0. The magnetic field parameter α is chosen in the range 0.1 to 0.2 to
give an acceptable ripple in the beam radius (see Section 7.6.1). The peak mag-
netic field and the period of the PPM stack can then be calculated.
vi) The plasma frequency reduction factor can now be calculated as a function
of frequency. The saturated efficiency of the tube at the synchronous point
is estimated using (14.56) multiplied by an estimate of the capture ratio (see
Figure 14.19). This figure which is based on the helix voltage must be multi-
plied the ratio Vh Va to give the estimated ratio of the RF power to the DC
power input to the tube. The preceding calculations can be repeated using this
revised estimate now, or deferred until a more accurate figure can be calculated
using a large-​signal disc model.
vii) The synchronous frequency is chosen to lie within the operating band of the
tube. It will usually lie close to the centre of the intended band, though, that
is not always the best choice. The synchronous point can be adjusted, if neces-
sary, to ensure that the maximum gain occurs at the centre frequency.
viii) The slow-​wave structure is designed to achieve the correct phase velocity at
the synchronous point. The sheath helix model can be used for initial calcula-
tions because the phase velocity is not seriously in error, and the chief effect of
errors in the coupling impedance is in the gain per unit length. Assumptions
must be made about the shield radius and the dielectric loading. Vane loading
is not necessary for a tube having an octave bandwidth. From this information
the pitch of the helix can be calculated together with the phase velocity, and
coupling impedance, at frequencies within the working band.
ix) When the preceding steps have been completed sufficient information is avail-
able for the small-​signal and large-​signal performance to be calculated. One
sever should be sufficient to ensure that the tube is stable. Initial estimates of
the lengths of the two sections can be made by using the two-​wave approxima-
tion to find the gain per unit length at the lowest frequency where the electrical
length of the tube is least. The minimum requirement of 33 dB of gain at the 1
dB band edges implies a saturated gain of at least 34 dB at the band centre. If
it is assumed that the launching loss, sever loss and gain compression are each
6 dB then the total small-​signal gain required is 52 dB at the band centre. This
can conveniently be achieved using two sections of equal length so that there
is sufficient gain in the output section to avoid loss of efficiency. The length of
the attenuator can be estimated by noting that tapered loss over three wave-
lengths is normally sufficient to ensure a good match. Thus the total length of
the central attenuator should be at least six wavelengths.
Table  14.3 shows the comparison between the tube parameters calculated
using the method described above (see Worksheet 14.3) and those given by [39].
The methods are essentially the same but there are some differences between the
 TWT Design 547

Table 14.3:  Comparison between the tube parameters given in [39] and those calculated
using Worksheet 14.3

Parameter Bliss [39] WS14.3 Revised Units

Estimated efficiency 15% 15% 26%

Gun perveance 1.96 1.9 1.9 µA V −1.5
Beam voltage (Va ) 7400 7200 5800 V
Beam current ( I 0 ) 1.25 1.16 0.84 A
Filling factor (b a ) 0.5 0.5 0.5
Synchronous helix voltage (Vh ) 6100 6100 4900 V
Normalised beam radius γ e b 0.75 0.7 0.7
Beam radius 1.65 1.73 1.55 mm
Helix radius 3.3 3.46 3.10 mm
Brillouin field 0.059 0.059 T
Peak magnetic field 0.09 0.09 0.09 T
Magnet period 17.8 19.7 18.0 mm
Active length of helix 178 192 172 mm
v p c at synchronism 0.14 0.134 0.120

design choices made. The estimated efficiency was 26% and the calculations were
repeated using this estimate which was found to be self-​consistent. The results of
the revised calculations are shown in the fourth column of the table. Figure 14.26(a)
shows the computed dispersion curves of the helix and the slow space-​charge wave.
Figure  14.26(b) shows the small-​signal gain of a single-​section tube 176  mm in
length calculated using the two-​wave model and assuming a fixed total of 12 dB for
the launching and sever losses.
The performance of a single-​section tube having the parameters in the fourth col-
umn of Table 14.3 was calculated at the centre frequency using Worksheet 14.2. The
saturated output power was 1.1 kW corresponding to an electronic efficiency of
23%. The same result is obtained from (14.56) if the capture ratio is reduced from
0.7 to 0.6. These figures ignore the effects of the cold loss of the helix, of mismatches
at the ends of the helix, and the radial motion of the electrons. Calculations using
Worksheet 14.2 with the data in the second column of Table 14.3 gave a saturated
efficiency of 22%. The design in [39] was for a tube with 33 dB gain at saturation. If
the input and output sections of the circuit are assumed to be of equal length then
the gain after the sever may be insufficient to allow the full efficiency to be reached.
The design assumption of 15% efficiency may reflect that expectation. The high
electronic efficiency of the tube means that little improvement could be achieved by
tapering the helix.
The overall efficiency could be increased by the collector depression using the
spent-​beam curve computed by the large-​signal disc model. A drift section can be
added at the end of the tube to allow the bunches to disperse under the influence
548 Travelling-Wave Tubes

Figure 14.26:  S-​band helix TWT: (a) dispersion curves for the helix and for the slow space-​
charge wave, and (b) small-​signal gain of a single section versus frequency.

of space-​charge forces. However, debunching does not affect the overall shape of
the spent-​beam curve very strongly [73]. In practice, the shapes of the spent-​beam
curves for TWTs vary with the design of the tube, and with frequency within the
operating bandwidth. Figure 14.27 shows the spent beam curve at saturation for
the tube whose parameters are in the fourth column of Table  14.3. At present
there is no simple model of the large-​signal behaviour of a TWT corresponding to
those used in this book for klystrons and IOTs. However, the maximum normalised
electron energy is typically 1.1 over a range of drive levels, and the spent-​beam
curve for helix TWTs is often approximately linear, as shown by the dashed line in
Figure 14.27 [73]. With these assumptions the power in the spent beam is

1  V 
Pspent = Pdc 1.1 + s  , (14.75)
2  Va 
where Vs is the energy of the slowest electrons in eV. The electronic efficiency is then

1 V 
ηe = 1 − 1.1 + s  (14.76)
2 Va 
or

Vs
= 0.9 − 2 ηe . (14.77)
Va
For the tube modelled above ηe = 0.23 at saturation so that from (14.77) the nor-
malised energy of the slowest electron is Vs Va = 0.44 , which is close to the value
shown in Figure 14.27. The energy of the slowest electron can be estimated in the
same way when the tube is operated with reduced output power. This information
 TWT Design 549

Figure 14.27:  Spent beam curve at saturation for the S-​band helix TWT.

can be used to construct an approximate model which reveals some of the features
of depressed collector operation of a TWT (see Worksheet 10.4).
An alternative way of estimating the overall efficiency of a TWT with a depressed
collector was suggested by Kosmahl [37]. The method is based on a trapezoidal
approximation to the spent-​beam curve such that the normalised energy of the
slowest electron is given by

Vs
≈ 1 − f ( K ) 3 ηe K , (14.78)
Va
where K is the microperveance and f ( K ) is an empirical function plotted in the
paper. Fitting a second-order function to the curve plotted gives

f ( K ) ≈ 0.073K 2 − 0.37 K + 1.25. (14.79)

Assuming that K = 1.9 and ηe = 0.23 in (14.78) gives Vs Va = 0.39 which is close to
the figure calculated above. Further information about the design of TWT collec-
tors can be found in [73–​77].

14.4.2 Millimetre-​Wave Helix TWTs

The design of millimetre wave TWTs with helix slow-​wave structures is discussed
in [36, 78–​80]. The discussion which follows examines the design of the 90 W, 34
GHz pulsed TWT described in [80]. The conversion efficiency at saturation of this
tube was 6.1% so that a DC beam power of 1.47 kW was required. The perveance
chosen was 0.08 µA V −1.5 , corresponding to a beam current of 115 mA at a volt-
age of 12.75 kV. The beam voltage is unusually high for a helix TWT and it has a
number of disadvantages: reduced coupling impedance; increased risk of voltage
breakdown; and increased problems with backward wave oscillations. It does, how-
ever, allow the radii of the beam and the helix to be greater than if a lower voltage
had been chosen. The normalised beam and helix radii were 0.7 and 1.4. These are
550 Travelling-Wave Tubes

a little larger than those suggested by Cutler for maximum efficiency [38] but are
close to the choice recommended in [39]. The Brillouin field for this beam is 0.124
T, while the RMS PPM field was 2.26 T leading to a much greater beam stiffness
than would normally be used at lower frequencies. The reason for this is that in
such a small beam the effects of transverse thermal velocities are much greater than
those of space-​charge forces [36, 78]. When thermal velocities are important, the
magnetic field required to confine a beam with equilibrium radius b is given by [81]
2 4
8m0 kT  rc   rc  2
B 2 = BB2 +   +   Bk , (14.80)
e 2 b2  b   b 
where BB is the Brillouin field in the absence of thermal velocities. This equation is
equivalent to (7.54) with the addition of the second term to account for the effects
of thermal velocities. This term becomes increasingly important when an electron
beam is scaled to higher frequencies, keeping the plasma frequency and the area con-
vergence of the gun constant, because of the presence of the beam radius in the
denominator. In the case of the tube described above the second term corresponds to
a magnetic field of 0.054 T at a cathode temperature of 1100 °K, which is comparable
with the Brillouin field. To minimise this term it is desirable to make the area conver-
gence of the gun as small as possible, consistent with acceptable cathode loading. In
this case the area convergence was 35:1 with a cathode loading of 2.4 A cm2. Thus the
maximum perveance of the electron gun in a millimetre wave tube is determined by
the current which can be collimated by the maximum practical magnetic field.
To obtain stable PPM focusing the ratio of the plasma wavelength to the PPM
period was 3.4. The 80 W, 30–​35 GHz pulsed helix TWT described by Ohtomo
et al. had very similar parameters [79]. At millimetre wavelengths the cold loss of
the helix has a major effect on the efficiency. For example, the efficiency of a 0.5
W, 54 GHz, TWT was increased by a factor of 3 by plating the molybdenum helix
with copper [78].
More recently the design of a 94 GHz pulsed TWT with saturated output
power greater than 150 W was described in [82]. This tube had an inter-​digital line
slow-​wave structure and a beam perveance of 0.058 µA V −1.5 , giving a conver-
sion efficiency of 5.5%. The data given in this paper suggests that its parameters
were approximately γ e b = 1.9 and ω q ω = 0.022 giving a theoretical efficiency of
8.4% corresponding to an effective capture ratio (including the effects of losses) of
0.65. The overall efficiency was increased to over 15% by the use of a single stage
depressed collector. Other work aimed at the development of TWTs at still higher
frequencies is reviewed in [83].

14.4.3 High Efficiency Helix TWTs

A very important application of helix TWTs is in the field of satellite communica-
tions both as uplink and downlink power amplifiers. The requirements for space
tubes are especially demanding since they must combine [5]:
 TWT Design 551

• Operating lifetime greater than 15  years in order to keep the number of spare
tubes carried to a minimum.
• High total electrical efficiency because of the limited power available on the sat-
ellite and the difficulty of dissipating waste heat in space.
• High linearity to avoid co-​channel interference arising from intermodulation
products.
• Low mass to minimise the cost of launching the satellite.
The advances in the performance of these tubes since they were first used in Telstar
1 in 1962 have been remarkable [5, 84–​86]. The state of the art is represented by
tubes giving continuous powers of 200 W at Ku-​band (10.7–​12.75 GHz) and 130 W
at Ka-​band (17–​22 GHz) with overall efficiencies of 68% and 66% respectively [5].
It should be noted that the operating bandwidths of these tubes are small compared
with the bandwidth of which helix TWTs are capable. A detailed discussion of the
design of space TWTs is given in [87]. Table 14.4 shows some examples of tubes
whose design has been described in the literature.
The lifetime of a TWT depends chiefly on the lifetime of the cathode, which is
determined by the point at which the production of free barium is no longer suf-
ficient to support the desired current density (see Section 18.5). M-​type dispenser
cathodes have achieved lifetimes greater than 100,000 hours (11.4 years) with con-
tinuous operation at up to 2 A cm−​2 [86]. The lifetime of a cathode depends not only
upon its construction but also on the current density. This is because a high current
density requires a high operating temperature that causes the evaporation rate of
the barium to increase. For this reason the electron guns used in space TWTs gener-
ally have low perveance because this allows the cathode loading to be small without
high area convergence. A further limitation is set by the PPM focusing which means
that the perveance must be less than about 0.5 µA V −1.5 , depending upon the type
of permanent magnets used.

Table 14.4:  Examples of TWTs for space communications

Frequency (GHz) Power(W) Overall efficiency Application Reference

4.0–​4.2 3.5 34% Downlink [88]

17.7–​21.2 75 45% Downlink [89]
27.5–​29.5 40 25% Uplink [90]
5.85–​6.425 3000 24% Uplink [91]
14.0–​14.5 600 26% Uplink [91]
17.7–​21.2 75 54% Downlink [92]
10.7–​12.7 65 60% Downlink [93]
59–​64 20 40% Downlink [94]
3.4–​4.2 70 70% Downlink [95]
10.9–​12.75 130 72% Downlink [95]
552 Travelling-Wave Tubes

The overall efficiency of a space TWT depends upon the design of the slow-​
wave structure and also on the performance of the multi-​element depressed col-
lector. We have seen that the efficiency of a tube with constant helix pitch depends
principally upon the beam perveance. For example, a tube with a beam perveance
of 0.2 µA V −1.5 and a uniform helix might have an electronic efficiency of about
11%. This could be increased by making the synchronous point below the centre
frequency but with the penalty of increased non-​linearity. The use of a simple taper
would increase the efficiency to around 15% but, again, with a reduction in linear-
ity. The design goals of high efficiency, and high linearity, are linked because it is
necessary to operate the tube some way below its saturated output power to achieve
acceptable linearity. Thus, to take an example, if output backoff of 3 dB is neces-
sary to achieve acceptable linearity, the operating efficiency is half the saturated
efficiency. A method for determining the operating parameters of a tube to give the
best efficiency for a given linearity is described in [87].
It has been found that it is possible to increase the efficiency and the linearity of a
TWT simultaneously by using the double taper profile as shown in Figure 14.28(a).
We have seen that when the electron velocity is reduced relative to the phase velocity
of the helix the maximum RF beam current increases, and the difference in the
phase between small-​signal conditions and saturation is reduced (see Figure 14.20).
The same thing happens when the phase velocity of the helix is increased, while
keeping the electron velocity constant. Thus, the effect of the section of helix with
increased phase velocity is to draw the electrons into the bunch more effectively, in
a manner similar to that of the detuned penultimate cavity in a klystron. It is found
that the section can be designed so that the phase lag introduced decreases with
increasing drive level. Hence the section can simultaneously increase the electronic
efficiency, by improving the bunching, and increase the linearity by reducing the
AM/​PM conversion, compared with a similar tube with a conventional negative
taper. The optimisation of a tube with a double taper is a complex multi-​variable
problem but a satisfactory initial design can be achieved systematically using a
large-​signal disk model [56, 96, 97]:
i) A uniform TWT is modelled to find the helix phase velocities giving maximum
gain (v p0 ), maximum bunching (v p1 ), and maximum efficiency (v p2 ) at the cen-
tre frequency of the band over which the tube is to work. This is normally
achieved by changing the pitch of the helix.
ii) A TWT is modelled at the centre frequency with the pitch profile shown in
Figure 14.28(b). The input signal level is chosen to be the drive at which the
tube is intended to saturate. The length z0 is chosen to give at least 20 dB of
gain so that the growing wave is fully established by the end of the section. The
position z1 is found within a second section of arbitrary length which maxim-
ises the amplitude of the first harmonic of the RF beam current.
iii) A TWT is modelled at the centre frequency with the pitch profile shown in
Figure  14.28(a), and the same drive level as before. The length of the third
section ( z12 ) is adjusted so that the phase of the bunch relative to the wave on
 TWT Design 553

Figure 14.28:  (a) Double-​taper profile used in high-​efficiency space TWTs, and (b) single-​step
taper used to design the taper in (a).

the helix is constant as the bunch loses energy. This is essentially a dynamic vel-
ocity taper (see Section 14.3.7). It has been found that a good initial approxi-
mation is to make this section at least three wavelengths long.
iv) The length of the fourth section is chosen so that the output power just reaches
saturation at the centre frequency.
v) Finally, the phase velocity of the first section may be adjusted so that the gain
is flattened across the frequency band, which may be up to 20% of the centre
frequency.
All modern space TWTs employ double tapers of this kind, together with high
efficiency multi-​element depressed collectors [5, 84, 85], to achieve high operating
efficiency at a working point which is backed off from saturation by a few decibels.
The efficiency can be increased still further by reducing the cold loss of the helix by
adding a high-​conductivity coating, and by using support rods with low dielectric
losses [92–​94].

14.4.4 Ultra-​Broad-​band TWTs
We have seen that a TWT having an octave bandwidth can be designed using a helix
slow-​wave structure. The bandwidth is then limited by the dispersion of the struc-
ture. However, when the dispersion of the structure is modified by vane or dielectric
loading (see Section 4.3.2) then synchronism between the beam and structure waves
can be maintained over a much greater bandwidth. Tubes have been built in this
way whose bandwidth exceeds three octaves [98]. The design of tubes of this kind
presents its own challenges [99]:
554 Travelling-Wave Tubes

Figure 14.29:  Double taper used in very wide bandwidth helix TWTs

(copyright 2009, IEEE. Reproduced, with permission, from [103]).

• The slow-​wave structure must have negative dispersion so that synchronism with
the slow space-​charge wave is maintained over a broad band.
• The gain in the output section must be equalised to avoid forward wave oscil-
lations at frequencies where the gain is greatest, and the section must be long
enough to give good efficiency at all frequencies.
• Means must be provided to suppress backward wave oscillations.
• The generation of harmonics must be suppressed.
The first method proposed for achieving a very broad bandwidth used a slow-​
wave structure in which a first section with positive dispersion was followed by a
section with negative dispersion [98]. The section with negative dispersion (vane
loading) was optimised for efficiency in the upper half of the band, while the
unloaded first section provided gain at the lower frequencies. It was found that
the harmonic performance was improved considerably if the relative phase veloci-
ties of the two sections were chosen correctly. This concept has recently been
re-​examined [100].
A uniform slow-​wave structure can be used if it is designed so that the increased
gain at high frequencies, caused by the greater electrical length, is compensated
by reduction in synchronism and/​or coupling impedance [101, 102]. Methods for
suppressing harmonics and backward-​wave oscillations have been reviewed in
Sections 14.3.5 and 14.3.8 respectively. Tubes designed in this way tend to suf-
fer from low efficiency at high frequencies because the electron bunch is divided
into slow electrons which are trapped by the wave, and fast electrons which are
not. The use of a positive phase velocity taper has been found to improve the
efficiency at the top of the operating band and, therefore, the useful bandwidth
of the TWT [55, 103]. This technique has also been found to reduce the output
power at the second harmonic. Further improvements in the uniformity of the
efficiency over a wide bandwidth have been obtained using the helix pitch profile,
shown in Figure 14.29, in which a section of helix with reduced pitch to improve
the bunching is followed by a short positive taper [103]. As an alternative to vari-
ation in the pitch of the helix the vane loading may be varied to achieve the same
velocity profile [104].
 TWT Design 555

14.4.5 Coupled-​Cavity TWTs
General reviews of the performance and design of coupled-​cavity TWTs are given in
[2, 14, 105]. The design procedure for coupled-​cavity TWTs follows the same general
pattern as that for helix TWTs. It is necessary to decide, first, whether a forward
fundamental or a space harmonic structure is required. The normalised beam radius
( γ e b) should be around 0.6 to ensure good efficiency. The point at which the slow-​
space charge wave is synchronous with the circuit wave is chosen so that the hot
band of the tube is centred within the cold pass-​band of the structure. Typically,
β0 p ~ 0.6 π for a forward fundamental tube, and 1.5π for a space-​harmonic tube.
The width of the cold pass-​band of the structure is usually around twice that of the
hot band, but it is possible for a tube to have useful output power outside the cold
pass-​band of the structure. The slow-​wave structure can then be designed, and the
final design checked, by small-​signal and large-​signal gain calculations using discrete
interaction models. If the tube is to be PPM focused, as is normal except at the
highest power levels, then the PPM system design must proceed in parallel with the
design of the slow-​wave structure to ensure that the two are mutually compatible.
Particular care is necessary to ensure that the tube is stable against oscillations at the
cut-​off frequencies of the structure, and in higher-​order modes (see Section 14.3.8).
The design of tapers for coupled-​cavity TWTs is discussed in [44]. In an unta-
pered tube the frequencies for maximum gain and maximum efficiency are differ-
ent. The gain at maximum efficiency may be too small for the tube to be saturated
with the available drive power. An important function of the taper in a coupled-​
cavity TWT is, therefore, to align the gain and efficiency bands. Figure 14.30 shows,
for a particular tube, the improvement in efficiency given by different step tapers at
the frequencies for maximum bunching ( fb ), and maximum gain ( f g ). The improve-
ment is greatest at maximum bunching, but also varies markedly with the change in
the phase velocity in the taper (v pt ). At maximum gain the improvement is smaller
and fairly insensitive to the change in phase velocity. At maximum efficiency ( fe )

Figure 14.30:  Increase in efficiency produced by a taper at the frequencies of maximum

bunching and maximum gain
(copyright 1991, IEE, reproduced, with permission of The IET, from [44]).
556 Travelling-Wave Tubes

Figure 14.31:  Saturated efficiency versus frequency curves for a tube without a taper, with
an 80% one-​step taper and with an ideal differential taper showing the conditions for
maximum bunching, gain and efficiency
(copyright 1991, IEE, reproduced, with permission of The IET, from [44]).

a taper produced no further improvement. Clearly an ideal taper would give maxi-
mum efficiency at every frequency within the band of the tube. Figure 14.31 shows
the effect on the efficiency of a single step taper, chosen to give the greatest effi-
ciency at the frequency of maximum bunching, and of an ideal differential taper.
In a coupled-​cavity TWT the dispersion diagram of the slow-​wave structure can
be changed by:
• changing the pitch of the structure;
• changing the resonant frequency of the cavities;
• changing the strength of the coupling between the cavities.
Figure 14.32(a) shows the dispersion diagram of an untapered space-​harmonic
structure, together with the dispersion line for an ideal differential taper in which
the change in phase velocity is optimum at each frequency. Figure  14.32(b)
shows  the dispersion diagrams corresponding to each method of tapering com-
pared with the standard structure. Evidently a dispersion diagram which is close to
the ideal curve can be produced by using a combination of the different possible
changes to the dimensions of the structure. This approach suggests that in design-
ing a coupled-​cavity TWT:
• The basic design should be such that fb and fe lie at the bottom and top of the
frequency band of the tube. Then f g will be in the centre of the band.
• The taper should be designed to give maximum efficiency at  fb.
It was found that good results were obtained by this method [44]. Table 14.5 shows
some examples of TWTs with coupled-​cavity, and other discrete interaction slow-​
wave structures, whose design has been described in the literature.

14.4.6 Hybrid Tubes
The bunched electron beams produced in both klystrons and TWTs are not ideal
because not all the electrons are grouped into the bunches, and because of the spread
 TWT Design 557

Figure 14.32:  Dispersion diagrams for a typical coupled-​cavity slow-​wave structure

showing: (a) the standard circuit and the ideal differential taper, and (b) the standard
circuit (solid line), cavity period taper (dotted line), cavity resonance taper (dashed line)
and slot length taper (chain dotted line)
(copyright 1991, IEE, reproduced, with permission of The IET, from [44]).

Table 14.5:  Examples of TWTs with coupled-​cavity and other discrete interaction slow-​wave structures

Frequency Bandwidth Power Efficiency Slow-​wave Reference

GHz % % structure

2.75 10 1.0 MW pk 13 Space harmonic [106]

2.6–​2.95 12 4.2 MW pk 37 Cloverleaf [107]
5.4–​5.9 9 4.1 MW pk 34 Cloverleaf [107]
9.0 12 1.2 MW pk 25 Centipede [108]
11.9–​12.1 14 240 W cw 30 Space harmonic [109]
3 14 160 kW pk 35 Stub-​supported ring [110]
5.25–​5.75 9 26 kW pk 16 Inverted [71]
space-​harmonic
90 20 100 W cw 2 Double ladder [111]
22.3–​24.8 7 1 kW 23 Space harmonic [112]
28.3–​30.0 2 500 W cw 31 Double staggered [113]
slot
90 1 400 W pk 15 Folded waveguide [114]
33.5 7 700 W pk 7 Folded waveguide [115]

of electron energies. A possible way of overcoming this is to create bunches which

have near ideal properties before extracting the energy from them in a standing-​wave,
or a travelling-wave, output structure. The IOT is a simple example of a device of
this kind which should have higher efficiency than a conventional klystron or a TWT.
This concept was investigated theoretically by Meeker and Rowe [116] who showed
558 Travelling-Wave Tubes

that high efficiencies should be attainable, but that they may be considerably reduced
by spreading of the bunches caused by space-​charge forces. In an experimental inves-
tigation Lichtenberg created bunches at 1 GHz using grid modulation, together with
an inductively tuned cavity [117]. The bunches were injected into a helix slow-​wave
structure which was also fed with an input signal. It was found that when the system
was optimised the peak efficiency was around 40%, which was double that of the
TWT on its own. The key finding was that it was necessary for the initial bunches to
have both current and velocity modulation to counteract the effects of space-​charge.
The propagation characteristics of a beam of this kind were investigated in [118].
It is difficult to form a prebunched beam using grid modulation at frequencies
much above 1 GHz because of transit-​time effects. A possible alternative is to com-
bine the properties of a klystron bunching section with a TWT output section.
We have seen that a very high degree of bunching with little velocity spread can
be achieved in a well-​design klystron. A  hybrid tube of this kind is known as a
Twystron®. This tube, operating at megawatt power levels, can achieve bandwidth,
efficiency and gain flatness superior to those which can be realised in either a klys-
tron or a TWT alone [105, 119–​121].
The development of cold cathode technologies has offered the possibility of pre-
modulating an electron beam at higher frequencies [122]. An experimental TWT of this
kind developed an output power of up to 280 mW in the range from 3.9 to 6.75 GHz
and this has been increased to around 50 W by subsequent development [123, 124].

References

[1]‌ S. O. Wallander, ‘Reflexions and gain ripple in TWT’s’, IEEE Transactions on Electron
Devices, vol. 19, pp. 655–​660, 1972.
[2]‌ J. T. Mendel, ‘Helix and coupled-​cavity traveling-​wave tubes’, Proceedings of the IEEE,
vol. 61, pp. 280–​298, 1973.
[3]‌ L. Sivan, Microwave Tube Transmitters. London: Chapman and Hall, 1994.
[4]‌ J. H. Booske et al., ‘Traveling-​wave tubes’, in R. J. Barker et al., eds, Modern Microwave
and Millimetre-​Wave Power Electronics. Piscataway, NJ: IEEE, pp. 171–​245, 2005.
[5]‌ G. Faillon et al., ‘Microwave Tubes’, in J. A. Eichmeier and M. K. Thumm, eds, Vacuum
Electronics: Components and Devices. Berlin: Springer-​Verlag, pp. 1–​84, 2008.
[6]‌ D. K. Abe and J. P. Calame, ‘Advanced materials technologies’, in R. J. Barker et al., eds,
Modern Microwave and Millimeter-​Wave Power Electronics. Piscataway, NJ: IEEE Press,
pp. 649–​689, 2005.
[7]‌ E. D. Maloney, ‘Developments in earth-​station and satellite tubes reflect information
demand’, Microwave Systems News, pp. 48, 50, 52–​55, March 1984.
[8]‌ A. S. Gilmour, Jr., Principles of Traveling Wave Tubes. Norwood, MA: Artech House,
1994.
[9]‌ D. K. Abe and J. P. Calame, ‘Advanced material technologies’, in R. J. Barker et al., eds,
Modern Microwave and Millimetre-​Wave Power Electronics. Piscataway, NJ: IEEE Press,
pp. 649–​689, 2005.
[10] B. R. Potter et al., ‘High-​power printed circuit traveling wave tubes’, in International
Electron Devices Meeting, pp. 521–​524, 1973.
 References 559

[11] D. M. Goebel et  al., ‘Gain stability of traveling wave tubes’, IEEE Transactions on
Electron Devices, vol. 46, pp. 2235–​2244, 1999.
[12] D. M. Goebel, ‘Theory of long term gain growth in traveling wave tubes’, IEEE
Transactions on Electron Devices, vol. 47, pp. 1286–​1292, 2000.
[13] D. Chernin et al., ‘“Power holes” and nonlinear forward and backward wave gain com-
petition in helix traveling-​wave tubes’, IEEE Transactions on Electron Devices, vol. 50,
pp. 2540–​2547, 2003.
[14] J. F. Gittins, Power Travelling-​Wave Tubes. London: English Universities Press, 1965.
[15] J. C. Slater, Microwave Electronics. New York: D. van Nostrand, 1950.
[16] N. H. Pond and R. J. Twiggs, ‘Improvement of traveling-​wave tube efficiency through
period tapering’, IEEE Transactions on Electron Devices, vol. 13, pp. 956–​961, 1966.
[17] G. S. Kino et al., ‘Small-​signal and large-​signal theories for the coupled-​cavity TWT’,
in 6th International Conference on Microwave and Optical Generation and Amplification,
Cambridge, UK, pp. 49–​53, 1966.
[18] P. K. Tien et al., ‘A large signal theory of traveling-​wave amplifiers’, Proceedings of the
IRE, vol. 43, pp. 260–​277, 1955.
[19] J. E. Rowe, Nonlinear Electron-​Wave Interaction Phenomena. New  York:  Academic
Press Inc., 1965.
[20] N. J. Dionne, ‘Harmonic generation in octave bandwidth traveling-​wave tubes’, IEEE
Transactions on Electron Devices, vol. 17, pp. 365–​372, 1970.
[21] D. Chernin et al., ‘A three-​dimensional multifrequency large signal model for helix trav-
eling wave tubes’, IEEE Transactions on Electron Devices, vol. 48, pp. 3–​11, 2001.
[22] V. Srivastava and S. N. Joshi, ‘Improved nonlinear model for multisignal analysis of
helix TWTs’, IEE Proceedings H, Microwaves, Antennas and Propagation, vol. 139, pp.
129–​134, 1992.
[23] V. Srivastava and R. G. Carter, ‘A fast large-​signal model for coupled-​cavity TWTs’,
IEEE Transactions on Electron Devices, vol. 35, pp. 2068–​2076, November 1988.
[24] D. J. Connolly and T. A. O’Malley, ‘Computer program for analysis of coupled-​cavity
traveling-​wave tubes’, NASA Lewis Research Center, Cleveland, OH, 1977.
[25] J. R. M. Vaughan, ‘Calculation of coupled-​cavity TWT performance’, IEEE Transactions
on Electron Devices, vol. 22, pp. 880–​890, 1975.
[26] S. K. Datta et al., ‘Nonlinear Eulerian hydrodynamical analysis of helix traveling-​wave
tubes’, IEEE Transactions on Electron Devices, vol. 45, pp. 2055–​2062, 1998.
[27] J. G. Wöhlbier et al., ‘The multifrequency spectral Eulerian (MUSE) model of a travel-
ing wave tube’, IEEE Transactions on Plasma Science, vol. 30, pp. 1063–​1075, 2002.
[28] N. Brealey, ‘Computer simulation of helix travelling wave tubes’, in High Power
Microwave Generation and Applications, SIF, Bologna, Italy, pp. 549–​556, 1992.
[29] C. L. Kory et al., ‘Traveling-​wave tube amplifier model to predict high-​order modulation
intersymbol interference’, in Third IEEE International Vacuum Electronics Conference,
Monterey, CA, pp. 308–​309, 2002.
[30] Z. Zhu et  al., ‘Particle-​ in-​
cell simulation of helix traveling-​ wave tube’, in IEEE
International Vacuum Electronics Conference, Rome, Italy, pp. 131–​132, 2009.
[31] V. Srivastava and S. N. Joshi, ‘One-​dimensional nonlinear model for helix TWTs’, IETE
Technical Review, vol. 6, pp. 500–​507, 1989.
[32] G. Dimonte and J. Malmberg, ‘Destruction of trapping oscillations’, Physics of Fluids,
vol. 21, pp. 1188–​1206, 1978.
[33] E. Buckingham, ‘On physically similar systems; illustrations of the use of dimensional
equations’, Physical Review, vol. 4, pp. 345–​376, 1914.
560 Travelling-Wave Tubes

[34] M. J. Schindler, ‘Can traveling-​wave tubes be scaled? ’, Microwave Journal, vol. 9, pp.
43–​47, 1966.
[35] C. C. Cutler and D. J. Brangaccio, ‘Factors affecting traveling wave tube power capac-
ity’, Transactions of the IRE Professional Group on Electron Devices, vol. PGED-​3, pp.
9–​24, 1953.
[36] W. E. Danielson et al., ‘Design of a 100 mW helix travelling-​wave amplifier at 50 Gc/​s’,
Proceedings of the IEE  –​Part B:  Radio and Electronic Engineering, vol. 105, pp.
405–​408, 1958.
[37] H. G. Kosmahl, ‘How to quickly predict the overall TWT and the multistage depressed
collector efficiency’, IEEE Transactions on Electron Devices, vol. 27, pp. 526–​529, 1980.
[38] C. C. Cutler, ‘The nature of power saturation in traveling wave tubes’, Bell System
Technical Journal, vol. 35, pp. 841–​876, 1956.
[39] E. E. Bliss, ‘Traveling-​wave tube design’, in R. S. Burnap, ed., Electron Tube Design.
Harrison, NJ: Radio Corporation of America, pp. 898–​928, 1962.
[40] A. H. W. Beck, Space-​Charge Waves and Slow Electromagnetic Waves. London: Pergamon
Press, 1958.
[41] J. J. Caldwell, Jr. and O. L. Hoch, ‘Large signal behavior of high power traveling-​wave
amplifiers’, IRE Transactions on Electron Devices, vol. 3, pp. 6–​17, 1956.
[42] O. Sauseng, ‘Efficiency enhancement of travelling wave tubes by velocity resynchro-
nisation’, in 7th International Conference on Microwave and Optical Generation and
Amplification, Hamburg, Germany, pp. 16–​29, 1968.
[43] L. Winslow, ‘Phase velocity dispersion shaping as a design parameter in traveling wave
tubes’, in International Electron Devices Meeting, pp. 350A–350C, 1977.
[44] V. Srivastava and R. G. Carter, ‘Design of phase velocity tapers in coupled-​cavity TWTs’,
IEE Proceedings H Microwaves, Antennas and Propagation, vol. 138, pp. 469–​474, 1991.
[45] R. W. Gerchberg and K. B. Niclas, ‘The positively tapered traveling-​wave tube’, IEEE
Transactions on Electron Devices, vol. 16, pp. 827–​828, 1969.
[46] A. I. Denisov, ‘Effect of sectionalisation on the saturation power of a traveling wave
tube’, Radio Engineering and Electron Physics, vol. 6, pp. 1416–​1417, 1961.
[47] A. I. Denisov, ‘The effects of sectionalization on saturation power of TWT for large
values of the space charge parameter’, Radio Engineering and Electron Physics, vol. 11,
pp. 617–​625, 1966.
[48] A. W. Scott, ‘Why a circuit sever affects traveling-​wave tube efficiency’, IRE Transactions
on Electron Devices, vol. 9, pp. 35–​40, 1962.
[49] V. Srivastava and R. G. Carter, ‘Determination of sever positions in a coupled-​cavity
TWTs’, IEE Proceedings H Microwaves, Antennas and Propagation, vol. 138, pp.
55–​60, 1991.
[50] A. Sangster, ‘Traveling-​wave interaction in structures with non-​zero impedance at har-
monics of the drive frequency’, in Proc. 6th Int. Conf. Microwave and Optical Generation
and Amplfication, pp. 125–​130, 1966.
[51] O. Sauseng et  al., ‘Reduction of intermodulation distortion with harmonic injection
for wideband traveling-​ wave tubes’, in International Electron Devices Meeting, pp.
411–​414, 1975.
[52] E. Ezura and T. Kano, ‘Measured and theoretical nonlinear phase distortion in traveling-​
wave tubes’, IEEE Transactions on Electron Devices, vol. 22, pp. 890–​897, 1975.
[53] J. G. Wöhlbier and J. H. Booske, ‘Mechanisms for phase distortion in a traveling wave
tube’, Physical Review E, vol. 69, p. 066502, 2004.
 References 561

[54] D. J. Bates and A. W. Scott, ‘The effect of circuit tapering on the efficiency bandwidth
characteristics of dispersive traveling-​ wave tubes’, IEEE Transactions on Electron
Devices, vol. 10, pp. 89–​94, 1963.
[55] S.-​S. Jung et al., ‘Positive phase-​velocity tapering of broadband helix traveling-​wave
tubes for efficiency enhancement’, Applied Physics Letters, vol. 80, pp. 3000–​3002,
2002.
[56] V. Srivastava et  al., ‘Design of helix slow-​wave structures for high efficiency TWTs’,
IEEE Transactions on Electron Devices, vol. 47, pp. 2438–​2443, December 2000.
[57] D. J. Connolly, ‘Efficiency enhancement of coupled-​cavity TWT’s through cavity reson-
ance tapering’, IEEE Transactions on Electron Devices, vol. 26, pp. 1576–​1580, 1979.
[58] L. Winslow, ‘RF loss as a design parameter in traveling wave tubes’, in International
Electron Devices Meeting, pp. 374–​376, 1974.
[59] H. Hirata, ‘Analysis of phase and intermodulation distortion of a travelling-​wave tube’,
International Journal of Electronics, vol. 83, pp. 249–​270, 1997.
[60] H. G. Kosmahl and J. C. Peterson, ‘A TWT amplifier with a linear power transfer char-
acteristic and improved efficiency’, NASA Lewis Research Center, 1984.
[61] B. Epsztein and G. Kantorowicz, ‘Suppression of backward-​wave oscillations in multi-
kilowatt helix TWT’s’, in 3rd European Microwave Conference, p. C.11.3, 1973.
[62] C. Hagström and O. Nilsson, ‘Start oscillation current in tapered BWO’S’, in 3rd
European Microwave Conference, p. C.11.5, 1973.
[63] J. L. Zhang et al., ‘Backward-​wave suppression for broadband helix traveling-​wave tubes
using the phase velocity variation of output circuit’, IEEE Transactions on Electron
Devices, vol. 59, pp. 2263–​2267, 2012.
[64] C. E. Hobrecht, ‘Resonant loss for helix traveling wave tubes’, in International Electron
Devices Meeting, pp. 348–​350, 1977.
[65] A. J. Sangster et al., ‘Backward-​wave suppression in a very wide-​band helix travelling-​
wave tube using a slow waveguide filter’, IEE Proceedings H, Microwaves, Antennas and
Propagation, vol. 138, pp. 79–​85, 1991.
[66] C. E. Hobrecht and J. L. Putz, ‘Traveling wave tube oscillation suppression’, Varian
Associates Inc., Palo Alto, CA, 1978.
[67] J. A. Ruetz, ‘Resonant circuit oscillations in travelling-​wave tubes’, in 4th International
Congress on Microwave Tubes, Eindhoven, Netherlands, pp. 94–​98, 1962.
[68] A. J. Bahr, ‘A coupled-​monotron analysis of band-​edge oscillations in high-​power
traveling-​wave tubes’, IEEE Transactions on Electron Devices, vol. 12, pp. 547–​556, 1965.
[69] E. Glass, ‘Suppression of spurious modes in high-​power traveling-​wave tubes’, IEEE
Transactions on Electron Devices, vol. 30, pp. 1798–​1806, 1983.
[70] C. C. Lo, Studies of the Effect of Circuit Tapering on TWT Performance. Stanford, CA:
Stanford University Microwave Lab, 1964.
[71] J. R. Frey and I. Tammaru, ‘A coupled-​cavity TWT operating in the inverted slot mode’,
in International Electron Devices Meeting, pp. 504–​506, 1981.
[72] A. Karp and G. T. Hunter, ‘Higher order modes and instabilities in coupled-​cavity
TWT’s’, IEEE Transactions on Electron Devices, vol. 33, pp. 1890–​1895, 1986.
[73] J. A. Dayton, Jr. et al., ‘Analytical prediction and experimental verification of TWT and
depressed collector performance using multidimensional computer programs’, IEEE
Transactions on Electron Devices, vol. 26, pp. 1589–​1598, 1979.
[74] A. C. Schram, ‘TWT efficiency improvement using multi-​stage collectors’, Microwave
Journal, vol. 18, pp. 31–​33, 81, 1975.
562 Travelling-Wave Tubes

[75] H. G. Kosmahl and P. Ramins, ‘Small-​size 81-​to 83.5-​percent efficient 2-​and 4-​
stage depressed collectors for octave-​ bandwidth high-​ performance TWT’s’, IEEE
Transactions on Electron Devices, vol. 24, pp. 36–​44, 1977.
[76] H. G. Kosmahl, ‘Modern multistage depressed collectors –​a review’, Proceedings of the
IEEE, vol. 70, pp. 1325–​1334, 1982.
[77] P. Ramins and T. A. Fox, ‘90-​to 93-​percent efficient collector for operation of a dual-​
mode traveling-​wave tube in the linear region’, IEEE Transactions on Electron Devices,
vol. 26, pp. 1662–​1664, 1979.
[78] H. L. McDowell et al., ‘A half-​watt CW traveling-​wave amplifier for the 5–​6 millimeter
band’, Proceedings of the IRE, vol. 48, pp. 321–​328, 1960.
[79] R. H. Ohtomo et al., ‘Recent advances in an EHF helix TWT’, in 1984 International
Electron Devices Meeting, pp. 502–​505, 1984.
[80] H. Sloley et al., ‘High power, high frequency helix TWT’s’, in Military Microwaves’ 86,
pp. 360–​365, 1986.
[81] G. Herrmann, ‘Optical theory of thermal velocity effects in cylindrical electron beams’,
Journal of Applied Physics, vol. 29, pp. 127–​136, 1958.
[82] W. Gerum et al., ‘94-​GHz TWT for military radar applications’, IEEE Transactions on
Electron Devices, vol. 48, pp. 72–​73, 2001.
[83] J. H. Booske et al., ‘Vacuum electronic high power terahertz sources’, IEEE Transactions
on Terahertz Science and Technology, vol. 1, pp. 54–​75, 2011.
[84] D. S. Komm et al., ‘Advances in space TWT efficiencies’, IEEE Transactions on Electron
Devices, vol. 48, pp. 174–​176, 2001.
[85] G. Kornfeld and E. Bosch, ‘From history to future of satellite TWT amplifiers’,
Frequenz, vol. 55, pp. 258–​262, 2001.
[86] J. D. Wilson et  al., ‘Advances in space traveling-​ wave tubes for NASA missions’,
Proceedings of the IEEE, vol. 95, pp. 1958–​1967, 2007.
[87] R. Strauss et al., ‘Traveling wave tubes for communication satellites’, Proceedings of the
IEEE, vol. 65, pp. 387–​400, 1977.
[88] M. Bodmer et al., ‘The satellite traveling-​wave tube’, Bell System Technical Journal, vol.
42, pp. 1703–​1748, 1963.
[89] J. Heney and R. Tamashiro, ‘A 20 GHz, 75 watt, helix TWT for space communica-
tions’, in Proceedings of the Sixteenth Annual Electronics and Aerospace Conference and
Exposition, pp. 69–​74, 1983.
[90] H. Hashimoto et  al., ‘A 30 GHz 40 watt helix traveling-​wave tube’, in International
Electron Devices Meeting, pp. 133–​136, 1983.
[91] M. Takahashi et al., ‘Non-​brazed helix TWT attained 3kW output at C-​band and 600W
at Ku-​band’, in International Electron Devices Meeting, pp. 167–​170, 1986.
[92] R. Tamashiro and S. Aldana, ‘60 percent efficient K-​band TWT using a new diamond
rod technology’, in International Electron Devices Meeting, pp. 187–​190, 1989.
[93] H. Safa and A. Pelletier, ‘Efficiency improvements in a 12 GHz-​50 W space TWT’, in
International Electron Devices Meeting, pp. 191–​194, 1989.
[94] G. K. Kornfeld et al., ‘60-​GHz space TWT to address future market’, IEEE Transactions
on Electron Devices, vol. 48, pp. 68–​71, 2001.
[95] W. L. Menninger et al., ‘70% efficient Ku-​band and C-​band TWTs for satellite down-
links’, IEEE Transactions on Electron Devices, vol. 52, pp. 673–​678, 2005.
[96] P. Thouvenin et al., ‘New helix tapers boost space TWT efficiency to 55%, broadband’,
in International Electron Devices Meeting, pp. 477–​480, 1987.
 References 563

[97] M. K. Alaria et  al., ‘Design and development of helix slow-​wave structure for Ku-​
band TWT’, IEEE Transactions on Plasma Science, vol. 39, pp. 550–​554, 2011.
[98] J. P. Rymer and M. J. Cascone, ‘Three octaves with one TWT’, in 1982 International
Electron Devices Meeting, pp. 30–​31, 1982.
[99] P. L. Walchli, ‘Multiple octave traveling wave tubes’, in 1981 International Electron
Devices Meeting, pp. 707–​707, 1981.
[100] S. K. Datta et al., ‘Analytical exploration of ultrawideband helix slow-​wave structures
using multidispersion phase velocity taper’, IEEE Transactions on Plasma Science, vol.
37, pp. 311–​316, 2009.
[101] H. J. Kim et al., ‘Experimental investigation of broadband vaned helix traveling-​wave
tube’, Japanese Journal of Applied Physics, vol. 45, pp. 292–​299, 2006.
[102] S.-​S. Jung et  al., ‘Wide-​band semivane and heavily dielectric loaded helix traveling-​
wave tubes’, IEEE Transactions on Plasma Science, vol. 30, pp. 1009–​1016, 2002.
[103] T. K. Ghosh et al., ‘Design of helix pitch profile for broadband traveling-​wave tubes’,
IEEE Transactions on Electron Devices, vol. 56, pp. 1135–​1140, 2009.
[104] A. B. Danilov et al., ‘A method for reducing the second harmonic level in wideband
traveling wave tubes’, Journal of Communications Technology and Electronics, vol. 58,
pp. 353–​356, 2013.
[105] A. Staprans et al., ‘High-​power linear-​beam tubes’, Proceedings of the IEEE, vol. 61,
pp. 299–​330, 1973.
[106] M. Chodorow et  al., ‘The design and characteristics of a megawatt space-​
harmonic traveling-​wave tube’, IRE Transactions on Electron Devices, vol. 6, pp.
48–​53, 1959.
[107] J. A. Ruetz and W. H. Yocom, ‘High-​power traveling-​wave tubes for radar systems’,
IRE Transactions on Military Electronics, vol. MIL-​5, pp. 39–​45, 1961.
[108] T. Roumbanis et  al., ‘A megawatt X-​band TWT amplifier with 18% bandwidth’, in
High Power Microwave Tubes Symposium, Fort Monmouth, NJ, pp. 114–​129, 1962.
[109] H. G. Kosmahl et  al., ‘High-​efficiency, 200-​watt, 12-​gigahertz traveling wave tube’,
NASA Lewis Research Center, Cleveland, OH, 1974.
[110] H. Desmur et  al., ‘160-​kW pulsed S-​band TWT’, Microwave Journal, vol. 17, pp.
53–​56, 1974.
[111] B. G. James, ‘Coupled-​cavity TWT designed for future mm-​wave systems’, Microwave
Systems News, vol. 16, September 1986.
[112] J. D. Wilson et  al., ‘A high-​efficiency ferruleless coupled-​cavity traveling-​wave tube
with phase-​ adjusted taper’, IEEE Transactions on Electron Devices, vol. 37, pp.
2638–​2643, 1990.
[113] J. R. Legarra et  al., ‘A 500-​W coupled-​cavity TWT for Ka-​band communication’,
IEEE Transactions on Electron Devices, vol. 52, pp. 665–​668, 2005.
[114] A. J. Theiss et al., ‘High-​average-​power W-​band TWT development’, IEEE Transactions
on Plasma Science, vol. 38, pp. 1239–​1243, 2010.
[115] H. Gong et  al., ‘Experimental investigation of a high-​power Ka-​band folded wave-
guide traveling-​ wave tube’, IEEE Transactions on Electron Devices, vol. 58, pp.
2159–​2163, 2011.
[116] J. E. Rowe and J. G. Meeker, ‘Interaction of premodulated electron streams with
propagating circuits’, International Journal of Electronics, vol. 9, pp. 439–​466, 1960.
[117] A. J. Lichtenberg, ‘Prebunched beam traveling-​wave tube studies’, IRE Transactions on
Electron Devices, vol. 9, pp. 345–​351, 1962.
564 Travelling-Wave Tubes

[118] J.-​G. Jin et al., ‘Propagation characteristics of a premodulated electron beam’, IEEE
Transactions on Plasma Science, vol. 26, pp. 794–​798, 1998.
[119] A. LaRue, ‘Development of a broadband S-​band amplifier’, Varian Associates Inc.,
Palo Alto, CA, 1972.
[120] P. S. Matthews et al., ‘Comparison of the performance of a travelling-​wave tube and a
hybrid tube designed for 12% bandwidth, multi-​megawatts output at L-​band’, in IEEE
International Electron Devices Meeting, pp. 134–​137, 1975.
[121] P. S. Matthews et  al., ‘The centipede structure in practical megawatt tubes’, in 8th
European Microwave Conference, Paris, France, pp. 753–​760, 1978.
[122] D. R. Whaley et al., ‘Application of field emitter arrays to microwave power ampli-
fiers’, in 27th IEEE International Conference on Plasma Science, p. 122, 2000.
[123] D. R. Whaley et al., ‘Experimental demonstration of an emission-​gated traveling-​wave
tube amplifier’, IEEE Transactions on Plasma Science, vol. 30, pp. 998–​1008, 2002.
[124] D. R. Whaley et al., ‘Operation of a low-​voltage high-​transconductance field emitter
array TWT’, in IEEE International Vacuum Electronics Conference, Monterey, CA, pp.
78–​79, 2008.
15 Magnetrons

15.1 Introduction

This chapter, and the next, are concerned with tubes which employ electron beams
in crossed electric and magnetic fields, as described in Chapter  8. Magnetron
oscillators are considered in this chapter, beginning with a review of the basic
principles of their operation. We shall refer to these devices as ‘magnetrons’
for convenience. Although both linear and circular geometries are possible (see
Figure 8.1), only the second of these is normally used because it minimises the
volume in which a uniform magnetic field must be maintained. In addition, all
high power devices are based on electron emission from a cylindrical cathode as
illustrated in Figure 8.1(d) because it is not possible to make stable, high current,
injected beams. The rotating cloud of electrons interacts with the RF field of a
slow-​wave structure which also forms the anode as shown in Figure 15.1. A num-
ber of different forms of anode are in use, and these are discussed in Section 15.3.
There follows a review of the properties of magnetrons, and a discussion of ways
of modelling them. The final section of the chapter is concerned with the design
of magnetrons. Crossed field amplifiers, which are closely related to magnetrons,
are dealt with in Chapter 16.
The magnetron was the first type of microwave tube to be used in systems, fol-
lowing its successful development as the power source for radar during World War
II. Modern tubes have the same basic features as the early tubes but are the result
of over seventy years of engineering development [1–​3]. An overall efficiency of
50% is typical for a pulsed tube but figures as high as 90% are achieved in CW
tubes designed specifically for high efficiency. In comparison with a linear beam
tube, at the same frequency and power level, the magnetron is smaller, simpler, and
cheaper, and works at a lower voltage. The tuning range is small, usually 2–​10%.
Magnetrons are, therefore, preferred to linear beam tubes when an oscillator is
suitable and only a narrow tuning range is needed. The main applications of mag-
netrons are in radar, in domestic and industrial microwave ovens, and in medical
linear accelerators [3–​5]. Relativistic magnetrons, which deliver powers up to sev-
eral GW with pulse durations of some tens of nanoseconds, lie outside the scope
of this book [6, 7].
566 Magnetrons

Figure 15.1:  General arrangement of a magnetron.

15.2 Basic Principles

Figure 15.1 shows the general arrangement of a magnetron. The electrons are emit-
ted from a cylindrical cathode. In most types of magnetron an indirectly heated
matrix cathode is used (see Section 18.5.3). The exceptions are tubes for domestic
or industrial microwave ovens in which the cathode is a directly heated thoriated
tungsten helix (see Section 18.5.1) [4]. The axial length of the interaction region
( La ) is normally small compared with the free-​space wavelength to avoid variation
of the interaction in the axial direction. The cathode end hats limit the flow of elec-
trons in the axial direction. The cathode is usually supported by a coaxial extension
(see Figure 15.25). The output power may be removed from the anode by probe,
or iris, coupling to a waveguide and window at right angles to the axis of the tube.
Alternatively, one or more of the anode vanes may be connected to an antenna on
the axis of the tube which radiates into the external waveguide though a domed
window (see Figure 15.26). The other possible arrangement, in which the cathode
is outside the anode, suffers from the disadvantage that the power density on the
anode is greater [8].
The magnetron is based upon the cut-​ off magnetron diode, discussed in
Chapter 8. Before oscillations commence much of the space between the cathode
and the anode is occupied by a rotating cloud of electrons. It is believed the cloud
is in a state of oscillation comprising azimuthal space-​charge modes with a back-
ground of broadband noise (see Section 8.6). However, the properties of magne-
trons can be understood by assuming that the cloud takes the form of a uniform
space-​charge hub whose outer radius is rb , as shown in Figure 15.1. A magnetron
 Basic Principles 567

is always operated at an anode voltage which is less than the Hull cut-​off voltage
(VH ) given by (8.59). Thus, no current can flow through the tube in the absence of
an interaction with the RF field of the anode. The properties of the space-​charge
hub are reviewed in Section 15.2.1.
The anode supports a number of resonant modes which can be decomposed
into pairs of counter-​rotating slow electromagnetic waves. These have radial and
tangential components of the electric field, as discussed in Section 15.2.2. If the
angular velocities of the electrons are approximately synchronous with one of the
travelling waves on the anode a collective interaction can take place. The modes
are initially excited by the fluctuations in the electron cloud. There is competition
between different possible modes of oscillation which grow at different rates. It is
therefore important to ensure that the desired mode eventually dominates. Because
the slow-​wave structure is a closed circle the device is an oscillator. The strength of
the interaction grows until it is limited by non-​linear effects.
The nature of the interaction between the electrons in the hub and the RF field
of the anode is illustrated in Figure 15.2. This shows the motion of the electrons in
a frame of reference which is moving with their steady velocity u0 which is greater
than the phase velocity of the RF electric field of the anode. Planar geometry has
been assumed for simplicity. In this frame of reference the steady electric and mag-
netic forces are in balance. Electrons, which are on the surface of the electron cloud,
move under the combined influence of the RF electric field, and the static magnetic
field, as shown by the arrows in Figure 15.2(a). Their new positions and the new
shape of the electron cloud are shown in Figure  15.2(b). The charge density in
the electron cloud is approximately constant so that the electrons are concentrated
regions of retarding field. The wave on the perturbed cloud resembles a slow space-​
charge wave on a Brillouin focused linear electron beam. Electrons which are in a
retarding phase of the tangential RF electric field of the anode transfer energy to it.
Because their tangential velocity is reduced there is an imbalance between the out-
ward electric force and the inward magnetic force on them and they drift outwards
towards the anode. In so doing they move into a region where their potential energy
is lower so that they gain kinetic energy and are able to maintain synchronism with

Figure 15.2:  Motion of electrons in a magnetron.

568 Magnetrons

the rotating RF field. Conversely, those electrons which are accelerated by the RF
field move inwards towards the cathode. The perturbation of the charge cloud, and
the growth in the magnitude of the RF electric field, are mutually reinforcing. It is
found that the electrons can only reach the anode as a result of the interaction if
the anode voltage is greater than the threshold voltage (VT ) which depends upon the
magnetic field. The RF power delivered to the anode is very small until this volt-
age is exceeded. Thus a working magnetron operates at an anode voltage such that
VT < Va < VH . The equation for the threshold voltage is derived in Section 15.2.3.
When stable oscillation has been established the electron bunches take the form
of ‘spokes’ of charge based on a space-​charge hub, as shown in Figure 15.3. A more
detailed view is shown in Figure 15.36. The spokes move with an angular velocity
which is synchronous with the wave, and the potential energy lost by the electrons
is converted directly into RF energy in the field of the anode. Those electrons which
return to the cathode may strike it with high energies and liberate secondary elec-
trons. The back-​bombardment of the cathode caused in this way means that the
emission from the cathode is a combination of thermionic and secondary electron
emission. It is difficult to determine the fraction of electrons falling into each cat-
egory, but it is known that this can have important effects on the properties of the
magnetron [9]. Electron bombardment also causes heating of the cathode, which
may be compensated by reducing the current in the cathode heater. In some cases it
is possible to operate a magnetron without any external heating of the cathode. The
way in which the electronic efficiency depends upon the magnetic field is explored
in Section 15.2.4.
The description of the operation of a magnetron in terms of an inner space-​
charge hub and spokes of charge is satisfactory as a basis for understanding the
operation of a magnetron and the principal features of its performance. But it is

Figure 15.3:  The space-​charge hub and spokes in an oscillating magnetron.

 Basic Principles 569

important to realise that the actual behaviour of the electrons is much more com-
plex than is assumed in this simple model. We saw in Chapter 8 that, in practice,
the electrons execute complex cycloidal motions about guiding centres which are in
uniform motion. However, this motion is usually rapid compared with the drift of
the guiding centres so that it is effectively averaged out and does not play a major
role in the interaction. This is the justification for using a uniform hub as a model.
It is assumed that the RF field of the anode does not act on the electrons within
the hub. Its properties are therefore unchanged when electrons are drawn from it to
form the spokes. A more detailed discussion of the interaction in a magnetron can
be found in Section 15.6.

15.2.1 Hub Model
The DC potential within the space-​charge hub is given by (8.68)
2
e  r2 
V (r ) = Bz2 r 2 1 − c2  . (15.1)
8m0  r 
We recall that this equation does not depend on any assumptions made about the
space-​charge distribution. The potential at the surface of the hub (Vb ) is found by
setting r = rb in this equation. Outside the hub the potential is given by (8.72)
2
e  r4   r e  r2 
V (r ) = Bz2 rb2 1 − c4  ⋅ ln   + Bz2 rb2 1 − c2  . (15.2)
4 m0  rb   rb  8m0  rb 
Then the radial component of the DC electric field outside the hub is

e  r4  1
E0 = − Bz2 rb2 1 − c4  . (15.3)
4 m0  rb  r
The radius of the hub can be determined in any particular case by numerical solu-
tion of (15.2) when V = Va at r = ra. The potential on the surface of the hub is then
found by substitution into (15.1). The angular velocity of an electron on the surface
of the hub is found to be

2 ( e m0 )Vb
ωb = (15.4)
rb
by using the principle of conservation of energy. It should be noted that this angu-
lar velocity is not normally equal to the angular velocity of the travelling electro-
magnetic wave on the anode.

15.2.2 Interaction Field
Although practical magnetrons have cylindrical geometry, it is helpful to examine
first the planar geometry shown in Figure 15.4 because that is sometimes used as an
570 Magnetrons

Figure 15.4:  Arrangement of a planar magnetron.

approximation to the cylindrical case. The design of magnetron anodes is discussed in

Section 15.3. For the present it is sufficient to note that the slow electromagnetic wave
is generated by the RF voltages between the tips of a regular series of metallic vanes.
Let us assume that a wave travelling in the x direction with frequency ω and
propagation constant β0 has uniform electric field amplitude E g between the vane
tips. Then, from (4.37), the tangential electric field of the mth space-​harmonic wave
on the surface of the anode can be written

w  sin (β m w 2 ) 
E xm ( x, h ) = E g exp j ( ωt − β m x ) , (15.5)
p  β m w 2 

where β m = β0 + 2 m π p and the other symbols are defined in Figure 15.4. Since the
phase velocity of the wave is much less than c, and the transverse dimensions are
small compared with the free-​space wavelength, we can use the quasi-​static approxi-
mation. Then the electrostatic potential of the mth space-​harmonic wave on the
surface of the anode is given by

Vm ( x, h ) = − ∫ E xm ( x, h ) dx
wE g  sin (β m w 2 ) 
= −j   exp j ( ωt − β m x ) . (15.6)
βm p  βmw 2 

Thus the amplitudes of the space-​harmonics of the potential wave on the surface
of the anode are

w  sin (β m w 2 ) 
Vsm = E g . (15.7)
β m p  β m w 2 

If the wave is a standing wave the amplitudes of the forward and backward waves
are E g = Vg 2w where Vg is the magnitude of the RF voltage between the tips of
the vanes. The total potential distribution is then the sum of the space-​harmonic
waves. For the special case of the π mode the m = 0 space harmonic of the forward
wave and the m = −1 space harmonic of the backward wave both have the same
propagation constant and the same amplitude. The amplitudes of all the other
 Basic Principles 571

space-​harmonics with positive phase velocities are smaller so that the RF potential
on the surface of the anode can be approximated by the sum of the equal forward
and backward waves given by

Vg  sin ( πw 2 p ) 
V± ( x, h ) = − j exp j ( ωt  πx p ) , (15.8)
π  πw 2 p 

since β0 p = π. Thus the amplitude of the standing potential wave is

2  sin ( πw 2 p ) 
Vs = Vg . (15.9)
π  πw 2 p 

Figure  15.5 shows a comparison between the RF potential on the surface of

the anode, modelled by the piecewise linear function which we have assumed, and
the potential of the standing sinusoidal wave, when w = p 2. It can be seen that in
this case, which is close to the situation in most magnetrons, the sinusoidal wave is
a good approximation. Thus the peak-​to-​peak amplitude of the wave is approxi-
mately equal to the gap voltage, as one would expect.
The amplitude of the quasi-​static potential of the standing wave in the π -​mode
must satisfy Laplace’s equation in two dimensions, hence

sinh ( π y p )
V ( x, y ) = Vs cos ( πx p ) . (15.10)
sinh ( πh p )
Then, taking the gradient of the potential

π sinh ( π y p )
E x ( x, y ) = Vs ⋅ sin ( πx p ) , (15.11)
p sinh ( πh p )
and

π cosh ( π y p )
E y ( x, y ) = −Vs ⋅ cos ( πx p ) . (15.12)
p sinh ( πh p )

Figure 15.5:  Comparison between the piecewise linear variation of electric potential

on the surface of a magnetron anode, and the sum of the first forward and backward
space-​harmonic waves.
572 Magnetrons

Figure 15.6:  R.F. electric field of the π-​mode of a planar magnetron.

Figure 15.6 shows the electric field of two adjacent cavities when the cavity pitch
is equal to the distance between the cathode and the anode. It should be noted that
the x component of the field is strongest close to the surface of the anode and that
it decays rapidly with increasing distance towards the cathode. This justifies the use
of a single term in the expansion of the field.
The analysis for circular geometry follows the same approach. Laplace’s equation
takes the form

1 ∂  ∂V  1 ∂ 2V
r + = 0. (15.13)
r ∂r  ∂r  r 2 ∂θ 2
If there are n wavelengths around the anode then, separating the variables, the
potential takes the form

V ( r, θ ) = Vr ( r ) cos ( nθ ) . (15.14)

Substituting for V in (15.13) gives

1 ∂  ∂Vr  n2
r  − Vr = 0. (15.15)
r ∂r  ∂r  r 2
It can be shown by substitution that the general solution of this equation is

Vr = Ar n + B r − n (15.16)
where A and B are constants. The relationship between the constants is found by
applying the boundary conditions V = 0 when r = rc , and V = Vs cos ( nθ ) when r = ra.
Substitution into (15.14) gives

 Rn − R−n 
V ( r, θ ) = Vs  n cos ( nθ ) (15.17)
 Ra − Ra− n 
where R = r rc and Ra = ra rc . Then

∂V nV  R n + R − n 
Er = − =− s n cos ( nθ ) (15.18)
∂r Rrc  Ra − Ra− n 
 Basic Principles 573

Figure 15.7:  Variation of the tangential electric field between the cathode and the anode
when Ra = 2 for three values of n.

and

1 ∂V nVs  R n − R − n 
Eθ = − = sin ( nθ ) (15.19)
r ∂θ Rrc  Ran − Ra− n 

Figure 15.7 shows the variation of the tangential electric field between the cathode
and the anode for n = 3, 4, 5 when Ra = 2. It can be seen that the tangential field falls
off rapidly with increasing distance from the surface of the anode and that the rate
of change increases with n. Figure 15.8 shows the arrow plot of the π-​mode field, for
one quadrant of an anode with 12 resonators, obtained from (15.18) and (15.19).
When the phase shift per cavity is π the standing-​wave field is the sum of equal
forward and backward space-​harmonics as in the planar case. The amplitude of the
standing potential wave on the surface of the anode is related to the RF gap voltage
by (15.9) where p = 2 πra Nv , Nv is the number of vanes and the gap between the
vane tips w = p − t . The amplitudes of the travelling waves are then

1  sin (wNv 4ra ) 

V1 = Vg . (15.20)
π  wNv 4ra 

15.2.3 Threshold Condition for Oscillations

The threshold voltage can be calculated by considering the conditions under which
electrons can just reach the anode. The RF field of a cylindrical anode is decom-
posed into a pair of equal waves propagating in opposite directions whose tan-
gential electric fields vary as exp j ( ωt ± nθ ) where n (an integer) is the number of
wavelengths around the anode. The condition for constant phase is

dθ ω
= ± = ± ω s (15.21)
dt n
574 Magnetrons

Figure 15.8:  RF electric field of a 12-​cavity magnetron anode operating in the π-​mode

when ra = 2rc .

so that a plane of constant phase rotates about the axis with angular velocity ω s . When
oscillations are just able to start, the RF field of the anode is negligible, the velocities
of the electrons are tangential to the surface of the anode and synchronous with the
wave. Then, using the principle of conservation of energy, the anode voltage is

m0 2 2
V0 = ra ω s . (15.22)
2e
This is the Characteristic Voltage which is the least anode voltage for which oscilla-
tions are possible. Since the diode is just cut off the corresponding magnetic field is
obtained by substituting V0 in (15.1) and setting r = ra to give

2 m0 ω s 1
B0 = ⋅ (15.23)
e (1 − rc2 ra2 )
This is known as the Characteristic Field. The point ( B0 ,V0 ) lies on the Hull cut-​off
curve (8.59) so that
2
e  r2 
V0 = B02 ra2 1 − c2  . (15.24)
8m0  ra 

Dividing (15.1) by (15.24) the Hull cut-​off curve can be expressed as

Vb Bz2
= , (15.25)
V0 B02
where Vb is the potential at the surface of the space-​charge hub. If the magnetic
field is increased, the diode is more strongly cut off, and the radius of the hub
decreases.
 Basic Principles 575

To show that oscillations can still start when Bz > B0 we consider the dynamics
of the electrons in the presence of the RF field of the wave on the anode. The equa-
tions of motion of an electron in cylindrical polar co-​ordinates are [9]

 d 2r  dθ 
2

m0  2 − r    = Fr (15.26)
 dt  dt  

d  dθ
 m0 r  = rFθ , (15.27)
2
dt  dt 
where Fr and Fθ are the components of the force acting on the electron. Multiplying
(15.26) by dr dt and (15.27) by d θ dt and adding gives

d  m0  dr  dθ  
2 2
2 
dr dθ
   + r     = Fr + Fθ r . (15.28)
dt  2  dt  dt   dt dt

This equation is a statement of the work-​energy theorem. The left-​hand side is the
rate of change of the kinetic energy of the electron with time and the right-​hand side
is the rate of working of the forces acting on it. If the forces are derived from a poten-
tial field the equation can be integrated to obtain the law of conservation of energy.
Unfortunately this is not possible in the present case because the forces vary with time
in a fixed coordinate system. This difficulty can be overcome by changing to a coord-
inate system which is rotating with the synchronous velocity, so that the RF electric
field of the wave can be derived from a static potential. To do this we define

θ ′ = θ − ω s t (15.29)
then (15.26) and (15.27) become

 d 2r  d θ′ 
2
d θ′ 
m0  2 − r   − 2rω s − rω 2s  = Fr (15.30)
 dt  dt  dt 
and

d  d θ′ 
 m0 r
2
+ m0 r 2 ω s  = rFθ . (15.31)
dt  dt 
Then, by the same procedure as before, we obtain the equation in the rotating frame
of reference, which is equivalent to (15.28).

d  m0  dr  d θ′   
2 2
2 
dr d θ′ d  1 
   + r    = Fr + Fθ r +  m0 ω 2s r 2  . (15.32)
dt  2  dt   dt    dt dt dt  2 

Now

 dθ   d θ′ 
Fr = −e  Er + r Bz  = −e  Er + r Bz + rω s Bz  (15.33)
 dt   dt 
576 Magnetrons

and

 dr 
Fθ = −e  Eθ − Bz  , (15.34)
 dt 
where the electric field, which is the superposition of the DC and RF electric fields
in the rotating system of coordinates, does not vary with time.
Substituting for Fr and Fθ in (15.32) we obtain, after simplification,

d  m0  dr  d θ′   
2 2
2 
dr dr d θ ′ d  m0 2 2 
−    + r     = Er + rω s Bz + Eθ r −  ω s r  .
dt  2e  dt  dt    d t dt dt dt  2e 
(15.35)
Integration during the time taken for an electron to move from the cathode to the
anode gives
2
 m  dr   d θ ′   
2 2 2
ra ra θa ′  m0 2 2 
−  0   + r 2 
 

 dt   
 = ∫rc E r dr + ω s Bz ∫rc r dr + ∫θc′ rE θ d θ ′ −  ω s r  .
 2e  dt 1 2e 1

(15.36)
At the cathode dr dt = 0 and d θ ′ dt = −ω s . To find the threshold condition for
oscillation we assume that an electron which just reaches the anode does so with
near-​zero radial and tangential velocities in the rotating frame. Substituting these
boundary conditions into the left-​hand side of (15.36) gives

1 m θa ′
VT = ω s Bz (ra2 − rc2 ) − 0 ω 2s ra2 + ∫ rEθ d θ ′, (15.37)
2 2e θc ′

where
ra
VT = − ∫ Er dr. (15.38)
rc

In the limit of a vanishingly small RF electric field the last term in (15.37) tends to
zero and

1 m
VT = ω s Bz (ra2 − rc2 ) − 0 ω 2s ra2 , (15.39)
2 2e
where VT is the threshold (or Hartree) voltage. This is the minimum DC voltage at
which electrons can reach the anode under the influence of the RF field, for a given
magnetic field. The threshold voltage can be given a simple physical explanation [10].
The first term on the right hand side of (15.39) is the voltage induced in the rotating
spoke, considered as a conductor. The second term is the kinetic energy in electron
volts of a synchronous electron which reaches the anode with zero radial velocity.
Combining (15.39) with (15.22) and (15.23) we obtain

VT 2 Bz
= − 1. (15.40)
V0 B0
 Basic Principles 577

Figure 15.9:  Threshold voltages for the n = 3, 4 and 5 modes of an eight-​cavity magnetron. The
axes are normalised to the characteristic voltage, and the characteristic field, for the n = 4 mode.

It is straightforward to show that (15.40) is a straight line which is tangential to the

cut-​off curve (15.25) at the point ( B0 ,V0 ) . The characteristic field is proportional
to ω s so that each possible mode has its own characteristic point and threshold
line. Figure 15.9 shows the normalised cut-​off curve for an eight-​vane magnetron,
together with the threshold lines for n = 4 (the π mode) and the adjacent modes.
If the magnetic field is greater than the characteristic field for the π mode no cur-
rent can flow until the anode voltage exceeds the threshold voltage for that mode.
Once the threshold voltage has been exceeded oscillations build up from the noise
already present in the electron hub. The operation of the magnetron as an oscillator
requires the anode voltage to remain below the cut-​off voltage for the given mag-
netic field. Thus the possible region of operation lies between the threshold voltage
line and the cut-​off voltage curve. The way in which the choice of operating point
affects the properties of the magnetron are discussed in Section 15.4.
The assumptions involved in the derivation of (15.39) can be questioned. This
is illustrated by Figure 15.10 which shows electron trajectories, in the synchronous
frame of reference, in a magnetron which is just conducting. The trajectories were
computed for a cylindrical magnetron but they are displayed in Cartesian coordi-
nates for convenience. A finite RF electric field was assumed. The tangential veloci-
ties of the electrons on the surface of the hub exceed the synchronous velocity so
that they are initially in the direction of positive θ ′ . The trajectory which finally
reaches the surface of the anode has already travelled through several RF periods.
It eventually reaches a position where the tangential retarding RF electric field is
strong enough to cause the electron to drift towards the anode. The continuing drift
of the electron to the right must be arrested before it leaves the retarding phase
of the field ( 0° − 180°). The drift motion in the synchronous frame of reference
is reversed and the electron then moves outward until it reaches the anode. The
tangential velocity on impact is equal to the synchronous velocity, but the calcula-
tions show that the impact energy is approximately 0.2 eVa. Thus the assumption
that the RF electric field, and the energies of electrons reaching the anode, are
578 Magnetrons

Figure 15.10:  Electron trajectories in a magnetron which is just conducting.

both vanishingly small at the threshold voltage is seen to be false. Nevertheless, the
threshold voltages computed from (15.39) are found to be close to the results of
experimental measurements. It must be concluded that the effect of the finite RF
electric field is balanced by the impact energy of the electron on the anode.
The requirement that the RF electric field should be finite means that the initiation
of oscillations cannot be explained by the theory described above. An alternative pos-
sibility is to consider the threshold condition in terms of the synchronism between the
slow space-​charge wave on the hub and the wave on the anode [11]. We have seen that,
in a circular magnetron, the angular velocity of the electrons on the surface of the
hub ( ω b ) exceeds the synchronous velocity. The angular velocity is synchronous at the
characteristic voltage and magnetic field, and the difference increases as the magnetic
field, and the anode voltage, are increased. In the example shown in Figure  15.10
where Va ≈ VT and Bz = 3.34 B0 it was calculated that ω b = 1.46 ω s.
Space-​charge waves on electron beams in crossed-​fields take the form of pertur-
bations of the surface of the electron beam with constant charge density. Thus they
resemble the space-​charge waves on linear beams with Brillouin focusing. It has
been shown that the propagation constants in a planar magnetron take the form
1

ω  ω  ω 2p 2
2

u0 β ± = ω − c   c  +  , (15.41)
2  2  1 + coth βd 

where d is the distance between the anode and the surface of the hub [12]. We will
assume that this equation is approximately correct for a cylindrical magnetron. The
plasma frequency on the surface of the hub is given by (8.67) so that (15.41) becomes
1

ω ω  2 (1 + rc4 rb4 )  2
u0 β ± = ω − c  c 1 +  . (15.42)
2 2  1 + coth (β ( ra − rb )) 
 Basic Principles 579

Figure 15.11:  Comparison between threshold voltages computed for the 4J50 magnetron
using: (a) (15.39) and (b) (15.44).

Making use of the synchronous condition

n
β− = (15.43)
rb
we find that the threshold voltage is the root of
1

ω ω  2 (1 + rc4 rb4 ) 2
ω − nω b − c + c 1 +  = 0 (15.44)
2 2  1 + coth ( n ( ra rb − 1)) 

for a given value of Bz , since ω b and rb are then functions of Va only. Figure 15.11
shows an example of the threshold voltage computed from this equation, compared
with that given by (15.39). There is excellent agreement between the two curves for
small values of the magnetic field. The difference at larger magnetic fields can be
attributed to the difference between the results for planar and cylindrical magne-
trons as the distance between the anode and the surface of the hub increases. We
saw in Section 8.6 that, in the absence of external RF fields, the space charge hub
is perturbed by oscillations that are strongest when there is an integral number of
wavelengths around the hub. If the anode voltage is set to the value given by (15.44)
then the slow space-​charge wave with mode number n is synchronous with the wave
on the anode. The currents induced in the anode by the space-​charge wave transfer
energy from the wave to the anode. There it is dissipated in resistive losses and in
the external load. Because the slow space-​charge wave carries negative energy its
amplitude increases through the transfer of energy until the point is reached where
the fields are strong enough for spoke formation. Thus space-​charge wave theory
explains why the threshold voltage is given correctly by (15.39) even when the RF
fields are vanishingly small.
580 Magnetrons

15.2.4 Electronic Efficiency

The electronic efficiency of a magnetron can be estimated by noting that the RF
output power is given by

Prf = Pdc − Pa − Pc , (15.45)

where the terms on the right-​hand side are, respectively, the DC input power and
the thermal power dissipated on the anode and on the cathode. Thus the electronic
efficiency is given by

Pa + Pc
ηe = 1 − . (15.46)
Pdc
The power dissipated on the cathode is typically less than 5% of the input power so
that it may usually be neglected in the first approximation (though it can be as great
as 50% at low output powers) [13]. Since the spokes are stationary in the rotating
frame the mean angular velocity of the electrons at the surface of the anode is ω s so
that that the tangential energy of the electrons is eV0 from (15.22). If it is assumed
that the radial energy of an electron striking the anode is approximately equal to
the tangential energy we can write (15.46) as

2V0
ηe ≈ 1 − . (15.47)
Va

Now a magnetron is normally operated at a voltage close to the threshold voltage

so setting Va ≈ VT

2V0
ηe ≈ 1 − . (15.48)
VT
Substituting for V0 VT from (15.40) we obtain

2 Bz − 3B0
ηe ≈ . (15.49)
2 Bz − B0
The graph of the efficiency given by (15.49) in Figure 15.12 shows that the efficiency
increases rapidly with Bz B0 and that high efficiency is possible if a high magnetic
field is used. The validity of the approximations made in deriving this result can be
examined using the theoretical models in Section 15.6.

15.3 Magnetron Anodes

Figure 15.13(a) shows a plan view of a magnetron anode which incorporates an

even number of regularly spaced cavity resonators. The detailed shape of the reso-
nators may vary but the vane-​type anode illustrated is the most common [14]. This
circuit is a modified two-​wire line in which the forward and backward waves are
 Magnetron Anodes 581

Figure 15.12:  Approximate dependence of the efficiency of a magnetron on the normalised

magnetic field.

Figure 15.13:  (a) Arrangement of a vane-​type magnetron anode and (b) detail of one cavity

(reproduced, with permission, from [15]).

strongly coupled together at frequencies close to the resonant frequencies of the

cavities.
The properties of this structure can be determined from the equivalent circuit
shown in Figure 15.14. The admittance at the tips of the vanes of the cavity, repre-
sented by L1 and C1, is

 1   ω2 
Ya ( ω ) = j  ωC1 −  = j ωC1 1 − 12  , (15.50)
 ωL1   ω 

where ω1 = 1 L1C1 is the resonant frequency of the cavity. The admittance pre-
sented at the same terminals by the space between the cathode and the anode is

j ωC0
Yk ( ω, φ ) = , (15.51)
2 (1 − cos φ )
582 Magnetrons

Figure 15.14:  Equivalent circuit of one cavity of a magnetron anode.

where C0 is the capacitance between the tip of a vane and the cathode and φ is the
phase difference between adjacent cavities. The resonant frequency of the cavity,
modified by the admittance of the inner space is given by

Ya ( ω ) + Yk ( ω, φ ) = 0. (15.52)

Substituting into this equation from (15.50) and (15.51) we obtain, after a little
rearrangement,

C0 1
cos φ = 1 − ⋅ 2 2 . (15.53)
2C1 ( ω1 ω − 1)
This is a low-​pass filter characteristic such that ω = 0 when φ = 0 and

ω1
ω= = ω π , (15.54)
1 + C0 4C1
when φ = π. Normally the capacitance between an anode vane and the cathode (C0 )
is small compared with the capacitance between the vanes (C1 ) so that ω π is slightly
less than ω1. Because the slow-​wave structure forms a closed circuit the only pos-
sible values of φ are those for which there are an integral number of wavelengths
around the circuit. Thus

2 nπ
φn = ; n = 0,1 Nv 2 , (15.55)
Nv
where Nv is the number of vanes in the anode and n is the mode number. Equation
(15.53) can be rearranged as
1
−
ω  C 1  2
= 1 + 0 ⋅ , (15.56)
ω1  2C1 1 − cos φ 
from which the dispersion curve can be calculated. Figure 15.15 shows examples of
dispersion curves calculated for a 12-​vane anode for two values of the ratio C0 C1 .
 Magnetron Anodes 583

Figure 15.15:  Dispersion curves of 12-​vane magnetron anodes.

Because the structure is periodic, the field can be expanded as a series of space
harmonics. Figure 15.15 shows the possible forward wave modes in the first two
Brillouin zones (see Figure  4.6). Thus, the section of each curve in the range
π ≤ φ ≤ 2 π represents the m = −1 space harmonic of the backward wave. The reso-
nances of a transmission line in the form of a circular loop can be represented
as linear combinations of modes having azimuthal variations cos nθ and sin nθ.
When the transmission line is periodic, as in the case of a magnetron anode,
pairs of modes exist in which the relationship between the phase of the mode and
the discontinuities in the line are different. In these modes, known as degenerate
modes, the nodes and antinodes of the field pattern are exchanged. Figure 15.16
shows, as an example, the two possible orientations of the radial field maxima
of the degenerate n = 4 modes of a 12-​resonator anode. The differing relation-
ships between the field patterns, and the positions of the vanes, mean that the
two modes have slightly different frequencies. The asymmetry introduced by the
output coupler means that one mode of the pair is strongly coupled to the output
waveguide and has a low Q factor. The other is weakly coupled and has a high Q
factor so that unwanted oscillations may be a problem. When φ = π the maxima
of the tangential electric field coincide with the centres of the gaps between the
vanes. This mode cannot be degenerate because the tangential electric field in the
alternative orientation is short-​circuited by the vanes. For this reason magnetrons
are always designed to operate using the π-​mode field of the anode. Figure 15.15
shows that the frequencies of other possible modes are very close to the frequency
of the π-​mode. It is therefore important in the design and operation of magnetrons
to ensure that the wrong mode is not excited. The most troublesome modes are
those for which n = Nv 2 − 1 and n = Nv 2 + 1. These are known as the ( π − 1) and
( π + 1) modes irrespective of the number of vanes. Because the output coupler has
584 Magnetrons

Figure 15.16:  Alternative orientations of the maxima of the radial electric field for the n = 4
mode of a 12-​resonator anode.

been designed with respect to the π-mode the excitation of other modes usually
shows itself as a reduction in the output power in addition to a shift in frequency.
The change in the interaction may also cause damage to the cathode and arcing
within the tube because the impedance is mismatched to that of the power supply.
Various techniques for ensuring operation in the correct mode are discussed in
Section 15.4.5.
The resonant frequency of the cavities in the anode in Figure  15.13(b) can be
estimated by assuming that the depth of each cavity is a quarter of a wavelength so
that λ = 4 ( r2 − r1 ). It is found that this formula overestimates the frequency by about
20%. A  more exact solution is obtained from the input admittance of a wedge-​
shaped cavity with open ends given by

ε 0 La  J 0 ( kr1 )Y1 ( kr2 ) − J1 ( kr2 )Y0 ( kr1 ) 

Y= j ⋅ ⋅  , (15.57)
µ 0 ψ ra  J1 ( kr1 )Y1 ( kr2 ) − J1 ( kr2 )Y1 ( kr1 ) 

where k = ω c is the free-​space propagation constant at frequency ω , La is the axial

length of the anode and ψ = 2 π Nv [15]. The length of the anode is normally less
than half of the free-​space wavelength to avoid the excitation of modes having axial
variation of the RF field strength. The radii r1 and r2 are measured from the point of
intersection of the projections of the vanes (see Figure 15.13(b)) so that

t
r1 = ra − , (15.58)
ψ
where ψ = 2 π Nv . At resonance, k = k0 = ω 0 c, Y = 0 and (15.57) reduces to

J 0 ( k0 r1 )Y1 ( k0 r2 ) − J1 ( k0 r2 )Y0 ( k0 r1 ) = 0. (15.59)

 Magnetron Anodes 585

This equation can be solved for k0 graphically or by numerical methods. The capaci-
tance of the resonator (Cr ) can be calculated from (15.57) since the input admit-
tance of the equivalent circuit is

  ω 2
Y = j ωCr 1 −  0   . (15.60)
  ω  
Differentiating (15.60) with respect to ω  yields

dY   ω 2
= jCr 1 +  0   (15.61)
dω   ω  
so that

j  dY 
Cr = −   , (15.62)
2c  dk  k = k0

where c is the velocity of light and the derivative of the Y with respect to k at res-
onance is calculated from (15.57). The value of the inductance L1 is then obtained
immediately from Cr  and ω 0.
This calculation of the resonant frequency assumes that the RF magnetic field
of the cavity is zero when r = r1. It therefore ignores the effects of the fringing fields
in the space between the anode and the cathode and at the ends of the anode. The
effects of the fringing field can be modelled by adding a capacitance C f in parallel
with the equivalent circuit of the cavity. Because the dimensions of the region of
fringing field are small compared with the free-​space wavelength it is possible to
determine the capacitance C f from an electrostatic solution. The fringing capaci-
tance at the tips of the vanes is given approximately by [16]

ε 0 La   4 
Cf = 1 + ln    , (15.63)
π   πµ  

where

Nv w
µ= . (15.64)
2 πra
The capacitance of the cavity including the correction for the fringing field is then

C1 = Cr + C f . (15.65)

The capacitance per vane between the anode and the cathode (C0 ) can be estimated
by finding upper and lower bounds [17]. An upper bound for C0 is obtained by
assuming that the equipotential surfaces are concentric cylinders to give

ε 0 La 2π
CU = ⋅ . (15.66)
ln ( ra rc ) Nv
586 Magnetrons

To find a lower bound we assume that the flux lines between the anode and the
cathode are confined to wedge-​shaped volumes which terminate on the tips of the
vanes so that

ε 0 La t
CL = ⋅ . (15.67)
ln ( ra rc ) ra
Finally C0 is taken to be the geometric mean of these two bounds so that

C0 = CLCU . (15.68)

Note that this calculation assumes that the cathode is a cylinder. In continuous
wave magnetrons the cathode commonly takes the form of a helical wire and the
capacitance calculated from (15.68) would need to be reduced to allow for this. We
have seen that the dispersion diagram is not strongly dependent on C0 for typical
values so this estimate is sufficiently accurate for our purposes. Further informa-
tion about the calculation of equivalent circuit parameters for resonators of other
shapes and about the effects of the fringing fields at the ends of the anode can be
found in [14].
The simple anode structure which has been discussed in this section suffers from
the disadvantage that the mode frequencies are too close together to ensure that the
π -​mode alone is excited, without contamination from the adjacent modes. It is evi-
dent that the problem gets more difficult as the number of resonators in the anode
is increased. A variety of solutions to this problem have been found in which the
properties of the anode are modified by changes in its design. These are discussed
in the sections which follow. In addition it has been found that the introduction of
Nv 2-​fold azimuthal periodicity into the magnetic field, the properties of the cath-
ode, or the shape of the anode, can help to ensure that the oscillations start in the
π -​mode [18–​21].
The output from a conventional cavity magnetron is generally made through a
slot in the outer wall of one of the resonators or by connecting the inner conductor
of a coaxial line to one, or more, of the vanes [3, 14]. In either case the coupling
is made into a rectangular waveguide through a vacuum window. Although the
magnetron is a fixed-​frequency oscillator, it is possible to vary the frequency over
a limited range by including metallic tuning elements that can be moved relative
to the anode block. Metal bellows are used to take the motion through the vac-
uum envelope. The tuning elements can perturb the capacitance, the inductance,
or both, of the resonant cavities. Pins inserted near the tips of the vanes alter the
cavity capacitance, while those inserted near the outer edge of the cavity alter the
inductance [3, 22–​24]. The frequency can be changed by up to 10% by a mechani-
cal tuner and various methods have been used to vary the frequency from pulse to
pulse [3]. Electronic tuning which can vary the frequency by a few percent can be
achieved using ferrites, diode switches or multipactor discharges [3, 25]. A greater
tuning range can be achieved by coupling the anode to a high Q external cavity as
described in Section 15.3.3.
 Magnetron Anodes 587

15.3.1 Strapped Anodes
In order to reduce the risk that a mode other than the π -​mode is excited, the anodes
in practical magnetrons are designed in such a way that the separation of these two
modes is increased. Historically, the first method used was the introduction of wires,
known as straps, which connect the vanes together. In modern magnetrons the straps
are generally connected to alternate vanes as shown in Figure 15.17. Two or three
straps are added to one, or both ends of the anode. Theoretically the straps do not
carry current in the π -​mode, since they are connected to vanes which are at the same
phase. They do carry current in the other modes whose resonant frequencies are there-
fore perturbed. The actual effect of strapping is a little different, as will be seen, but it
achieves the purpose of increasing the separation between the π and the π −1 modes.
The geometrical complexity of the strapping makes it difficult to determine their
effects accurately except by cold test measurements, or from computational elec-
tromagnetics software. However, useful insight can be gained by considering the
straps as TEM transmission lines which are connected to alternate vanes, as shown
in Figure 15.18, where the admittance of each cavity is Y = Ya + Yk from (15.50) and
(15.51) [26].
Let us assume that the voltages on the tips of the vanes are given by

V = V0 exp ( jqφ ) , (15.69)

where φ is the phase shift per cavity and q = 0, ±1, ±2  is the index number of the
vanes such that q = 0 when x = 0. Now let the voltage on strap 1 relative to the cath-
ode in the range 0 ≤ x ≤ 2 p be

V1 = V+ exp ( − jkx ) + V− exp ( jkx ) (15.70)

whereV+ andV− are the amplitudes of the forward and backward waves, and k = ω c .

Figure 15.17:  Strapped magnetron anode.

588 Magnetrons

Figure 15.18:  Equivalent circuit for a strapped anode.

When x = 0

V1 = V0 = V+ + V− (15.71)

and, when x = 2 p

V1 = V0 exp( 2 j φ ) = V+ exp ( −2 jkp ) + V− exp ( 2 jkp ) . (15.72)

Substituting for V− in (15.72) from (15.71) gives

V0 exp( 2 j φ ) − exp ( 2 jkp ) = −2 jV+ sin ( 2 kp ) . (15.73)

Similarly, substituting for V+  gives

V0 exp( 2 j φ ) − exp ( −2 jkp ) = 2 jV− sin ( 2 kp ) . (15.74)

The current flowing in the positive x direction when x is just greater than zero is
given by

I + = Yc (V+ − V− ) , (15.75)

where Yc is the characteristic admittance of the transmission line formed by strap

1. Substituting for V+ and V− in (15.75) from (15.73) and (15.74) gives

jYcV0
I+ =
sin ( 2 kp )
(exp(2 j φ) − cos (2kp)). (15.76)
Similarly, the current flowing in the negative x direction when x is just less than zero
is given by

jYcV0
I− =
sin ( 2 kp )
(exp( −2 j φ) − cos (2kp)). (15.77)
 Magnetron Anodes 589

The current flowing to the strap from the vane is the sum of these two currents

 2 cos( 2φ ) − 2 cos ( 2 kp ) 
I = jYcV0   . (15.78)
 sin ( 2 kp ) 
This current can also be written in terms of the loop current circulating in the
cavity as

I = i (1 − exp ( − j φ )) . (15.79)

The voltage driving the loop current into the transmission line is

V = V0 (1 − exp ( j φ )) . (15.80)

Hence the admittance presented to the cavity by the straps is

i  cos( 2φ ) − cos ( 2 kp ) 
Ys ( ω, φ ) = = jYc   . (15.81)
V  sin ( 2 kp ) (1 − cos ( φ )) 
This result is not the same as that given by Yu and Hess [26]but it is believed to be
correct.
The dispersion curve of a strapped anode is found from the solutions of

Ya ( ω ) + Yk ( ω, φ ) + Ys ( ω, φ ) = 0. (15.82)

The characteristic admittance of the straps (Yc ) is unknown, but it can be used as an
adjustable parameter to fit the dispersion curve to experimental data. Figure 15.19
shows, as an example, the dispersion curves, with and without straps, for the anode
of the 4J50 magnetron whose dimensions are given in [27] (see Worksheet 15.1). This

Figure 15.19:  Dispersion curves for the anode of the Litton 4J50 magnetron with, and
without, straps.
590 Magnetrons

anode has 16 vanes and two strapping rings at each end. The characteristic admit-
tance of the straps was adjusted to give the correct π -​mode frequency [13]. The fre-
quency of the π −1 mode then agreed with the figure given in [13] to three significant
figures. It can be seen that the effect of the straps is to reduce the frequency of the
π -​mode and to increase the separation between the π and π −1 modes. The admit-
tance of the straps is capacitive at the π -​mode so that the frequency of this mode
is lower than in an unstrapped anode. It should be noted that the strapped anode
is a bi-​periodic structure so that the π -​mode corresponds to a phase shift of 2 π in
one period (two cavities). The dispersion diagram has a forward-​wave characteristic
when the phase shift per cavity is small. This must be the case because the structure
is a heavily modified form of two-​wire line whose dispersion curve therefore passes
through the origin. At larger values of phase shift the characteristic becomes back-
ward wave. The technique is limited to frequencies below 16 GHz by manufacturing
difficulties. It is not suitable at the highest lower power levels because of problems
with voltage breakdown between the straps and the vanes.
We have seen that the modes other than the π -​mode exist as pairs whose orienta-
tions are determined by the position of the output coupler. If a break is made in the
straps at an angle of 45° to the output waveguide, then the orientations of the two
modes are determined by it. Both modes are then coupled to the output waveguide to
some extent, the Q factors of the modes are reduced, and oscillations are less likely.

15.3.2 Rising Sun Anodes

An alternative anode structure which also makes use of bi-​periodicity to increase
the separation between the π and π −1 modes is the Rising Sun structure, shown in
Figure 15.20. This structure is useful at high frequencies and high powers where
construction of straps is difficult, or they suffer from voltage breakdown. The
ratio of the depths of the two resonators d1/​d2 (where d = (r2 –​ r1) is known as the
circuit ratio.
The equivalent circuit of the rising sun structure is shown in Figure  15.21.
Analysis of this circuit shows that the dispersion curves are given by

Z1 + 2Z0 −Z0 (1 + exp ( 2 j φ ))

= 0, (15.83)
−Z0 (1 + exp ( −2 j φ )) Z2 + 2 Z0

where φ is the phase shift per cavity,

Z1 = ( j ωC1 + 1 j ωL1 ) , (15.84)

−1

Z2 = ( j ωC2 + 1 j ωL2 ) , (15.85)

−1

and

Z0 = 1 j ωC0 . (15.86)
 Magnetron Anodes 591

Figure 15.20:  Arrangement of a rising sun magnetron anode

(reproduced, with permission, from [15]).

Figure 15.21:  Equivalent circuit of a rising sun anode.

Equation (15.83) can be expanded to give

 Z  Z 
cos 2 φ = 1 + 1  1 + 2  ,
 2 Z0   2 Z0  (15.87)
= cos φ1 cos φ2
where φ1 and φ2 are the phase shifts per cavity of uniform structures in which the
one set of cavities is replaced by a short circuit, as shown in Figure  15.22. This
is not quite accurate because short-​circuiting a set of cavities also makes a small
change in C0 . Each structure has half as many elements as the full anode and the π
-​modes of these structures therefore correspond to the π 2 modes of the full anode.
Figure 15.23 shows an example of the coupled, and uncoupled, dispersion curves
(see Worksheet 15.2). The coupling between the modes produces a pair of pass-​
bands with a stop band between them. The upper pass-​band is cut-​off at φ = 0
and φ = π with a frequency somewhat above the resonant frequencies of the large
592 Magnetrons

Figure 15.22:  Uniform structures which are combined in a rising sun anode.

Figure 15.23:  Dispersion curves of a typical rising sun anode, normalised to the frequency
of the larger cavities (solid lines), together with the dispersion curves of the uniform
uncoupled structures (chain dotted lines), and the frequency of the smaller cavities (dashed
line) ( N v = 18, d 2 d1 = 1.78).

cavities. The anode is designed to operate in the π -​mode of the upper pass-​band
and the steep dispersion curve close to that point ensures good mode separation.
Figure 15.24 shows how the mode separation increases with circuit ratio d 2 d1 for
an anode where Nv = 18. The use of a large circuit ratio is superficially attractive, but
it is found to lead to problems with the excitation of the φ = 0 mode. At the π -​mode
resonance the fields are different in the large and the small cavities, and this gives
rise to a component of the tangential field which is in phase for all cavities. This
component increases as the difference between the depths of the cavities increases.
When interaction with this mode occurs it causes heavy back-​bombardment of the
cathode and serious loss of efficiency. An electron which is synchronous with the π -​
mode sees the φ = 0 mode as an RF field at the signal frequency. If the cyclotron fre-
quency coincides with the signal frequency then a resonance occurs. It is therefore
vital to ensure that the cyclotron frequency is well away from the signal frequency
in a magnetron with a rising sun anode. It is usual also to avoid values of Nv which
are divisible by four because it is found that they can lead to problems associated
 Magnetron Anodes 593

Figure 15.24:  Dependence of the π and π − 1 mode frequencies of a rising sun anode with 18
vanes on the circuit ratio d 2 d1 .

with the mode for which φ = π 2. Further information about the properties of rising
sun anodes can be found in [16].

15.3.3 Coaxial Anodes
Conventional magnetrons have mode and frequency stabilities which are inad-
equate for some purposes. Their tuning ranges are also rather restricted. These
problems can be overcome by coupling the anode to another resonator having a
very high Q factor [28–​30]. The additional cavity is placed around the anode result-
ing in the coaxial magnetron, shown in Figure 15.25. The coaxial cavity is coupled
to alternate cavities by slots, so that the π -​mode of the anode is coupled to the TE011
(circular electric) mode of the coaxial cavity. The energy stored in the cavity is much
greater than that stored around the vanes. This means that the operating frequency
is determined by the resonant frequency of the cavity and can be tuned over a
greater range than that of a conventional magnetron. Because the TE011 mode does
not have any current flow across the end walls of the cavity, it is possible for the tun-
ing plunger to be made non-​contacting. The space behind the tuner can hold lossy
material to damp out unwanted modes. A comparison between a conventional and
a coaxial magnetron is shown in Table 15.1. It is seen that the latter is superior in its
tuning range and stability (frequency pushing and pulling are discussed in Sections
15.4.2 and 15.4.3). The disadvantages of the coaxial magnetron are its large size,
for a given frequency, and the slow rise of the anode voltage pulse needed to allow
time for the energy stored in the coaxial cavity to build up as oscillation commences.

15.3.4 Long Anodes
The power output of a conventional magnetron is limited by heat dissipation in
the cathode. To increase the output power it is necessary to increase the diameters
594 Magnetrons

Table 15.1:  Comparison between a coaxial and a conventional X-​band magnetron

(copyright 1961, Academic Press, reproduced, with permission, from [28])

Coaxial magnetron Conventional magnetron

Peak power 250 kW 250 kW

Circuit efficiency 85% 70%
Electronic efficiency 65% 60%
Tuning range 20% 10%
Frequency pulling 3 MHz 14 MHz
Frequency pushing 0.08 MHz/​A 0.4 MHz/​A

Figure 15.25:  Arrangement of a coaxial magnetron.

of the anode and the cathode and, therefore, the number of vanes. This brings
with it problems of reduced mode separation. These problems can be overcome by
using a magnetron with a much longer anode (known as a long anode magnetron)
[31–​33]. By making the anode length comparable with the free-​space wavelength,
the cathode area can be increased without increasing its diameter and the number
of vanes can be small. These magnetrons are normally made without straps and
use the method of output connection to select the correct mode. A long anode can
be regarded as a length of waveguide of unusual cross-​section which propagates a
TE mode in the axial direction. The anode then has resonances at the cut-​off fre-
quency of each mode, and also at higher frequencies where there are longitudinal
standing waves determined by the boundary conditions at the ends of the anode.
The notation for the modes is H n , H n′ , H n′′ , etc. where n is azimuthal mode number
and the number of primes represents the number of null planes along the length of
the anode.
 Magnetron Anodes 595

Figure 15.26:  Simplified diagram of a long anode magnetron.

The output connection is made by connecting the inner conductor of a coaxial

line symmetrically to alternate vanes as shown in Figure 15.26. The strength of
the coupling is greatest when the connections are made close to the tips of the
vanes, and it decreases as the pitch circle diameter of the connections is increased.
The effect of the symmetry of the coupler is to couple the modes together in
pairs so that the H n mode is coupled to the H(Nv 2) − n mode. For example, in a
12 cavity anode the n = 0 mode is coupled to the n = 6 mode, the n = 1 mode to
the n = 5 mode, and so on. These coupled modes are denoted by H(0,6) , H(1,5) etc.
since the H n and H(Nv 2) − n modes do not exist independently. Where one of the
component modes is strongly cut off, it represents local storage of energy close
to the output coupler, and the axial variation of the electric field strength is
determined by the other mode. It is then possible to simplify the notation still
further. Thus, if the H 0 mode is strongly cut off, the H(0,6) mode can be referred
to as the 6 mode or π mode. This mode is strongly coupled to the TEM mode of
the output coaxial line. The line propagates all frequencies so that all the π mode
resonances ( π, π ′, π ′′ etc.) are strongly coupled to the output and have loaded Q
values determined by it. The non-​π modes couple to the higher-​order modes of
the coaxial line which are cut off, and these modes therefore have high Q fac-
tors determined by the losses in the anode. The symmetry of the output coupler
allows pure π -​modes to be established when the frequency separation from the
adjacent modes is as low as 1% so that straps are not necessary. The effect of
adding straps was discussed in [34]. The output is coupled directly into a rectan-
gular waveguide via a capacitive probe enclosed by a hemispherical dielectric
window. Further information about the design of long-​anode magnetrons can
be found in [31–​33].
596 Magnetrons

15.4 Magnetron Properties

We saw in Section 15.2 that the behaviour of a magnetron can be described in terms
of the cut-​off and threshold voltages. At the characteristic voltage and field (V0 , B0 )
these two voltages coincide. However, it can be seen from Figure 15.12, that a tube
must be operated appreciably above the characteristic field in order to achieve sat-
isfactory conversion efficiency. Usually B B0 ≥ 3 and the magnetron is operated at
an anode voltage just above the threshold voltage, and well below the cut-​off volt-
age (see Figure 15.9). The threshold voltage for each mode can be considered as
the anode voltage at which oscillation in that mode begins. The oscillations in the
various possible modes start from the noise in the circulating electron current and
grow competitively until one becomes dominant. When the wrong mode becomes
dominant the magnetron is said to be moding. This is revealed by a change in the
frequency, reduced output power and excessive heat dissipation on the cathode.
The anode current remains small until the oscillation is fully established. Therefore
the impedance presented to the power supply by the magnetron varies during the
start-​up of the oscillation. This section reviews the practical aspects of the perform-
ance of magnetrons.

15.4.1 The Performance Chart

Manufacturers of magnetrons commonly supply information about their char-
acteristics in the form of a performance chart like the one shown in Figure 15.27.
The diagram shows the anode voltage and current, the magnetic field, and the
power output and efficiency. The lines for constant magnetic field are commonly
called Gauss lines. Gauss line discontinuities are associated with discontinuous
changes in other properties, as described in Section 15.4.4. The projection of a
Gauss line intersects the vertical axis close to the threshold voltage for that mag-
netic field. For the example shown, the intersections lie a little below the theoreti-
cal threshold voltages. Better agreement is obtained by linear projections which
are tangents to the Gauss lines at the right-​hand side of the diagram. Regions
of operation to be avoided because of poor mode stability or excessive heat dis-
sipation may be marked. Note that the axes of this figure are exchanged, com-
pared with the usual presentation of characteristic curves. For most magnetrons
with strapped or rising sun anodes the magnetic field is provided by a permanent
magnet whose strength is set by the manufacturer. Where the magnetic field is
provided by an electromagnet (e.g. long anode magnetrons) it may be varied to
some extent by the user. The efficiency increases as the magnetic field is increased.
For a fixed magnetic field there is a region of optimum efficiency, as shown in
Figure 15.27. It is found that the graph of the RF power against the DC current
is approximately a straight line. A small current flows when the power is zero. This
offset is caused by current which does not take part in the interaction, mainly
emission from the cathode end-​caps.
 Magnetron Properties 597

Figure 15.27:  Performance chart of the 4J50 magnetron

(reproduced, with permission, from [35]).

Figure 15.28:  Typical frequency pushing curve of a magnetron.

15.4.2 Frequency Pushing

The frequency of oscillation of a magnetron varies with the anode current, as shown
in Figure 15.28 where f0 is the resonant frequency of the anode in the absence of
any electrons. The frequency at the peak of the curve is depressed relative to the
unloaded resonant frequency of the anode by electron loading. The change in fre-
quency ( ∆f ) is given approximately by

∆f 0.6
~ , (15.88)
f0 QL
where the loaded Q of the anode (QL ) is typically 250 for conventional magne-
trons, and 1000 for coaxial magnetrons. The shape of the curve is determined by the
598 Magnetrons

changing phase of the electron spokes with respect to the RF field of the anode (see
Section 15.6.7). Tubes are normally designed to be operated on the flattest part of
the curve. The residual variation of frequency with current (~100 kHz/​A) is called
the frequency pushing figure of the magnetron [36, 37].

15.4.3 Frequency Pulling

The effect of the output match on the performance of a magnetron is displayed by
plotting the frequency and output power against the match on a Rieke diagram, as
shown in Figure 15.29. It should be noted that the phase of the mismatch is referred
to the output flange of the tube. Thus the line showing zero frequency shift does
not normally coincide with the real axis of the impedance chart. The point of max-
imum power output does not usually coincide with the centre of the chart. This to
ensure that the performance of the tube is stable throughout the permitted range
of mismatch, that thermal limits are not exceeded, and that internal arcing does
not occur. It is usual for magnetrons, other than low power tubes, to be protected
from excessive variations in the output match by the inclusion of a circulator in the
output waveguide.
To investigate the effect of a mismatched load we note that the admittance of the
unloaded resonator is

  ω ω0  
YU = GU 1 + jQU  −  , (15.89)
  ω 0 ω  

where GU is the shunt conductance of the resonator and QU its unloaded Q. The
external load may be represented by a transmission line whose characteristic

Figure 15.29:  Rieke diagram of the 4J50 magnetron

(reproduced, with permission, from [35]).
 Magnetron Properties 599

admittance is GE so that the external loading conductance is GE when the transmis-

sion line is matched. When the line is terminated by a mismatch, the external admit-
tance loading the resonator may be written

S + j tan θ
YE = GE , (15.90)
1 + jS tan θ
where S is the VSWR on the line and θ its electrical length. It is usual to specify
the effect of a mismatched load for S = 1.5 , and it is then found that the imaginary
part of (15.90) is maximum when θ = ±0.588 and YE = (1.083 ± 0.417 j )GE . When
the external load is connected to the resonator the imaginary part of YU + YE is zero
at resonance so that

 ω ω0 
QU GU  − ± 0.417GE = 0. (15.91)
 ω 0 ω 

Now QU GU = QE GE so that (15.91) may be written

2 ∆ω 0.417
=± , (15.92)
ω0 QE
where ω = ω 0 + ∆ω . The frequency pulling figure ( f p ) is defined as the change in the
frequency when a load for which S = 1.5 is varied through all possible phases. Thus

f0
f p = 0.417 . (15.93)
QE
This derivation assumes that there is only one possible frequency at which the sum
of the imaginary parts of the admittances is zero. When the transmission line con-
necting the load to the magnetron is long, as may be the case when an antenna is
connected directly to the tube, multiple solutions are possible. The frequency of
oscillation may then be unstable or show hysteresis effects. These long line effects
can be avoided by careful design, or by placing an isolator between the magnetron
and the transmission line [38].

15.4.4 Spectrum
The frequency spectrum of a magnetron normally includes some output at har-
monics of the fundamental frequency. Since the electronic charge does not vary
sinusoidally with angle, but is concentrated in the spokes, the RF current flowing
to the anode has a high harmonic content. The extent to which this is reflected in
the output from the tube depends upon the coincidence between any higher-order
modes of the anode with the harmonics of the signal frequency, and also on how
well such modes are coupled to the output waveguide. Since these frequencies are
tuned by changes in the output match it may be expected that those changes will
also affect the harmonic output. It is possible to suppress harmonic emissions by
600 Magnetrons

including a filter in the output waveguide either inside, or outside, the tube [39].
However, since the filter reflects the harmonic signals back into the anode, there
may be unexpected changes in the output at the fundamental frequency.
The spectrum of a pulsed magnetron is normally determined by the Fourier
Transform of the RF pulse. It is found, however, that the spectrum can be much
broader if the magnetron is operated at a low anode current where there is strong
frequency pushing. This effect can be explained by the variation in the frequency of
oscillation during the pulse. The spectrum of continuous wave cooker magnetrons
is found to vary with the operating current. A study of magnetrons, operating at
2.45 GHz, for use in domestic microwave ovens showed three distinct regimes [40].
At low currents (~20% of normal operating current) broadband random noise was
observed with a bandwidth of up to 600 MHz. The noise showed distinct peaks
corresponding to modulation at a frequency in the range 150 to 300 MHz, which
increased with anode current. This low frequency noise can be conducted, and radi-
ated, by the high-​voltage cathode leads and its harmonics can cause interference
with UHF TV reception. At an anode current around 50% of the normal operating
current the noise could take the form of discrete sidebands at frequencies about
300 MHz above and below the fundamental, with power about 30 dB below it. This
condition, involving spurious oscillation, was associated with Gauss line discon-
tinuities where a discontinuity in the characteristic curve corresponds to an abrupt
change in the spectrum. Finally, at high currents, the spectrum showed no noise
above the noise floor of the spectrum analyser. All cooker magnetrons show low-​
current, high-​noise, and high-​current, low-​noise, regions with a discontinuous tran-
sition between them. Where spurious oscillation is associated with the transition it
is strongly affected by the external VHF impedance between the cathode and the
anode. In particular, it is possible to lock the frequency to a VHF resonant circuit
in the cathode leads. The spectrum of a magnetron tends to deteriorate with age
showing increasing levels of broadband noise. This change is presumably correlated
with decreasing thermionic emission from the cathode and consequent changes in
the relative proportions of thermionic and secondary emission.
Some pulsed magnetrons exhibit a phenomenon known as twinning in which the
spectrum shows two distinct peaks, and two different levels of RF output power
and anode current, as shown in Figure 15.30 [41, 42]. This phenomenon is clearly
distinguishable from moding (see Section 15.4) because the frequency difference
between the two modes is much smaller (typically only a few MHz). It is still not
completely understood, but appears to be related to changes in the space-​charge
cloud that alter the electronic loading of the anode. The changes probably arise
from changes in the proportions of thermionic and secondary electrons emitted by
the cathode. A detailed study of twinning in pulsed magnetrons showed that the
two states do not coexist, but that the tube may switch from one to the other during
the pulse. On occasion multiple transitions were observed [42]. It was found that
the conditions under which a tube twinned could be altered by changes in the out-
put match, the magnetic field, the heater power, and the cathode coating. However,
none of these would completely eliminate the problem from tubes which showed
 Magnetron Properties 601

Figure 15.30:  Spectrum of a magnetron showing twinning.

(copyright 2000, IEEE, reproduced, with permission, from [42]).

this tendency. Twinning is one of those phenomena, known to, and frustrating to,
all tube engineers that may appear or disappear between batches of tubes as a result
of minute changes in the production process. The instability appears to be corre-
lated with Gauss line discontinuities. A survey of a wide range of different types of
magnetrons showed that twinning is most likely to occur when the ratio of the elec-
tron cyclotron frequency to the operating frequency lies in the range 0.75 to 1.25.

15.4.5 Mode Selection, Priming, and Locking

The typical characteristic curves of a magnetron oscillator, in Figure 15.27, show
that for a given magnetic field the current remains low until the threshold voltage is
reached. It then rises rapidly over a small change of variation of the anode voltage,
and settles at an operating point determined by the load line of the power supply.
When the magnetron is pulsed the shape of the pulse must be designed to ensure
that the correct mode is excited [43, 44]. Figure  15.31 shows typical pulse wave-
forms for the anode voltage and anode current. For a discussion of pulse modula-
tors see Section 20.3.
The important features of the voltage pulse are:
• The rate of rise of the voltage can be fast as far as VT but must thereafter be slow
enough to ensure that the oscillation is fully established in the π mode before the
threshold voltage of the π −1 mode is reached. Typically this takes of the order of
50–​100 RF cycles. During the initial rise of the pulse the current consists mainly
of the charging current of the capacitance between the anode and the cathode. If
the oscillation is not fully established by the time the voltage reaches VT then the
anode current is low, the magnetron impedance is high, the voltage overshoots
and π −1 mode oscillation is probable.
602 Magnetrons

Figure 15.31:  Typical voltage and current pulses of a magnetron.

• The pulse top must be flat because, as shown in Figures 15.27 and 15.28, a small
voltage swing results in a large current swing, strong frequency pushing, and fre-
quency modulation during the pulse.
• The voltage pulse should fall fast to VT and must fall fast enough thereafter to
ensure that none of the modes beyond the π mode is excited.
If the output port of the tube is connected to a circulator, then a signal can be
injected at the π mode frequency. A signal with an amplitude 40 dB below the final
output level has the effect of giving that mode a head start over the competing
modes. The result is greater certainty that the correct mode will be excited. The
technique, known as Priming, also determines the phase of the oscillation and pro-
vides pulse-​to-​pulse phase coherence.
The phase and frequency of an oscillator, such as a magnetron, can be locked
by an injected signal [45]. The injection power required increases with the square
of the frequency difference between the injected signal and the natural frequency
of the oscillator. The bandwidth of a free-​running magnetron is normally so large
that the injected signal must be around 10 dB below the output power to achieve
locking [46]. However, if the bandwidth can be reduced by suitable selection of the
operating conditions, or by an external control loop, the power required is much
less [47–​49]. In the same way the frequencies of two or more tubes can be locked to
one another [46, 50].
 Particle in Cell Magnetron Models 603

15.5 Particle in Cell Magnetron Models

To model a magnetron fully it is necessary to find self-​consistent solutions for the

electron motions and the RF fields (including the space-​charge field). A  three-​
dimensional model is required to include the effects of longitudinal motion of the
electrons, longitudinal variations of the fields, and the effects of fields in the end
spaces. It is also necessary to model the combination of thermionic and secondary
electron emission from the cathode (which may well be non-​uniform), and to model
the build-​up of oscillations from the initial noise level to saturation. The additional
complexity, compared with a model of a klystron or a TWT, requires the use of a
fine computational mesh close to the surface of the cathode, and continuation of
the calculations over some fifty or more RF cycles. Thus, the computational time is
very long, even on a very powerful computer [51–​53].
As with klystrons and TWTs, it is possible to achieve some economy in the com-
putational effort: by using a two-​dimensional model which ignores axial variations;
by pre-​calculating the static magnetic field; and by representing the properties of
the anode by an equivalent circuit. This avoids the need to re-​compute the electro-
magnetic fields outside the interaction region at each time step. However, because
the electrons are collected on the anode it is desirable to include at least part of the
space between the vanes in the interaction region. Calculations of this kind can
provide valuable insight into the complex processes in a magnetron [54–​57]. Three
states have been investigated [56, 57]:
• In the first the cathode current is dominated by primary emission, and the DC
electric field is zero at the surface of the cathode. The electron trajectories are
then similar to first-order Slater trajectories (see Section 8.3.2) and the thickness
of the space-​charge hub is close to that given by the numerical solution of (15.2).
In order to maintain the space-​charge hub the current emitted from the cathode
must be between three and four times the anode current. The fraction of the DC
power dissipated on the surface of the cathode is around 2%. The spokes are nar-
row as shown in Figure 15.32(a) and the base of the spoke is an unstable region
with fluctuating behaviour. High-​efficiency magnetrons for microwave ovens and
industrial heating, which have thoriated tungsten cathodes with low secondary
emission coefficients, are believed to operate in this condition.
• In the second case the cathode current is dominated by secondary emission.
Because the distribution of returning electrons is not uniform on the surface of
the cathode, the shapes of the spokes vary with time, and regions of low-​charge
density move through the hub. These fluctuations are considerably greater than
those in the primary emission state. The spokes are much thicker in this case as
shown in Figure 15.32(b) and the cathode dissipation is of the order of 10% of
the DC power. Most high power pulsed magnetrons, which have oxide or dispen-
ser cathodes with high secondary emission coefficients, are believed to operate
in this condition.
604 Magnetrons

Figure 15.32:  Space-​charge distribution in a 16-​cavity magnetron where the emission is

dominated by: (a) thermionic emission, and (b) secondary emission. The spokes rotate in
the clockwise direction
(copyright 2004, IEEE, reproduced, with permission, from [57]).

• In the third case there is no secondary emission and the cathode current falls
below around three times that required to supply the anode current. It is then
limited by the primary emission from the cathode. This state is characterised by
a complete absence of fluctuations. It appears to correspond to the low-​noise
state observed in microwave oven magnetrons when the heater power is reduced.
Table 15.2 shows a summary of the results obtained for a 4J52 magnetron [56]
with experimental data for comparison [35]. The computed results differ consider-
ably from the experimental data so that conclusions drawn from the model must be
treated with caution.

15.6 Simple Magnetron Models

Because the modelling of magnetrons using PIC codes is very time-​consuming

it is useful to have simpler models which can give approximate results quickly.
The images in Figure 15.32, and the discussion in the previous section, illustrate
the complexity of the problem of modelling a magnetron, and the danger of
making assumptions about, for example, the shapes of the spokes. It has been
observed that magnetrons may exhibit chaotic behaviour in the sense that very
small changes in the operating conditions produce appreciable changes in per-
formance. Twinning is an example of this (see Section 15.4.4). Despite these
 Simple Magnetron Models 605

Table 15.2:  Properties of a 4J52 magnetron ( Bz = 4900 G) , computed for different emission regimes
(copyright IEEE, 1998, reproduced, with permission, from [56]), compared with experimental data [35]

Primary Secondary Primary Experiment

emission emission emission
dominated dominated limited

Anode voltage (kV) 16.5 15.5 17.5 16

Va VT 1.15 1.08 1.22
Anode current (A) 15.8 20.3 27.8 27
Output power (kW) 140.6 130.6 145.7 180
Efficiency (%) 53.9 41.4 29.9 42
Anode dissipation (%) 20.9 29.8 51.8
Cathode dissipation (%) 2.1 11.1 5.4
Circuit loss (%) 23.1 17.7 12.8
Frequency shift (MHz) –22.1 –17.9 –25.0
Thermionic current 220 2.2 55.0
available (A)
Total cathode current (A) 54 77.6 53.5
Current fraction 0.71 0.74 0.48
returned (α )
Wave voltage /​DC voltage 0.28 0.27 0.28

difficulties it is possible to gain useful insight into the factors affecting the per-
formance of magnetrons using simple methods [13, 37, 58–​61]. These models are
based upon the motion of the electrons in simplified fields with the following
assumptions:
• The electrons are drawn into the spokes from a space-​charge hub whose proper-
ties are determined by static theory.
• The electric field of the anode is represented by a travelling wave and the RF mag-
netic field is neglected.
• The effects of all space-​harmonics, other than the one synchronous with the
spokes, are assumed to average to zero.
• The space-​charge field of the spokes is neglected.
• The current flowing to the anode through the spokes is determined from the
properties of the hub and the proportion of its perimeter from which electrons
are drawn.
The properties of the space-​charge hub, including the effects of secondary electron
emission from the cathode are discussed in the next section. Following that, three
different ways of modelling the spokes are reviewed and compared. It is then shown
how the performance chart, the Rieke diagram, and the frequency pushing charac-
teristics of a magnetron can be computed.
606 Magnetrons

15.6.1 The Space-​Charge Hub

The conditions affecting the electron emission can be illustrated by a simple discus-
sion [62]. Let us assume that the total current emitted by the cathode is I c and that a
fraction α of this current returns to the cathode so that the anode current is

I a = (1 − α ) I c . (15.94)
If the average secondary electron emission coefficient of the cathode is δ then the
available secondary emission current is
I s = α δ I c . (15.95)
McDowell found that α was typically in the range 0.65–​0.75, except when the cur-
rent was limited by primary emission [56]. Thus if the product αδ is greater than
unity the total current will be a self-​balancing combination of primary and second-
ary emission. The secondary emission dominated state observed by McDowell will
only arise if the heating of the cathode by the heater, and by back bombardment, is
insufficient to generate enough primary emission. Thus a magnetron model which
assumes that the current is uniform and space-​charge limited should correspond
to the operating conditions in many magnetrons. However, this kind of model is
unable to represent the fluctuating secondary emission from the cathode observed
by McDowell when the secondary emission is dominant.
We saw in Section 8.5 that, theoretically, only zero or first-order trajectories can
form a stable hub in a cylindrical magnetron. Either of these may be used as a
model for the hub. For a given anode voltage and magnetic field the hub radius
(rb ) is obtained by numerical solution of (15.2). The electrostatic potential on the
surface of the hub is found by setting r = rb in (15.1), and the angular velocity of
electrons on the surface of the hub is given by (15.4). These calculations assume a
zero-order hub but, as was shown in Section 8.5, the properties of a first-order hub
are the same to within a few percent.
In a zero-order (Brillouin) hub the electrons move in circles concentric with the
cathode [13, 37, 60, 61]. The electric field at the surface of the cathode is zero, and
the space-​charge density at the surface of the hub, obtained from (8.67), is

ε 0 m0 2  rc4 
ρb = ω c 1 + 4  , (15.96)
2e  rb 
where ω c = eBz m0 is the cyclotron frequency. It is assumed that the electrons form-
ing the spokes are drawn from the surface of the space-​charge hub with constant
charge density. Since the spokes are synchronous with the field of the anode we
assume, from (15.34), that
dr Eθ
≈ (15.97)
dt Bz
at the base of the spoke. The tangential electric field is obtained by setting R = Rb
in (15.19).
 Simple Magnetron Models 607

Then anode current is given by

θ′
⌠ Eθ
2

I a = nρb rb La  d θ ′ (15.98)
⌡θ1′ Bz

where the integration is carried out over the angles for which current can flow to a
spoke. Three different methods for determining this range of angles are described
in the sections which follow. This model suffers from the difficulty that, because
there is no radial current within a zero-order hub, the electrons must be assumed
to diffuse through the hub from the surface of the cathode. It also assumes that
the charge density in the hub is not perturbed by the RF field of the anode. The
calculation in (15.98) avoids the assumptions about the shape of the spoke made
in [13].
An alternative model of the space-​charge hub, suggested by the results obtained
by McDowell [56], is to assume that the electrons in the hub follow first-order
Slater orbits which start from the cathode and return to it. The radius of the hub
and the angular velocities of the electrons on its surface are close to those of a
zero-order hub. However, the radial electrostatic field is reduced so that the elec-
trons experience a net inward force. When their tangential velocities are reduced,
by interaction with the RF electric field, they can move outwards into the spokes.
If it is assumed that the total cathode emission is the maximum possible for a
given ratio ( rb rc ), then the cathode current in a cut-​off diode is approximately
0.35 of the space charge limited current in the absence of a magnetic field (see
Figure  8.13). When the diode is just conducting, the fraction is 0.7 because no
electrons return to the cathode and the cathode current is doubled for given space
charge. Thus, if the space-​charge hub is essentially the same as that in a diode at
the point of cut off, the charge density and the electrostatic potential are cylindri-
cally symmetrical, so we can write

(1 − α ) I c + 2αI c ≈ 0.7ICL , (15.99)

where ICL is determined using the Child–​Langmuir law (5.70). Note that this
assumes that the current density has azimuthal variation because the hub is cut off
between the spokes but not at their bases. The multiplying constant on the right-​
hand side can be determined more precisely from rb rc , if necessary (see Figure 8.13
with ra set to the hub radius). The fraction of the cathode current which flows to
the anode (1− α ) is obtained from the range of angles over which current can enter
the spokes so that

n (θ2′ − θ1′ )
α = 1− . (15.100)
2π
From (15.99) the current drawn from the cathode is

0.7
Ic ≈ ICL . (15.101)
1+ α
608 Magnetrons

We note that this tends to the expected values in the limits α → 0 and α → 1. The
anode current is found by substitution in (15.94) to be

1− α
Ia =  0.7 ICL (15.102)
 1 + α 
and the back bombardment current striking the cathode is

 α 
Ib =  I . (15.103)
 1 − α  a
Thus the anode current can be obtained directly from α, the hub radius, and the
anode voltage. The variation in current with changes of the RF voltage is deter-
mined solely by changes in α.

15.6.2 The Rigid Spoke Model

The simplest approach to finding the properties of the spokes was suggested by
Welch [37] and subsequently developed by Hull and Vaughan [13, 59]. A necessary,
but not sufficient, condition for electrons to enter the spokes is that they should be
in the retarding phase of the tangential RF electric field. However, if they are to
be able to reach the anode, then the sum of the DC and RF voltages on the anode
must also exceed the threshold voltage given by (15.39), as shown in Figure 15.33.
Then, in the rotating frame of reference which is synchronous with the wave on
the anode,

Va + V1 cos nθ ′ ≥ VT , (15.104)

where V1 is the amplitude of the synchronous RF wave, which is a travelling wave in

the stationary frame of reference, and a stationary wave in the synchronous frame.
Equation (15.104) shows that the tube can conduct when the anode voltage is below

Figure 15.33:  Criterion for the conduction angle of a magnetron proposed by Welch

(copyright 1953, IEEE, reproduced, with permission, from [37]).
 Simple Magnetron Models 609

the threshold voltage if the inequality in (15.104) is satisfied. The conduction angle
within which an electron can reach the anode ranges from θ1′ = 0 to

1  V − Va 
θ2′ = cos −1  T , (15.105)
n  V1 
and the fraction of the electrons returned to the cathode is

1  V − Va 
α = 1− cos −1  T . (15.106)
2π  V1 
Using the data in Table 15.2 the fractions of the current returned to the cathode
given by (15.106) are 0.67, 0.71, and 0.64, respectively, for the three cases. The first
two figures are close to those in the table so the rigid spoke model is a valid approxi-
mation when the current is space-​charge limited.
When a zero-order hub is assumed, the anode current, obtained from (15.19) and
(15.98), is

θc′
n2 ρb LaV1  Rbn − Rb− n 
Ia =
Bz  R n − R − n 
a a
∫ sin ( nθ′ ) d θ′ (15.107)
0

Evaluating the integral we find that

nρb LaV1  Rbn − Rb− n 

Ia = (1 − cos nθ2′ ) . (15.108)
Bz  Ran − Ra− n 
For a first-order hub, the anode current is given by (15.102), where α is given by
(15.106), and the Child–​Langmuir current has been calculated from (5.70) for a
diode with inner radius rc, outer radius rb , length La, and applied voltage Vb. The tip
of the spoke is assumed to reach the anode at the angle θ s′ = θ2′ 2.

15.6.3 Guiding Centre Orbits

A more accurate model of the spokes is provided by the motion of the guiding cen-
tres of the electron trajectories [60, 61, 63]. In this model the rapid cycloidal motion
of the electrons in the spokes, and the space-​charge field, are ignored. The papers
cited deal only with the case of a planar magnetron but the same approach can also
be used for circular geometry. It is assumed that, in the synchronous frame of refer-
ence, the radial acceleration and the angular velocity are small so that the equations
of motion (15.30) and (15.31) can be rewritten using (15.33) and (15.34) as

 d θ′  d θ′ e
ω c − 2ω s − r =− Er − {ω c − ω s } rω s , (15.109)
 dt  dt m0

and

 d θ ′  dr e
ω c − 2 ω s − 2  = Eθ , (15.110)
 dt  dt m0
610 Magnetrons

where the cyclotron frequency ω c = eBz m0 . Now, typically ω s and d θ ′ dt are

around 10% of ω c so that the products of these terms can be neglected in the first
approximation. Then (15.109) and (15.110) become

d θ′ E
= − r − ω s , (15.111)
dt Bz r
where Er is the sum of the DC and RF fields given by (15.3) and (15.18), and

dr Eθ
= . (15.112)
dt Bz
We note that (15.112) is identical to (15.34) if the tangential force Fθ is negligible.
Now the electric field can be derived from the sum of the DC and RF electrostatic
potentials in the rotating frame of reference so that

d θ′ 1 ∂  m0 
= V − ω s ω c r 2  (15.113)
dt Bz r ∂r  2e 

and

dr 1 ∂V
=− ⋅ . (15.114)
dt Bz r ∂θ ′
These equations can be expressed in terms of a total potential defined by

m0
U =V − ω c ω s (r 2 − rc2 ) , (15.115)
2e
which is zero on the surface of the cathode so that

d θ′ 1 ∂U
= ⋅ (15.116)
dt Bz r ∂r
and

dr 1 ∂U
=− ⋅ . (15.117)
dt Bz r ∂θ ′
From (15.116) we see that on the surface of the cathode, where V = 0, the angular
velocity is

d θ′
= −ω s (15.118)
dt
as expected. Substituting for the DC and RF potentials in the rotating frame of
reference from (15.1), (15.2) and (15.17) we can write (15.115) as

m0
U ( r, θ ′ ) = V0 ( r ) + V1 ( r ) cos ( nθ ′ ) − ω c ω s (r 2 − rc2 ) . (15.119)
2e
 Simple Magnetron Models 611

The kinetic energy of the electrons is

m0 2 2  2
T ( r, θ ′ ) =
2e
( )
r + r θ ′ (15.120)

and the Hamiltonian for the electron trajectories is

H = T + V . (15.121)
This quantity is a constant for the motion of an electron so that the trajectory of an
electron which is initially on the surface of the hub at ( rb , θ0′ ) is given by

H ( r, θ ′ ) = H ( rb , θ0′ ) . (15.122)

It is found that, typically, T ∼ 0.1U so that it can be neglected in the first approxi-
mation. Then, from (15.119)

1  m0 
cos ( nθ ′ ) = V0 ( rb ) − V0 ( r ) + V1 ( rb ) cos ( nθ0′ ) + ω c ω s ( r 2 − rb2 ) .
V1 ( r )  2e 
(15.123)
This equation can be used to calculate the trajectories of the guiding centres of
the electrons forming the spokes. Figure 15.34 shows a typical set of trajectories in
the retarding phase of the tangential electric field. The radius is normalised to the
radius of the hub and the trajectories shown intersect the hub at 30° intervals. The
electrons move outwards from the hub because the field is retarding. Their initial
tangential velocity is greater than the synchronous velocity so that they move for-
wards in phase. In this example, electrons with initial phases between 0° and 120°
can reach the anode, while those with phases in the range 150° to 180° move into the
accelerating phase before reaching the anode. The limiting phase is represented by
a trajectory which is moving directly outwards when its phase is 180°. Trajectories
lying above, and to the right of, this line are non-​physical because they can never

Figure 15.34:  Typical guiding centre trajectories which cross the surface of the hub at 30°
intervals, and the limiting trajectory (broken line).
612 Magnetrons

be reached by electrons starting from the cathode. The trajectories are plotted in
Cartesian coordinates for convenience so that motion to the right corresponds to
anticlockwise rotation of the spokes. Setting d θ ′ dt = 0 in (15.111), substituting for
the radial electric field, and setting nθ ′ = π  gives

e Bz rb2  rc4  nVs  r n + r − n 

⋅
4 m0 r 2 1 − r 4  − B r 2  r n − r − n  − ω s = 0, (15.124)
b z a a

which can be solved for the radius r at which nθ ′ = π . Equation (15.123) can then be
solved for θ2′ corresponding to the limiting trajectory which is plotted as a dashed
line in Figure 15.34. The conduction angle is usually greater than that assumed in
the rigid spoke model. The anode current can be calculated for zero- and first-order
hubs using (15.108) and (15.102). The empirical correction made by Riyopoulos,
that the charge density in a zero-order hub should be approximately half of that
given by (15.96) in order to get agreement with experimental data, has not be used
here. An empirical model of the space-​charge in the spokes is described in [60, 61].

15.6.4 Electron Trajectory Model

The electron trajectories can be calculated directly by numerical integration of
the equations of motion (15.30) and (15.31). The trajectories are initially equally
spaced on the surface of the space-​charge hub with angular velocities given from
(15.1) and (15.4) by

d θ ′ 1  rc2 
= ω c 1 − 2  − ω s . (15.125)
dt 2  rb 
The space-​charge of the spokes, and modification of the hub by the RF fields, are
ignored and the integration is continued until the trajectories end on either the
cathode or the anode. The fraction of the current returned to the cathode and the
limiting initial angles between which trajectories can reach the anode can be deter-
mined directly.
Figure  15.35 shows an example of the trajectories calculated together with
the RF voltage and the threshold voltage normalised to the anode voltage (see
Worksheet 15.3). Figure 15.36 shows a detail of the trajectories in one RF cycle,
plotted using Cartesian coordinates for convenience. The spoke assumed by the
rigid spoke model, and the spoke calculated from the guiding centre model, are
shown, for comparison. These calculations are for the 4J52 magnetron, at the oper-
ating point shown in Figure 15.27. The DC and RF voltages were computed using
the rigid spoke model with a zero-order hub. It can be seen that the spoke calculated
by the guiding centre model is similar to the computed trajectories. The difference
is attributable to the approximations made in the derivation of the guiding centre
model. The phase range within which electrons can reach the anode, and the phase
of the tip of the spoke in the rigid spoke model, are similar to those from the guid-
ing centre model. The principal difference is that the rigid spoke model ignores the
initial angular velocity of the electrons at the base of the spoke.
 Simple Magnetron Models 613

Figure 15.35:  Spokes computed for the 4J52 magnetron at its normal operating point with
the RF voltage and the threshold voltage normalised with respect to the anode voltage. The
rotation of the spokes is counter-​clockwise. The position of the spoke tip given by the rigid
spoke model is shown for comparison.

Figure 15.36:  Detail of the trajectories of the 4J52 magnetron in one RF cycle plotted using
Cartesian co-​ordinates for convenience, together with the spokes predicted by the simple
and the guiding centre models. Note that counter-​clockwise rotation of the spokes is
represented by motion to the right in this figure.
614 Magnetrons

15.6.5 Calculation of the Output Power

The energy with which the electrons strike the anode is

m0 2
Via =
2e
( )
r + ra θ 2 . (15.126)

Substituting for the radial and angular velocities at the tip of the spoke, from
(15.112) and (15.111), and multiplying by the anode current, gives the power dissi-
pated on the anode. This calculation avoids the arbitrary assumption about the tan-
gential velocity made by Vaughan [13]. It should be noted that the angular velocity
of the electrons at the tip of the spoke is normally slightly greater than ω s . It is
not possible to calculate the power dissipated on the cathode in the same manner.
Vaughan [13] proposed the empirical expression

0.04 I aVa
Pc = (15.127)
cos ( nθ )
on the basis that this power is typically around 5% of the input power at the nor-
mal operating point, but it may be up to 50% at very low inputs when nθ s → π 2.
Because the power dissipated on the cathode is usually small, errors in its estima-
tion do not result in major errors in the calculated performance. In the trajectory
calculations the impact energies on the anode and the cathode can be calculated
directly. Since the anode current is known, the RF power transferred to the anode,
and the electronic efficiency, can be calculated from (15.45) and (15.46) as functions
of the anode voltage, the RF wave voltage, and the magnetic field.
A second expression for the RF power delivered to the anode is

1
Pa = (GU + GE )Vg2 (15.128)
2
where GU and GE are the conductances of the anode, and of the external load, and
Vg is the magnitude of the RF voltage as measured between adjacent vane tips.
Now Vg is related to the wave voltage by (15.20) and therefore (15.128) gives the RF
power delivered to the anode as a function of V1. The operating point can then be
determined by solving the simultaneous equations (15.45) and (15.128) for Va and
V1, given the values of I a and Bz . Figure 15.37 shows the performance chart calcu-
lated for the 4J50/​4J52 magnetron for comparison with the experimental results
in Figure  15.27. The rigid spoke model with a zero-order hub was used for this
calculation. Note that the magnetic fields represented by the Gauss lines are not
identical in the two figures, also that this tube differs from the Litton 4J50 modelled
by Vaughan [13]. Overall, the rigid spoke model produces results which are close
enough to those obtained from experiment for initial exploration of the properties
of a proposed design.
Table 15.3 shows a comparison between the results given by these models (see
Worksheet 15.3). Overall the results from both models with zero-order hubs
agree with experiment to a similar extent. Both sets are closer than the results
 Simple Magnetron Models 615

Table 15.3:  4J52 magnetron: Comparison between the results given by the models discussed in the
text, and experimental data from [35]

Hub order Zero First Zero First Experiment

Spoke Rigid Rigid Guiding Guiding

centre centre
Anode voltage (kV) 16.5 18.2 13.2 17.3 16
Va VT 1.15 1.26 0.92 1.20
Anode current (A) 27.0 27.0 27.0 27.0 27
Output power (kW) 198 220 174 248 180
Efficiency (%) 44 45 49 53 42
Anode dissipation 32 32 26 20
(%)
Cathode dissipation 5 5 4 4
(%)
Circuit loss (%) 19 19 20 23
Current fraction 0.69 0.64 0.50 0.62
returned (α )
Wave voltage /​DC 0.36 0.35 0.42 0.39
voltage

Figure 15.37:  Calculated performance chart for the 4J50/​4J52 magnetron for comparison
with Figure 15.27.

of PIC simulation in Table 15.2. Trajectory calculations, based on the electric fields

determined using the rigid spoke model with a zero-order hub, give α = 0.52 . The
reason that this figure is smaller than the one in the table is that some electrons
for which θ1′ < 0 can enter the spokes. Those electrons are accelerated and move
forwards just below the surface of the hub until they enter the retarding phase of
the field. It can be argued that the guiding centre model gives similar results if tra-
jectories lying inside the hub are permitted. Strictly speaking, the Welch criterion
616 Magnetrons

(see Figure 15.33) determines the range of angles from which electrons leaving the
cathode can reach the anode. However, because the cathode current is zero in a
zero-​order hub, it is necessary to make some assumption about the extent of the
bases of the spokes in order to determine the current in them. In addition, the
method for determining the current assumes that the radial velocities of the elec-
trons are zero inside the hub, and non-​zero at the base of the spoke. Thus there are
a number of objections to theories based on the zero-​order hub. In principle, the
use of a first-order hub could overcome these difficulties. In the rigid spoke model
the current drawn from the cathode can be determined directly using the Welch
criterion without making any assumptions about the shape of the spoke. The guid-
ing centre model effectively replaces the Welch criterion. The assumption that the
RF fields are zero inside the hub means that the conduction angle calculated using
the guiding centre model also applies on the surface of the cathode. At present the
rigid spoke model based on a zero-order hub appears to give the best agreement
with experiment.

15.6.6 The Rieke Diagram

If the load is varied we can write the external admittance as

 1 + ρ exp j φ 
YE = GE 0  = GE + jBE , (15.129)
 1 − ρ exp j φ 
where ρ is the voltage reflection coefficient of the load and φ its phase angle.
Substituting this expression into (15.128), and applying the same method as before,
we can plot the contours of constant power on a Rieke diagram. Also, using
(15.129) in combination with (15.89), the frequency shift as a function of reflection
coefficient and phase angle is

f0
∆f = BE (15.130)
2QE
from which the contours of constant ∆f can be plotted. Figure  15.38 shows the
Rieke diagram computed for the 4J50 magnetron, operating with a constant cur-
rent of 8.8A, for comparison with Figure 15.29 (see Worksheet 15.3). This current
was chosen so that the output power into a matched load was the same as in the
experimental data. It should be remembered that the zero of the phase is arbitrary
because of the convention that the experimental data is referred to the plane of the
output flange.

15.6.7 Frequency Pushing

The RF current in the anode circuit is given by

  ω ω0  
I1 = VgGL 1 + jQL  − . (15.131)
  ω 0 ω  
 Simple Magnetron Models 617

Figure 15.38:  Calculated Rieke diagram of the 4J50 magnetron for comparison with
Figure 15.29.

Thus the phase angle of the current relative to the voltage is given by

 ω ω0  2 ∆ω
tan φ = QL  −  ≈ QL . (15.132)
 ω0 ω  ω0

Frequency pushing arises because the frequency must change to match the phase
of the RF anode current so that (15.132) is satisfied. The RF current flowing into
the anode comprises two components: the conduction current intercepted by the
anode, and the induced current from the transverse motion of the charge in the
spokes. Reference to Figure 15.36 shows that, at the moment when the RF voltage
is maximum on the vane at θ = 0, the position of the tip of the spoke is θ = θ s and
moving in the positive direction of θ. To a first approximation we may assume that
this is also the effective position of the spoke for calculating the induced current.
The power transferred to the anode by the induced current is maximum when the
phase of the spoke coincides with the position at which the tangential electric field
is retarding (θ = π 2n ). Thus, the phase of the spoke relative to the tangential field is

π
φ s = nθ s − . (15.133)
2
The power transferred to the anode by the conduction current is maximum when
the phase of the spoke coincides with the vane on which the voltage has its greatest
negative value (θ = π n ). Hence the phase of the conduction current is

φc = nθ s − π. (15.134)
Comparing (15.134) with (15.133) we see that the conduction current lags the
induced current by 90°. Thus the currents can be represented by the phasor
618 Magnetrons

Figure 15.39:  Phasor diagram for calculating frequency pushing.

diagram in Figure 15.39 where I1θ is the induced current, and I1r is the conduc-
tion current. The phasor sum of these currents ( I1 ) is the net current flowing into
the anode.
It is difficult to find the induced current directly but, fortunately, this is not neces-
sary. The conduction current can be approximated by a rectangular pulse of phase
duration 2∆ and amplitude ( π ∆ ) I a where ∆ is the width of the tip of the spoke.
Then, using Fourier analysis, it can be shown that the amplitude of the first har-
monic of the conduction current is

sin ∆
I1r = 2 I a . (15.135)
∆
Now, from Figure 15.36 we see that ∆ is typically of the order of 10° so that I1r = 2 I a
to a very good approximation. The power transferred to the travelling voltage wave
on the anode by the conduction current is

1
Pr = I1rV1 cos φc . (15.136)
2
This can also be expressed in terms of the associated component of the current ( I gr )
flowing in the anode circuit

1
Pr = I grVg cos φc (15.137)
2
so that

V1
I gr = 2 I a , (15.138)
Vg

where V1 Vg is given by (15.20). The component of the RF current which is in phase

with the gap voltage is

I g cos φ = Vg (GU + GE ) . (15.139)

Then, from the phasor diagram in Figure 15.39,

Vg (GU + GE ) = I gθ cos φ s + I gr sin φ s (15.140)

 Magnetron Design 619

Figure 15.40:  Calculated frequency pushing curves for the 4J50/​4J52 magnetron.

or

Vg (GU + GE ) − I gr sin φ s
I gθ = , (15.141)
cos φ s
and

I gθ sin φ s − I gr cos φ s
tan φ = , (15.142)
Vg (GU + GE )

so that the frequency pushing curve can be calculated by substitution in (15.132).

Figure 15.40 shows the frequency pushing curves calculated for the 4J50/​4J52 mag-
netron for various values of the magnetic field.

15.7 Magnetron Design

The conceptual design of a magnetron is based on the theory described above. For
further information about the design of magnetrons see [1, 64]. Discussion of the
design of magnetrons of different types can also be found in [29, 33, 65–​67].

15.7.1 Dimensionless Parameters

Magnetrons, like other tubes, are commonly designed by scaling from existing
designs. We have seen in previous chapters that this means that geometric similarity
and dynamic similarity must both be maintained. The first requirement is that all
dimensions, and the wavelength at the π mode, are scaled by the same factor. The
second requirement means that the tubes must be described by the same normalised
parameters. The anode voltage and the magnetic field can be normalised by divid-
ing them by the characteristic voltage (V0 ) given by (15.22) and the characteristic
field ( B0 ) given by (15.23). We recall that the point defined by these values lies on
620 Magnetrons

the cut-​off curve of the magnetron diode. A  scale factor for the current can be
defined as the current which flows in a diode when it is just not cut off [9]. We have
seen that this current is approximately 0.7 times the Child–​Langmuir current, where
the constant is a function only of the ratio ra rc provided that the emission from
the cathode is great enough to ensure that the current is limited by space-​charge
(see Section 8.5). The scale factors for power and impedance can then be defined
as P0 = I 0V0 and G0 = I 0 V0 . The operating point on the performance chart of a
tube is uniquely defined by any two of: the magnetic field; the anode voltage; and
the anode current. Thus, a universal performance chart can be drawn in terms of
dimensionless parameters. This applies to all geometrically similar magnetrons hav-
ing the same normalised values of GE G0  and GU G0 .
An alternative set of dimensionless parameters, which assumes geometrical simi-
larity, is [64]

ηVa
v= , (15.143)
ra2 ω 2s

ηBz ω c
b= = , (15.144)
ωs ωs
and

ηI a
i= . (15.145)
ε 0 ra3 ω3s

15.7.2 Design Parameters

The basic specification of a magnetron includes the frequency, the power output,
and the duty cycle. Because the current in a magnetron does not obey the 3/​2 power
law the concept of perveance is not useful. It is usual, instead, to use estimates of
the efficiency, and the DC impedance, to find initial figures for the DC current and
voltage. Table 15.4 shows typical values for different types of magnetron based on
manufacturers’ data sheets. The circuit efficiency is typically 90–​95% at frequencies
up to 3 GHz, falling to 70–​80% at 9 GHz, and 50–​60% at 24 GHz.

15.7.3 Design Case Study

This section illustrates the conceptual design of a 500 kW, 915 MHz, CW magne-
tron for microwave heating described by Shibata in [66] (see Worksheet 15.4). The
principal steps of the design calculation below are not identical to those used by
Shibata but the parameters have been chosen to give very nearly the same results.
i) Since this is an exceptionally high-​power tube we estimate, conservatively, from
Table 15.4 that the efficiency will be 80%. In order to keep the anode voltage as
 Magnetron Design 621

Table 15.4:  Typical impedances and efficiencies for different types of magnetron. The output powers and anode
currents for pulsed tubes are peak values

Anode Duty Frequency Output power Anode Impedance Overall

(GHz) (kW) current ( kΩ) Efficiency
(A) (%)

Strapped CW 0.9 30–​100 3–​6 3.0–​5.0 80–​90

Strapped CW 2.45 0.2–​30 2–​40 3.0–​7.0 65–​70
Strapped pulse 2–​4 5–​4500 4–​130 0.5–​1.5 35–​50
Strapped pulse 4–​8 200–​1000 25–​80 0.5–​1.0 30–​50
Strapped pulse 8–​11 1–​2000 2–​100 0.4–​1.4 30–​50
Strapped pulse 11–​18 20–​250 8–​28 0.8–​1.2 20–​45
Rising sun pulse 25–​35 40–​125 13–​28 0.7–​1.0 20–​25
Coaxial pulse 8–​10 10–​350 4–​27 0.8–​1.1 30–​50
Coaxial pulse 14–​17 1–​180 2–​25 0.8–​1.9 20–​40
Coaxial pulse 32–​36 25–​130 10–​20 1.0–​1.4 20–​30
Long anode pulse 3 2000–​5500 50–​200 0.2–​0.5 45–​55

low as possible the DC impedance is taken to be 2.9 kΩ. The anode voltage and
current are then found to be 44.0 kV and 15.2 A.
ii) The normalised magnetic field is related to the electronic efficiency by (15.49).
However, this equation does not take account of the power lost by cathode
back-​bombardment and by RF heating of the anode. If the circuit efficiency
is assumed to be 90%, and the cathode back-​bombardment power is ignored,
then Bz B0 = 6.5 and, from (15.40), VT V0 = 12.0.
iii) Figure 15.27 shows that the working anode voltage of a magnetron is usually
close to the threshold voltage. Therefore, we assume that Va VT = 1 so that the
characteristic voltage V0 = 3.7 kV. The cut-​off voltage is found from (15.25) to
be 155 kV.
iv) From (15.22) the anode radius is given, as function of the number of vanes, by

2 ( e m0 )V0
ra = Nv . (15.146)
2ω
The choice of the radius of the space-​charge hub is very important for correct
magnetron operation. Since the RF fields in a planar magnetron and a cylin-
drical magnetron are similar it is possible to deduce the design criteria for a
cylindrical tube from a consideration of the simpler, planar, geometry. If the
normalised depth of the space-​charge hub in a planar magetron ( d h ) is too
small there is a large gap between the surface of the space-​charge layer and the
anode, and the field of the π mode at the surface of the hub is smaller than that
of the π −1 mode. There is, then, a serious risk of π −1 mode excitation. If d h is
622 Magnetrons

too large then the potential difference between the surface of the space charge
layer and the anode is small and the tube efficiency is low.
In the π mode the normalised tangential RF field on the surface of a planar
hub obtained from (15.11) is

E x ( x, d ) sinh ( πd p )
= ≈ exp ( π ( d − h ) p ) . (15.147)
E x ( x, h ) sinh ( πh p )

The approximation in (15.147) is valid if d p > 2. Then a figure of merit (the

Slater ratio) can be defined by

E x ( x, d ) π ( h − d )
R = − ln = . (15.148)
E x ( x, h ) p

We expect that empirical evidence will suggest the best value of R for optimum
interaction. Eliminating Bz between (8.12) and (8.25) gives

Va d  d
=  2 −  . (15.149)
VH h  h
Solving for d h gives

d V
= 1 − 1 − a . (15.150)
h VH
Then, substituting into (15.148), we have

πh V
R= 1 − a . (15.151)
p VH
In this equation the inaccessible variable d has been replaced by the ratio of
the anode voltage to the cut-​off voltage. It is not possible to find an equivalent
closed-​form expression for circular geometry, so (15.151) is adapted by assuming
that the planar parameters can be defined in terms of the circular parameters by

h = ra − rc , (15.152)

and

π ( ra + rc )
p= . (15.153)
Nv
This definition of p assumes that it is the mean circumference of the interaction
region which matters. Note that p is not the pitch of the circular anode. The fig-
ure of merit for a cylindrical magnetron, known as the Modified Slater Factor,
is therefore given by

r −r  V
R ′ = Nv  c a  1 − a . (15.154)
 rc + ra  VH
 Magnetron Design 623

Experience suggests that R ′ should have a value of about 2 for a wide range of
designs. Making use of (15.146) and (15.154) we can tabulate the anode and
cathode radii for a range of values of Nv . We recall that the length of the anode
should be small, compared with the free-​space wavelength, to avoid the excita-
tion of longitudinal modes. Since this is an exceptionally high-​powered tube we
choose La = 0.16 λ = 53 mm. Then the ratio of the anode current to the surface
area of the cathode can be calculated. If it is assumed that 75% of the electrons
are returned to the cathode then the cathode current is four times the anode
current so that the cathode current density can be calculated. The results of this
calculation are shown in Table 15.5.
In a very high power CW magnetron the design is limited by the maximum
temperature of the vane tips rather than by the cathode current density. If the
power dissipated on the cathode is neglected the power dissipated by the is
anode is given by

Pa = Pdc − Prf , (15.155)

and the vane tip temperature is

Lth
Tv = Pa + Tr , (15.156)
κ Av
where Lth is the thermal path distance to the cooling channel, κ is the thermal
conductivity of copper (401 W m−​1 K−​1),Av is the total cross-​sectional area of
the vanes normal to the heat flow, and Tr is temperature of the cooling channel.
The thickness of the vanes (t ) is chosen to that the ratio of the thickness to the
pitch is approximately 0.5. The pitch calculated from (15.146) is 20 mm, so let
us choose the vane thickness to be 12 mm. To cool the anode effectively water
cooling channels must be embedded in the vanes. Let us suppose that the effec-
tive thermal path is 10 mm. The vane tip temperatures calculated assuming that
Tr = 293 K are shown in Table 15.5.
All the numbers of vanes considered result in cathode currents which are com-
fortably within the capability of thoriated tungsten [66, 68]. This is the preferred
choice of cathode material because a back-​bombardment power of around 14 kW
is anticipated. The experimental tube used a pure tungsten wire cathode which
was pre-​heated to 1870 °C and operated at 2100 °C. The cathode terminals were
water-​cooled to reduce the heating by back-​bombardment. The maximum anode
temperature must be less than 700 °C to avoid re-​crystallisation of the oxygen-​
free high-​conductivity copper of which it is made. To allow a safety margin to
cover local hot-​spots the maximum temperature should be 500 °C (773 °K). It
is desirable to keep the number of vanes as small as possible to reduce problems
with excitation of the π −1 mode. Therefore we choose a 14-​vane anode.
v) Once the cathode and anode radii have been determined, the characteristic field
can be found from (15.23) and, hence, the magnetic flux density required. The
results are: B0 = 0.019 T and Bz = 0.121 T. We note that Bz is well below the sat-
uration flux density for soft iron (see Section 19.4.1), and therefore no problems
624 Magnetrons

Table 15.5:  Design choices for a 500 kW magnetron

Nv ra (mm) rc (mm) Jc (A cm–2) Tv (K)

8 25.2 13.7 1.34 1117

10 31.4 19.4 0.95 952
12 37.7 25.3 0.73 842
14 44.0 31.3 0.59 764
16 50.3 37.3 0.49 705
18 56.6 43.4 0.42 659
20 62.9 49.6 0.37 622
22 69.2 55.7 0.33 593
24 75.5 61.9 0.30 568

with non-​uniformity of the field should occur. The magnetic field of the experi-
mental tube described in [66] was provided by a solenoid magnet. The ratio of
the cyclotron frequency to the frequency of oscillation is 3.7 so that the differ-
ence is much greater than the figure of 20% which is desirable.
vi) The threshold voltage for the π −1 mode can be found from (15.40) after calcu-
lating V0 and B0 with n = 6. The result is 50.6 kV which is comfortably above the
working voltage.
The experimental magnetron had a coaxial anode to increase the circuit efficiency
and the overall efficiency. No details of the properties of the anode are given in [66].
Although they could be calculated we shall infer them from simulation of the tube
using the rigid spoke model with a zero-​order hub. Table 15.6 shows the leading
parameters and performance characteristics of this tube.

15.7.4 Other Considerations

The analysis and design methodology described above involve a number of sim-
plifying assumptions which may not always be valid. It is found that magnetrons
do not always behave as expected, for reasons which are not understood. The phe-
nomenon of twinning (see Section 15.4.4) is just one example. These problems can
occur as a result of tiny differences in manufacture between different batches of
tubes. They may also depend upon the history of operation, and have a big effect on
the satisfactory operation of the tube and its lifetime. Possible causes include: vari-
ations in the primary and secondary emission properties of the cathode; changes
in the resistance of the heater; higher-order resonances of the anode and the end
spaces; the external impedance presented by the power supply; and the presence of
residual gas. These problems are generally solved by empirical ‘fixes’ and further
research is needed to explain and understand them. Cooker magnetrons, for exam-
ple, require a filter box connected to the cathode, to control the impedance between
the anode and the cathode, if they are to work correctly. It has been found difficult
to scale CW magnetrons to work at frequencies above 2.45 GHz [69].
 References 625

Table 15.6:  Leading parameters and performance characteristics of a 500 kW magnetron

Parameter Calculated Calculated [66] Measured [66]

Va (kV) 44.0 44 44
Ia (A) 15.2 15 16
Bz (T) 0.121 0.10 0.10
V0 (kV) 3.7 3.7
B0 (T) 0.019 0.0153
VT V0 11.9 12.0
Bz B0 6.4 6.5
Nv 14 14 14
ra rc 1.41 1.59
rc (mm) 31 27.7
ra (mm) 44 44
La (mm) 53 (50)
t (mm) 12 (12.1)
Y0 (1/​Ω) 0.47
QU 1000
Prf (kW) 510 kW 421
ηO 81% 72.5%

References

[1]‌ E. Okress, Ed., Crossed-​Field Microwave Devices. New York: Academic Press, 1961.

[2]‌ A. S. Gilmour, Jr., Klystrons, Traveling Wave Tubes, Magnetrons, Crossed-​ Field
Amplifiers and Gyrotrons. Norwood, MA: Artech House, 2011.
[3]‌ L. Sivan, Microwave Tube Transmitters. Kluwer Academic Publishers, 1994.
[4]‌ W. C. Brown, ‘The microwave magnetron and its derivatives’, IEEE Transactions on
Electron Devices, vol. 31, pp. 1595–​1605, 1984.
[5]‌ G. Faillon et al., ‘Microwave tubes’, in J. A. Eichmeier and M. K. Thumm, eds, Vacuum
Electronics: Components and Devices. Berlin: Springer-​Verlag, pp. 1–​84, 2008.
[6]‌ S. H. Gold and G. S. Nusinovich, ‘Review of high-​power microwave source research’,
Review of Scientific Instruments, vol. 68, p. 3945–3974, 1997.
[7]‌ J. Benford et al., High Power Microwaves. CRC Press, 2015.
[8]‌ J. F. Hull, ‘Inverted Magnetron’, Proceedings of the IRE, vol. 40, pp. 1038–​1041, 1952.
[9]‌ J. C. Slater, Microwave Electronics. New York: D. van Nostrand, 1950.
[10] J. F. Skowron, ‘The continuous-​ cathode (emitting-​ sole) crossed-​field amplifier’,
Proceedings of the IEEE, vol. 61, pp. 330–​356, 1973.
[11] W. A. Smith, ‘A wave treatment of the continuous cathode crossed-​field amplifier’, IRE
Transactions on Electron Devices, vol. 9, pp. 379–​387, 1962.
[12] R. G. E. Hutter, Beam and Wave Electronics in Microwave Tubes. Princeton, NJ: D. van
Nostrand, 1960.
[13] J. R.  M. Vaughan, ‘A model for calculation of magnetron performance’, IEEE
Transactions on Electron Devices, vol. 20, pp. 818–​826, 1973.
626 Magnetrons

[14] G. B. Collins, Microwave Magnetrons. New York: McGraw-​Hill, 1948.

[15] N. Kroll, ‘The unstrapped resonant system’, in G. B. Collins, ed., Microwave Magnetrons.
New York: McGraw-​Hill, pp. 49–​82, 1948.
[16] S. Millman and W. V. Smith, ‘The resonant system’, in G. B. Collins, ed., Microwave
Magnetrons. New York: McGraw-​Hill, pp. 460–​502, 1948.
[17] R. G. Carter, Electromagnetism for Electronic Engineers. Fredriksberg, Denmark: Ventus
Publishing, 2010.
[18] R. M. Gilgenbach et al., ‘Cathode priming of magnetrons for rapid startup and mode-​
locking’, in The Joint 30th International Conference on Infrared and Millimeter Waves
and 13th International Conference on Terahertz Electronics, vol. 2, pp. 535–​536, 2005.
[19] V. B. Neculaes et al., ‘Magnetic perturbation effects on noise and startup in DC-​operating
oven magnetrons’, IEEE Transactions on Electron Devices, vol. 52, pp. 864–​871, 2005.
[20] V. B. Neculaes et al., ‘Magnetic priming effects on noise, startup, and mode competition
in magnetrons’, IEEE Transactions on Plasma Science, vol. 33, pp. 94–​102, 2005.
[21] J. I. Kim et al., ‘Reduction of noise in strapped magnetron by electric priming using
anode shape modification’, Applied Physics Letters, vol. 88, p. 221501, 2006.
[22] W. V. Smith, ‘Mechanical tuning’, in G. B. Collins, ed., Microwave Magnetrons.
New York: McGraw-​Hill, pp. 561–​591, 1948.
[23] M. J. Bernstein and N. M. Kroll, ‘Mechanically tuned rising-​sun magnetrons’, in
E. Okress, ed., Crossed-​Field Microwave Devices, vol. 2. New  York:  Academic Press,
pp. 149–​153, 1961.
[24] R. B. Nelson, ‘Methods of tuning multiple-​cavity magnetrons’, Proceedings of the IRE,
vol. 36, pp. 53–​56, 1948.
[25] A. Pickering et  al., ‘Electronically tuned pulse magnetron’, in IEEE International
Electron Devices Meeting, pp. 145–​148, 1975.
[26] S. P. Yu and P. N. Hess, ‘Slow-​wave structures for M-​type devices’, IRE Transactions on
Electron Devices, vol. 9, pp. 51–​57, 1962.
[27] J. F. Hull, ‘Crossed field electron interaction in space charge limited beams’, Doctor of
Electrical Engineering, Polytechnic Institute of Brooklyn, New York, 1958.
[28] J. Feinstein and R. J. Collier, ‘The circular electric mode magnetron’, in E. Okress, ed.,
Crossed-​Field Microwave Devices, vol. 2. New York: Academic Press, pp. 123–​134, 1961.
[29] A. J. Bamford et al., ‘A 1-​MW L-​band coaxial magnetron with separate cavity’, IEEE
Transactions on Electron Devices, vol. 14, pp. 844–​851, 1967.
[30] T. E. Ruden, ‘Design and performance of a one megawatt, 3.1–​3.5 GHz coaxial magne-
tron’, in 9th European Microwave Conference, pp. 731–​735, 1979.
[31] A. H. Pickering, ‘Further developments of long anode magnetrons’, in E. Okress, ed.,
Crossed Field Microwave Devices, vol. 2. New York: Academic Press, pp. 275–​290, 1961.
[32] H. A. H. Boot, ‘Long anode magnetrons’, in E. Okress, ed., Crossed Field Microwave
Devices, vol. 2. New York: Academic Press, pp. 261–​274, 1961.
[33] H. A. H. Boot et al., ‘A new design of high-​power S-​band magnetron’, Proceedings of
the IEE –​Part B: Radio and Electronic Engineering, vol. 105, pp. 419–​425, 1958.
[34] J.-​J. Feng et  al., ‘Simulation of a long anode magnetron resonant system using
MAFIA’, in International Conference on Microwave and Millimeter Wave Technology,
pp. 748–​751, 1998.
[35] A. G. Smith, ‘Typical magnetrons’, in G. B. Collins, ed., Microwave Magnetrons.
New York: McGraw-​Hill, pp. 739–​796, 1948.
 References 627

[36] C. R. Schumacher, ‘Frequency pushing’, in E. Okress, ed., Crossed-​Field Microwave

Devices, vol. 2. New York: Academic Press, pp. 401–​422, 1961.
[37] H. W. Welch, ‘Prediction of traveling wave magnetron frequency characteristics:  fre-
quency pushing and voltage tuning’, Proceedings of the IRE, vol. 41, pp. 1631–​1653, 1953.
[38] W. L. Pritchard, ‘Long-​ line effect and pulsed magnetrons’, IRE Transactions on
Microwave Theory and Techniques, vol. 4, pp. 97–​110, 1956.
[39] R. Yokoyama and A. Yamada, ‘Development status of magnetrons for microwave
ovens’, in Microwave Power Symposium, pp. 132–​135, 1996.
[40] J. M. Osepchuk, ‘The cooker magnetron as a standard in crossed-​field device research’,
in 1st Int. Workshop on Crossed-​ Field Devices, University of Michigan, USA,
pp. 159–177, 1995.
[41] C. R. Schumacher, ‘Spectrum shape’, in E. Okress, ed., Crossed-​Field Microwave
Devices, vol. 2. New York: Academic Press, pp. 457–​471, 1961.
[42] R. G. Carter et  al., ‘Magnetron frequency twinning’, IEEE Transactions on Plasma
Science, vol. 28, pp. 905–​909, June 2000.
[43] ‘Preamble: Magnetrons’, Chelmsford, UK: EEV Ltd., 1974.
[44] Varian, ‘Technical Manual: Installation, Operation, Maintenance, Care and Handling
Instructions, General:  Microwave Tubes, Magnetron Tubes, Electron Tubes’, 1
October 1979.
[45] R. Adler, ‘A study of locking phenomena in oscillators’, Proceedings of the IRE, vol. 34,
pp. 351–​357, 1946.
[46] H. L. Thal and R. G. Lock, ‘Locking of magnetrons by an injected RF signal’, IEEE
Transactions on Microwave Theory and Techniques, vol. 13, pp. 836–​846, 1965.
[47] I. Tahir et  al., ‘Noise performance of frequency-​and phase-​locked CW magnetrons
operated as current-​controlled oscillators’, IEEE Transactions on Electron Devices, vol.
52, pp. 2096–​2103, September 2005.
[48] I. Tahir et al., ‘Frequency and phase modulation performance of an injection-​locked
CW magnetron’, IEEE Transactions on Electron Devices, vol. 53, pp. 1721–​ 1729,
July 2006.
[49] W. C. Brown, ‘The sophisticated properties of the microwave oven magnetron’, in IEEE
MTT- S International Microwave Symposium Digest, vol. 3 , pp. 871–874, 1989.
[50] E. E. David, Jr., ‘Phasing by RF signals’, in E. Okress, ed., Crossed-​Field Microwave
Devices, vol. 2. New York: Academic Press, pp. 375–​399, 1961.
[51] H. Kim and J. Choi, ‘Characterization of a 16-​vane strapped magnetron oscillator by
three-​dimensional particle-​in-​cell code simulations’, Current Applied Physics, vol. 6,
pp. 66–​70, 2006.
[52] A. D. Andreev and K. J. Hendricks, ‘ICEPIC simulation of a strapped nonrelativistic
high-power CW UHF magnetron with a solid cathode operating in the space-charge
limited regime’, IEEE Transactions on Plasma Science, vol. 40, pp. 1551–​1562, 2012.
[53] Y. Alfadhl et al., ‘Advanced computer modelling of magnetrons’, in 2010 International
Conference on the Origins and Evolution of the Cavity Magnetron (CAVMAG),
pp. 67–​70, 2010.
[54] G. E. Dombrowski, ‘Simulation of magnetrons and crossed-​field amplifiers’, IEEE
Transactions on Electron Devices, vol. 35, pp. 2060–​2067, 1988.
[55] G. E. Dombrowski, ‘Computer simulation study of primary and secondary anode load-
ing in magnetrons’, IEEE Transactions on Electron Devices, vol. 38, pp. 2234–​2238, 1991.
628 Magnetrons

[56] H. L. McDowell, ‘Magnetron simulations using a moving wavelength computer code’,

IEEE Transactions on Plasma Science, vol. 26, pp. 733–​754, 1998.
[57] H. L. McDowell, ‘Magnetron simulations using a multiple wavelength computer code’,
IEEE Transactions on Plasma Science, vol. 32, pp. 1160–​1170, 2004.
[58] H. W. Welch and W. G. Dow, ‘Analysis of synchronous conditions in the cylindrical
magnetron space charge’, Journal of Applied Physics, vol. 22, pp. 433–​438, 1951.
[59] J. F. Hull, ‘Crossed-​field electron interaction of the distributed-​emission space-​charge-​
limited type’, IRE Transactions on Electron Devices, vol. 8, pp. 309–​323, 1961.
[60] S. Riyopoulos, ‘Magnetron theory’, Physics of Plasmas, vol. 3, pp. 1137–​1161, 1996.
[61] S. Riyopoulos, ‘New improved formulas for magnetron characteristic curves’, IEEE
Transactions on Plasma Science, vol. 26, pp. 755–​766, 1998.
[62] E.-​Q. Zhang, ‘On the magnetron cathode’, IEEE Transactions on Electron Devices, vol.
33, pp. 1383–​1384, 1986.
[63] J. Feinstein, ‘Planar magnetron theory and applications’, in E. Okress, ed., Crossed-​
Field Microwave Devices, vol. 1. New York: Academic Press, 1961.
[64] A. M. Clogston, ‘Principles of design’, in G. B. Collins, ed., Microwave Magnetrons.
New York: McGraw-​Hill, pp. 401–​459, 1948.
[65] J. R. G. Twisleton, ‘Twenty-​kilowatt 890 Mc/​s continuous-​wave magnetron’, Proceedings
of the Institution of Electrical Engineers, vol. 111, pp. 51–​56, 1964.
[66] C. Shibata et al., ‘High-​power (500 kW) c.w. magnetron for industrial heating’, Electrical
Engineering in Japan, vol. 111, pp. 94–​100, 1991.
[67] A. M. Cripps and P. A. Jerram, ‘X-​band linear accelerator magnetrons’, in 19th European
Microwave Conference, pp. 1086–​1090, 1989.
[68] G. R. Feaster, ‘The cathode’, in E. Okress, ed., Crossed-​Field Microwave Devices, vol. 1.
New York: Academic Press, pp. 113–​140, 1961.
[69] J. M. Osepchuk, ‘Private communication’, 2016.
16 Crossed-​Field Amplifiers

16.1 Introduction

Broad-​band crossed-​field amplifiers (CFAs), based on electron flow in crossed

fields, are analogous to travelling-​wave tubes. In essence, a CFA is a magnetron in
which the slow-​wave structure is not a closed loop but has been broken so that the
two ends can be connected to external waveguides. CFAs share with the magnetron
the advantages of small size, low weight, and high efficiency. However they have
the disadvantages of low gain, high noise levels and non-​linearity compared with
TWTs. The applications of CFAs are mainly in military radar systems. For this rea-
son the available literature is sparse.
In theory many possible types of crossed-​field tube exist, but not all of these have
been found to be practical devices. The different types can be classified by consid-
ering the options for the tube geometry and for the type of beam and slow-​wave
structure used. The slow-​wave structure fulfils the same role as the anode in a mag-
netron1. The geometry may be either linear or circular, and the electron beam may
be emitted from the whole area of the cathode, as in a magnetron, or injected from
an electron gun (see Figure 8.1). Cylindrical geometry has the advantage of com-
pactness and better use of the permanent magnet material. Planar geometry offers
the possibility of greater interaction length and gain, but it has only been used in
experimental tubes [1–​4]. Figure 16.1 shows the arrangements of circular emitting
cathode, and injected beam, CFAs.
The slow-​wave structure, like that in a TWT, can have either forward-​wave or
backward-​wave characteristics, as shown in Figure  16.2. With a forward-​wave
structure the interaction takes place instantaneously over a band of frequencies.
With a backward-​wave structure the interaction is narrow band, and can be tuned
over a range of frequencies by changing the electron drift velocity. It should be
noted that the drift velocity corresponds to the voltage on the surface of the elec-
tron hub, which is normally around an order of magnitude less than the anode
voltage. Because the slow-​wave structure is periodic it has forward and backward-​
wave characteristics in successive Brillouin zones. Thus, the same tube can be
operated as either a forward-​wave or a backward-​wave amplifier by changing the
anode voltage, and the magnetic field, to achieve synchronism with the appropriate

1
In the older literature the cathode and the slow-​wave structure are called the sole, and the line, respectively.
630 Crossed-Field Amplifiers

Figure 16.1:  Arrangement of: (a) emitting cathode CFA, and (b) injected beam crossed-​field
amplifiers
(copyright 1973, IEEE, reproduced, with permission, from [5]).

Figure 16.2:  CFA dispersion diagrams, normalised to the upper cut-​off frequency, showing
the electron velocity line intersecting at π 2 phase shift per cell: (a) forward wave, and
(b) backward wave.

space-​harmonic [6]. The circular geometry means that care must be taken to avoid
positive feedback through RF coupling between the ends of the slow-​wave struc-
ture, by allowing sufficient separation between them.
The interaction in a CFA resembles that in a magnetron. The electrons retarded
by the RF electric field move outwards towards the slow-​wave structure, while those
accelerated move inwards towards the cathode. The resemblance is especially close
 Introduction 631

in an emitting cathode CFA where spokes of charge are formed, and all the elec-
trons are collected on either the slow-​wave structure or the cathode.

16.1.1 Emitting Cathode CFAs

All crossed-​field amplifiers which are of continuing importance are of the circular,
re-​entrant, emitting cathode type and the abbreviation CFA is now generally taken
to refer to them [5, 7–​11]. The detailed discussions, in sections from 16.2 onwards,
refer exclusively to tubes of this kind. The slow wave structure can be either forward-​
or backward-​wave, but both have the arrangement shown in Figure  16.1(a), and
it is convenient to discuss them together. The input and output connections are
made according to the structure type and the direction of rotation of the electrons.
A drift region may be included to control the feedback caused by the use of a re-​
entrant electron beam. In backward-​wave tubes the drift region is sometimes omit-
ted, though the result is to produce frequencies within the band at which the output
power is low. Backward-​wave emitting-​cathode CFAs are sometimes known by the
proprietary name ‘Amplitron’. The electron beam is demodulated to some extent as
it passes through the drift space but some modulation is carried forward to the start
of the interaction.
Emitting cathode CFAs typically provide average powers ranging from 400 kW
at 2.8 GHz to 100 W at 30 GHz, with efficiency in the range 35–​60% and bandwidth
10–​15% [12]. The mean output power is limited by heat dissipation in the slow-​wave
structure. However, peak powers from hundreds of kilowatts to megawatts, with
duty cycles around 0.001, may be obtained with good efficiencies. The gain is low
(8–​15 dB) because of the limited length of the slow-​wave structure, and the need to
lock out noise-​generating mechanisms. Where high gain is required a chain of tubes
is used. Comparisons with systems employing TWTs should be made with whole
systems, and not just the power output tubes. The final CFA of a chain can be made
‘transparent’ so that the system has both high and low output power capabilities.
Low power output is obtained by leaving the output tube switched off so that the
power of the penultimate tube passes through it with little attenuation. The CFA
is particularly suited for use as the final amplifier for mobile radar applications
because it is smaller, lighter, more efficient, and operates at a lower voltage, than an
equivalent coupled-​cavity TWT. CFAs are non-​linear amplifiers at all power levels.
They are therefore unsuited to multi-​carrier operation because of the generation
of intermodulation products. In radar applications the bandwidth can be used to
achieve frequency agility. For this purpose either forward-​or backward-​wave inter-
actions may be used though the latter requires a more complicated modulator to
adjust the anode voltage for each frequency.
When making comparisons between the efficiencies of different tube types it is
important to be careful about the way in which the efficiency is defined. In linear
beam tubes the heater power, and any solenoid power, need to be included. In CFAs
the low gain means that the RF input power is important. The best approach is to
compare the efficiencies of complete amplifier chains having the same overall gain
632 Crossed-Field Amplifiers

Figure 16.3:  Efficiencies of emitting cathode CFAs.

Figure 16.4:  Peak power output of emitting cathode CFAs.

and output power. Figure 16.3 shows the efficiencies of emitting cathode CFAs, as
far as they can be ascertained from published data. For this purpose the efficiency
has been defined as the ratio of the RF output power to the DC input power. The
state of the art of the peak power of emitting cathode CFAs is shown in Figure 16.4.
The signal to noise ratio of a CFA is typically −45 to −55 dBc/​MHz, which is
comparable with that of a TWT [12]. This noise is thought to arise from turbulence
within the space-​charge hub. It has been reported that the noise level can be reduced
by shaping the magnetic field, or by incorporating a slow-​wave structure in the cath-
ode (see Section 16.2.3), but no modern CFAs use these techniques [13].
An emitting-​cathode CFA in which the electrons are collected at the end of the
interaction region, in either a linear or a circular tube, is known by the proprietary
name DEMATRON [1, 14]. Although such an amplifier has theoretical attractions,
it has been found that gain is lower than expected because of the variation in the
electron velocity caused by the space-​charge in the spokes. This can be compensated
 CFA Construction 633

for by tapering the DC electric field, the magnetic field, or the phase velocity of the
slow-​wave structure. It has not been found possible to build practical non-​re-​entrant
CFAs of this kind whose performance is competitive with other types of tube.

16.1.2 Injected Beam CFAs

Injected-​ beam CFAs have been found to be inferior to linear-​ beam tubes in
terms of power and bandwidth, and they are now chiefly of historical interest [7].
Figure  16.1(b) shows the arrangement of an injected beam CFA. Forward-​wave
amplifiers and backward-​wave oscillators have essentially the same construction.
The beam is in the form of a flat strip generated by a Kino gun (Figure 9.20). The
electron velocity can be controlled by the accelerator voltage, and by the poten-
tial difference between the slow-​wave structure and the cathode. As a result of the
RF interaction some of the electrons are collected on the slow-​wave structure. The
remainder are collected on a collector electrode which may be depressed to increase
the overall efficiency of the tube. It is important to ensure that secondary electrons
are not emitted from the cathode. This is achieved by the use of grooves, or low
secondary emission coatings. The geometry of the interaction region must be con-
trolled very accurately for satisfactory operation.
The crossed-​field backward-​wave oscillator (M-​type Carcinotron) resembles the
injected beam backward-​wave amplifier shown in Figure 16.1(b). The input port is
replaced by a matched termination inside the vacuum envelope [15–​20]. The attenu-
ator provides a good RF match at the end of the slow-​wave structure remote from
the output. Tubes of this kind have been used in electronic counter-​measures sys-
tems with CW output powers up to a kilowatt and efficiencies around 50%. Peak
powers up to 230 kW at S-​band with 50% efficiency, and 50 kW at X-​band with 15%
efficiency, and tuning ranges around 15% have been reported [17].
Circular injected beam crossed-​field amplifiers can provide average powers of 1
kW and pulsed power of 5 kW at X-​band with octave bandwidth and 20 dB gain
[21]. The feasibility of generating 10 MW peak power at L-​band has been demon-
strated, but the efficiency of injected beam CFAs is not high enough to make them
competitive with tubes of other types [5]. The theory of injected beam CFAs is dis-
cussed in [22–​25]. They can have high noise output, and considerable research has
been undertaken to find ways of reducing this [8]. Linear injected-​beam CFAs have
been used for experimental purposes [2, 3, 26, 27].

16.2 CFA Construction

The review of emitting-​cathode CFAs by Skowron [5] remains the best available
introduction to the subject, and much of this chapter is based on that paper.
Further information is given in [28]. The close resemblance of a CFA to a mag-
netron means that the principles of operation and the characteristics of the two
types of device are very similar. The electron beam is assumed to be drawn from a
634 Crossed-Field Amplifiers

space-​charge hub where the electrons on the surface of the hub have a tangential
velocity approximately equal to the phase velocity of the wave on the slow-​wave
structure.

16.2.1 Slow-​Wave Structures

The slow-​wave structure of a CFA, like the anode of a magnetron, must com-
bine suitable RF properties with the ability to dissipate the heat of the electrons
intercepted by it. The heat dissipation is not uniform, as in the magnetron, but is
greater close to the output. It has been found by experimental observation that a
few vanes close to, but not at, the output receive power around twice the average.
Allowing for variation in the axial direction, a 3:1 variation in power density may
occur. The structure is normally designed so that the phase shift is around π 2 per
cell. The pitch of the structure is then approximately half that of a comparable
magnetron. The first practical tubes used a strapped-​vane structure in which the
input and output waveguides were connected to the straps [6]. This structure has
a backward-​wave characteristic when the phase shift per cell is around π 2 (see
Figure 15.19).
Figure  16.5 shows some typical structures used in emitting-​ cathode CFAs.
Forward-​wave structures having good bandwidth, and flat dispersion, are often
based on the helix or the meander line [5]. Examples of helix-​derived structures
are the helix-​coupled vane structure (Figure 16.6(a) and (b)) and the stub-​mounted
helix. In the latter a flattened helix is supported by quarter-​wave stubs whose load-
ing effects are small. Ceramic-​supported meander lines are suitable for use at higher
frequencies [5, 21]. Backward-​wave structures include the strapped ladder line
(Figure 16.6(d)), the inter-​digital line (Figure 16.6(c)), and the split folded wave-
guide [14, 29]. Strapped-​vane, ring-​bar and dielectric-​supported helix structures are
described in [30].

16.2.2 Cathodes
CFAs designed for peak power of less than about 10 kW employ thermionic cath-
odes. Back-​bombardment liberates secondary electrons so that the total current is
a combination of both primary and secondary electrons in a proportion which is
self-​adjusting. The emission is space-​charge limited, as in the magnetron. At higher
power levels cold cathodes may be used so that all the electrons are produced by
secondary emission. This is possible because the high RF electric fields produced
by the input signal are sufficient to initiate emission using seed electrons produced
by field emission, or by ionisation of the residual gas in the tube. The process
requires the simultaneous application of the DC voltage and the RF pulse and the
emission builds up in less than 10 ns. The current obtained from a secondary emit-
ting cathode is constant during the pulse, whereas that of a primary emitter may
decay during the pulse.
 CFA Construction 635

Figure 16.5:  Slow-​wave structures for emitting-​cathode CFAs: forward wave (a) helix-​

coupled vane and (b) double helix-​coupled vane; backward wave (c) inter-​digital line and
(d) strapped ladder line (copyright Thales; courtesy of Thales).

The cathode in a cold-​cathode tube may be either an oxide cathode or a pure

metal. The oxides of elements such as barium, aluminium, thorium, magnesium,
and beryllium have secondary electron emission coefficients greater than 2.5. Pure
metals have coefficients below 1.9, with platinum having the highest value at 1.85.
Pure metal cathodes do not exhibit any wear-​out mechanism, apart from erosion,
so that very long tube lifetimes are possible. Oxide cathodes wear out because of
depletion of the oxide caused by back-​bombardment. However, it is possible to
replenish it by, for example, including an oxygen dispenser within the vacuum enve-
lope [31]. CFAs with oxide cathodes can have lives greater than 40,000 hours.

16.2.3 Cathode-​Driven CFAs
One of the chief disadvantages of the CFA is its low gain which results from the
fall-​off of the RF electric field between the slow-​wave structure and the surface of
the electron hub. The gain can be increased appreciably by incorporating a slow-​
wave structure in the surface of the cathode and applying the input signal to it. This
structure is much closer to the surface of the hub and the gain can be increased by
10 to 20 dB. The power output of such a tube is comparable with that of a conven-
tional CFA and the signal to noise ratio is reportedly improved by at least 20 dB
[32]. If the RF input is applied to the anode slow-​wave structure of a tube of this
kind the signal to noise ratio is improved by around 10 dB compared with a con-
ventional CFA. The design of a millimetre wave high gain CFA is described in [33].
636 Crossed-Field Amplifiers

16.3 Basic Principles

The resemblance of a CFA to a magnetron means that much of the theory of

magnetrons can be used, or adapted, to understand them. For the present we shall
assume that the electrons are drawn from a uniform electron hub described by the
equations in Section 15.2.1. We shall see later that detailed numerical modelling
shows that this model is not strictly correct for cold cathode tubes because the sec-
ondary electron emission from the cathode is non-​uniform.
Because the electron flow in a CFA is re-​entrant the angular velocity of the
spokes must be given by

ω
ωs = , (16.1)
M
where ω is the frequency of operation and M is the number of spokes. Since the vel-
ocity ω s is determined by the anode voltage it follows that a CFA may be expected
to exhibit a number of modes of operation corresponding to different values of M.
There is evidence from both experiment and simulation that (16.1) is satisfied for
synchronous operation of both forward-​and backward-​wave CFA’s [34–​39]. It has
been suggested that all the spokes are then identical [19] and the limited evidence
available from the papers cited supports this, even in the presence of a drift region.
Simulations using a moving wavelength code have shown that debunching effects in
the drift region are not sufficient to eliminate the spokes completely.
When the operating point is moved away from synchronism, by changing either
the frequency or the anode voltage, the residual spokes emerging from the drift
region are no longer in phase with the RF voltage on the slow-​wave structure as
they re-​enter it. Figure 16.6 shows the results of simulation of a forward-​wave CFA
when the operating voltage is slightly below that for synchronism. It can be seen
that the spokes are only well-​formed for the second half of the interaction region.
Calculations based on the angles of these spokes show that M in (16.1) is no longer
an integer. Thus the view expressed in early work [6, 34] that M remains an inte-
ger, even when the operation is not synchronous, is seen to be incorrect. When the
operation of the CFA is less efficient, the spokes may be erratic and less well defined
[37, 40].
If a uniform periodic slow-​wave structure is assumed then each section subtends
an angle 2 π N a on the axis where N a is an integer. The phase shift per section is θ for
synchronous operation, and the structure may have either forward-​or backward-​
wave characteristics. In a tube without a drift region N a is the number of cells in
the structure. In general, the number of active cells is reduced by those which are
replaced by the drift region. At synchronism the angular velocity of the wave on
the structure is

2 πω
ωs = . (16.2)
Naθ
 Basic Principles 637

Figure 16.6:  Spoke formation in a typical forward wave CFA (courtesy of Dr D.P.

Chernin [41]).

Equating (16.1) and (16.2) gives

M
θ = 2π . (16.3)
Na
Note that this differs from the expression given in [6] because of a difference of π in
the definition of θ. As a consequence of (16.3) the possible synchronous values of
θ are fixed by the values of M and N a, as shown in Figure 16.7, for a structure with
N a = 11 and a backward-​wave characteristic in the first Brillouin zone. The syn-
chronous spoke velocity line is shown for M = 4. It is evident from Figure 16.7 that
the choice of an odd integer for N a means that no interaction is possible at the π
mode cut-​off of the structure where the interaction impedance is high, and the tube
would be likely to oscillate. The threshold voltage at synchronism in a CFA is given
by (15.39) where ω s is determined from (16.1). Now suppose that the tube whose
dispersion diagram is shown in Figure 16.7 is operating in the M = 4 mode. If the
anode voltage is increased then, at some point, it will exceed the threshold voltage
for the M = 3 mode. The operation of the tube may then move discontinuously to
the new mode [35]. Since the anode current increases with increasing anode voltage,
in the same way as a magnetron, it may be limited by the onset of a different mode
[5]. The difference between the velocity of the spokes and the phase velocity on the
slow-​wave structure for non-​synchronous operation means that the synchronous
relation (16.3) is not strictly enforced. Some phase slippage is permissible at the
input end of the structure [6].
638 Crossed-Field Amplifiers

Figure 16.7:  Dispersion diagram of a backward-​wave CFA ( N a = 11) showing synchronous

points corresponding to different numbers of spokes.

Although the interactions in forward-​wave and backward-​wave tubes are simi-

lar there are some differences in detail. Consideration of Figure 16.2 shows that,
if two tubes have slow-​wave structures with the same cold bandwidth, then the
change in the synchronous electron velocity for a given change in frequency is
greater for the backward-​wave tube. The anode voltage increases with increasing
frequency in a backward-​wave CFA, whereas the opposite is true for a forward-​
wave tube. The properties of each spoke are chiefly determined by the RF field on
the slow-​wave structure at the tip of the spoke. Thus the spokes are most strongly
formed close to the output of the tube. This means that in a forward-​wave tube
the charges carried by the spokes entering the drift region, and the debunching in
the drift region, are both greater than in an equivalent backward-​wave tube. This
difference can be seen in ­figures 7 and 3 in [37]. A further difference between the
two types of tube arises because of the effects of the transit time of the electrons
from the surface of the hub to the anode. The current flowing in a spoke is deter-
mined by the RF field experienced by electrons at its base. At high signal levels,
close to the tube output, the transit time is typically small compared with the
synchronous period. The current in a spoke is then determined by the strength of
the local RF field. However, at low signal levels, near to the tube input, the transit
time is typically greater than the synchronous period. Thus the current reaching
the anode is determined by the fields one or more synchronous periods away [42].
In a forward-​wave amplifier the spokes are moving in the direction of increasing
RF field strength and the currents in the spokes close to the input are smaller than
would be determined by the RF field at the spoke tip. The opposite is true in a
backward-​wave amplifier.
 CFA Characteristics 639

16.4 CFA Characteristics

16.4.1 Performance Chart
The performance chart of an emitting cathode CFA operated at a fixed frequency
and RF drive level is like that of a magnetron (see Figure 16.8). For a given mag-
netic field no current flows until the anode voltage exceeds the threshold voltage.
Thereafter the current increases, the anode voltage remains approximately con-
stant, and the output power increases with the current. The magnetic field is nor-
mally provided by a permanent magnet and is therefore fixed by the manufacturer.
The operating curve varies slightly with frequency as shown in Figure  16.9. At
constant current the voltage increases with frequency in a backward-​wave CFA
and decreases with frequency in a forward-​wave tube, as discussed in the previous
section. Figure  16.10 shows how the performance of a backward-​wave amplifier
tube depends upon the load line of the modulator for constant voltage, constant
current, and constant impedance operation. Constant voltage operation produces
an unacceptably large variation in power output and efficiency across the frequency
band of the tube. Constant current operation gives essentially constant output
power. These two conditions are approximated by hard tube and line type modu-
lators respectively (see Section 20.3) [5]. The change in the operating curve with
frequency is smaller in a forward-​wave amplifier and constant voltage operation is
possible.

Figure 16.8:  Typical performance chart of a CFA at 1190 MHz with 5 kW peak input power
(copyright 1967, SFD, reproduced, with the permission of CPI, from [28]).
640 Crossed-Field Amplifiers

Figure 16.9:  Variation in the operating curve of a CFA with frequency: (a) backward-​wave

amplifier, and (b) forward-​wave amplifier
(copyright 1967, SFD. Reproduced, with the permission of CPI, from [28]).

Figure 16.10:  Backward-​wave CFA characteristics showing the effect of different modulator

load lines
(copyright 1967, SFD, reproduced, with the permission of CPI, from [28]).
 CFA Characteristics 641

16.4.2 Modulation
Figure 16.11 shows typical pulses for operation of a pulsed CFA. If a cold cathode
tube is operated from a line type modulator the RF power must be applied before
the voltage pulse. This ensures that current is drawn from the modulator as soon as
the voltage is applied to the tube. Otherwise, the tube would present an open circuit
to the modulator, resulting in a doubling of the voltage, leading to internal arcing
and possible damage to the tube. If the tube has a hot cathode with appreciable
thermionic emission, the application of the voltage pulse before the RF pulse leads
to noise generation from the uncontrolled space-​charge cloud. As a rule of thumb,
the voltage pulse is designed to reach 50% of its peak value when the RF input
power is 90% of the full drive level. A similar situation arises with any type of CFA
if the RF input is removed before the end of the voltage pulse. It is desirable for the
rise-​and fall-​times of the voltage pulse to be as short as possible to reduce the risk
of oscillations at band edges, or in other modes where the gain is high.
Some CFAs are designed to be operated with cold cathodes and with DC volt-
ages applied to the electrodes. The current pulse is then triggered by the RF pulse.
When the RF pulse is removed the electrons continue to circulate and would, if
unchecked, lead to the generation of noise. For this reason a control electrode is
incorporated, as shown in Figure 16.12. At the end of the RF pulse a positive pulse
is applied to the control electrode to remove the circulating electrons. Figure 16.13

Figure 16.11:  Typical pulse nesting for a cold cathode CFA

(copyright 1967, SFD, reproduced, with the permission of CPI, from [28]).
642 Crossed-Field Amplifiers

Figure 16.12:  Arrangement of a CFA with a control electrode

(copyright 1973, IEEE. Reproduced, with permission, from [5]).

Figure 16.13:  Typical pulse nesting for a CFA with a DC potential between the cathode and
the anode
(copyright 1967, SFD. Reproduced, with the permission of CPI, from [28]).

shows the pulses for a cold-​cathode CFA with a DC voltage applied to the cathode.
Electron emission commences within a few nanoseconds of the start of the RF
pulse. The proportion of electrons collected by the control electrode varies from
zero, when it is unbiased, to 100% with sufficient positive bias. Thus, it is possible to
find an intermediate level of DC bias where sufficient electrons continue to circulate
for the tube to operate in the presence of RF drive but not when it is absent. It has
been shown that RF keying can be achieved in this manner [5].

16.4.3 Transfer Characteristics

The amplitude transfer characteristics of a CFA may be displayed on either linear
or logarithmic scales, as shown in Figure 16.14. It can be seen that the tube does
not have a linear region of operation, nor does it exhibit saturation. The low gain
 CFA Characteristics 643

Figure 16.14:  Typical CFA amplitude transfer curves: (a) linear scales, and

(b) logarithmic scales
(copyright 1967, SFD, reproduced, with the permission of CPI, from [28]).

of the tube means that the input RF power is a significant fraction of the total
power input to the tube. Thus the RF input power is added to the (approximately
constant) RF power generated by the interaction within the tube. If the RF drive of
a tube with a primary emitting cathode is increased from low levels it is found that,
initially, the output at the drive frequency is proportional to the input and accom-
panied by noise. At higher drive levels the input signal suppresses the noise and the
output is determined by the DC operating conditions. For this reason it is best to
regard the CFA as a locked amplifier which is always in a non-​linear condition. The
minimum drive level for stable operation may be indicated on the AM/​AM curves.
The bandwidth of a CFA is determined by the range of frequencies over which
stable operation is possible. It may also be specified as the range of frequencies for
which the output is constant within some limits (e.g. ± 0.5 dB). Because, as we have
seen, the output power is affected by the load line of the modulator, it is important
to specify the conditions under which the bandwidth is defined. For a given fre-
quency, and given DC and RF input powers, it is found that a tube can operate over
a range of currents. The upper limit is set by the maximum current available from
the cathode, the onset of a competing oscillation, or the inability of the RF input
power to suppress noise. The lower limit is set by the start of oscillation in a mode
with a lower threshold voltage [5].
The phase of the output of a CFA is determined by the DC voltage and current,
the RF drive level, and the variation of the phase length of the tube with frequency.
Figure  16.15 illustrates the dependence on the anode voltage, and the RF drive
level, of the phase of the output of an X-​band backward-​wave CFA, relative to the
644 Crossed-Field Amplifiers

Figure 16.15:  Dependence of the relative phase of the output of a CFA on the anode voltage
and the RF drive leve l
(copyright 1967, SFD, reproduced, with the permission of CPI, from [28]).

input. The figure shows typical phase variation for a 1% variation in anode voltage,
and three different drive levels. The phase sensitivity to voltage is proportional to
the gain of the tube. It is, therefore, least at high drive levels where the gain is least
(see Figure 16.14). The sensitivity of the phase of a CFA to changes in anode volt-
age is about an order of magnitude less than that of a TWT. The variation of phase
with drive power at constant voltage is usually measured with small changes in the
input power around the normal operating point. It is typically of the order of 1 or
2 degrees/​dB.
The phase length of a CFA is 1 to 10 wavelengths for a backward-​wave tube,
and 15–​20 wavelengths for a forward-​wave tube. These are less than those of a
linear-​beam tube, and the resulting variation of phase with frequency typically less
than ±5°. The phase length of a CFA is affected by the reactive loading of the elec-
trons (compare frequency pushing in a magnetron). The hot phase length is around
20–​40° greater then the cold phase length, and decreases with increasing anode
current. Because the CFA is a transparent amplifier, the phase variation in prac-
tical cases is strongly affected by the RF match presented to the input and output
of the tube. If the matches of the source and the load are perfect then the devia-
tions from linearity are small, and determined by mismatches within the tube. The
phase variation caused by external mismatches can be estimated from the circuit in
Figure 14.1 as

∆φ = ρS ρL AL (16.4)

where ρS and ρL are the input and output voltage reflection coefficients combining
internal and external mismatches, and A and L are the amplification and cold loss
of the tube expressed as voltage ratios.
 Theoretical Models of CFAs 645

16.4.4 Signal Growth and Anode Dissipation

Additional information about the properties of CFAs is provided by a detailed
experimental study of an S-​band forward-​wave tube [43]. Experimental versions
of this tube were fitted with RF probes and thermal sensors so that the RF power,
and the vane temperatures, could be measured for a range of operating conditions.
Typically, the temperature of the slow-​wave structure varied approximately linearly
from 20–​30 °C, at the input, to 70–​80 °C, at the middle of the interaction region.
It then remained approximately constant until the output. The RF power on the
slow-​wave structure increased slowly over the first 20–​30% of the length of the
interaction region. Thereafter it increased approximately linearly until the output.
It was concluded that the initial growth in the RF power was based on the recircu-
lating space-​charge above the electron hub, and that the linear growth made use of
current extracted directly from the hub.

16.5 Theoretical Models of CFAs

16.5.1 PIC Codes
CFAs can be modelled by PIC codes in much the same way as magnetrons and,
in some cases, the same code has been used for both types of tube. The properties
of the slow-​wave structure are usually modelled by an equivalent network. The
accurate modelling of secondary electron emission is essential for cold-​cathode
tubes. When the whole tube is modelled large computing resources and long CPU
times are required [37–​40, 44–​46]. Moving window codes model the interaction
with a single spoke as it develops through time [47–​53]. Where electrons move
out of the moving wavelength they are re-​injected at the opposite boundary, as in
the TWT simulations described by Vaughan [54]. This assumption is not strictly
valid because the electron motion is not periodic in space. Additional guard wave-
lengths to overcome this problem do not appear to have been used in CFA mod-
els. The computer codes commonly employ rectangular Cartesian coordinates
with a correction factor to account for the curvature of the interaction space [48].
In these codes it is necessary to track the interaction over multiple passes through
the tube. It is found that the computations do not converge to a stable solution
but show pass-​to-​pass fluctuations which may be an indication of the level of
noise generation within the tube [49]. Stable results are obtained by taking the
average over several passes, and the main features of the tube performance can
be modelled with good accuracy. The models also show the details of spoke for-
mation, including the debunching in the drift region. When secondary emission is
the dominant source of electrons it is found that regions of low density propagate
through the space-​charge hub. The method has been extended to tubes with a
slow-​wave structure on the cathode [51]. Whole tube simulations have been used
to give additional information about spoke formation in CFAs, and to study the
646 Crossed-Field Amplifiers

conditions for low noise operation that have been observed experimentally [38,
44, 45].

16.5.2 Soliton Theory
We have seen that considerable progress can be made in understanding the proper-
ties of linear-​beam tubes by using models in which the non-​linear equations describ-
ing the interaction are linearised by the small-​signal approximation. This approach
has been extended into the non-​linear region of operation by the addition of more
terms in the expansions. An alternative approach is to find non-​linear solutions to
weakly non-​linear equations using soliton theory [55, 56]. A  soliton is a solitary
wave that maintains its shape while it propagates at a constant velocity. Solitons
occur when the non-​linearity, and dispersion, of a medium counteract each other
to prevent the break-​up of the wave. The application of soliton theory to CFAs
shows that growing solitons can only exist for a limited range of anode voltages.
For a particular tube this range was found to be from the threshold voltage (VT )
to 1.13 VT . This behaviour was found to be in good agreement with experimental
results. At the upper limit the amplification ends abruptly. It was shown that the
true synchronous voltage is slightly greater than the threshold voltage because of
differences between the electron motion and the idealised Brillouin hub. However,
the difference between the synchronous voltage and the threshold voltage was small
in the typical operating range of the tube studied. The model was also able to pre-
dict the dependence of the phase shift on anode voltage [56].

16.5.3 Guiding Centre Theory

The guiding centre theory, whose application to magnetrons is described in Section
15.6.3, was applied to modelling CFAs in [42, 57, 58]. The electrons were assumed
to be drawn from a uniform hub whose surface was taken to be approximately that
of the theoretical Brillouin hub. The electron density in the hub was not constant
but was found by considering the balance between secondary emission of electrons
from the cathode and their motion out of the hub into the spokes. Thus the electron
density and the spoke current varied according to the local RF field. The model
included the effect of the finite transit time of the electrons from the hub to the
anode. The evolution of the spokes with time was computed by successive passes
through the interaction region. The electron density at the start of a pass was taken
to be a fixed fraction of that at the end of the previous pass. The electron stream
was assumed to be unmodulated at the start of each pass. The results converged
after only a few iterations and the Gauss lines which were computed showed some
agreement with experimental results, including their dependence on the frequency.
It is not clear whether the parameters of the model were adjusted to achieve agree-
ment with measurements. The model also showed the spatial growth of the electron
density in the hub and of the RF voltage on the slow-​wave structure. The electron
density in the hub was typically about 0.8 of the theoretical hub density. When the
 Theoretical Models of CFAs 647

secondary electron emission coefficient of the cathode was increased above a cer-
tain level it was found that the output power did not settle to a steady value but
oscillated from pass to pass indicating the generation of noise [59].

16.5.4 Non-​Linear Fluid Mechanics

The application of the methods of fluid mechanics to the modelling of crossed-​
field devices has been described in a large number of papers including [60–​63]. In
this approach the non-​linear, cold-​fluid, equations have been expanded into static
and first-​order time-​varying parts, corresponding to the electron hub and spokes
assumed in the guiding centre model. It has been shown that the two parts of the
solution are not independent of one another, as in simpler models, but that the
evolution of the spokes has an effect on the static electron density distribution.
In particular, the electron density at the position of the theoretical hub surface is
around half of that at the cathode, when space-​charge limited emission is assumed.
Although this work provides some insights into the behaviour of CFAs it does not
appear to be useful for engineering modelling of their performance.

16.5.5 Rigid Spoke Model

We have seen that the rigid spoke model of the magnetron gives good agree-
ment with experimental results, despite problems with the validity of some of the
assumptions on which it is based. It is therefore of interest to consider whether a
similar model can be constructed for CFAs. A simple model of this kind, described
in [19], assumed that fully formed spokes were present throughout the interaction
region, and that the current induced in the slow-​wave structure was nearly constant.
The induced current was divided equally between forward and backward travel-
ling waves. The difference in the phase relationships means that the induced cur-
rents add constructively in one direction, and cancel in the other. Since each spoke
induces the same power in the circuit it follows that the signal growth with distance
is linear. Experimental measurements have shown that this is true over most of the
interaction region by (see Section 16.4.4). Increasing the gain of a CFA by a factor
of 10 therefore requires a tenfold increase in the length of the interaction region. In
contrast the exponential growth in a TWT only requires the length of the interac-
tion region to be doubled. The gain of a CFA can be estimated by assuming that the
power induced in unit length of the slow-​wave structure is equal to that of a mag-
netron having the same parameters and output power. The input power can then
be calculated from the length of the structure, making allowance for the cold loss.
The efficiency of the tube is found from the total power delivered to the structure.
The application of this approach to an experimental tube gave good agreement for
the efficiency. Further information on the parameters and performance of that tube
can be found in [64–​66].
A straightforward improvement to the model described above uses a method
developed for modelling non-​reentrant CFAs [14]. The assumption that all spokes
648 Crossed-Field Amplifiers

deliver equal power to the slow-​wave structure is replaced by one in which the
power delivered to each section is calculated from the properties of a magnetron
equivalent to that section. The rigid-​spoke magnetron model described in the pre-
vious chapter can be used for this purpose. The synchronous angular velocity and
the number of spokes are calculated from the frequency and the cold phase shift
per cell of the slow-​wave structure. The threshold voltage and the properties of
the electron hub are calculated in exactly the same manner as for a magnetron.
These values do not depend on the RF field of the slow-​wave structure and they
are therefore the same throughout the interaction region. The electric field acting
on the electrons, and the conduction angle of the spokes, can then be calculated as
functions of the anode voltage, the RF wave voltage on the slow-​wave structure,
and the magnetic field. The RF wave voltage (V1 ) is related to the RF power flow in
the structure ( P1 ) by

V1 = 2Zc P1 , (16.5)

where Zc is the coupling impedance of the circuit. The energies of the electrons
striking the anode and the cathode are computed in the same way as for a magne-
tron. The DC current flowing to the anode is computed from the charge density in
the hub, and the electric and magnetic fields at the base of the spoke. The impact
energies of the electrons, and the current in each spoke, are functions of V1. They
therefore vary with position in the CFA. The RF power delivered to the anode of
the equivalent magnetron is given by

Prf = (Va − Via − Vic ) I 0 , (16.6)

where Va and I 0 are the DC anode voltage and current and Via and Vic are the impact
energies of the electrons on the anode and the cathode. Note that Vic is the actual
impact energy of electrons on the cathode, multiplied by the ratio of the fraction
of electrons striking the cathode to the fraction striking the anode. In (16.6) all the
terms on the right-​hand side, apart from Va are functions of the anode voltage, the
magnetic field, and the RF power flow on the structure. The RF power delivered to
the nth cell of the structure is then

∆Pn = Prf ( Pn ) N a , (16.7)

where Pn is the power in the growing wave in the nth cell and N a is the perimeter of
the anode divided by the length of a cell. The anode voltage and the magnetic field
are the same for all the cells. The power in the next cell is given by

Pn +1 = ( Pn + ∆Pn ) (1 − Ac ) , (16.8)

where Ac is the fraction of the power lost per cell. This is related to the cold loss of
the slow-​wave structure in decibels by

LdB = 10 N c log ( Ac ) , (16.9)

 Theoretical Models of CFAs 649

where N c is the number of active cells. If the RF input power is known then (16.8)
can be applied repeatedly to find the power flow in each cell, and the output power
of the tube, at the specified anode voltage and magnetic field. If we assume that
the power in the CFA grows linearly to this level over N c cells, whilst that in the
magnetron is constant over N a cells, we find that the power output of the CFA is
approximately

Nc
PCFA ≈ PEM , (16.10)
2N a

where PEM is the output power of the equivalent magnetron. The total anode cur-
rent is given by

I a = ∑ n =1 ( I n N a ), (16.11)
Nc

where I n is the anode current of the equivalent magnetron for the nth cell. The total
heat dissipation on the anode and the cathode can be found in a similar manner
by summing the contributions of the equivalent magnetrons (see Worksheet 16.1).
This model was applied to the data for an S-​band forward-​wave CFA given in
[49] and there was good qualitative agreement with the typical behaviour of a CFA,
described in Section 16.4. This tube was identical, or very similar, to the experi-
mental tube studied in [43]. Figure 16.16 shows the Gauss lines computed for three
different magnetic fields at the synchronous frequency (3.3 GHz) of the tube. These
curves are similar to those in Figure 16.8. Figure 16.17 shows the Gauss lines at
three different frequencies, when B = 0.3 T, with the experimental curves given
in [49] for comparison. The properties of the slow-​wave structure, which are not
given in that paper, were estimated using the data for a similar tube in [47]. It can

Figure 16.16:  Gauss lines computed for the baseline tube at 3.3 GHz and 7 kW RF
input power.
650 Crossed-Field Amplifiers

Figure 16.17:  Gauss lines computed for the baseline tube at Bz = 0.3 T and 7 kW RF input
power, with experimental data from [49] for comparison.

Figure 16.18:  RF output power plotted against anode current for the baseline tube at
Bz = 0.3 T and 7 kW RF input power, with experimental data from [49] for comparison.

be seen that the anode voltage increases with decreasing frequency, as shown in
Figure  16.9(b). For frequencies other than the synchronous frequency the num-
ber of spokes calculated from (16.3) was not an integer. This is not inconsistent,
because the interaction is assumed to be restarted from an unmodulated hub after
the drift region.
Figure 16.18 shows the dependence of the output power on the anode current
for three different frequencies compared with experimental data from [49]. It can be
seen that the output power and efficiency predicted by the model are rather greater
than the measured values. Possible reasons for this are explored below.
 Theoretical Models of CFAs 651

Figure 16.19:  RF output power plotted against RF input power for the baseline tube at
3.3 GHz and Bz = 0.3 T. The dashed line shows the minimum RF input power for which
spokes are formed.

Figure 16.20:  Electron trajectories for RF input powers: (a) just below the minimum for
spoke formation, and (b) just above the minimum for spoke formation.

Figure 16.19 shows the AM/​AM curves computed for three different anode volt-
ages. These may be compared with Figure  16.14(a). At low values of the input
power the RF field of the slow-​wave structure is not sufficient to enable spokes to
form, and the only output is noise. Figure 16.20 shows examples of the electron
trajectories just below and just above the limiting input power. Below the limiting
power no trajectories reach the anode and the modulation of the hub does not
induce sufficient current in the slow-​wave structure to enable the wave to grow. The
figure also shows the spoke profiles for the rigid spoke and guiding centre approxi-
mations. Below the limit for spoke formation these models are incorrect because
they predict non-​zero anode currents.
Table  16.1 shows a detailed comparison between the results of calculations
using the model described with those in [49]. It can be seen that the anode current,
652 Crossed-Field Amplifiers

Table 16.1:  Comparison between calculated and experimental results for McDowell’s baseline forward wave CFA
(copyright 2002, IEEE, reproduced, with permission, from [49])

Moving Rigid spoke Rigid spoke Experiment

wavelength (corrected)

Voltage (kV) 13.25 13.25 13.25 13.25

Current (A) 22.0 23.9 20.0 22.0
Power (kW) 156.9 203 168 155
Power added efficiency (%) 54.7 63 61 52
Gain (dB) 13.5 14.6 13.8 13.5
RF power added /​DC input (%) 51.4 63.2 60.9
Anode dissipation /​DC input (%) 25.9 15.1 20.1
Cathode dissipation /​DC input (%) 10.6 5.3 7.3
Circuit loss /​d.c input (%) 10.5 13.9 12.0

output power, efficiency and gain are over-​estimated by the rigid spoke model
while the anode and cathode dissipation are underestimated. The differences in
the figures for the dissipation may be a result of the different assumptions about
electron emission from the cathode. The rigid spoke model assumes that the cur-
rent flow in the hub is space-​charge limited whilst the moving wavelength code
models secondary electron emission from a cold cathode. Table 15.2 shows that,
in a magnetron, this can have a large effect on the anode and cathode dissipation.
Because the rigid spoke model uses the dissipation to calculate the RF output
power, underestimation of the former leads automatically to over-​estimation of
the latter. The results are, in any case, very sensitive to changes in the magnetic
field. McDowell adjusted the magnetic field to compensate for the use of a lin-
ear, rather than a circular model of the tube. It is not clear whether the value
quoted in Table II in [49] was adjusted in this way. He observed that the agree-
ment between his results and those of experiment was, in any case, somewhat
fortuitous because the calculations ignored the effects of penetration of electrons
into the space between the vanes.
Further understanding of the difference between the two models is obtained by
plotting the forward power as a function of distance along the slow-​wave structure,
as shown in Figure 16.21. The low rate of growth close to the input calculated by
the moving wavelength model is confirmed by experimental results [43]. It can be
seen that both models predict very similar linear rates of growth and that better
agreement could be obtained, quite simply, by moving the curve for the rigid spoke
model to the right by around eight cells. This can be explained in terms of the finite
time taken for electrons to move from the base to the tip of the spoke. A magnetron
operates in a steady state, and the transit time can be ignored. In a CFA, on the
other hand, the RF fields experienced by the spokes change in amplitude as they
move through the tube.
 Theoretical Models of CFAs 653

Figure 16.21:  Growth in the forward RF power with distance.

The transit time can be estimated as a function of the wave voltage in the equiva-
lent magnetron by integrating (15.113), and making use of the expression for the
tangential electric field in (15.19), to give
ra ra

⌠B B r ⌠  rn − r−n  1
∆t =  z dr = z a   an a− n  dr. (16.12)
⌡ θ
E nV 1 ⌡  r − r  sin ( nθs )
rb rb

This is the transit time for an electron at the centre of the spoke, assuming that its
angular position, relative to the wave, is approximately constant during its motion
from the hub to the anode. It can be seen that ∆t is inversely proportional to the
wave voltage V1. The transit time can be expressed in terms of the number of cells
traversed by the electron in its motion given by ns = ω s ∆t φc . For the tube under
consideration this was found to vary from 2 at the input to around 1 at the out-
put. However, this calculation underestimates the transit time close to the input.
Figure 16.20(b) shows that, when the spokes are only just formed, the angular pos-
ition of the electrons varies considerably. For part of the motion they are in a region
of very small tangential electric field. Thus the transit time is increased, so that
the electrons move through a greater number of cells. It is difficult to determine
the number of cells precisely because the electron dynamics at the input are very
dependent on the properties of the electron hub as it emerges from the drift region.
But this argument suggests that most of the difference between the two models can
be explained in terms of transit time effects.
The rigid spoke model is based on some rather crude approximations, so the
effort involved in including transit time effects correctly is not justified (though they
have been included in a guiding centre model [42]). However, the shift of the curve
in Figure 16.21 can be achieved using a piecewise constant model for the transit
time, so that no current reaches the anode in the first eight cells. Current is drawn
654 Crossed-Field Amplifiers

from the surface of the hub along the whole length of the slow-​wave structure, but
it is concentrated along a shorter length of the anode because of transit time effects.
The results obtained when this correction is made are shown in the fourth column
of Table 16.1. Because the distribution of the current along the anode is not known
the power dissipated on the anode and the cathode has been calculated by scaling
to achieve an overall power balance. Overall it may be concluded that, with this
modification, the rigid spoke model gives a good qualitative understanding of the
properties of CFAs.

16.6 CFA Design

The very limited information on CFAs which is available in the literature, the com-
plexity of their behaviour, and the scarcity of good theoretical models means that
they must usually be designed by scaling from existing tubes, using dimensionless
parameters similar to those used for magnetrons (see Section 15.7.1) [49].
An initial estimate of the performance of a CFA may be made using the sim-
plifying assumption that the power generated per unit length is constant [19].
Figure 16.21 shows that this is true for the greater part of the slow-​wave structure.
The gain can then be written

Pin + Pgen Pgen

G= = 1+ , (16.13)
Pin Pin
where Pgen is the RF power generated within the tube. The power added efficiency
is given by

Pout − Pin  Pout − Pin   Pgen 

ηa = =  ⋅  = ηc ηe , (16.14)
Pdc  Pgen   Pdc 
where ηc is the circuit efficiency and ηe is the electronic efficiency.
The electronic efficiency of the equivalent magnetron, given approximately by
(15.50), depends only on the dimensions of the tube and on the magnetic field.
Thus, to this level of approximation, it is constant for the growing wave through-
out the CFA. The normalised magnetic field of the baseline tube in [49] was
Bz B0 = 4.6 , giving the efficiency of the equivalent magnetron as approximately
75%. Gewartowski and Watson multiplied this figure by the fraction of the perim-
eter of the tube occupied by the slow-​wave structure. That adjustment is not correct
because the RF power and the anode current are both proportional to the length
of the slow-​wave structure. However, these quantities are affected differently by the
electron transit time. Thus the electronic efficiency is reduced by the fraction of the
slow-​wave structure that contributes to the signal growth. We saw, above, that this
could be represented by reducing the active length of the slow-​wave structure by
8 cells, in a total of 62, giving a reduction in efficiency of 0.87. Thus the estimated
electronic efficiency of the CFA is 65%.
 References 655

To find an expression for the circuit efficiency we note that the output power is
given by
Lc
Pgen
Pout = Pin exp ( −2α Lc ) + exp ( −2α ( Lc − x )) dx
Lc ∫0

Pgen
= Pin exp ( −2α Lc ) +
2α Lc
(1 − exp ( −2αL )) ,
c (16.15)

where Lc is the length of the slow-​wave structure and α is the attenuation coefficient
which is related to the cold loss of the structure by

LdB = −20 α Lc log ( e ) . (16.16)

Thus the circuit efficiency is

 1 P 
ηc =  − in  (1 − exp ( −2α L )) . (16.17)
 2α L Pgen 
The baseline tube in [49] had 13 dB gain and 2 dB cold loss so that 2α Lc = 0.46,
Pin Pgen = 0.05 and ηc = 0.78. Thus, the estimated overall efficiency of the tube is
51%, which is close to the experimental figure in Table 16.1. Figure 16.3 shows the
efficiencies of CFAs for which data is available.
Under synchronous conditions this tube had 19 spokes. An equivalent mag-
netron would therefore have 34 cavities, which is much too great for satisfactory
separation of the modes. That problem does not occur in a CFA, and the anode
and cathode radii can be much greater than in an equivalent magnetron (see
Table  15.5). The ratio of the anode length to the free-​space wavelength is 0.13
which is very similar to that in a magnetron. The modified Slater ratio is 1.54
because the gap between the anode and cathode is smaller than it would be in a
typical magnetron. This could be to ensure that the RF electric field at the desired
input power was strong enough to ensure satisfactory cold-​cathode operation.
In this tube the cathode was offset from the mechanical centre because that had
been found experimentally to give improved performance. It was also shown that
similar improvements could be made by tapering the pitch of the slow-​wave struc-
ture [43, 47]. This is probably a result of the reduction in the tangential velocity of
the spokes, caused by space-​charge potential depression close to the output, where
the spoke charge is greatest [14]. Some additional information about the design of
CFAs can be found in [8, 33, 67].

References

[1]‌ J. R. M. Vaughan, ‘Beam buildup in a DEMATRON amplifier’, IEEE Transactions on

Electron Devices, vol. 18, pp. 365–​373, 1971.
[2]‌ R. G. Lock, ‘A non-​reentrant crossed-​field amplifier with cycloiding injected beam’,
IEEE Transactions on Electron Devices, vol. 16, pp. 986–​995, 1969.
656 Crossed-Field Amplifiers

[3]‌ J. Z. Ye et al., ‘An experimental investigation of the end-​hat effects in a crossed-​field

amplifier via three-​dimensional electron density measurements’, IEEE Transactions on
Electron Devices, vol. 41, pp. 258–​265, 1994.
[4]‌ D. Chen, ‘An emitting-​sole linear magnetron amplifier’, in E. Okress, ed., Crossed-​Field
Microwave Devices, vol. 2. New York: Academic Press, pp. 223–​246, 1961.
[5]‌ J. F. Skowron, ‘The continuous-​ cathode (emitting-​ sole) crossed-​field amplifier’,
Proceedings of the IEEE, vol. 61, pp. 330–​356, 1973.
[6]‌ W. C. Brown, ‘The Platinotron: Amplitron and Stabilotron’, in E. Okress, ed., Crossed-​
Field Microwave Devices, vol. 2. New York: Academic Press, pp. 165–​209, 1961.
[7]‌ W. C. Brown, ‘The microwave magnetron and its derivatives’, IEEE Transactions on
Electron Devices, vol. 31, pp. 1595–​1605, 1984.
[8]‌ R. M. Gilgenbach et  al., ‘Crossed-​field devices’, in R. J. Barker et  al., eds, Modern
Microwave and Millimetre-​ Wave Power Electronics. Piscataway, NJ:  IEEE Press,
pp. 289–​342, 2005.
[9]‌ A. S. Gilmour, Jr., Microwave Tubes. Dedham, MA: Artech House, 1986.
[10] L. Sivan, Microwave Tube Transmitters. London: Chapman and Hall, 1994.
[11] B. Epsztein, ‘Cross-​field tubes’, in B. L. Smith and M.-​H. Carpentier, eds, The Microwave
Engineering Handbook, vol. 1. London: Chapman & Hall, pp. 65–​79, 1993.
[12] V. Granatstein et  al., ‘Vacuum electronics at the dawn of the twenty-​first century’,
Proceedings of the IEEE, vol. 87, pp. 702–​716, 1999.
[13] N. J. Dionne et al., ‘Methods for enhancing low-​noise CFA performance (crossed-​field
amplifiers)’, in International Electron Devices Meeting, pp. 529–​532, 1990.
[14] G. E. Pokorny et  al., ‘The DEMATRON  –​a  new crossed-​ field amplifier’, IRE
Transactions on Electron Devices, vol. 9, pp. 337–​345, 1962.
[15] R. R. Warnecke et al., ‘The M-​type carcinotron tube’, Proceedings of the IRE, vol. 43,
pp. 413–​424, 1955.
[16] G. Boucher, ‘Technology of carcinotrons for short and long wavelengths’, Proceedings
of the IEE –​Part B: Radio and Electronic Engineering, vol. 105, pp. 897–​899, 1958.
[17] M. Favre, ‘Results obtained on cross-​ field carcinotrons under pulsed oper-
ation’, Proceedings of the IEE  –​Part B:  Radio and Electronic Engineering, vol. 105,
pp. 533–​537, 1958.
[18] O. Doehler, ‘The M-​Carcinotron’, in E. Okress, ed., Crossed-​Field Microwave Devices,
vol. 2. New York: Academic Press, pp. 11–​34, 1961.
[19] J. W. Gewartowski and H. A. Watson, Principles of Electron Tubes. Princeton, NJ: D. van
Nostrand, 1965.
[20] G. Faillon et al., ‘Microwave Tubes’, in J. A. Eichmeier and M. K. Thumm, eds, Vacuum
Electronics: Components and Devices. Berlin: Springer-​Verlag, pp. 1–​84, 2008.
[21] R. J. Espinosa and R. R. Moats, ‘Broad-​band injected-​beam crossed-​field amplifiers’,
IEEE Transactions on Electron Devices, vol. 24, pp. 13–​21, 1977.
[22] W. Sobotka, ‘Computation of nonreentrant CFA characteristics’, IEEE Transactions on
Electron Devices, vol. 17, pp. 622–​632, 1970.
[23] R. H. Levy, ‘Diocotron instability in a cylindrical geometry’, Physics of Fluids (1958–​
1988), vol. 8, pp. 1288–​1295, 1965.
[24] R. W. Gould, ‘Space charge effects in beam-​type magnetrons’, Journal of Applied
Physics, vol. 28, pp. 599–​605, 1957.
[25] J. Feinstein and G. S. Kino, ‘The large-​signal behavior of crossed-​field traveling-​wave
devices’, Proceedings of the IRE, vol. 45, pp. 1364–​1373, 1957.
 References 657

[26] J. Browning et  al., ‘A low-​frequency crossed-​field amplifier for experimental investi-
gations of electron-​radio frequency wave interactions’, IEEE Transactions on Plasma
Science, vol. 19, pp. 598–​606, 1991.
[27] J. Browning et al., ‘Electron plasma and wave measurements in a crossed-​field amplifier
and comparison with numerical simulation’, IEEE Transactions on Electron Devices,
vol. 39, pp. 2401–​2407, 1992.
[28] SFD, Introduction to Pulsed Crossed-​Field Amplifiers. Union, NJ:  SFD Laboratories
Inc., 1967.
[29] H. Steyskal, ‘Studies on C.W. magnetron amplifiers with continuous cathode and re-​
entrant beam’, in 4th International Congress on Microwave Tubes, Scheveningen, The
Netherlands, pp. 168–​172, 1962.
[30] M. J. Clark, ‘Slow wave circuits for miniature crossed field amplifiers’, in 11th European
Microwave Conference, pp. 645–​649, 1981.
[31] A. Dallos et al., ‘Effects of ionized oxygen on primary and secondary emission, and total
current of a CFA’, IEEE Transactions on Electron Devices, vol. 34, pp. 1201–​1208, 1987.
[32] G. H. MacMaster, ‘Current status of crossed-​field devices’, in International Electron
Devices Meeting, pp. 358–​361, 1988.
[33] G. H. MacMaster and L. J. Nichols, ‘Millimeter-​wave high gain crossed-​field amplifier’,
in International Electron Devices Meeting, pp. 963–​966, 1990.
[34] G. E. Dombrowski, ‘Theory of the Amplitron’, IRE Transactions on Electron Devices,
vol. 6, pp. 419–​428, 1959.
[35] W. A. Smith, ‘A wave treatment of the continuous cathode crossed-​field amplifier’, IRE
Transactions on Electron Devices, vol. 9, pp. 379–​387, 1962.
[36] H. Steyskal, ‘Continuous cathode crossed-​ field amplifiers’, IEEE Transactions on
Electron Devices, vol. 10, pp. 95–​96, 1963.
[37] D. MacGregor, ‘Computer modeling of crossed-​field tubes’, Applications of Surface
Science, vol. 8, pp. 213–​224, 1981.
[38] D. P. Chernin, ‘Computer simulations of low noise states in a high-​power crossed-​field
amplifier’, IEEE Transactions on Electron Devices, vol. 43, pp. 2004–​2010, 1996.
[39] L. D. Ludeking et  al., ‘An examination of the performance of backward wave
CFA’s in simulation and experiment’, in Pulsed Power Plasma Science, vol. 1,
pp. 229–​231, 2001.
[40] D. M. MacGregor, ‘Computer simulation of the backward-​wave distributed-​emission
crossed-​field amplifier’, IEEE Electron Device Letters, vol. 1, pp. 134–​136, 1980.
[41] D. Chernin, ‘Private communication’, ed, 2014.
[42] S. A. Riyopoulos et al., ‘Guiding center fluid model of the crossed-​field amplifier’, IEEE
Transactions on Electron Devices, vol. 39, pp. 1529–​1542, 1992.
[43] G. K. Farney, ‘CFA design improvement program: Final Report. Volume 1: Instrumented
CFA studies’, Varian Associates, Beverly, MA, 1978.
[44] D. Chernin and A. Drobot, ‘Computer simulations of re-​entrant crossed-​field ampli-
fiers’, in International Electron Devices Meeting, pp. 521–​524, 1990.
[45] D. Chernin et al., ‘Computer studies of noise generation in crossed-​field amplifiers’, in
International Electron Devices Meeting, pp. 593–​596, 1991.
[46] G. E. Dombrowski, ‘Simulation of magnetrons and crossed-​field amplifiers’, IEEE
Transactions on Electron Devices, vol. 35, pp. 2060–​2067, 1988.
[47] H. L. McDowell, ‘CFA design improvement program:  Final Report. Volume
2: Computer modeling studies’, Varian Associates, Beverly, MA, 1978.
658 Crossed-Field Amplifiers

[48] H. L. McDowell, ‘CFA computer modeling using a moving wavelength code’, in Proc.
1st Int. Workshop Crossed-​Field Devices, pp. 15–​16, 1995.
[49] H. L. McDowell, ‘Crossed-​field amplifier simulations using a moving wavelength com-
puter code’, IEEE Transactions on Plasma Science, vol. 30, pp. 962–​979, 2002.
[50] S. Yu et al., ‘Time-​dependent computer analysis of electron-​wave interaction in crossed
fields’, Journal of Applied Physics, vol. 36, pp. 2550–​2559, 1965.
[51] G. I. Churyumov and G. I. Sergeev, ‘Simulation and modeling of self-​modulated re-​
entrant beam crossed-​field amplifier’, IEEE Transactions on Electron Devices, vol. 46,
pp. 1063–​1069, 1999.
[52] K. Eppley, ‘Numerical simulation of cross field amplifiers SLAC-​PUB-​5183’, Stanford
Linear Accelerator Center, Menlo Park, CA (USA), 1990.
[53] K. Eppley and K. Ko, ‘Design of a high power cross field amplifier at X band with
an internally coupled waveguide’, presented at the SPIE Symposium on High Power
Lasers, Los Angeles, CA, 1991.
[54] J. R. M. Vaughan, ‘Calculation of coupled-​cavity TWT performance’, IEEE Transactions
on Electron Devices, vol. 22, pp. 880–​890, 1975.
[55] G. E. Thomas, ‘The nonlinear operation of a microwave crossed-​field amplifier’, IEEE
Transactions on Electron Devices, vol. 28, pp. 27–​36, 1981.
[56] G. E. Thomas, ‘Soliton-​voltage and phase characteristics of a microwave crossed-​field
amplifier’, IEEE Transactions on Electron Devices, vol. 29, pp. 1210–​1218, 1982.
[57] S. Riyopoulos, ‘Simulations of crossed-​field amplifier operation using guiding center
dynamics’, in International Electron Devices Meeting, pp. 525–​528, 1990.
[58] S. Riyopoulos, ‘Guiding center theory and simulations of the crossed-​field amplifier’,
Physics of Fluids B: Plasma Physics (1989–​1993), vol. 3, pp. 3505–​3516, 1991.
[59] S. A. Riyopoulos, ‘Feedback-​induced noise in crossed field devices’, IEEE Transactions
on Plasma Science, vol. 20, pp. 360–​369, 1992.
[60] D. Kaup and G. E. Thomas, ‘Density profile and current flow in a crossed-​field ampli-
fier’, Journal of Plasma Physics, vol. 58, pp. 145–​161, 1997.
[61] D. Kaup and G. E. Thomas, ‘Stationary operating density profiles in a crossed-​field
amplifier’, Journal of Plasma Physics, vol. 59, pp. 259–​276, 1998.
[62] D. J. Kaup, ‘Theoretical modeling of crossed-​field electron vacuum devices’, Physics of
Plasmas (1994-​present), vol. 8, pp. 2473–​2480, 2001.
[63] D. J. Kaup, ‘Initiation and stationary operating states in a crossed-​field vacuum electron
device’, Proceedings of SPIE, vol. 4720, pp. 67–​74, 2002.
[64] R. J. Collier, ‘Gain measurements on a forward wave crossed-​field amplifier’, Proceedings
of the IRE, vol. 49, p. 372, 1961.
[65] R. J. Collier, ‘Bi-​ signal amplification by a forward wave crossed-​ field amplifier’,
Proceedings of the IRE, vol. 49, p. 646, 1961.
[66] J. Feinstein and R. J. Collier, ‘Waveguide-​ coupled crossed-​ field amplifier’, in E.
Okress, ed., Crossed-​ Field Microwave Devices, vol. 2.  New  York:  Academic Press,
pp. 211–​222, 1961.
[67] T. E. Ruden, ‘The Amplitron as a high power, efficient, RF power source for long pulse,
high resolution linear accelerators’, IEEE Transactions on Nuclear Science, vol. 12,
pp. 169–​173, 1965.
17 Fast-​Wave Devices

17.1 Introduction

The purpose of this chapter is to provide an introduction to the fast-​wave tubes,

especially to gyrotron oscillators and amplifiers which are important sources of
RF power at millimetre and sub-​millimetre wavelengths. The scope of the chapter
is limited to those devices up to a frequency of around 1 THz which can be classed
as vacuum tubes. Many fast-​wave devices are described as masers or lasers (from
the acronym Microwave/​Light Amplification by Stimulated Emission of Radiation)
and some theoretical treatments regard them as quantum devices [1]. However, at
least for the frequency range considered in this book, they can be regarded as clas-
sical devices in which the energy is transferred from a bunched electron beam to a
synchronous electromagnetic wave. We shall see that, in many cases, the theory is
analogous to that of the TWT. Many devices employ the electron cyclotron maser
(ECM) interaction which is reviewed in Section 17.2. Examples, including gyrotron
oscillators, gyro-​klystrons, and gyro-​TWAs, are discussed in Sections 17.3 and 17.4.
Peniotrons and ubitrons are considered at the end of the chapter.
We have seen in the preceding chapters that, above some frequency, the output
power and efficiency of linear-​beam and crossed-​field tubes decreases very rapidly
with increasing frequency. This is caused chiefly by the very rapid fall-​off in the
RF electric field from the surface of the electromagnetic structure as discussed in
Section 4.1.1. The transverse variation of the interaction field is as I 0 ( γ e r ) in cylin-
drical geometry so that, as a tube is scaled to higher frequencies the parameters
γ e b and γ e a must be kept constant to maintain the strength of the interaction. The
beam plasma frequency is related to the strength of the magnetic focusing field by
(7.55) which may be written
1 e Bz
ωp = ⋅ ⋅ , (17.1)
2 m0 m
where Bz is the axial magnetic field and m is the ratio of the magnetic field to the
Brillouin field. If a solenoid with iron pole-​pieces is used to provide the magnetic
field, then the maximum value of Bz is limited to around 2T by saturation of the
iron. Higher fields can be provided by normal, or superconducting, solenoids which
do not incorporate any iron. But it is then more difficult to achieve the desired pro-
file of Bz as a function of the axial position. The maximum value of ω p is also likely
660 Fast-Wave Devices

to be reduced as the frequency increases by the need to increase the beam stiffness,
as the beam and tunnel radii are reduced. We have also seen (in (13.51)) that the
beam perveance is approximately proportional to (ω p ω ) when γ e b is constant.
2

Thus the maximum possible perveance of a magnetically focused beam decreases

rapidly with increasing frequency. If the beam voltage is kept constant, then the
beam current and power are inversely proportional to the square of the frequency.
The RF output power is further reduced by the decreases in electronic efficiency
and circuit efficiency as the frequency increases. Additional limitations may be set
by the need to reduce the beam filling factor to ensure good beam transmission.
Hence, the limitation in performance of a TWT or crossed-​field tube at millimetre
and sub-​millimetre wavelengths arises directly from the need to reduce the phase
velocity of the electromagnetic wave to achieve synchronism with the velocity of
the electrons. Similar considerations apply to klystrons because of the need to
keep the normalised interaction gap lengths (βe g ) small to achieve useful coupling
between the beam and the cavities. These limitations on performance can be miti-
gated to some extent by the use of sheet or annular electron beams, and of higher
beam voltages. But the essential problem, of keeping an electron beam of sufficient
power very close to the surface of the electromagnetic structure, remains [2].
We saw in Chapter 2 that, for frequencies above cut-​off, the phase velocity of the
waves in a smooth metallic waveguide circular waveguide is greater than the velocity
of light, and strong RF electric fields can exist throughout the guide. Thus, if an
electron beam can interact with the electromagnetic field of a uniform waveguide,
it does not have to be close to the surface of the guide. In addition, the volume of
space which the beam can occupy is much greater so that the possible beam cur-
rent and power are increased. It is, of course, still necessary to achieve synchronism
between the electrons and the electromagnetic wave if a useful interaction is to
occur. In a folded-​waveguide TWT synchronism is achieved by making the struc-
ture periodic so that the interaction with a uniform beam is also periodic. That
suggests the alternative strategy of using a periodic electron beam and a uniform
structure. In principle this could be achieved in a number of ways, but practical
devices have linear beams controlled by either a uniform axial magnetic field or a
periodic magnetic field. These devices are classed as fast-​wave devices because the
phase velocity of the wave in the electromagnetic structure is greater than the vel-
ocity of light [3]. Fast-​wave devices have been discussed by many authors, and the
reader is referred to the literature for detailed information [4–​14]. Much of the pio-
neering work was done in the former Soviet Union, and a summary can be found
in [15]. The sources cited in this chapter are English because they are most readily
accessible, but work in Russia continues to be important.

17.2 Electron Cyclotron Masers

The electron cyclotron maser (ECM) interaction is the basis for a number of types
of tube of which gyrotrons (see Section 17.3) are an important example. The
 Electron Cyclotron Masers 661

principle of operation of the ECM interaction can be understood by considering

Figure 17.1 which shows an electron moving under the influence of a uniform lon-
gitudinal magnetic field in the presence of a uniform transverse RF electric field of
frequency ω . The electron moves in a circular orbit with an angular velocity equal
to the relativistic cyclotron frequency
eBz
ωc = , (17.2)
m0 γ
where
1
γ= (17.3)
1 − u2 c2
is the relativistic factor and
u 2 = uθ2 + uz2 , (17.4)

where uθ and uz are the tangential and axial components of the electron velocity and
it is assumed that the radial component of velocity is negligible. It can be seen that
the cyclotron frequency decreases as the electron velocity increases.
If the axial motion of the electrons is ignored, for the moment, then the motion
of an electron is in synchronism with the RF field when ω = ω c . A  synchronous
electron at position 1 experiences a retarding field. When it reaches position 2 the
direction of the RF electric field has reversed and the motion of the electron is fur-
ther retarded. Similarly, an electron starting at position 2 is accelerated and then
accelerated again when it reaches position 1.  The accelerated electrons fall back
in phase relative to the RF field, because of the change in the cyclotron frequency
with velocity. Similarly, retarded electrons move forward in phase so that electron
bunches are formed. If these bunches fall in the retarding phase of the RF field then
there is a net transfer of energy from the electrons to the wave, as required. When
relativistic effects are ignored the cyclotron frequency is constant and no bunching

Figure 17.1:  Principle of the electron cyclotron maser interaction.

662 Fast-Wave Devices

takes place. Hence a necessary condition for the ECM interaction is that the elec-
tron velocity should be high enough for relativistic effects to be important.
The radius of the orbit is the Larmor radius given by

uθ
rL = , (17.5)
ωc
which increases as the electron velocity increases. Because the RF electric field is
uniform in space, the change in the Larmor radius does not affect the field acting
on the electrons. This situation changes if the electric field is non-​uniform because
a net transfer of energy to the wave is possible, even when the velocities are low.
This is the peniotron interaction which is discussed in Section 17.5. It may also
occur in ECM devices if the RF electric field is non-​uniform in the region of the
electron beam.
The electron beam is normally generated by a magnetron injection gun so that
the energy of the electrons can be written using (1.4) as

 1 
uθ2 + uz2 = c 2 1 −  , (17.6)
( )
2
 1 + eV m c 2
 
  a 0  
where Va is the potential of the anode relative to the cathode. The ratio of the trans-
verse velocity to the axial velocity is denoted by the pitch factor

uθ
α= . (17.7)
uz
This parameter is normally greater than unity. Because only the azimuthal velocity
is changed by the interaction it follows that only the azimuthal energy can be con-
verted into RF power. Thus the electronic efficiency cannot exceed

ηe =
(uθ2 + uz2 ) − uz2 =
α2
. (17.8)
u +u
2
θ
2
z 1 + α2
In practice it is not possible to reduce the tangential velocities of all the electrons to
zero and the electronic efficiency is smaller that given by (17.8).
When the axial velocity of the electrons is taken into account the synchronous
condition is slightly altered. The time taken for an electron to make one circuit of
its orbit is

2π
t= . (17.9)
ωc
In that time the electron moves through an axial distance

2 πuz
z = uz t = . (17.10)
ωc
 Electron Cyclotron Masers 663

For synchronism, the phase of the electromagnetic wave as seen by the electron must be
ωt − βz = 2 sπ, (17.11)
where the wave propagates as exp j ( ωt − βz ) and s is a positive integer. Substituting for
t and z in (17.11) from (17.9) and (17.10) the condition for synchronism is found to be
ω = βuz ± sω c , (17.12)
where the first term on the right-​hand side represents the Doppler shifting of the
frequency caused by the longitudinal motion of the electron. Now the dispersion
curve of any mode in a uniform waveguide is given by (2.11), which may be written
1
β= ω 2 − ω 2mn , (17.13)
c
where ω mn is the cut-​off frequency of the mode with mode numbers ( m, n ). Note
that this notation has been used to avoid confusion with the use of ω c for the cyclo-
tron frequency. Equation (17.13) can be written in normalised form as
ω β2
= 1 + 2 , (17.14)
ω mn β mn
where β mn = ω mn c. Using the same normalisations the dispersion equation for
waves on the beam (17.12) can be written
ω β uz sω c
= ⋅ ± . (17.15)
ω mn β mn c ω mn
Figure 17.2 shows a typical uncoupled dispersion diagram plotted using equa-
tions (17.14) and (17.15). It can be seen that there are normally two points of
intersection between the waveguide mode and the electron cyclotron wave (marked
A and B in the diagram). The different types of ECM device correspond to different
choices of these points [10]:
• Gyrotron oscillator and gyro-​klystron amplifier: The two points of intersection
coincide so that the cyclotron wave line is tangential to the waveguide line close to
the waveguide cut-​off frequency on the right-​hand side of the diagram.
• Gyro-​TWT: The two points of intersection lie close to one another on the right-​
hand side of the diagram so that a travelling-​wave interaction is possible over a
band of frequencies.
• Gyro-​BWO: Point A lies on the left-​hand side of the diagram so that the beam
interacts with the backward wave in the waveguide.
• Cyclotron auto-​resonance maser (CARM): Point B lies well to the right-​hand side
of the diagram so that the phase velocity is close to the velocity of light. When
energy is extracted from the beam, the electron velocity decreases but the relativistic
cyclotron frequency increases so that the resonant condition (17.12) is maintained.
These types of tube are discussed in greater detail below.
664 Fast-Wave Devices

Figure 17.2:  Uncoupled dispersion diagram for electron cyclotron mode

interactions (uz c = 0.264, α = 1.5, sω c ω mn = 1.2 ).

17.2.1 Small-​Signal Theory of ECM Interactions

From coupled-​mode theory we expect the small-​signal dispersion equation, when
the modes shown in Figure 17.2 are coupled, to be of the form

(ω 2
− c 2 (β2 + β2mn )) ( ω − βuz − sω c ) = K f ( ω, β ) , (17.16)
2

where K is a coupling constant which depends upon the properties of the waveguide
and of the electron beam, and f is a function of ω and β. The squaring of the second
term on the left-​hand side of the equation is suggested by considering (11.137) in
the limit of low space-​charge. When K → 0 the solutions of (17.16) are the uncou-
pled modes shown in Figure  17.2, as expected. Equation (17.16) is a dispersion
equation having the general form

D ( ω, β ) = 0, (17.17)

which is very similar to the dispersion equation for the travelling-wave tube (see
(11.137)). The properties of a device represented by (17.17) are investigated by
searching for the roots of this equation (compare Sections 11.5 and 11.7) [16, 17].
In general both ω and β may be complex.
In an ECM the RF electric field is provided by a guided electromagnetic wave.
The simplest way of doing this is to use a TE0 n mode of a circular waveguide in
which the electric field lines form concentric circles (see Figure 2.14). The tangential
component of the electric field varies with radius as

Eθ ( r ) = E0 J1 (βr ) , (17.18)
 Electron Cyclotron Masers 665

Figure 17.3:  Radial variation of the tangential electric field in the TE 0 n modes of a circular
waveguide.

as shown in Figure 17.3. The tangential electric field must be zero on the wall of the
waveguide so that
J1 (β0 n a ) = 0, (17.19)

where a is the radius of the waveguide and n = 1, 2, 3. When n = 1 β01a = 3.832 and
the maximum of the curve is β01r = 1.841 so that r a = 0.48. Thus, for the best inter-
action the electrons should be in a thin annular beam whose mean radius is centred
on the peak of the field as shown in Figure 17.4. The Larmor radius ( rL ) should be
small compared with the mean radius r0 .
The dispersion equation is then found to be [18, 19]

 (ω 2 − β2 c 2 )
− νc 2 uθ2 ω − βuz 
ω 2 − c 2 (β2 + β20 n ) =  ⋅ R − Qs ,
πγ a 2 J 02 (β0 n a )  ( ω − βuz − sω c )2 c 2
s
ω − βuz − sω c 

(17.20)
where J 0 is the Bessel function of the first kind,
Ne 2
ν= (17.21)
ε 0 m0 c 2
is a dimensionless beam density parameter in which N is the number of electrons
per unit length of the beam,
2
Rs =  J s (β0 n r0 ) J s ′ (β0 n rL ) , (17.22)
 
and Qs (β0 n r0 , β0 n rL ) is defined in [18]. The second term in the brackets in (17.20)
is small, except when the transverse velocity uθ is small and it may, therefore, be
neglected so that the equation becomes
2
− νuθ2  J s (β0 n r0 ) J s ′ (β0 n rL ) 
ω − c (β + β
2 2 2 2
) [ω − βuz − sωc ] 2
= ⋅  (ω 2 − β 2 c 2 ) .
aJ 0 (β0 n a )
0n
πγ  
(17.23)
666 Fast-Wave Devices

Figure 17.4:  Arrangement of the electron beam in an ECM.

We note that this equation has the form expected in (17.16). The parameter ν can
be written
 cI Z
ν = ( γ − 1)   0 0 , (17.24)
 uz  Va
where Z0 = µ 0 ε 0 . The analogy with (11.137) is made more explicit if we write
β0 = ω c, βc = ω c c and rearrange the right-​hand side to give

[β20 − β2 − β20 n ] β0 − β (uz c ) − sβc  = − K (β02 − β2 ) β20 n , (17.25)

2

where
2
αuz ( γ − 1) I 0 Z0  J s (β0 n r0 ) J s ′ (β0 n rL ) 
K= ⋅ ⋅  . (17.26)
πγ c Va  β0 n a J 0 (β0 n a ) 
It can be seen that the coupling between the beam and the waveguide increases as
the beam current and the parameter α increase. Equation (17.25) is derived using
the Vlasov equation which is more familiar to plasma physicists than it is to elec-
tronic engineers. The derivation requires some major assumptions including the
representation of the beam by an infinitely thin annulus in which all the electrons
have the same initial velocity [20]. An alternative derivation based on an azimuthal
perturbation of the charge density in a thin annular beam gives essentially the
same result [21]. It should be noted that both derivations neglect the azimuthal
component of the space-​charge field. This can be justified on the grounds that
the dominant component of the space-​charge field is radial. It is found that RF
space-​charge effects are not important in gyrotron oscillators, but they can have
an important influence of the bunching in drift regions of gyro-​amplifiers [12].
Because space-​charge effects have been neglected, the two cyclotron modes are
 Electron Cyclotron Masers 667

Figure 17.5:  The real and imaginary parts of the propagation constants of an ECM
computed for a range of real frequencies.  (Va = 70 kV, I 0 = 2 A, α = 1.5, ω c ω 0 n = 0.945).

degenerate (i.e. they have the same propagation constants at a given frequency)
like space-​charge waves in the limit ω p → 0. The resemblance between (17.25) and
(11.137) has led some authors to follow Pierce [22] in approximating the equation
as a cubic so that the solutions could be expressed in terms of Pierce parameters
familiar to TWT engineers [19, 23]. This approach is now unnecessary because it
is straightforward to find the solutions to the full quartic equation, as shown in
Worksheet 17.1.
The roots of (17.25) can be computed for a given set of parameters. Figure 17.5
shows the real and imaginary parts of the propagation constant computed for
real frequencies, using data from [23] and a beam current of 2 A. The axes are
normalised to the cut-​off frequency of the waveguide and the free-​space propa-
gation constant at that frequency. Below the cut-​off frequency the forward and
backward waveguide modes (a)  are evanescent. The propagation constants of
the negative and positive energy cyclotron modes (b and c) are real, but per-
turbed by coupling with the waveguide, so that the degeneracy of the uncoupled
modes is removed. It should be noted that the negative energy mode (b) has a
higher phase velocity than the positive energy mode (c). This is the reverse of the
behaviour of longitudinal space-​charge waves. Above the cut-​off frequency, the
propagation constants of the backward waveguide mode (d), and the positive
energy cyclotron mode (c), are real. The coupling between the negative energy
cyclotron mode, and the forward waveguide mode, yields a complex conjugate
pair of roots (e)  over a range of frequencies. Similarly, when the equation is
solved for real values of the propagation constant, the result is a complex con-
jugate pair of roots for the frequency over the same range. Thus the interaction
satisfies the criteria given by Briggs for a convective instability analogous to that
in a TWT [17]. This can be used in a gyro-​TWT to produce gain over a band of
frequencies (see Section 17.4).
When the beam current is increased, the frequency range over which a complex
conjugate pair of roots exists increases. The point P at which β is real moves towards
668 Fast-Wave Devices

Figure 17.6:  The real and imaginary parts of the propagation constants of an ECM
computed for a range of real frequencies.  (Va = 70 kV, I 0 = 4 A, α = 1.5, ω c ω 0 n = 0.945)

Q. At a current of approximately 4A (for the example chosen) these points coincide,
as shown in Figure 17.6 so that all the roots are real at that point and a double root
exists [23]. This satisfies Briggs’ criterion for the onset of a non-​convective instabil-
ity [17, 23]. If the beam current is increased still further the tube may oscillate under
conditions determined by (17.17) and by
d
D ( ω, β ) = 0, (17.27)
dβ
where both ω and β are complex (compare Section 11.7). The term Qs (β0 n r0 , β0 n rL )
in (17.20) has the opposite sign to Rs and tends to increase the threshold at which a
non-​convective instability can occur. The threshold current is found to be increased
by the wall losses of the waveguide but not by velocity spread in the electron beam
[23]. The onset of oscillations is also affected by the length of the interaction
region, as in the case of a backward-​wave oscillator. Unlike the linear-​beam BWO
an ECM can also oscillate when the instability is convective because of reflections
at the ends of the interaction region. From this discussion it can be concluded that a
ECM oscillator may be based on either a convective, or a non-​convective, instability
according to the design and the operating conditions. In an ECM amplifier, how-
ever, it is necessary to ensure that the tube is stable against both the non-​convective
instability and feedback oscillations.

17.3 Gyrotron Oscillators

A gyrotron is an ECM oscillator. This type of tube has been the subject of
much research and development effort, and nearly all the ECMs manufactured
 Gyrotron Oscillators 669

Figure 17.7:  Output power of commercial gyrotron oscillators: (a) pulsed tubes, and

(b) CW tubes.

Figure 17.8:  Efficiency of commercial gyrotron oscillators.

commercially are gyrotrons. Figure  17.7 shows the power output as a func-
tion of frequency for both pulsed and CW tubes based on manufacturers’ data
sheets. Figure  17.8 summarises the efficiencies of both pulsed and CW tubes
as a function of frequency. From these figures it can be seen that the majority
of tubes operate at frequencies above about 20 GHz where they are superior
to linear-​beam tubes. Pulsed and CW powers up to 1 MW are achieved with
efficiencies greater than 30% up to a frequency around 200 GHz. Tubes exist
with lower output power and efficiency up to 1.3 THz. Further information is
available in [4, 24].
670 Fast-Wave Devices

Figure 17.9:  (a) Arrangement of a gyrotron oscillator, (b) Magnetic field profile, and (c) RF
electric field profile
(copyright 1977, IEEE, reproduced, with permission from [15]).

Figure 17.9(a) shows the basic arrangement of a gyrotron oscillator. A hollow

electron beam generated by a magnetron injection gun is injected into a cylindri-
cal cavity resonator which is operated close to the cut-​off frequency for longi-
tudinal travelling waves. The beam is controlled by the axial magnetic field of a
solenoid which also produces the desired orbits at the cyclotron frequency. For
an 80 kV beam the relativistic cyclotron frequency is 44 GHz at a field of 1.8 T,
which is around the maximum field which can readily be produced by a normal
conducting magnet [25]. This limit is set by the saturation of iron at around 2.1
T. At higher frequencies it is necessary to use a superconducting solenoid. As the
beam leaves the cavity it enters a circular waveguide of larger diameter which
allows electromagnetic waves to propagate out of the cavity. The magnetic field
decreases in this region, as shown in Figure  17.9(b), so that the spent electron
beam can be collected on the walls of the waveguide. The RF output power passes
through a circular output window into an external waveguide. The resonant
modes employed are normally either TE0 n circular waveguide modes, or TE m, p
whispering gallery modes (where m  1 and m > p), as shown in Figure 17.10. The
longitudinal variation of the azimuthal component of the RF electric field in the
cavity is a standing wave whose profile is approximately Gaussian as shown in
Figure 17.9(c). Methods for determining the eigenmodes of open cavity resona-
tors are discussed in [10]. Variations from this basic arrangement are described
briefly in Section 17.4.
 Gyrotron Oscillators 671

Figure 17.10:  Examples of the electric fields of cavity modes used in gyrotron oscillators:
(a) symmetric (TE 0 n ) modes, and (b) whispering gallery (TE m, p ) modes (m  1 and m > p )
(copyright 1997, IEEE, reproduced, with permission, from [12]).

17.3.1 Large-​Signal Interaction Model

The small-​signal theory of the gyrotron oscillator has been discussed in Section
17.2.1. A different approach is required to find the properties of a tube under large-​
signal conditions. Particle in Cell (PIC) codes can be used for this purpose but large
computer resources and long run-​times are needed [26]. Results can be obtained
more quickly using codes written specifically for modelling gyrotrons, such as
MAGY [27]. A code of this kind finds self-​consistent solutions of the electromag-
netic fields and the dynamics of the electrons. Since the purpose of this chapter is
to provide basic understanding of fast-​wave tubes it is convenient to use a simpli-
fied approach in which the electromagnetic fields are represented by an approxi-
mate function. The problem is then reduced to the solution of the equations of
motion of the electrons in the prescribed field [28]. The principles of the method are
described below and it is implemented in Worksheet 17.1.
The relativistic equations of motion of an electron in electromagnetic fields are
d
( γ m0 c 2 ) = −e u ⋅ E, (17.28)
dt
where γ is defined by (17.3), and
dp e
= −eE − v × B, (17.29)
dt c
where
p = γ m0 u (17.30)

is the momentum of an electron. These equations can be transformed using nor-

malised variables to give

dw Eφ
= p⊥ sin θ (17.31)
dZ sBz
672 Fast-Wave Devices

and
dθ E (1 − w )
= δ0 − w + r cos θ, (17.32)
dZ Bz p⊥
where θ is the angular position of the electron about its gyro-​centre,
γ
w = 1− , (17.33)
γ0

ω
Z= z, (17.34)
uz

1
p⊥ = (β2⊥ 0 − 2w + w 2 ) 2 , (17.35)

and the detuning parameter is given by

sω c 0
δ0 = 1 − . (17.36)
ω
In these equations the subscript 0 is used to indicate the value of the parameter at
the entrance to the resonator, β ⊥ = uθ c and β = uz c. A  further transformation
introduces the independent variable
β2⊥ 0 β2 z
ζ≡ Z = π ⊥ 0 ⋅ , (17.37)
2 β 0 λ
where λ = 2 πc ω , and the dependent variable

2 2c 2  γ 
υ≡ w = 1 − γ  . (17.38)
β2⊥ 0 uθ20 0

Note that the symbol υ has been used here for the normalised electron energy par-
ameter to avoid confusion with the use of u for velocities. The normalised inter-
action length is defined as

β2⊥0 L
µ≡π ⋅ , (17.39)
β 0 λ
where L is the physical length of the cavity and the detuning parameter is
transformed to

2δ 0 2  sω c 0 
∆≡ = 1 −  . (17.40)
β2⊥ 0 β2⊥ 0 ω 
Then equations (17.31) and (17.32) become

dυ  2s s !  p
= 2  s s −1  F f (ζ ) ⊥ J s′ ( sp⊥ ) sin θ (17.41)
dζ  s β⊥0  β⊥0
 Gyrotron Oscillators 673

and

dθ  2s s !  β ⊥ 0 (1 − β2⊥ 0 υ 2 )
= ∆ − υ − s  s s −1  F f (ζ ) J s ( sp⊥ ) coss θ. (17.42)
dζ  s β⊥0  p⊥2
Here F is the normalised amplitude of the RF electric field given by

E0 s − 4  s s −1 
F≡
Bz
β⊥0  J (x r a ) , (17.43)
 s ! 2 s −1  m ± s m, p 0
where xm, p is the pth non-​vanishing zero of J m′ ( x ) and the azimuthal component of
the RF electric field is

Eφ = E0 f ( z ) J m′ ( xm, p r a ) exp ( − jmφ ) . (17.44)

The field used is that of a rotating mode and the plus and minus signs in the
Bessel function subscript in (17.43) correspond to the two possible directions of
rotation.
The working equations (17.41) and (17.42) can be integrated numerically for a
number of electrons with the initial conditions υ = 0, and θ distributed uniformly
over ( 0, 2 π ). Most actual axial field profiles are approximated quite closely by a
Gaussian distribution. It is therefore assumed that

f (ζ ) = exp − ( 2ζ µ ) . (17.45)
2

The radius of the orbit of an electron is found from (17.5) to be

rL uγ
= θ . (17.46)
rL0 uθ 0 γ 0
Rearranging (17.3) we find that

uθ 1
= 1 − β20 − 2 , (17.47)
c γ
since it is assumed that the axial component of the electron velocity is unchanged
by the interaction. Substitution into (17.46) gives

rL
=
(1 − β ) γ
2
0
2
−1
. (17.48)
rL0 (1 − β ) γ
2
0
2
0 −1

But, from (17.38)

 β2 υ 
γ = γ 0 1 − ⊥ 0  (17.49)
 2 
so that the radius of the orbit of each electron as the interaction proceeds can be
calculated from the values of υ which have been computed.
674 Fast-Wave Devices

The electronic efficiency is computed from the change in the kinetic energy of the
electrons as they move through the cavity so that

γ0 − γ β2⊥ 0
ηe ≡ = η⊥ , (17.50)
γ 0 − 1 2 (1 − 1 γ 0 )
where η⊥ is the value of υ at the output plane averaged over all the electrons. The
beam current can be found by making use of the energy balance equation

F 2 = η⊥ I , (17.51)

where I is a normalised current parameter defined by

I0
I= , (17.52)
IR
in which I 0 is the beam current in amps and I R is a reference current given (in amps) by

J m2 ± s ( xm, p r0 a )
−1
  λ   ss 
2
β 2 ( s − 3) 
I R = 0.238 × 10 −3 ×    s  ⋅ ⊥ 0 QL  , (17.53)
  L   s s !  ( xm, p − m ) J m ( xm, p ) γ 0
2 2 2

where QL is the loaded Q of the resonator. It can be seen that I R is a constant, inde-
pendent of the operating point, for given tube parameters and mode numbers. The
overall efficiency is obtained by multiplying the electronic efficiency by the circuit
efficiency given by
QL
ηc = , (17.54)
QE
where QL and QE are the loaded and external Q factors of the cavity. In the gyrotron
literature it is usual to call these the total (QT ) and diffractive (QD ) Q factors, and
the unloaded (QU ) is known as the ohmic Q. For consistency the terminology used
in the rest of this book is used here. The ohmic losses are typically less than 10% of
the output power. The model described above provides a convenient way of inves-
tigating the properties, and design trade-​offs, of gyrotrons (see Worksheet 17.1).
It can be seen that equations (17.41) and (17.42) which describe the motion of an
electron depend upon the three parameters F, µ and ∆. If the first two of these are held
constant then it is found that there is a value of ∆ for which the transverse efficiency
( η⊥ ) is maximum. If, in addition, the beam is only weakly relativistic ( sβ2⊥ 0 2  1)
then the transverse efficiency does not depend upon the electron energy. Thus the
optimised transverse efficiency depends only on F, µ and the cyclotron harmonic s
[28–​30]. Figure 17.11 shows the plot for the fundamental ( s = 1) cyclotron interaction.
When the data from Figure 17.11 is substituted into (17.51) then the result is a plot
showing the dependence of transverse efficiency on the normalised current and the
interaction length, as shown in Figure 17.12. These plots are useful tools for gyrotron
design. Corresponding plots for cyclotron harmonics up to s = 5 are given in [28, 30].
We saw in Section 17.2 that a gyrotron will not oscillate until the beam current
exceeds a starting value. That analysis ignored the effects of the length of the cavity,
 Gyrotron Oscillators 675

Figure 17.11:  Contours of the transverse efficiency of a gyrotron oscillator optimised with

respect to ∆ plotted against the normalised field strength F and normalised interaction
length µ for s = 1
(copyright 1985, MIT. Reproduced, with permission, from [30]).

Figure 17.12:  Contours of the transverse efficiency of a gyrotron oscillator optimised with respect
to ∆ plotted against the normalised current I and normalised interaction length µ for s = 1
(copyright 1986, AIP, reproduced, with permission, from [28]).

and of the variation of the interaction field along it, so it cannot provide an accur-
ate value for the starting current. An alternative approach, using the large-​signal
theory above, seeks the limiting current when the field strength parameter F tends
to zero. Thus, from (17.51), the normalised starting current is given by

 F2 
I st = Lim   . (17.55)
F →0  η 
⊥
676 Fast-Wave Devices

This can be calculated for any value of ∆ by finding the solution for a sufficiently
small value of F. For a given mode, and fixed beam parameters, the value of µ is
constant. Then the least possible starting current is given by the value of I min that
corresponds to the value of ∆ for which η⊥ is maximum. The dashed line marked
I = I min in Figure 17.12 shows the locus of this current as µ is varied. Below this line
it is not possible for oscillations to start.
If ∆ is varied while µ is held constant it can be seen from (17.55) that the value of
I st must be greater than I min . The same result is obtained directly using the equation
for the starting current given in [28]

 4  exp (2 x 2 )
2

I st =   ⋅ , (17.56)
 πµ  µx − s
where
µ∆
x≡ . (17.57)
4
Figure 17.13 shows a graph of I st against ∆, for three values of µ , for the operat-
ing mode of a 1 MW, 140 GHz CW gyrotron [31]. This graph is independent of the
cavity mode and of the properties of the beam. The starting current in amps for
each mode can be plotted against the magnetic field using (17.40) and (17.52). This
plot is useful for determining which mode is excited first for a given magnetic field.
It should be noted that there is normally a difference between the starting currents
of co-​and contra-​rotating modes having the same mode numbers. In general the
co-​rotating mode (negative sign in the Bessel function subscript in (17.53)) is found
to have the best coupling to the beam [10].
The normalised starting current can be computed from (17.56) for any point in
Figure 17.12. The dashed line marked I = I st shows the contour along which that
condition is satisfied. In the region below this line the operating current I is greater

Figure 17.13:  Normalised starting current of a gyrotron as a function of normalised

detuning parameter.
 Gyrotron Oscillators 677

than the starting current. This region is referred to as the soft-​excitation region
where oscillations will start if the beam current exceeds the starting current. Above
the line I = I st lies the hard-​excitation region. A  tube whose intended operating
point lies in this region must be started at a point in the soft-​excitation region. The
intended operating point is then reached by changing the current, voltage or mag-
netic field [10, 12, 28].

17.3.2 Case Study: A 140 GHz, 1 MW CW, Gyrotron

To illustrate the use of the theory outlined above let us consider the 140 GHz, 1
MW CW, gyrotron whose design parameters are given in Table 17.1 [31, 32]
The values of the normalised parameters of this tube for start-​up and normal
operating conditions are given in Table  17.2. In both cases the value of F was
adjusted to give a beam current of 40 A. Note that the tube is not operating at opti-
mum efficiency at start-​up. The computed output power is 895 kW at start-​up and
1 MW under normal operating conditions. When the operating point is plotted on
Figure 17.12 it can be seen that it lies in the hard excitation region. Potential depres-
sion in the beam is important in short-​pulse tubes, and in the start-​up of long-​pulse
and CW tubes (see Section 17.3.3). In CW and long pulse operation it is found that
the beam space-​charge is compensated by the accumulation of ions, after around
100 ms [31]. The reduction of the effective beam voltage at start-​up in this tube,
caused by space-​charge potential depression, ensures that the start-​up point lies in
the soft excitation region.
Figure  17.14 shows the phases of the electrons relative to the RF field along
the length of the interaction region calculated using the normalised parameters in
Table 17.2. The beam is modelled by 60 sample electrons, having the same initial

Table 17.1:  Design parameters of a 140 GHz, 1 MW CW, gyrotron [31]

(* corrected values [32])

Frequency 140 GHz

Cavity mode TE 28,8

Power, pulse length 1 MW, CW (1800 s)
Electron beam radius 10.1 mm*
Beam voltage 81 kV (74.74 kV at start-​up)
Beam current 40 A
Velocity ratio α 1.3 (1.5 at start-​up *)
Magnetic field 5.56 T
Cavity radius 20.48 mm
Cavity length (Lc) 14.5 mm
Q0 48650*
QL 1150*
678 Fast-Wave Devices

Table 17.2:  Normalised operating parameters of a 140

GHz CW gyrotron oscillator [31]

Start-​up Operating

∆ 0.35 0.49
µ 13.0 11.1
F 0.090 0.103
I 0.019 0.020
Ist 0.007 0.028

Figure 17.14:  Phase of sample electrons plotted against axial position, normalised to the
length of the cavity, for a 1 MW, 140 GHz CW gyrotron.

angular velocity, that are uniformly distributed in phase at the input to the cavity.
At the centre of the cavity the beam is strongly bunched, but the absence of azi-
muthal space-​charge forces means that there is a large velocity spread so that the
bunch disperses by the end of the cavity. A more detailed view of the bunching is
shown in Figure 17.15, which shows the azimuthal and radial positions of the elec-
trons at six equally-​spaced planes along the length of the cavity. Figure 17.16 shows
the evolution of the transverse and electronic efficiencies along the cavity. Note that
the maximum transverse efficiency is close to the values at the normal operating
point in Figures 17.11 and 17.12.
The non-​linear model described above gives results which are useful for explor-
ing the initial design of a gyrotron. Its chief limitation lies in the assumption of
a Gaussian standing-​wave field in the cavity. In practice the field usually differs
appreciably from the Gaussian profile at the output end of the cavity. In this region
the wave becomes a travelling wave so that the phase is no longer constant [10, 33].
To get more accurate results it is necessary to find a self-​consistent solution for the
RF fields and the electron dynamics [12, 27].
 Gyrotron Oscillators 679

Figure 17.15:  Phases and radial positions of sample electrons at equally-​spaced planes

( z1 = z/Lc ) within the cavity of a 1 MW, 140 GHz CW gyrotron.

Figure 17.16:  Evolution of the transverse and electronic efficiencies plotted against axial
position, normalised to the length of the cavity, for a 1 MW, 140 GHz CW gyrotron.

17.3.3 Design of Gyrotron Oscillators

The design of gyrotrons is discussed in detail in the literature and only the principal
points are considered here [10, 12, 29, 34, 35]. An example of a design study for a
high power gyrotron can be found in [36] and a detailed description of the develop-
ment of a 1 MW CW, 140 GHz, gyrotron is given in [31, 37, 38].
Factors which are important in the design of a gyrotron are ohmic heating of
the cavity walls and stability of the operating mode. The power handling capabil-
ity of the output window, which used to be a constraint, is of less concern fol-
lowing the adoption of CVD diamond for windows. Gyrotrons are typically
designed and operated with parameters in the high efficiency region. It can be
seen, from Figures  17.11 and 17.12 that this corresponds to a detuning param-
eter ∆  0.45 − 0.55 and a cavity length parameter µ  15 − 20. This lies in the hard
680 Fast-Wave Devices

excitation region and programmed voltage rise during start-​up must be used to
access the operating point. The choice of operating point is constrained by a num-
ber of factors, especially the average ohmic power density on the cavity walls and
the external (diffractive) Q [29]. The average power density should be less than
2 kW cm −2 . The external Q has been shown to be proportional to ( L λ ) for power
2

extraction from the end of the cavity. Thus, reducing the power density on the walls
by increasing the length of the cavity leads to a reduction in the circuit efficiency as
more power is trapped within the cavity.
Because the operation of a gyrotron depends upon the resonance between the
cyclotron frequency and a resonant frequency of the cavity it is possible to design
tubes which employ high-​order cavity modes, while still achieving stable single-​
mode operation. It is found that a large-​amplitude mode suppresses its neigh-
bours so that single-​frequency operation is possible, even when mode density is
high. Symmetric modes (TE0 n ) have low wall losses and require small cavity radii.
Whispering-​gallery modes (TE m, p ; m  1 and m > p ) have superior mode stability
and are preferred for the highest power devices. It can be shown that ( ν2mp − m2 )
is a measure of the stored energy in a cavity, and it is desirable for this parameter
to be as large as possible. However, as the mode numbers are increased, the dens-
ity of modes increases, leading to problems of mode selection. It is also found
that the external Q increases, producing a decrease in the circuit efficiency. For
the 1 MW tube considered above ν2mp − m2 = 53 [31]. The excitation of possible
competing modes can be examined by plotting curves of their starting currents
against magnetic field. It is desirable for the operating point to be in a region rela-
tively free from competing modes and for it to have the lowest starting current.
The choice of beam voltage and current depends upon the overall efficiency
anticipated. The optimum beam radius for operation in the first cyclotron har-
monic is given from (17.44) by

xm ±1,1 λ
r0 = a = xm ±1,1 , (17.58)
xm, p 2π

where the plus and minus signs refer to the direction of rotation of the wave [36]. In
short-​pulse tubes it is necessary to consider the space-​charge potential depression
in the beam and the limit which that sets on the maximum beam current. The volt-
age depression for a thin hollow beam is given by [36]

I0  a 
∆V ≈ 60 Ω × ln . (17.59)
β  r0 

If the beam thickness is too big, the variation in space-​charge potential produces
variations in γ , in the detuning parameter, and in F, so that η⊥ cannot be optimised
for all the electrons. Finally, it also increases the risk of multimode operation. Since
energy is only extracted from the azimuthal motion of the electrons the transverse
efficiency is reduced as shown by (17.50). The ratio of transverse to longitudinal
velocity is generally chosen to lie in the range 1 < α < 1.5 to achieve good electronic
 Gyrotron Oscillators 681

Figure 17.17:  Gyrotron output coupling arrangements: (a) over-moded circular waveguide,

and (b) quasi-​optical mode converter
(copyright 2014, IEEE, reproduced, with permission, from [34]).

efficiency. It is desirable for the velocity spread in the electron beam to be as small
as possible. However, though an RMS spread of 6% in the transverse velocity has
little effect on efficiency, a10% spread causes an efficiency drop of the order of 20%.
A 10% spread in the electron energy was found to cause a 50% drop in efficiency
[12]. The electron beam in a gyrotron is usually provided by a magnetron injection
gun with a temperature-​limited cathode. This permits independent control of the
beam current. The development of the gun relies heavily on computer modelling to
achieve the required performance.
Figure  17.17 shows alternative output coupling methods used in gyrotrons. In
both cases the electron beam is compressed as it leaves the electron gun until it
reaches a point where the surrounding waveguide is cut off at the intended operat-
ing frequency. Care is taken to suppress parasitic oscillations in this region [34].
The diameter of the waveguide is increased until it reaches the start of the cav-
ity. It is normally constant along the length of the cavity but variations are some-
times used to increase mode stability. At the end of the cavity there is an increasing
taper so that the higher-order mode employed can propagate downstream. Because
discontinuities in over-​moded waveguides can cause unwanted mode conversion,
the junctions between the cavity and the tapers at each end are made gradual to
minimise mode conversion [10, 34, 39]. Early gyrotrons operating in TE0 n modes
682 Fast-Wave Devices

employed the output coupling scheme, shown in Figure 17.17(a), in which the col-
lector is a section of over-​moded waveguide long enough for all the electrons to
be collected before the output window is reached. External converters are then
used to convert the unpolarised output mode into a linearly polarised plane wave.
Modern high power gyrotrons employ high-order whispering-​gallery modes which
cannot be converted using conventional waveguide mode converters. Instead, the
mode is converted into the fundamental, Gaussian, free space mode (TEM 0,0 ) by
an asymmetrical waveguide launcher and an arrangement of mirrors, as shown in
Figure  17.17(b). This is known as a quasi-​optical mode converter. The RF win-
dow is then on the side of the tube, and the spent beam expands into a collector,
which can have a large surface area. This is possible because the conflict between
the requirements of the electron collector, and of the output coupler, has been
removed. The overall efficiency of the tube can be increased by collector depression
[34]. In either case it is important that the spent electrons are collected on the side
of the cylindrical collector and do not impinge on the window. A detailed descrip-
tion of the design of the collector for a high power gyrotron is to be found in [40].
For information about the design of depressed collectors for gyrotrons see [41–​44].
It is difficult to extend the operation of conventional whispering-​gallery mode
gyrotrons to higher powers and frequencies because the maximum average power
dissipated in the walls of the cavity is determined by the cooling technology avail-
able. In theory the average thermal loading can be reduced by going to even higher
radial mode indices, but there are then severe problems with mode competition.
The use of a coaxial cavity offers a solution to this problem because some of the
possible modes are short-​circuited by the inner conductor so that the total number
of modes is reduced. The modes used are TE m, p modes, with m ≈ 2 p , that are found
to give lower ohmic losses. Further information about coaxial gyrotrons is to be
found in [45–​47].
The need to provide a higher magnetic field as the frequency increases can be
offset by working at harmonics of the cyclotron frequency. Theoretical and experi-
mental studies have shown that useful efficiencies should be attainable even at quite
high harmonics [28, 33]. The difficulty is, again, that of mode selection because
of the possibility that modes involving lower-order harmonics will be excited [35].
This means that, in practice, it is difficult to operate a conventional gyrotron above
the second harmonic. One solution to this problem is to employ an electron beam
in which the guiding centres lie on the axis of the resonator so that electron orbits
encircle the axis. The result is known as a Large Orbit Gyrotron in which only co-​
rotating modes whose azimuthal indices are equal to the harmonic number can be
excited ( m = s ) [35, 48, 49]. The number of possible modes is then greatly reduced.
In this way it has been possible to generate 10 to 20 kW at 410 GHz using the fifth
cyclotron harmonic [14]. A further possibility is to replace the cylindrical cavity by
one in which the wave is trapped between a pair of opposed, curved mirrors [50,
51]. The result is a TE1, p mode which can be considered as a generalisation of the
modes in a cylindrical resonator which has been divided into two halves by axial
cuts. This resembles the kind of open resonator used in lasers [52–​54]. Tubes of this
 Gyrotron Oscillators 683

Figure 17.18:  Operating points for cyclotron auto-​resonance masers (A), and gyro-​backward-​
wave oscillators (B).

kind, known as Quasi-​Optical Gyrotrons, do not appear to have progressed beyond

the experimental stage.

17.3.4 Cyclotron Auto-​Resonance Masers

A way of achieving high frequency operation without a very high magnetic field is
the Cyclotron Auto-​resonance Maser (CARM) which operates at a point such as A
in Figure 17.18. The frequency at this point is well above both the cut-​off frequency
of the guide and the cyclotron frequency. The interaction is convective and high
efficiency can be achieved. However it is very sensitive to the velocity spread in the
beam if the full theoretical efficiency is to be obtained [10, 55].

17.3.5 Tuneable Gyrotrons

The frequency of oscillation of a gyrotron oscillator is very close to the cut-​off
frequency of the cavity mode selected. Some tuning is possible by changing the
magnetic field to excite different cavity modes. Step-​tuned gyrotrons with power
outputs up to 100 W in the frequency range 75–​600 GHz have been developed for
plasma diagnostics and the study of nonlinear effects in materials research [56–​59].
These tubes employ both first and second harmonic cyclotron resonances. Smooth
tuning over a frequency range has been achieved by excitation of successive axial
modes. Power of the order of a few watts has been generated over a range of more
than 1 GHz at 460 GHz [33, 60]. If the operating point is moved into the backward-​
wave region, as shown at B in Figure 17.18, the interaction is non-​convective as in
684 Fast-Wave Devices

Figure 17.19:  Arrangement of gyro-​amplifiers: (a) gyro-​klystron and (b) gyro-​TWA

(copyright1997, IEEE, reproduced, with permission, from [12]).

a conventional backward-​wave oscillator. Smooth variation of the frequency of

operation can be achieved by changing either the beam velocity, or the magnetic
field. The frequency of oscillation is below the cyclotron frequency so that a high
magnetic field is required for high frequency operation [10, 61–​63].

17.4 Gyro-​Amplifiers

We saw in Section 17.2 that the interaction between the electron beam and the elec-
tromagnetic wave takes the form of a convective instability when the beam current
is below the threshold value for oscillation. The dispersion diagram in Figure 17.5
bears a strong resemblance to that for a TWT (see Figure 11.19). It is therefore pos-
sible to build gyrotron amplifiers which are analogous to klystrons and TWTs. The
development of these tubes is more difficult than the development of oscillators
and only a few commercial devices exist. It is essential that they are stable in the
absence of RF input and the development of overmoded input couplers is chal-
lenging. They must also achieve performance defined by factors such as bandwidth,
gain, phase stability, and noise which are not relevant for oscillators [7, 12]. The
principal application of these tubes is in radar at 35 and 94 GHz where the attenu-
ation of waves passing through the atmosphere is low.
In a gyro-​klystron the electron beam passes through a series of cavities separated
by cut-​off drift tubes as shown in Figure 17.19(a). A practical, five-​cavity, tube pro-
duced a peak output power of 100 kW at 94 GHz with a 3 dB bandwidth of 700
MHz, saturated gain of 33 dB and efficiency of about 25% at 10% duty cycle [64–​66].
In a gyro-​TWA the operating point is moved further above the cut-​off frequency of
the waveguide so that the growing wave has a group velocity which is not close to zero.
 Peniotrons 685

Figure 17.20:  Principle of operation of the peniotron.

Figure 17.19(b) shows a schematic view of a gyro-​TWA. Practical tubes employing a

helically corrugated waveguide have been developed [67–​69]. A recent example gen-
erated more than 1 kW of continuous power in the range 30–​31 GHz, and efficiency
greater than 5% [70]. In general, like their linear-​beam counterparts, gyro-​klystrons
have the highest efficiency and gyro-​TWAs the greatest bandwidth [12].

17.5 Peniotrons

A different interaction between a fast wave and a periodic electron beam can exist
where the RF electric field is not uniform, as shown in Figure 17.20. It is not neces-
sary for the electrons to have relativistic velocities, and the cyclotron frequency is
constant. Let us suppose that the signal frequency is twice the cyclotron frequency.
An electron which experiences a retarding field at A moves in an orbit with smaller
radius, as shown by the inner dashed circle. After half a cyclotron period it reaches
B where it encounters an accelerating field. However, because the orbit is smaller
than before, this field is weaker than the retarding field at A. There is therefore a
net transfer of energy from the electron to the field. Similarly, an electron which is
accelerated by the field at A moves in a larger orbit and encounters a retarding field
at C that is stronger than the field at A. Once again there is a net transfer of energy
to the RF wave. This process continues in successive cyclotron periods so that the
guiding centres drift away from the original centre of the beam. A device operating
on this principle is known as a Peniotron.
The peniotron is a cyclotron resonance device, like the gyrotron, but with the
important difference that energy is taken from all the electrons, and no phase
bunching takes place. Thus, it is possible for the interaction to be highly efficient.
Also, the phase of the RF field seen by an electron varies by 2 π as it moves once
around its orbit. Thus the resonant condition becomes

ωt − βz = ( s ± 1) 2π, (17.60)
686 Fast-Wave Devices

which may be compared with the corresponding equation for the gyrotron (17.11).
Hence, substituting for t and z using (17.9) and (17.10) we obtain

ω − βuz = ( s ± 1) ω c . (17.61)

Then, by analogy with (17.16) we expect the small-​signal dispersion relation to take
the form

(ω 2
)
− c 2 (β2 + β2m,n ) ( ω − βuz − ( s ± 1) ω c ) = K f ( ω, β ) . (17.62)
2

Hence the dispersion diagram for a peniotron is identical to that for a gyrotron,
shown in Figure 17.2 and the gyrotron and peniotron mechanisms may compete
with one another in the same device [71].
The original experimental peniotron employed an axis-​encircling hollow beam
within a double-​ridged waveguide propagating the TE10 mode [72–​75]. More recent
tubes have employed cavities with longitudinal vanes which resemble unstrapped
magnetron cavities. This type of cavity is more suited to high harmonic peniotron
operation because it is smaller, permitting lower voltage operation, and because it
provides better mode selection. Large signal modelling of both gyrotron and peni-
otron interactions in a magnetron type cavity show that the fast peniotron interac-
tion (the positive sign in (17.61)) has very poor efficiency. Very good efficiency is
possible for the slow peniotron harmonics (( s − 1) , ( s − 2 ) ) but it decreases rapidly
with increasing spread in the positions of the guiding centres [76, 77]. The peniotron
is therefore chiefly of interest because of its potential for delivering high efficiency
at high harmonic numbers and, therefore, for high-​frequency operation without the
need for a superconducting magnet. It has been the subject of a number of theoreti-
cal and experimental studies (see for example: [78–​84]) but the experimental results
were generally disappointing. A  35 GHz third harmonic oscillator, which gener-
ated 6.9 kW of RF output power, demonstrated the achievement of a theoretical
electronic efficiency of 75% but the conversion efficiency was only 39% because of
high circuit losses [83]. The same paper reported results for operation in the tenth
cyclotron harmonic at 100 GHz. The highest electronic conversion efficiency was
2.3% at an output power of 160 W, corresponding to an electronic efficiency of 6%.
At the time of writing these seem to have been the best results obtained.

17.6 Ubitrons (Free Electron Lasers)

The properties of an electron beam can also be made periodic by passing it through
a periodic magnetic field, as shown in Figure 17.21 [85]. Figure 17.21(a) shows a
sheet electron beam within a rectangular waveguide. The periodic transverse mag-
netic field causes the electron trajectories to follow undulating paths in the y-​z plane.
This motion can be coupled to the RF electric field of the TE10 mode of the wave-
guide. A microwave tube employing this principle is known as an Ubitron (undulat-
ing beam interaction) [86]. Figures 17.21(b) and (c) show alternative arrangements
 Ubitrons (Free Electron Lasers) 687

Figure 17.21:  Arrangements of ubitrons: (a) planar, (b) coaxial and (c) cylindrical

(copyright 1960, IRE, reproduced, with permission, from [86]).

of such a tube where the beam is focused by a periodic permanent magnet field. The
arrangement of electro-​or permanent magnets to produce an undulating beam is
known as a wiggler (or undulator).
If space-​charge forces are ignored, the propagation constants of the space-​
harmonics of the waves on the beam are given by
ω 2 νπ
βν = ± , (17.63)
uz λw
where λ w is the wavelength of the wiggler and ν = 0, 1, 2, etc. The strongest inter-
action with the field of the waveguide is that of the ν = −1 space harmonic for which
the dispersion equation takes the form

(ω 2
)
− c 2 (β2 + β2m,n ) ( ω − (β + βw ) uz ) = K f ( ω, β ) , (17.64)
2

where βw = 2 π λ w . The dispersion diagram is then, as shown in Figure 17.22, where

βw and uz have been chosen so that the space-​harmonic line is tangential to the wave-
guide dispersion curve. It can be shown that the condition for tangential operation is

1 1
βw = ω m, n − 2 . (17.65)
uz c
2
688 Fast-Wave Devices

Figure 17.22:  Uncoupled dispersion diagram for an ubitron.

The magnitude of the term under the square root is set by the beam velocity. The fre-
quency of operation is slightly above the cut-​off frequency of the waveguide mode
(ω m,n ), so the wavelength of the wiggler decreases as the frequency increases. The
maximum frequency of operation at a given voltage is determined by the minimum
practical period of the wiggler. Since the field of the wiggler is essentially governed
by Laplace’s equation, the transverse distance between the pole-​pieces needs to be
less than their axial separation, in order to maintain a high transverse component
of the magnetic field. Thus, the transverse space available for the beam decreases
as the frequency increases. For example, at a beam voltage of 80 kV ( uz c  0.5),
and a cut-​off frequency of 50 GHz, it is found that λ w  3.5 mm. This is probably
around the minimum practical value. Operation at higher frequencies requires the
use of higher beam voltages.
The first experimental ubitron used the arrangement in Figure  17.21(a) and
achieved a maximum gain of 13 dB at about 2.6 GHz with a beam voltage of 135
kV [86]. The peak output power was 1.2 MW at a beam voltage of 170 kV with
an efficiency of 10%. Subsequent tubes, using the arrangement in Figure 17.21(c),
delivered 1.65 MW pk at 15.75 GHz with a solid beam, and 150 kW pk at 54 GHz
with a hollow beam [85]. Another tube gave a peak output power of 4.2 MW at
16.6 GHz with 17.5% efficiency, 29 dB gain and 22% instantaneous bandwidth [87].
Ubitrons tend to suffer from poor beam transmission, problems with excitation of
lower modes, and high sensitivity to changes in the beam voltage and the magnetic
field. Information about the design, and in some cases, the performance of a num-
ber of other tubes can be found in [88–​92].
 References 689

Ubitrons work at relatively low velocities where the interaction between the
electrons and the waves is analogous to that in a TWT. It is interesting to note that
computer modelling of this interaction shows axial bunching of the beam similar
to that in a TWT. If the velocities of the electrons approach the speed of light,
then they lose appreciable energy by radiation, as a result of transverse accelera-
tion within the wiggler. When this process takes place in a strong RF field then the
radiation emitted is phase locked to that field and the device becomes a coherent
power source. A device which works in this way is known as a Free Electron Laser
(FEL). However, despite its name, the FEL is not a quantum device because the
electrons are free, and therefore do not have quantised energies. The operation of
an FEL can be explained entirely by classical theory. In an FEL the beam is pro-
vided by an accelerator, and coherent radiation can be generated at wavelengths
from the infra-​red to the X-​ray region of the electromagnetic spectrum. These
devices lie outside the scope of this book but further information can be found
in [4, 6, 8].

References

[1]‌ M. I. Petelin, ‘One century of cyclotron radiation’, IEEE Transactions on Plasma

Science, vol. 27, pp. 294–​302, 1999.
[2]‌ J. H. Booske et al., ‘Vacuum electronic high power terahertz sources’, IEEE Transactions
on Terahertz Science and Technology, vol. 1, pp. 54–​75, 2011.
[3]‌ J. Feinstein and K. Felch, ‘Status review of research on millimeter-​wave tubes’, IEEE
Transactions on Electron Devices, vol. 34, pp. 461–​467, 1987.
[4]‌ G. Faillon et al., ‘Microwave tubes’, in J. A. Eichmeier and M. K. Thumm, eds, Vacuum
Electronics: Components and Devices. Berlin: Springer-​Verlag, pp. 1–​84, 2008.
[5]‌ V. L. Granatstein and I. Alexeff, High-​Power Microwave Sources. Boston, MA: Artech
House, 1987.
[6]‌ J. A. Pasour, ‘Free-​electron lasers’, IEEE Circuits and Devices Magazine, vol. 3, pp.
55–​64, 1987.
[7]‌ K. L. Felch et  al., ‘Characteristics and applications of fast-​ wave gyrodevices’,
Proceedings of the IEEE, vol. 87, pp. 752–​781, 1999.
[8]‌ H. P. Freund and G. R. Neil, ‘Free-​electron lasers:  vacuum electronic generators of
coherent radiation’, Proceedings of the IEEE, vol. 87, pp. 782–​803, 1999.
[9]‌ B. G. Danly et al., ‘Gyro-​amplifiers’, in R. J. Barker et al., eds, Modern Microwave and
Millimetre-​Wave Power Electronics. Piscataway, NJ: IEEE Press, pp. 247–​287, 2005.
[10] M. V. Kartikeyan et  al., Gyrotrons:  High-​ Power Microwave and Millimeter Wave
Technology. Berlin: Springer-​Verlag, 2010.
[11] K. Chu, ‘The electron cyclotron maser’, Reviews of Modern Physics, vol. 76, p. 489, 2004.
[12] V. L. Granatstein et  al., ‘A quarter century of gyrotron research and development’,
IEEE Transactions on Plasma Science, vol. 25, pp. 1322–​1335, 1997.
[13] G. Nusinovich et  al., ‘The gyrotron at 50:  historical overview’, Journal of Infrared,
Millimeter, and Terahertz Waves, vol. 35, pp. 325–​381, 2014.
[14] M. Y. Glyavin et al., ‘Terahertz gyrotrons: state of the art and prospects’, Journal of
Communications Technology and Electronics, vol. 59, pp. 792–​797, 2014.
690 Fast-Wave Devices

[15] V. A. Flyagin et  al., ‘The gyrotron’, IEEE Transactions on Microwave Theory and
Techniques, vol. 25, pp. 514–​521, 1977.
[16] P. A. Sturrock, ‘Kinematics of growing waves’, Physical Review, vol. 112,
pp. 1488–​1503, 1958.
[17] R. J. Briggs, Electron-​Stream Interaction with Plasmas.  Cambridge, MA: MIT Press,
1964.
[18] K. R. Chu et al., ‘Characteristics and optimum operating parameters of a gyrotron trav-
eling wave amplifier’, IEEE Transactions on Microwave Theory and Techniques, vol. 27,
pp. 178–​187, 1979.
[19] A. J. Sangster, ‘Small-​signal analysis of the travelling-​wave gyrotron using Pierce param-
eters’, IEE Proceedings I: Solid-​State and Electron Devices, vol. 127, pp. 45–​52, 1980.
[20] P. Lindsay, ‘Gyrotrons (electron cyclotron masers):  different mathematical models’,
IEEE Journal of Quantum Electronics, vol. 17, pp. 1327–​1333, 1981.
[21] Y. Y. Lau, ‘Simple macroscopic theory of cyclotron maser instabilities’, IEEE
Transactions on Electron Devices, vol. 29, pp. 320–​335, 1982.
[22] J. R. Pierce, Traveling-​Wave Tubes. Princeton, NJ: D. van Nostrand, 1950.
[23] K. R. Chu and A. T. Lin, ‘Gain and bandwidth of the gyro-​TWT and CARM ampli-
fiers’, IEEE Transactions on Plasma Science, vol. 16, pp. 90–​104, 1988.
[24] M. Thumm, State-​ of-​
the-​
Art of High Power Gyro-​ Devices and Free Electron
Masers. Update 2015 (KIT Scientific Reports; 7717), vol. 7717.  KIT Scientific
Publishing, 2016.
[25] D. Borodin and M. Einat, ‘Copper solenoid design for the continuous operation of a
second harmonic 95-​GHz gyrotron’, IEEE Transactions on Electron Devices, vol. 61,
pp. 3309–​3316, 2014.
[26] J. Neudorfer et al., ‘Efficient parallelization of a three-​dimensional high-​order particle-​
in-​cell method for the simulation of a 170 GHz gyrotron resonator’, IEEE Transactions
on Plasma Science, vol. 41, pp. 87–​98, 2013.
[27] M. Botton et al., ‘MAGY: a time-​dependent code for simulation of slow and fast micro-
wave sources’, IEEE Transactions on Plasma Science, vol. 26, pp. 882–​892, 1998.
[28] B. Danly and R. J. Temkin, ‘Generalized nonlinear harmonic gyrotron theory’, Physics
of Fluids, vol. 29, pp. 561–567, 1986.
[29] K. E. Kreischer et  al., ‘The design of megawatt gyrotrons’, IEEE Transactions on
Plasma Science, vol. 13, pp. 364–​373, 1985.
[30] B. Danly and R. J. Temkin, Generalized Nonlinear Harmonic Gyrotron Theory, MIT,
Cambridge MA 1070–​6631, 1985.
[31] M. Thumm et al., ‘EU megawatt-​class 140-​GHz CW gyrotron’, IEEE Transactions on
Plasma Science, vol. 35, pp. 143–​153, 2007.
[32] M. Thumm, ‘Private communication’, 2015.
[33] M. K. Hornstein et al., ‘Second harmonic operation at 460 GHz and broadband con-
tinuous frequency tuning of a gyrotron oscillator’, IEEE Transactions on Electron
Devices, vol. 52, pp. 798–​807, 2005.
[34] M. Thumm, ‘Recent advances in the worldwide fusion gyrotron development’, IEEE
Transactions on Plasma Science, vol. 42, pp. 590–​599, 2014.
[35] V. Bratman et al., ‘Review of Subterahertz and Terahertz Gyrodevices at IAP RAS and
FIR FU’, IEEE Transactions on Plasma Science, vol. 37, pp. 36–​43, 2009.
[36] M. V. Kartikeyan et  al., ‘Possible operation of a 1.5-​ 2-​
MW, CW conventional
cavity gyrotron at 140 GHz’, IEEE Transactions on Plasma Science, vol. 28,
pp. 645–​651, 2000.
 References 691

[37] G. Dammertz et al., ‘Development of a 140-​GHz 1-​MW continuous wave gyrotron for
the W7-​X stellarator’, IEEE Transactions on Plasma Science, vol. 30, pp. 808–​818, 2002.
[38] S. Alberti et al., ‘European high-​power CW gyrotron development for ECRH systems’,
Fusion Engineering and Design, vol. 53, pp. 387–​397, 2001.
[39] M. K. Thumm and W. Kasparek, ‘Passive high-​power microwave components’, IEEE
Transactions on Plasma Science, vol. 30, pp. 755–​786, 2002.
[40] K. T. Nguyen et  al., ‘Electron gun and collector design for 94-​GHz gyro-​amplifiers’,
IEEE Transactions on Plasma Science, vol. 26, pp. 799–​813, 1998.
[41] M. E. Read et al., ‘Depressed collectors for high-​power gyrotrons’, IEEE Transactions
on Electron Devices, vol. 37, pp. 1579–​1589, 1990.
[42] G. P. Saraph et  al., ‘Design of a single-​ stage depressed collector for high-​ power,
pulsed gyroklystron amplifiers’, IEEE Transactions on Electron Devices, vol. 45,
pp. 986–​990, 1998.
[43] R. L. Ives et  al., ‘Design of a multistage depressed collector system for 1 MW CW
gyrotrons. II. System consideration’, IEEE Transactions on Plasma Science, vol. 27,
pp. 503–​511, 1999.
[44] A. Singh et al., ‘Design of a multistage depressed collector system for 1-​MW CW gyro-
trons. I. Trajectory control of primary and secondary electrons in a two-​stage depressed
collector’, IEEE Transactions on Plasma Science, vol. 27, pp. 490–​502, 1999.
[45] M. E. Read et al., ‘Design of a 3-​MW 140-​GHz gyrotron with a coaxial cavity’, IEEE
Transactions on Plasma Science, vol. 24, pp. 586–​595, 1996.
[46] R. Advani et al., ‘Experimental investigation of a 140-​GHz coaxial gyrotron oscillator’,
IEEE Transactions on Plasma Science, vol. 29, pp. 943–​950, 2001.
[47] O. Dumbrajs and G. S. Nusinovich, ‘Coaxial gyrotrons:  past, present, and future
(review) ’, IEEE Transactions on Plasma Science, vol. 32, pp. 934–​946, 2004.
[48] V. L. Bratman et al., ‘Large-​orbit gyrotron operation in the terahertz frequency range’,
Physical Review Letters, vol. 102, p. 245101, 2009.
[49] V. L. Bratman et al., ‘Moderately relativistic high-​harmonic gyrotrons for millimeter/​
submillimeter wavelength band’, IEEE Transactions on Plasma Science, vol. 27,
pp. 456–​461, 1999.
[50] A. W. Fliflet et al., ‘Review of quasi-​optical gyrotron development’, Journal of Fusion
Energy, vol. 9, pp. 31–​58, 1990.
[51] B. Levush and W. M. Manheim, ‘Generation of high-​frequency radiation by quasi-​
optical gyrotron at harmonics of the cyclotron frequency’, IEEE Transactions on
Microwave Theory and Techniques, vol. 32, pp. 1398–​1401, 1984.
[52] R. G. Carter, ‘Synthesis of the fields of barrel open resonators’, Nuclear Instruments
and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and
Associated Equipment, vol. 657, pp. 1–​5, 2011.
[53] R. N. Clarke and C. B. Ronseberg, ‘Fabry-​Perot and open resonators at microwave
and millimetre wave frequencies, 2–​ 300 GHz’, J. Phys. E.:  Sci. Instrum., vol. 15,
pp. 9–​24, 1982.
[54] H. Kogelnik and T. Li, ‘Laser beams and resonators’, Applied Optics, vol. 5,
pp. 1550–​1567, 1966.
[55] V. Bratman et  al., ‘FEL’s with Bragg reflection resonators:  cyclotron autoresonance
masers versus ubitrons’, IEEE Journal of Quantum Electronics, vol. 19, pp. 282–​296, 1983.
[56] G. F. Brand and M. Gross, ‘A tunable source of linearly-​polarized, near-​millmeter wave
radiation’, International Journal of Infrared and Millimeter Waves, vol. 10, pp. 121–​136,
1989.
692 Fast-Wave Devices

[57] K. D. Hong et al., ‘Submillimeter wave generation by second harmonic operation of

tunable gyrotrons’, International Journal of Infrared and Millimeter Waves, vol. 13,
pp. 215–​227, 1992.
[58] K. Hong et  al., ‘A 150–​600 GHz step-​tunable gyrotron’, Journal of Applied Physics,
vol. 74, pp. 5250–​5258, 1993.
[59] T. Idehara et  al., ‘Development of a high-​ frequency, second-​ harmonic gyrotron
tunable up to 636 GHz’, Physics of Fluids B:  Plasma Physics (1989–​1993), vol. 5,
pp. 1377–​1379, 1993.
[60] A. C. Torrezan et al., ‘Operation of a continuously frequency-​tunable second-​harmonic
CW 330-​ GHz gyrotron for dynamic nuclear polarization’, IEEE Transactions on
Electron Devices, vol. 58, pp. 2777–​2783, 2011.
[61] S. Y. Park et al., ‘Experimental study of a Ka-​band gyrotron backward-​wave oscillator’,
IEEE Transactions on Plasma Science, vol. 18, pp. 321–​325, 1990.
[62] S. V. Samsonov et al., ‘Frequency-​tunable CW gyro-​BWO with a helically rippled oper-
ating waveguide’, IEEE Transactions on Plasma Science, vol. 32, pp. 884–​889, 2004.
[63] B.-​T. Liu et al., ‘Experimental study of a Ku-​band gyrotron backward-​wave oscillator
with a single stage depressed collector’, IEEE Transactions on Plasma Science, vol. 35,
pp. 1065–​1069, 2007.
[64] B. G. Danly et al., ‘Development and testing of a high-​average power, 94-​GHz gyroklys-
tron’, IEEE Transactions on Plasma Science, vol. 28, pp. 713–​726, 2000.
[65] M. Garven et al., ‘Experimental studies of a four-​cavity, 35 GHz gyroklystron ampli-
fier’, IEEE Transactions on Plasma Science, vol. 28, pp. 672–​680, 2000.
[66] B. G. Danly, ‘Gyro-​amplifiers for high power millimeter wave radar’, in Third IEEE
International Vacuum Electronics Conference, Monterey, CA, pp. 361–​362, 2002.
[67] A. Cross et al., ‘Helically corrugated waveguide gyrotron traveling wave amplifier using
a thermionic cathode electron gun’, Applied Physics Letters, vol. 90, p. 253501, 2007.
[68] V. Bratman et al., ‘High-​gain wide-​band gyrotron traveling wave amplifier with a helic-
ally corrugated waveguide’, Physical Review Letters, vol. 84, p. 2746, 2000.
[69] G. Denisov et al., ‘Gyrotron traveling wave amplifier with a helical interaction wave-
guide’, Physical Review Letters, vol. 81, p. 5680, 1998.
[70] S. V. Samsonov et al., ‘CW Ka-​band kilowatt-​level helical-​waveguide gyro-​TWT’, IEEE
Transactions on Electron Devices, vol. 59, pp. 2250–​2255, 2012.
[71] G. Dohler and D. Gallagher, ‘The small-​ signal theory of the cyclotron maser
and other gyrotron-​type devices’, IEEE Transactions on Electron Devices, vol. 35,
pp. 1730–​1745, 1988.
[72] G. Dohler et  al., ‘The peniotron:  fast wave device for efficient high power mm-​wave
generation’, in International Electron Devices Meeting, pp. 400–​403, 1978.
[73] G. Dohler and B. Wilson, ‘A small signal theory of the peniotron’, in International
Electron Devices Meeting, pp. 810–​813, 1980.
[74] G. Dohler et al., ‘Peniotron oscillator operating performance’, in International Electron
Devices Meeting, pp. 328–​331, 1981.
[75] G. Dohler et al., ‘Peniotron amplifier results’, in International Electron Devices Meeting,
pp. 845–​848, 1984.
[76] P. S. Rha et  al., ‘Self-​consistent large theory and simulation of high harmonic gyro-
tron and peniotron oscillators operating in a magnetron-​type cavity’, in International
Electron Devices Meeting, pp. 535–​538, 1985.
 References 693

[77] P. S. Rha et al., ‘Self-​consistent simulation of harmonic gyrotron and peniotron oscilla-
tors operating in a magnetron-​type cavity’, IEEE Transactions on Electron Devices, vol.
36, pp. 789–​801, 1989.
[78] H. Shimawaki et al., ‘2nd cyclotron harmonic peniotron experiments’, in International
Electron Devices Meeting, pp. 787–​790, 1991.
[79] G. Dohler et al., ‘Harmonic high power 95 GHz peniotron’, in International Electron
Devices Meeting, pp. 363–​366, 1993.
[80] T. Ishihara et  al., ‘Experiments of 10th cyclotron harmonic peniotron oscillator’, in
International Electron Devices Meeting, pp. 367–​370, 1993.
[81] A. K. Ganguly et  al., ‘Nonlinear theory of harmonic peniotron and gyrotron inter­
actions in a rising-​ sun slotted waveguide’, IEEE Transactions on Plasma Science,
vol. 22, pp. 902–​912, 1994.
[82] T. Ishihara et  al., ‘Space harmonic peniotron in a magnetron waveguide resonator’,
IEEE Transactions on Electron Devices, vol. 43, pp. 827–​833, 1996.
[83] T. Ishihara et al., ‘Highly efficient operation of space harmonic peniotron at cyclotron
high harmonics’, IEEE Transactions on Electron Devices, vol. 46, pp. 798–​802, 1999.
[84] D. B. McDermott et al., ‘Efficient Ka-​band second-​harmonic slotted peniotron’, IEEE
Transactions on Plasma Science, vol. 28, pp. 953–​958, 2000.
[85] R. M. Phillips, ‘History of the ubitron’, Nuclear Instruments and Methods in Physics
Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment,
vol. 272, pp. 1–​9, 1988.
[86] R. M. Phillips, ‘The ubitron, a high-​power traveling-​wave tube based on a periodic
beam interaction in unloaded waveguide’, IRE Transactions on Electron Devices, vol. 7,
pp. 231–​241, 1960.
[87] D. E. Pershing et al., ‘Amplifier performance of the NRL ubitron’, Nuclear Instruments
and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and
Associated Equipment, vol. 358, pp. 104–​107, 1995.
[88] H. Bluem et al., ‘Demonstration of a new free-​electron laser harmonic interaction’, in
International Electron Devices Meeting, pp. 791–​794, 1991.
[89] A. J. Balkcum et  al., ‘250-​MW X-​band TE01 ubitron using a coaxial PPM wiggler’,
IEEE Transactions on Plasma Science, vol. 24, pp. 802–​807, 1996.
[90] A. J. Balkcum et al., ‘High-​power coaxial ubitron oscillator: theory and design’, IEEE
Transactions on Plasma Science, vol. 26, pp. 548–​555, 1998.
[91] H. P. Freund et al., ‘Designs for W-​band free-​electron masers’, IEEE Transactions on
Plasma Science, vol. 27, pp. 243–​253, 1999.
[92] H. P. Bluem et  al., ‘A compact, high-​ power THz source’, in IEEE International
Conference on Plasma Science, p. 7B-7, 2012.
18 Emission and Breakdown
Phenomena

18.1 Introduction

This chapter reviews the variety of emission and breakdown phenomena which
occur, intentionally or unintentionally, in vacuum tubes. The underlying physics
of each process is summarised but more specialised texts (e.g. [1]) should be con-
sulted for the details. In order for electrons to be released from a solid surface into
a vacuum they must acquire sufficient energy to overcome a potential barrier. This
energy can be provided by heat, by the impact of electrons or ions, or by irradi-
ation by short-​wavelength light. The nature of the potential barrier depends upon
the material and also upon the external electric field. These processes are import-
ant for the cathode, which is the source of the electrons employed in a tube, and
also because of their role in initiating DC and RF voltage breakdown in vacuum.
Secondary electrons, liberated by the impact of primary electrons, affect the per-
formance of multi-​element depressed collectors (see Chapter 10). They also play
an important role in crossed-​field tubes (see Chapters 15 and 16), and in sustaining
multipactor discharges (Section 18.8). Outside a tube it is important to avoid break-
down across high voltage insulators, and in waveguides. Bombardment of solid
surfaces by high energy electrons leads to the production of X-​rays which can be a
hazard in the operation of high power tubes.

18.2 Emission of Electrons from Metal Surfaces

The discussion of the emission of electrons from metals in a vacuum begins with
a review of the free electron model of a metal [1]. The conduction electrons, which
are free to move within the metal, are considered to occupy a large number of
closely spaced, quantised energy states, within a potential well. According to the
Pauli Exclusion Principle no two electrons can have the same quantum numbers
(including spin). Thus, each state can contain at most two electrons with oppos-
ing spin. At the absolute zero of temperature (T = 0 K ) the electrons fill the lowest
possible states up to a maximum energy known as the Fermi Energy ( E F ). All the
states above the Fermi Energy are then empty. If the electron energy of the bottom
of the potential well is taken to be zero, then the Fermi Energy is of the order of a
 Emission of Electrons from Metal Surfaces 695

Figure 18.1:  Free electron model of a metal.

few electron volts for typical metals. The Fermi Energy lies somewhat below that
of stationary free electrons outside the metal ( E s ), as shown in Figure 18.1. This
energy difference, known as the Work Function ( φ ), depends upon the metal. It is
typically in the range 2 to 6 eV but is strongly dependent upon the condition of the
surface, including any contamination. Tables of work functions for metals can be
found in [2, 3].
In the free electron model the electrons are assumed to be trapped in a rect-
angular potential well, as shown by the dashed line in Figure 18.1. If an electron
escapes from the well, and its distance from the surface ( x ) is large compared with
the atomic spacing, it moves in the potential of a positive image charge. This poten-
tial, referred to the potential at the surface of the metal in the absence of emission,
is given by

e
V= . (18.1)
16 πε 0 x
The energy of the electron in this potential is shown by the solid curve in Figure 18.1.
At a distance of 1 µm (3000–​4000 atomic spacings) from the surface the potential
is 0.36 mV below the external potential. This model does not hold very close to the
surface where the atomic structure becomes important.
In general, the probability that a quantum state with energy E is occupied is given
by the Fermi-​Dirac distribution function

1
p (E ) = , (18.2)
 E − EF 
1 + exp 
 kT 

where T is the absolute temperature and k is Boltzmann’s constant

(k = 1.3806 × 10 −23 J K −1 ). At T = 0 the function defined by (18.2) has the property
that p ( E ) = 1 when E < E F and p ( E ) = 0 when E > E F so that all the energy states
below the Fermi energy are filled, and all those above it are empty, as required. The
density of states is of the order of 1023 cm−​3 so that, at normal temperatures, the
energy difference between adjacent states is small compared with thermal energies.
As the temperature rises, some of the electrons have sufficient energy to move from
696 Emission and Breakdown Phenomena

Figure 18.2:  Probability of occupation of states given by the Fermi-​Dirac distribution

function at 300 °K and 1300 °K when E F = 3.8 eV.

states just below the Fermi energy to states just above it. Figure  18.2 shows the
probability of occupation of states at 300 °K and 1300 °K for barium ( E F = 3.8 eV ).
If electrons are to escape from a metal, then the difference between their ener-
gies and the Fermi energy must exceed the work function. The energy required to
enable electrons to escape may be supplied thermally, by photons, or by electron or
ion bombardment. The height of the potential barrier can be reduced by a strong
external electric field. These processes are discussed in the sections which follow.

18.2.1 Thermionic Emission

We have seen that work functions are typically of the order of a few electron volts.
It is therefore possible to approximate (18.2) for energies of the order of E F + φ by

 E − EF 
p ( E ) = exp  −  , (18.3)
 kT 
because E − E F  kT . At 17 C for example kT = 0.025 eV. The probability that
an electron has just enough energy to escape from the metal is exp ( −φ kT ). For
example, the probability that an electron can escape from the surface of tungsten
(φ = 4.54 eV ) is 2 × 10 −76 at 300 °K but is 4 × 10 −17 at 1300 °K. Since the density
of conduction electrons is of the order of 1023 cm−​3 it can be seen that no emis-
sion is possible at 300 °K but that appreciable emission may occur at the higher
temperature.
The current density of thermionic emission is governed by the Richardson-​
Dushman equation [1]

J = A0T 2 exp ( −φ kT ) . (18.4)

 Emission of Electrons from Metal Surfaces 697

The theoretical value of the constant A0 is

4 π m0 ek 2
A0 = = 120 A cm −2 K −2 (18.5)
h3
where h is Planck’s constant ( h = 6.626 × 10 −34 J s ). The probability that an elec-
tron will be reflected at the surface is negligible because of the smooth variation of
potential outside the metal (see Figure 18.1). The work functions of metals depend
upon temperature, and this can be modelled by rewriting (18.4) as

J = A0T 2 exp − ( φ + αT ) kT , (18.6)

where α is the temperature coefficient of the work function. Taking the logarithm
of (18.6) we obtain

 J  α φ φ
ln  2  = ln A0 − − = ln AR − , (18.7)
T  k kT kT

so that a plot of ln ( J T 2 ) against 1 T is a straight line from which the constants AR

and φ can be determined. The current density measured is affected by a number of
factors:
• The apparent work function increases if there is negative space-​charge close to
the surface. This is the case with space-​charge limited emission when a virtual
cathode is formed (see Section 5.3).
• The apparent work function decreases with increasing external field strength (see
Section 18.2.2).
• The value of φ depends of the crystallographic plane of the emitting surface
• The work function is strongly affected by small quantities of adsorbed gases [4].
• The effective area of the emitting surface is not normally equal to the actual
surface area.
Thus considerable care must be exercised in making experimental measurements
of the properties of thermionic emitting surfaces. Nevertheless it is found that the
properties of nearly all surfaces can be represented by (18.4) when A0 is replaced
by values of AR determined by experiment [2, 5]. Practical thermionic cathodes are
discussed in Section 18.5.

18.2.2 Field-​Enhanced Emission (the Schottky Effect)

When the emitting surface forms the cathode of a plane-​parallel diode then the
potential outside the metal is given by

e x
V= + Va , (18.8)
16 πε 0 x d
698 Emission and Breakdown Phenomena

Figure 18.3:  Reduction in the effective work function by an external electric field (the
Schottky effect).

where d is the diode spacing and Va is the anode voltage relative to the cathode, as
shown by the solid line in Figure 18.3. The work function is then reduced by

eVa
∆φ = . (18.9)
4 πε 0 d

It should be noted that the change in the work function has been exaggerated in
Figure 18.3 for clarity. In practice the change is small. If the electric field is 1 MV m−​1,
for example, then ∆φ = 0.038 V. For field-​enhanced emission (18.7) becomes

 J  φ e eVa
ln  2  = ln AR − + . (18.10)
T  kT kT 4 πε 0 d

A plot of ln ( J ) against Va at constant temperature gives a straight line for voltages

greater than those for which the emission is space-​charge limited. Extrapolation of
the line back to Va = 0 gives the zero field current (the saturation current) at that
temperature. If the measurements are repeated at different temperatures then the
plot of (18.7) gives the constants AR and φ [6].

18.2.3 Field Emission

When the external electric field is greater than about 5 MV mm −1 it is possible for
electrons to pass through the potential barrier by quantum mechanical tunnelling
without the need for the surface to be heated (see Figure  18.4) [7, 8]. The pro-
cess of tunnelling is analogous to the attenuation of waves as they pass through a
short section of cut-​off waveguide. Field emission, typically, takes place at sharp
points, edges, and surface irregularities, where the electric field is enhanced by the
geometry. A special case which is important in vacuum tubes is triple point emis-
sion. This occurs at the intersection between a metallic surface, a dielectric surface
and vacuum where strong enhancement of the electric field is produced by the
geometry [9].
 Emission of Electrons from Metal Surfaces 699

Figure 18.4:  Cold field emission of electrons by tunnelling through the energy barrier.

The emitted current density is given by the Fowler-​Nordheim equation in the

form [8]

J = aV 2 exp ( −b V ) , (18.11)

where J is the current density in A cm−2, V is the potential difference between the
emitting surface and a distant anode, and a is defined by

β2  10.4 
a ≅ 1.5 × 10 −6 exp   , (18.12)
φ  φ

where φ is the work function in eV, b is defined by

3
φ2
b ≅ 6.44 × 107 . (18.13)
β
β is a geometric factor which depends on the electrode configuration, such that
the electric field over the emitting area is E = βV . For a given material and geometry
the parameters a and b are constants and a plot of ln ( J V 2 ) against 1 V (a Fowler-​
Nordheim plot) should be a straight line. It is found that this is generally the case
over many orders of magnitude of current. Changes in the operating conditions
(temperature, residual gas composition and pressure, and the current being drawn)
can produce large changes in the plot. These changes are usually due to changes in
the surface conditions and in the shape of the emitter. They may, or may not, be
reversible.

18.2.4 Photo-​Electric Emission

The energy of a photon is hc λ, where λ is its wavelength and h is Planck’s constant.
If hc λ > e φ then an electron may be ejected from the surface with kinetic energy
given by

1 hc
m0 u 2 = − e φ. (18.14)
2 λ
700 Emission and Breakdown Phenomena

Thus, for a given material, the maximum wavelength of light for which photo-​
electric emission is possible is determined by the work function [2, 10]. The energies
of photons in the visible region of the spectrum are in the range 1.6 to 3.3 eV. The
work functions of metals are mostly greater than 4 eV so it is necessary to irradiate
them with ultra-​violet light to obtain photo-​electric emission. The optical power
incident on the surface is

n p hc
Popt = , (18.15)
λ
where n p is the number of photons per second. The ratio of the number of electrons
emitted to the number of incident photons is known as the quantum efficiency ( ηq )
of the surface. The electron current is then

 λe 
I = n p ηq e = Popt ηq   . (18.16)
 hc 

The quantum efficiencies of clean metal surfaces are of the order of 10–​4 to 10–​5 but
values as high as 0.3 have been obtained from compound surfaces.
In order to get an electron current high enough to be useful in an electron tube it
is necessary to illuminate the surface with a high power laser. Since the light from
the laser can be in very short pulses it is possible to produce a bunched stream
of electrons. Photo-​cathodes are used as electron sources in particle accelerators
but are not yet suitable for use in microwave tubes. Unwanted photo-​electric emis-
sion can occur in a microwave tube as a result of the bombardment of surfaces by
X-​rays produced within the tube (see Section 18.4). Further information on photo-​
emission can be found in [10–​13].

18.3 Secondary Electron Emission

When electrons strike a solid surface in a vacuum they may be reflected elastically,
or penetrate the surface and exchange energy with the electrons in the material [1].
An incident electron entering a solid material makes a series of collisions losing
energy of about 30 eV each time. The electrons which gain energy through colli-
sions may make further collisions so that a cascade of secondary electrons moves
through the material as a result of the impact of a primary electron. One or more
secondary electrons may reach the surface and escape. In addition the primary elec-
tron may re-​emerge with reduced energy. Secondary electrons can also be generated
by ion bombardment [14]. Secondary electron emission is possible from metals, and
from insulating and semi-​conducting materials. The ways in which the phenom-
enon is important in vacuum tubes include:
• the influence on the performance of a magnetron of the relative proportions of
primary and secondary emission from the cathode (see Chapter 15);
 Secondary Electron Emission 701

• the use of secondary electron emission from cold cathodes in some crossed-​field
amplifiers (CFAs) (see Section 16.2.2);
• the effect of secondary electron emission on the performance of multi-​element
depressed collectors (see Section 10.4);
• voltage breakdown initiated by secondary emission from the surface of insulators
(see Section 18.7.3);
• resonant RF discharges (multipactor) are sustained by the emission of secondary
electrons (see Section 18.8).

Figure 18.5 shows the energy distribution of electrons emitted from a gold surface
for three different primary energies [15]. This behaviour is typical of that found
with other surfaces [16–​18]. For incident energies greater than 100 eV it is conveni-
ent to divide the electrons emitted into three categories:

• elastically reflected primaries;

• inelastically reflected (re-​diffused) primaries (energies > 50 eV);
• true secondary electrons (energies < 50 eV).

The energy of the boundary between the second and third groups is arbitrary, but
50 eV is commonly used in the literature [18–​20]. The total number of true second-
ary electrons emitted may exceed the number of incident primary electrons. For
primary energies greater than 100 eV the energy distribution of the true secondary
electrons is essentially independent of the primary energy [20]. It is characterised
by the most probable energy ( E sm ), and full width at half maximum (HW), of the
distribution, as shown in Figure 18.5. Typically, for metals, 1 eV < E sm < 5 eV and
3 eV < HW < 15 eV . The full width at half maximum is much smaller for insulators
than for metals, and it depends strongly on very thin surface layers. It is found that
the angular distribution of true secondary electrons from polycrystalline metals is
independent of the angle of incidence of the primary electron, to a good approxi-
mation, and that it is proportional to the cosine of the angle between the direction
of emission and the normal.
The fraction of elastically reflected electrons falls rapidly with increasing primary
energy, as shown, for oxidised beryllium, in Figure 18.6. However, at primary ener-
gies greater than 50 keV the fraction of all (elastic and inelastic) reflected primaries
starts to increase again [1]. Note that the absence of re-​diffused electrons below a
primary energy of 50 eV in Figure 18.6 is a consequence of the definition of sec-
ondary electrons. For primary energies greater than 100 eV the true secondaries are
the dominant group [15, 17, 21, 22]. The secondary electrons that are able to escape
from the surface are generated in a very thin surface layer which is of the order of
5 nm in metals, and 75 nm in insulators. Beyond that depth the secondary emission
no longer depends upon the properties of the bulk material [20].
The secondary electron emission coefficient (secondary electron yield) of a mater-
ial ( δ ) is the ratio of the number of secondary electrons emitted to the number of
incident primary electrons. For all materials it is found that the dependence of δ on
the energy of normally incident primary electrons can be represented approximately
702 Emission and Breakdown Phenomena

Figure 18.5:  Typical distributions of the energies of secondary and reflected primary

electrons from a gold surface for various primary electron energies
(copyright1996, Elsevier, reproduced, with permission, from [15]).

Figure 18.6:  Fraction of elastically-​reflected, rediffused and secondary electrons as a function

of primary electron energy for oxidised beryllium
(copyright 1994, IEEE, reproduced, with permission, from [17]).

by a universal curve having the shape shown in Figure 18.7 for θi = 0. This curve
is defined by the maximum value of the secondary emission coefficient ( δ m ) and
the primary electron energy at which it occurs ( E pm ). The value of E pm usually
lies between 200 and 1000 V [23, 24]. When the angle of incidence of the primary
electrons is not zero it is found that the coefficient of secondary electron emission
increases, and the peak of the curve moves to higher primary energies, as shown in
Figure 18.7. The effect of increasing the angle of incidence is that the primary elec-
trons stay closer to the surface for longer so that the secondary electrons generated
can escape more readily.
 Secondary Electron Emission 703

Figure 18.7:  Variation of the normalised secondary electron emission coefficient with

normalised primary electron energy for three different angles of incidence.

The secondary electron emission properties of materials depend upon the prep-
aration and conditioning of the surfaces, and upon any surface layers present [25,
26]. The preparation may include cleaning, coating, and texturing of the surface
before assembly. Conditioning may include heating, and exposure to ion, or elec-
tron, bombardment after evacuation of the tube. In addition to surface layers
added deliberately, the surface may be contaminated or oxidised. The contamin-
ation of the surface may occur from exposure to air before the tube is evacuated.
After evacuation there may be contamination from residual gases or evaporation of
materials such as barium (from the cathode). For this reason it is necessary to treat
the extensive data in the literature with some caution. For example, a comparison
of measured data for aluminium from 13 different sources showed consistent values
of the E pm (  400 eV ), but values of δ m ranging from 0.5 to 3.3 [23]. The first figure
may be the coefficient for a clean surface, while higher figures probably represent
the effects of various thicknesses of oxide layer. A study of the properties of tech-
nical copper found δ m ~ 2.3 for material as received from the supplier. This figure
was reduced to about 1.8 by baking the sample to 300 °C, and reduced to about 1.4
after exposure to an argon glow discharge [25]. The secondary electron emission
coefficients of insulators are usually greater than those of metals. However, the
measurement of secondary emission from insulators is made more complicated by
the build-​up of surface charge. Tables of the secondary electron emission properties
of pure materials are given in [20, 23, 24, 27], but it should be noted that the fig-
ures given show considerable differences. The properties of some compounds and
insulators are given in [20, 28]. The conditions under which the measurements of
secondary electron yield are made usually differ from those existing in a vacuum
tube. It is therefore difficult to have reliable knowledge of the properties of a surface
within a tube.
The secondary electron emission coefficient of a surface can be adjusted by the
choice of material, by the application of surface coatings, and by changing the
704 Emission and Breakdown Phenomena

texture of the surface. Where a high secondary electron yield is required, as for the
cathodes of some crossed-​field amplifiers, the surface of a metal may be deliber-
ately oxidised [17, 29]. In most other cases the aim is to reduce secondary emission
as far as possible. In particular, it is desirable to reduce the possibility of avalanche
multiplication of the number of electrons by ensuring that the range of primary
energies for which δ > 1 is as small as possible. This is often achieved by the addition
of surface coatings such as titanium nitride and diamond-​like carbon [30–​32]. An
alternative approach modifies the surface of the material so that the ability of sec-
ondary electrons to escape from it is reduced. The modification can take the form
of grooves in the surface which tend to trap the electrons emitted [33]. The emis-
sion of secondary electrons can also be reduced by adding microscopic textures to
surfaces [34–​36]. Further details of these techniques, and their applications, can be
found by searching for papers citing those referenced here.

18.3.1 Modelling Secondary Electron Emission

Models of secondary electron emission are needed to represent its effects in mag-
netrons, in multi-​element depressed collectors, and in multipactor discharges (see
Section 18.8). Theoretical studies of secondary electron emission give values of
δ m which bear some resemblance to measured values [37]. But the large scatter in
the experimental data available makes the validation of the theory uncertain. The
strong dependence of δ m on the nature of the emitting surface means that measured
values should be used in modelling, if possible. We have noted that E pm is subject to
smaller variations so that typical values can be obtained from the literature.
The shape of the curve for normal incidence in Figure  18.7 is much the same
for all materials and empirical functions have been suggested by many authors [1,
15, 18, 20, 38, 39]. One approach is based on simple models of the penetration of
the primary electrons into the material, and the subsequent escape of the second-
ary electrons generated. The results, for normal incidence, can be expressed in the
form [24]

1− k
   E p   
k
δ 1  Ep  
= 
1 − exp − xm     , (18.17)
δ m 1 − e − xm  E pm    E   
  pm

where E p is the energy of the primary electrons, k is adjusted to fit the curve to
experimental data and xm is the solution of

 1
xm = 1 −  (e xm − 1) . (18.18)
 k
Equation (18.17) is identical to those given by other authors who assumed fixed
values for k in the range 1.35 to 1.67 [23, 40–​43]. However, the fit to experimen-
tal data can be considerably improved by using k as an adjustable parameter. The
curves generated by (18.17) in this way give an excellent fit to the best available
 Secondary Electron Emission 705

Figure 18.8:  Normalised secondary electron emission curves for technical copper compared
with the curves calculated using equations (18.17) and (18.19).

experimental data for 18 different elements from beryllium to lead [24]. The values
of k ranged from 1.25 to 1.69. The mean absolute relative error of the fit to experi-
mental data was better than 5% in nearly every case. It seems, therefore, that the
shape of the curve is not truly universal, as was assumed in earlier work.
An alternative, simpler, equation adopted by Furman and Pivi that appears to
give a better fit to data for technical materials is [18]

δ s ( E p E pm )
= , (18.19)
δ m s − 1 + ( E p E pm )s

where s is a parameter which is chosen to give the best fit to experimental data. The
value of this parameter was 1.54 for copper, and 1.81 for stainless steel. Figure 18.8
shows a comparison between experimental results for technical copper taken from
[25] and curves calculated from (18.17) with k = 1.4, and (18.19) with s = 1.54.
When the angle of incidence of the primary electrons is not zero it is found that
the secondary electron emission curve can be derived from the curve for normal
incidence. The maximum value of δ , and the primary energy at which it occurs, can
be expressed in terms of empirical functions of the angle of incidence. Furman and
Pivi [18] proposed the relations

δ m (θi ) = δ m ( 0 ) 1 + 0.66 (1 − cos 0.8 θi ) (18.20)

and

E pm (θi ) = E pm ( 0 ) 1 + 0.7 (1 − cos θi ) . (18.21)

These functions, which apply to the unpolished, polycrystalline surfaces commonly

found in vacuum tubes, are substituted into (18.19) to give the secondary electron
706 Emission and Breakdown Phenomena

emission coefficient as a function of the energy of the primary electrons and the
angle of incidence

δ s ( E p E pm (θi ))
= . (18.22)
δ m (θi ) s − 1 + ( E p E pm (θi ))s

Alternative expressions, given by Vaughan [44], are

 θ2
δ m (θi ) = δ m ( 0 ) 1 + ksδ i  (18.23)
 2π 
and

 θ2
E pm (θi ) = E pm ( 0 ) 1 + ksV i  , (18.24)
 2π 
where ksδ and ksV are constants which depend upon the surface roughness. The
default value of both constants is unity for normal machined surfaces. The range
is from zero, for deliberately roughened surfaces, such as textured carbon, up to 2
for exceptionally smooth, clean, surfaces [44, 45]. The curves given by these two
equations are close to those from (18.20) and (18.21) if the constants are set to 1.6
and 1.9, respectively. The curves in Figure 18.7 were computed using (18.19) and
(18.22). The effect of surface layers was studied by Yu et al. who were able to show
agreement with experimental results for layers of silicon dioxide of different thick-
nesses on silicon substrates [46]. In any particular case the model used should, if
possible, be validated by reference to experimental measurements of the properties
of the surface. It is also desirable that it should correctly model all three groups of
electrons leaving the surface in both energy and angular distribution [18, 19].

18.4 X-​ray Emission

X-​rays are generated whenever electrons with energies greater than about 5 keV
strike solid surfaces. Tubes which are designed to generate X-​rays are outside the
scope of this book, but X-​ray generation is of more general importance because of
its implications for safety [47]. The rapid deceleration of the electrons as they enter
the material generates a broad spectrum of brehmsstrahlung (brake radiation) as
shown in Figure 18.9. The spectrum has a sharp cut-​off at the photon energy (eVp )
equal to that of the incident electrons. In addition there are spectral lines, arising
from transitions between the electron energy levels of atoms within the material.
The wavelength of X-​rays is related to the photon energy (Vp ) by

hc
λ= . (18.25)
eVp
 X-ray Emission 707

Figure 18.9:  X-​ray emission spectrum for a silver target with 40 keV electrons
(copyright 2017, AMPTEK Inc., reproduced with permission, from [48]).

where h is Planck’s constant. The fraction of the power carried by the electrons that
is converted into X-​rays is given approximately by

F = 7 × 10 −4 ZVa , (18.26)

where Z is the atomic number of the target and Va is the accelerating voltage. The
remainder of the power is dissipated as heat. Thus the X-​ray intensity increases
with increasing electron energy, and with increasing current striking the surface.
When X-​rays with energies less than 1 MeV pass through materials they experi-
ence coherent, and incoherent, scattering and photo-​electric absorption. Additional
loss mechanisms apply at higher energies. The intensity of the radiation ( I ) as a
function of depth into the material is given by

I = I 0 exp ( −µx ) , (18.27)

where I 0 is the initial intensity, x is the thickness of the material, and µ is the attenu-
ation coefficient. Equation (18.27) applies to the attenuation of monochromatic
X-​rays incident normally on a plane sheet of material. The attenuation coefficient
increases rapidly with atomic number. It decreases with increasing photon energy,
as shown in Figure 18.10, which is based on data from [49]. The absorption edges, at
which the value of µ is discontinuous, correspond to the lines in the spectra of the
elements. It can be seen that, at a photon energy of 50 keV, the thickness of copper
required to produce a given attenuation is about four times that of lead. Similarly
the thickness of aluminium must be 24 times that of copper. Equation (18.27) can
also be written

  µ 
I = I 0 exp  −   (ρx ) , (18.28)
  ρ 
708 Emission and Breakdown Phenomena

Figure 18.10:  Comparison between the X-​ray attenuation coefficients of lead, copper, and
aluminium.

where ρ is the density of the material. The parameter ( µ ρ) is the mass attenuation
coefficient and (ρx ) is the mass per unit area of the sheet of material. This form is
more convenient for calculations of X-​ray shielding. Tables of the mass attenuation
coefficient as a function of photon energy for many materials are given in [49]. It
should be noted that in practical cases, it is necessary to consider the spectrum of
the X-​rays, and the geometry of the material, when computing the attenuation [50].
The implications of X-​ray generation for the safe operation of tubes are discussed
in Section 20.8.

18.5 Thermionic Cathodes

All the tubes discussed in this book require sources of electrons for their oper-
ation. In the great majority of cases this source is a thermionic cathode. The
cathode is a critical component because failure of emission is one of the chief
causes of tube failure. According to (18.6) the current density depends upon the
temperature of the cathode and the work function of the surface. The active
material of the surface of the cathode is gradually evaporated at a rate which
increases with temperature. Thus there is a trade-​off between the current density
and the life of the cathode. The normal end of life of a tube is determined by the
failure of the cathode emission, which is commonly taken to be a reduction of
10% in the current density [51]. The temperature of a cathode is normally meas-
ured by optical pyrometry in which the brightness of the surface is compared
with that of the filament of a standardised lamp. The temperature measured is
the brightness temperature. This is the temperature at which the radiation from
a black body is as bright as the cathode. The brightness temperature is always
lower than the true temperature which must be obtained by applying a conver-
sion formula [6].
 Thermionic Cathodes 709

The current which can be obtained from a cathode surface is reduced by:

• Bombardment by positive ions generated by ionisation of the residual gas in

the tube. Material is removed from the surface of the cathode by sputtering
leading to an increase in the work function. This problem arises if the residual
gas pressure in a tube is too high. It is more likely to occur in continuously
operated tubes, where the ions accumulate in the potential well caused by the
space-​charge in the beam. In pulsed tubes there is less time for the ions to accu-
mulate, and they are dispersed by their space-​charge in the intervals between
the pulses.
• Absorption of oxidising gases such as oxygen, carbon dioxide, and water vapour
that poison the cathode and increase its work function.

The emitted current density from a damaged cathode can sometimes be restored
if the source of damage is removed, but the final current density does not always
return to the previous level [4, 52, 53].
The life expectancy of the cathode is a major factor in tube design, and manufac-
turers are understandably wary of the premature introduction of new technologies
until they have been subjected to intensive life testing. An introduction to the main
types of thermionic cathode is given below. These have been established for many
years but they are still the subject of research and of an extensive literature. Further
information about thermionic cathodes can be found in [2, 6, 13, 51, 54–​56].

18.5.1 Metal Emitters

Pure metals have relatively high work functions and they must be operated at high
temperatures to give useful emission. The metal, in the form of a wire, is heated by
the passage of an electric current through it. The cathode is, typically, constructed
in the form of a cage, or a helix. The choice of metals is limited to those which have
high melting points of which tungsten ( φ = 4.54 eV, MP = 3640 K ) is the one most
commonly used. The high operating temperature means that surface contamina-
tions are rapidly evaporated and a tungsten emitter can operate in a poor vacuum,
such as that in an unbaked demountable system. The current density is typically
0.4 A cm−​2 at a temperature around 2,500 °K. Over time the wire becomes thinner
because of evaporation of the tungsten. The life of a tungsten emitter is assumed
to end when 10% of the material has evaporated because hot spots then tend to
form [57].
The addition of a monolayer of thorium to a tungsten wire reduces the work
function to 2.7 eV. Thus, a directly heated thoriated tungsten wire can deliver higher
current densities, at a lower temperature, than pure tungsten. At a typical operat-
ing temperature of 2000 °K such an emitter can deliver up to 4 A cm−​2 with a life
in excess of 10,000 hours. The thorium layer is readily evaporated if the wire tem-
perature is too high, leading to premature failure of emission. It has been found
that this problem can be reduced, to some extent, by carburising the top layer of
710 Emission and Breakdown Phenomena

the tungsten to form tungsten carbide [2, 57]. The tungsten is doped with small
grains of thorium oxide, which react with the tungsten carbide to replenish the free
thorium on the surface of the filament. This type of emitter is used in triodes and
tetrodes and in magnetrons for industrial heating and microwave ovens [58].

18.5.2 Oxide Cathodes

Cathodes having lower work functions than metallic emitters and, therefore,
lower operating temperatures can be made using oxides of certain elements.
Oxide cathodes are made by painting, or spraying, a coating onto a metal base.
The coating is usually a mixture of the carbonates of barium, strontium, and cal-
cium with a molar ratio (50–​58%):(38–​45%):(4–​8%) [2, 59]. The surface is acti-
vated during the evacuation of the tube by heating it to convert the carbonates to
oxides, giving a surface whose work function is between 1.2 and 1.8 eV. The DC
current density of a conventional oxide cathode, with a life in excess of 10,000 h,
is up to 1 A cm−2 but higher currents can be obtained for short-​pulse, low duty
cycle operation [13]. The DC current density is limited by the high resistivity of
the oxide layer. Oxide cathodes are easily poisoned and must be kept in high vac-
uum conditions once they have been activated. Problems may also be caused by
poor adhesion of the oxide layer to the metal base. Because oxide cathodes are
simple and inexpensive to manufacture they are a continued subject of research
[2, 13, 59].

18.5.3 Dispenser Cathodes

The cathodes generally used in microwave tubes are dispenser cathodes in which
the active materials are held within a porous metal matrix [2, 6]. In these cathodes
the active material stored in the matrix can diffuse to the surface and replace that
lost by evaporation. Dispenser cathodes have work functions of 1.8–​2.1 eV but
do not suffer from the problem of high resistivity which limits the usefulness of
oxide cathodes. Many different types of dispenser cathode have been invented and
details can be found in the literature. The three main arrangements are shown in
Figure 18.11. The essential component in all three is a porous metal matrix which
is commonly made by sintering tungsten powder. The simplest arrangement is the
impregnated cathode in Figure 18.11(a) in which the matrix is impregnated with a
BaO:CaO:Al2O3 mixture in proportions such as 5:3:2 (the Phillips B cathode), or
4:1:1 (the S cathode). The life of the cathode can be greatly increased by adding a
reservoir of BaO behind the tungsten plug to form the reservoir cathode shown in
Figure 18.11(b). This arrangement is especially useful for space tubes where very
long life is essential. In the third arrangement a surface layer having a thickness of
about 0.5 µm is added to the tungsten plug as shown in Figure 18.12(c). This layer,
which is commonly an osmium–​ruthenium alloy, reduces the work function of the
surface allowing operation at a lower temperature and thus extending the cathode
life (the M cathode).
 Thermionic Cathodes 711

Figure 18.11:  Arrangements of dispenser cathodes: (a) impregnated cathode, (b) reservoir

cathode, and (c) impregnated cathode with a thin surface film.

The way in which the saturated current available from a dispenser cathode
changes with life can be modelled by supposing that it depends upon the relative
proportions of the surface in which the emission is limited by space-​charge, and
by temperature (see (5.9)). Where there is space-​charge limited emission, a low
work function is maintained by a thin layer of the active material on the surface
of the matrix. In an impregnated cathode this layer can be thought of as a balance
between the rate of evolution of active material from the matrix and the rate of
evaporation. It has been shown that this model gives a good fit to experimental
data [60]. The balance is strongly affected by the geometry of the diode in which
the measurements are made. Lifetime measurements made using a closely spaced
parallel diode are not representative of the lifetime achieved when the same cathode
is incorporated in an electron gun. This is because evaporated barium is refluxed
to the cathode in a closely spaced parallel diode so that the active layer is main-
tained [61]. In an M-​type cathode the diffusion of tungsten, from the matrix, into
the surface layer changes the properties of the layer over time. With this addition,
the model is a good fit to experimental data. An interesting feature of the model is
that it predicts that at temperatures exceeding 1400 °K the saturated current falls
with increasing temperature. This is because the rate of evaporation exceeds that at
which the surface layer can be replenished from the matrix. The shape of the curve
showing the transition between space-​charge limited and temperature limited emis-
sion depends upon the parameter α in (5.9). This parameter, which is related to the
uniformity of the emitting surface, decreases with time so that knee of the curve in
Figure 5.2 becomes more rounded. The rate at which α decreases is increased as the
temperature of the cathode is raised [61].
Dispenser cathodes can deliver DC current densities up to 4 A cm−2 at around
1300 °K with lifetimes of over 100,000 hours. The addition of Sc2O3 to the barium
mixture in a dispenser cathode has been found to decrease the work function and
increase the current density (the scandate cathode). Although these cathodes have
delivered current densities of 60 A cm−2 at around 1250 °K it has been found that
they are sensitive to ion bombardment, from which they do not readily recover [52].
712 Emission and Breakdown Phenomena

Because the thin surface layer in Figure 18.12(c) is also readily eroded by ion bom-
bardment, cathodes have been developed in which the coating material is added to
the tungsten matrix. The result is a mixed-​matrix (MM) cathode [62]. Fuller infor-
mation about all these types of cathode, including details of their manufacture, can
be found in the references cited.

18.6 Field Emission Cathodes

A thermionic cathode requires a heater which has a finite warm-​up time and adds
to the power consumption of a tube. Hence there is continuing interest in the possi-
bility of using cold cathodes employing field emission. The most successful embodi-
ment of this concept is the Spindt cathode shown in Figure 18.12 [13, 63, 64]. This
cathode comprises an array of field emitter tips, made using micro-​electronic tech-
nology, arranged in holes in a gate electrode. The emitting tips are made from metals
with high melting points, such as tungsten and molybdenum, from semiconductors,
or from carbon nanotubes [13]. The presence of a gate electrode allows the electron
beam to be pulsed, and gives the potential for modulation of the beam at micro-
wave frequencies. This has been demonstrated in experimental tubes, but the cath-
ode lifetime is short [65–​67]. In one example a cathode with 50,000 tips delivered a
current of 120 mA for several days at a current density of 13 A cm −2. The reliability
is limited by local overheating and voltage breakdown leading to short-​circuiting
of the structure. Research into improved reliability is continuing [68, 69]. This type

Figure 18.12:  The Spindt cathode

(copyright 1991, IEEE, reproduced, with permission, from [63]).
 Voltage Breakdown 713

of cathode also lends itself to incorporation in vacuum micro-​electronic devices [8,

64]. For further information on this and other types of cold cathode see [12, 13].

18.7 Voltage Breakdown

Voltage breakdown sets limits to the design of tubes. It may take place either inside,
or outside, the vacuum envelope, or through insulators, with both DC and RF
voltages. The physical processes involved are discussed below. A  voltage break-
down dissipates stored energy which may cause damage to the tube, and alter its
performance. In extreme cases the damage may be so severe that the tube can no
longer operate. For this reason high power tubes are provided with protection cir-
cuits which are designed to turn the tube off when a breakdown takes place (see
Chapter 20). Whenever a tube ‘trips out’ it ceases operation for the time taken to
restore normal operating conditions. Thus the objective of the designer is to elimin-
ate voltage breakdown as far as possible to increase the reliability of a tube.

18.7.1 Voltage Breakdown in Vacuum

The primary cause of voltage breakdown between metal electrodes in vacuum is
field emission from whiskers and from microscopic dielectric inclusions in the sur-
faces. The latter are sites for triple point emission as described in Section 18.2.3.
However, it is found that breakdown occurs at lower voltages than can be explained
by field emission alone. The additional emission is believed to be in the form of sec-
ondary electrons produced by ion bombardment. If it is assumed that the second-
ary electron emission is linearly proportional to the ion energy then the threshold
for breakdown is given by

 K 
WE 2 exp  − 1  ≤ K 2 , (18.29)
 E
where W is the maximum possible ion energy, E is the electric field on the surface
of the negative electrode and K1 = 17 MV m −1 and K 2 = 1.8 MV3 m −2 are empirical
constants. This is Kilpatrick’s criterion, which is commonly used as a guide to DC,
pulsed DC, and RF vacuum breakdown [70, 71].
For DC voltages W is equal to the potential difference between the electrodes.
Thus, for example, the breakdown voltage between plane electrodes with a spacing
of 10 mm given by (18.29) is 100 kV. When the electrodes are not flat the electric
field on the surface of the negative electrode may be increased and the breakdown
voltage reduced. The effect of departures of the field from uniformity can be rep-
resented by field enhancement factors defined as the ratio of the maximum electric
field to that between plane electrodes with the same spacing and potential diffe-
rence [72]. The field enhancement factor may be as high as 10 for typical geometries,
but figures up to 100 occur for tiny protuberances on the surfaces. The breakdown
714 Emission and Breakdown Phenomena

voltage also depends upon surface finish, the metal from which the electrodes are
made, and on any contamination present. It is usual for a high-​voltage system to
be conditioned by gradually raising the voltage and allowing breakdown events to
remove the sources of emission. The criterion in (18.29) applies when the voltage is
pulsed with a pulse length is 1 ms or more. For shorter pulses the breakdown volt-
age increases by a factor of 3 when the pulse length is reduced to 1 μs [73]. When
the ion energy is greater than 100 keV there is a marked decline in secondary elec-
tron yield and the breakdown voltage is greater than that predicted by (18.29) [71].
For an RF voltage (V1 ) the maximum ion energy (W ) in (18.29) is reduced by
transit time effects. A graph for the ion transit time correction for higher voltages
is given in [70] as a plot of W V1 against V1 V * where the normalising voltage is
defined by

2
 2π g   M 0 c 2 
V* =  , (18.30)
 λ   πe 

where g is the gap between plane electrodes, λ is the free space wavelength and M 0
is the rest mass of a hydrogen atom. If V1 V * ≤ 1 the correction can be expressed as

W V
 0.63 1* . (18.31)
V1 V

Substituting for W in (18.29) and rearranging gives

f (MHz ) = 1.63E 2 exp ( − 8.5 E ) , (18.32)

where the electric field E is in MV m−1 [71]. The maximum electric field is found
to be proportional to the square root of the frequency up to about 1 GHz [74].
At higher frequencies the increase is more gradual. In the range 20 GHz to 40
GHz the maximum attainable surface field has been found to be independent of
frequency [75].
The breakdown process described above may lead to the generation of a full vac-
uum arc [72, 76, 77]. An arc may be initiated in various ways including the explosion
of a microscopic protrusion on the negative electrode, and the presence of loose
particles within the gap. The material of the negative electrode is vaporised and ion-
ised so that a plasma cloud is formed, which provides a conducting path between
the electrodes. On the negative electrode the arc terminates at one, or more, small
spots a few micrometres in diameter. The intense local heating causes thermionic
emission of electrons, liberates adsorbed gases, and releases metal vapour, all of
which sustain the plasma in the arc. At arc currents up to a few hundred amps the
anode spot is the passive collector of the cathode plasma jet, and is therefore fixed.
The cathode spots, however, move rapidly over the surface producing a char-
acteristic pattern of erosion. The arc voltage is usually 10–​30 volts and it varies
little with the current. The extent to which material is removed from the negative
electrode by the arc depends upon the current, the duration of the arc, and the
 Voltage Breakdown 715

material from which the electrode is made. The erosion is generally greater in mate-
rials having low boiling points and low thermal conductivity. Tube manufacturers
sometimes specify the maximum permissible arc energies for their products beyond
which irreversible damage may be caused.
A DC arc commonly has a lifetime of between 1 msec. and 1 sec. after which it
extinguishes spontaneously. The lifetime depends upon the material of the cath-
ode and increases rapidly with the current. For example, the duration of an arc on
copper is about 1 ms at a current of 5 A but becomes less than 1 µs if the current is
reduced below about 3 A [76]. If the current in the arc is reduced by the action of
protection circuitry (see Chapter 20) then it will extinguish very fast. The high rate
of change of current with time is sufficient to produce dangerously high voltages on
the wires connected to the tube. It is therefore very important to ensure that their
self-​inductances are as low as possible. This applies especially to the connection
between the tube body and ground.

18.7.2 Voltage Breakdown in Gases

DC voltage breakdown in gases depends upon the type of gas, and its pressure, and
also upon the shapes and surface condition of the electrodes [11, 78–​80]. The pro-
cesses are complex and only a brief introduction is given here. Figure 18.13 shows
a typical graph of current against the voltage applied between a pair of electrodes
surrounded by a gas. This graph illustrates the properties of the Townsend dis-
charge [81]. Some of the gas molecules are ionised by cosmic radiation and radio-
activity in the ground. When a voltage is applied to the electrodes the free electrons
and ions can drift towards them so that a small current flows. The energy gained
by the electrons is not sufficient to cause further ionisation of the gas. The cur-
rent saturates when the formation of new electron-​ion pairs balances the rate at
which they are removed by the applied field (region T0 in Figure 18.13). When the
voltage is increased, the electrons can gain enough energy from the electric field
to produce some additional ionisation and the current starts to rise (region T1).
The positive ions drift towards the cathode. A further increase in the voltage gives
the positive ions sufficient energy to liberate secondary electrons from the negative
electrode (cathode). The electron current reaching the positive electrode (anode)
then increases more rapidly (region T2). The electric discharge through the gas in
these three regions relies on an external source of electrons, and the flow of current
ceases if that source is removed. As the current increases, a point is reached where
the generation of secondary electrons, by ions striking the cathode, is sufficient to
sustain the flow of current without any external source of electrons. At the point
of transition from a non-​self sustaining discharge to a self-​sustaining discharge the
voltage reaches a peak. This is the breakdown (or sparking) voltage. Once electrical
breakdown has occurred the current is determined by the impedance of the exter-
nal circuit.
It has been found that experimental values of the breakdown voltage, for a
given gas and arrangement of electrodes, depend only upon the product of the
716 Emission and Breakdown Phenomena

Figure 18.13:  Current –​voltage characteristics of the Townsend discharge

(copyright 1941, Dover Publications, reproduced, with permission, from [78]).

Figure 18.14:  The Paschen curves for air, nitrogen (N2 ) and sulphur hexafluoride (SF6)
(copyright 1982, IEEE, reproduced, with permission, from [82]).

gas pressure and the separation of the electrodes. This result is known as Paschen’s
Law. Figure 18.14 shows measured Paschen curves for air, nitrogen ( N 2 ) and sul-
phur hexafluoride (SF6 ) for plane electrodes [82]. The results for plane electrodes
are a useful guide to other cases, if the field enhancement factor is not too great. For
example, the curve for air in Figure 18.14 gives a breakdown voltage of 30 kV for
a 1 cm gap at standard pressure (100 kPa). This is close to that commonly quoted
for breakdown between spheres in dry air. The breakdown voltage of air is reduced
by humidity, and by the presence of dust particles, and it is good practice to use
a safety factor of three for design purposes. The DC breakdown voltage may be
used as an estimate of the RF breakdown voltage to calculate the maximum power
which can be passed down coaxial lines and waveguides (see Section 2.3).
For typical electrode spacings the minimum of the Paschen curve normally
occurs well below atmospheric pressure. In the region of the minimum the break-
down voltage is particularly affected by the material of the electrodes. To the right
of the minimum the breakdown voltage of a fixed gap increases with pressure. This
 Voltage Breakdown 717

result is sometimes used in high voltage systems by pressurising them to reduce the
risk of voltage breakdown. High power waveguide systems may also be pressurised
for the same reason. The breakdown voltage can also be increased by using a gas
such as SF6 which readily attaches electrons to form negative ions, so reducing the
build-​up of the discharge. This gas is inert and non-​toxic but it breaks down into
components which are corrosive and highly toxic when subjected to sparking. It is
also expensive and it has been found that results which are nearly as good can be
obtained by adding a small fraction of SF6 to N 2 [79]. Moreover, it is a greenhouse
gas that is extremely long-​lived in the atmosphere and its use is therefore subject to
restrictions. It should be noted that the breakdown voltage decreases as the pressure
is reduced, and this is important in the design of systems for use at high altitudes.
For further information on voltage breakdown in gases see [78, 83, 84].
To develop an expression for the Paschen curve we consider a pair of plane par-
allel electrodes with separation d, and potential difference V between them, sur-
rounded by a uniform gas at pressure P. Let us suppose that electrons are liberated
from the surface of the negative electrode (cathode) at a steady rate n0 per second.
The actual cause of emission (themionic emission, photo-​electric emission etc.) is
not important to the argument. These electrons are accelerated towards the anode
and collide with the gas molecules. Some of these collisions will result in positive
ionisation of the molecules so that the number of electrons moving through the gas
increases with distance towards the anode. If the rate of electron flow is n at a plane
distant x from the cathode then the number of additional electrons produced in a
distance dx is proportional to n so that

dn
= α n, (18.33)
dx
where α is the fraction of electrons which make ionising collisions in unit distance.
Integrating (18.33) with respect to x from the cathode to the anode gives the num-
ber of electrons per second reaching the anode as

na = n0 e αd . (18.34)

Thus the current reaching the anode is

I a = I 0 e αd , (18.35)

where I 0 is the initial electron current leaving the cathode. The positive ions formed
by the collisions drift towards the cathode under the influence of the electric field.
In the steady state the total current must be the same at the cathode and the anode.
Thus the number of positive ions striking the cathode per second must be n0 (e αd − 1).
The impact of the ions on the cathode may liberate secondary electrons, which are
added to the initial electron current. Thus, the total electron current leaving the
cathode is

I c = I 0 + γ I c (e αd − 1) , (18.36)
718 Emission and Breakdown Phenomena

where γ is the secondary electron emission coefficient for impact by the positive
ions. The dependence of γ on the energy of the incident ions is neglected. The elec-
tron current leaving the cathode is increased by ionisation so that the anode current
becomes

I 0 e αd
I a = I c e αd = . (18.37)
1 − γ (e αd − 1)

The parameters α and γ are the first and third Townsend ionisation coef-
ficients. The second coefficient (β ) , representing ionisation caused by the
impact of ions on neutral molecules, is effectively zero for discharges of
interest here.
Now, let us consider the effect of the collisions between electrons and gas mol-
ecules on the energies of the electrons. For the moment it will be assumed that that
these collisions do not result in ionisation, and that an electron loses all its energy as
a result of a collision. If the fraction of electrons making a collision in unit length is
1 λ then the number of electrons which have yet to make a collision decreases with
position so that

dn n
= − (18.38)
dx λ

and therefore

n = n0 e − x λ . (18.39)

The mean distance travelled by an electron before making a collision is

1 ⌠ ∞ nx
x=  dx. (18.40)
n0 ⌡0 λ

Substituting for n from (18.39) and performing the integration we find that x = λ .
Thus λ is the mean free path of the electrons.
If an electron starts from rest, and moves in a uniform electric field E x , then it has
sufficient energy to ionise a molecule if

E x x ≥ Vi or x ≥ Vi E x , (18.41)

where Vi is the ionisation potential of the molecule. Thus, electrons which have trav-
elled less than this distance do not have sufficient energy to cause ionisation. From
(18.39) the fraction of the electrons which are able to produce ionisation is

n  V 
= exp  − i  . (18.42)
n0  λEx 
 Voltage Breakdown 719

The fraction of electrons per unit length which cause ionisation ( α ) is then the
product of the fraction which make collisions in unit length (1 λ ) and the fraction
which have sufficient energy to cause ionisation from (18.42). Hence

1  V 
α= exp  − i  . (18.43)
λ  λEx 
Now the mean free path is inversely proportional to the density of the gas and,
therefore, inversely proportional to the pressure. Thus we can write (18.43) as

 BP 
α = AP exp  − (18.44)
 E x 
where P is the gas pressure and A and B = AVi are constants for the gas. This equa-
tion gives results which agree with measurements for values of E x P from 450 to
7500 V kPa−1 cm−1 [82].
The condition for electric breakdown is found by setting the denominator in
(18.37) to zero so that

 1
αd = ln 1 +  . (18.45)
 γ
We note that this condition does not depend upon the initial current I 0 . Also, from
(18.44),
BP
Ex = . (18.46)
ln ( AP α )
Setting E x = Vb d , where Vb is the breakdown voltage, and substituting for α from
(18.45) we find that the breakdown voltage is given by
BPd
Vb = . (18.47)
ln  APd ln (1 + 1 γ )
In this equation the constants A and B depend only on the properties of the gas
while γ depends also on the properties of the cathode surface. We note that, if all
these are constant, then the breakdown voltage depends only on the product Pd, in
conformance with Paschen’s Law. The graph of Vb against Pd has a minimum when
e
Pd min = ln (1 + 1 γ ) (18.48)
A
and
eB
Vmin = ln (1 + 1 γ ) . (18.49)
A
Then (18.47) can be expressed in normalised form as

Vb Pd ( Pd )min
= . (18.50)
Vmin ln  Pd ( Pd )min  +1
720 Emission and Breakdown Phenomena

Figure 18.15:  Theoretical Paschen curve.

Figure 18.15 shows the curve defined by (18.50). This curve has the same general
form as the curves in Figure 18.14, but the prediction that all cases can be repre-
sented by the same normalised curve is not confirmed. This is a consequence of the
simplifying assumptions which were made in the derivation of (18.50). However,
the derivation does give insight into the physics of voltage breakdown. To the right
of the minimum of the Paschen curve the breakdown voltage increases because the
mean free path decreases, and fewer electrons gain sufficient energy to cause ion-
isation. To the left of the minimum the breakdown voltage increases because the
number of molecules available for ionisation decreases. It should be noted that the
equations derived above do not agree well with experimental results for breakdown
in narrow gaps. This is because the electric field can be high enough to cause field
emission, which has been ignored in the derivation.

18.7.3 Voltage Breakdown on Insulators

Any vacuum tube has one or more insulators which form part of the vacuum
envelope. These insulators, which are usually ceramic, include those separating
the cathode from the anode, and collector elements from one another and from
the tube body. Voltage breakdown on the vacuum side of these insulators occurs
at a lower voltage than for an equivalent vacuum gap [79]. This breakdown is ini-
tiated by the impact of electrons on the surface of the ceramic. These may come
from any source but a particular problem is caused by triple junction emission.
If the ceramic is attached to the metal of the tube body by brazing, then some
of the brazing alloy usually spreads a little way onto the surface of the ceramic,
providing a sharp metallic edge. The risk of triple junction emission at that edge
can be reduced by shaping the electrodes at the ends of the insulator to shield
the brazed joints from the electric field. The aim is to ensure that any breakdown
 Multipactor Discharges 721

takes place directly between the electrodes, rather than along the surface of the
insulator.
Ceramics have high secondary electron emission coefficients. Thus the impact
of an electron on the surface of a ceramic may generate a number of secondary
electrons and leave behind a positive surface charge. The secondary electrons are
accelerated across the surface of the ceramic by the external field, and drawn back
to it by the surface charge, so that they make further collisions and an electron ava-
lanche develops. The growing discharge liberates adsorbed gas molecules from the
surface of the ceramic. These may become ionised, increasing the risk of break-
down. A breakdown can damage the surface of the insulator so that the breakdown
voltage is reduced. In extreme cases the insulator may be punctured or may fracture,
leading to loss of vacuum. It is important that the surface of the insulator should be
free from contamination, and that adsorbed gases are driven off, as far as possible,
by baking when the tube is evacuated. The breakdown voltage is strongly dependent
on the shape of the insulator. It can be increased by adding corrugations to increase
the path length. A detailed discussion of surface breakdown on insulators in vac-
uum is given in [85].
To reduce the risk of breakdown on the air side of an insulator it is important to
keep the surface clean. The surface may be corrugated to increase the path length.
The risk may be also be reduced by pressurisation, or by immersing the insulator in
a dielectric fluid, such as transformer oil [79].

18.8 Multipactor Discharges

Multipactor is resonant radiofrequency vacuum discharge which is sustained by

secondary electron emission [86, 87]. The basic principle can be explained by con-
sidering a pair of parallel metal plates in vacuum with an RF voltage between them,
as shown in Figure  18.16. Note that the horizontal axis is time. A  free electron
between the plates at A at a moment when the potential of the upper plate is posi-
tive is accelerated towards the upper plate. When it strikes the plate at B it may lib-
erate secondary electrons. If the potential of the upper plate is now negative these
electrons are accelerated downwards so that they strike the lower plate at C. There
they may liberate further secondary electrons. If the time taken for the electrons
to cross from one plate to the other is just half of an RF cycle then the process is
repeated and the number of electrons increases each time. In order for a two-​surface
multipactor discharge to be sustained the RF voltage between the electrodes must
give impact energies in the range for which the secondary emission coefficient ( δ ) of
at least one of the surfaces is greater than unity (see Figure 18.7). It is found that
electrons, which are emitted from a surface before the optimum phase, experience
a smaller accelerating force than those which are emitted after the optimum phase.
Thus, the electrons are grouped into a smaller phase range when they arrive at the
opposite plate. This phenomenon is known as phase focusing. As the intensity of
722 Emission and Breakdown Phenomena

Figure 18.16:  Principle of the two-​surface multipactor discharge.

the discharge increases the process is limited by space-​charge debunching so that

the current saturates [88]. In resonant RF structures saturation may also be caused
by detuning of the resonator [87]. A  two-​surface multipactor discharge typically
involves currents of less than 1 A and voltages of a few hundred volts. The theory
of two-​surface multipactor is discussed in Section 18.8.1.
When a magnetic field is present the electron trajectories are bent by the field
so that the impacts made on the surfaces are oblique. This type of discharge
can involve impacts on either one or two surfaces. When the primary electrons
strike a surface obliquely the peak value of δ is greater than for normal incidence
and the range of energies over which δ is greater than unity is increased (see
Figure 18.7). Thus crossed-​field multipactor discharges can occur at much higher
energies than the simple multipactor and they are, in consequence, potentially
much more damaging. The theory of multipactor in crossed fields is discussed
in Section 18.8.3. Because strong magnetic fields are used to focus linear-​beam
tubes it is quite possible for the conditions for crossed-​field multipactor to exist
somewhere within the tube. The manufacturer will normally have taken steps to
ensure that this is not the case. However, if the magnetic field around the tube is
disturbed in any way, for example by the field of a circulator, then it is possible
for a destructive discharge to occur. It is also possible for multipactor discharges
to occur on ceramic surfaces such as vacuum windows where surface charging
produces a static electric field [87]. A discharge of this kind changes the match
of the window and the energy transferred may be sufficient to crack the ceramic.
It is probable that multipactor discharges occur in most vacuum tubes. They
have been observed in klystrons [86, 89] and magnetrons [90]. They also occur in
transmission lines of many kinds including rectangular waveguides [91, 92] and
coaxial lines [93, 94], and in components such as windows [95, 96] and filters [97].
Signs of multipactor discharges include heating, and the emission of light and
X-​rays. They are sources of noise, and can produce appreciable changes in the
RF performance of a tube by loading the RF circuit. The energy transferred to the
surfaces by the impact of electrons can cause window failure, and local outgassing,
which may lead to the formation of arcs. Multipactor discharges can sometimes
be suppressed by changing the shapes of surfaces, by surface coatings to reduce
the secondary electron yield, and by the imposition of static electric and magnetic
fields [86, 87, 98].
 Multipactor Discharges 723

18.8.1 Theory of Multipactor Discharges between Parallel Plates

The theory of the multipactor discharge between parallel metal plates has been
discussed by a number of authors. The analysis presented here is based on that
given in [86, 99]1. It provides insight into the factors contributing to the existence
of multipactor discharges.
Consider two plates located at y = 0 and y = d with an RF voltage V1 sin ( ωt )
between them. The equation of motion for an electron is

d 2 y eV1
= sin ( ωt ) . (18.51)
dt 2 m0 d
If the electron is injected at x = 0 with velocity v0 when ωt = θ1 then its velocity at
time t is

dy eV1
= (cos θ1 − cos ωt ) + v0 . (18.52)
dt m0 ωd
A second integration gives the position of the electron as a function of time

y eV1
=
d m0 ω 2 d 2
{(ωt − θ1 ) cos θ1 − sin ωt + sin θ1 } + ωv0d (ωt − θ1 ). (18.53)
If the electron reaches the second plate when ωt = θ2 then

eV1 v
1=
m0 ω d
2 2
{(θ2 − θ1 ) cos θ1 − sin θ2 + sin θ1 } + 0 (θ2 − θ1 ) . (18.54)
ωd
In order for the discharge to be resonant, the phase of the RF voltage when the
electron arrives at the second plate (θ2 ) must differ from the initial phase by an odd
multiple of π . Thus

θ2 = θ1 + nπ; n = 1, 3, 5 (18.55)
Substituting this condition into (18.54) gives the RF voltage as a function of the
initial phase

m0 ωd ( ωd − nπv0 )
V1 = ⋅ . (18.56)
e nπ cos θ1 + 2 sin θ1
We note that v0 only appears in the numerator of (18.56) and θ1 only in the denomi-
nator. Thus, the minimum RF voltage for which multipactor can occur is given by
the value of θ1 which maximises nπ cos θ1 + 2 sin θ1 regardless of the value of v0 . This
occurs when

2
tan θ1 = . (18.57)
nπ

1
I am indebted to my colleague Dr Amos Dexter for making his lecture notes on this subject available to me.
724 Emission and Breakdown Phenomena

If the RF voltage is greater than the minimum, electrons can reach the second elec-
trode with a range of initial phases. An electron whose initial velocity (v0 ) is zero
can only reach the second plate if it is emitted in the positive half-​cycle of the RF
voltage. If v0 > 0 then the electron can still reach the second plate if it is emitted into
a slightly retarding field, and the limiting phase, beyond which the electron cannot
escape from the surface, is negative.
A sustained discharge can only exist if there is phase focusing as described
above. Thus two electrons which leave the first plate at slightly different times arrive
at the second plate with a smaller time interval between them. This is expressed
mathematically by

d θ2
< 1. (18.58)
d θ1
If this condition is not satisfied then the electrons drift away from the resonant
condition over a number of cycles. Differentiating (18.54) with respect to θ1 gives

eV1  d θ 2  dθ   dθ 
 − 1 cos θ1 − (θ2 − θ1 ) sin θ1 − 2 cos θ2 + cos θ1  + v0  2 − 1 = 0.
 m0 ωd   d θ1  d θ1   d θ1 
(18.59)
When d θ2 d θ1 = 1 the limiting phase satisfies

cos θ1 = (θ2 − θ1 ) sin θ1 + cos θ2 . (18.60)

Substituting the resonant condition from (18.55) we find that this is identical
to the phase given by (18.57). The second limit for the phase is found by setting
d θ2 d θ1 = −1 to give

eV1

m0 ωd
{− cos θ1 − (θ2 − θ1 ) sin θ1 + cos θ2 } = 2v0 . (18.61)
Applying the condition for resonance we obtain

2 m0 ωd v0
2 cos θ1 + nπ sin θ1 = − . (18.62)
eV1
To eliminate V1 (18.56) is rewritten as

m0 ωd ( ωd − nπv0 )
nπ cos θ1 + 2 sin θ1 = . (18.63)
eV1
Dividing (18.63) by (18.62), and rearranging gives

4 nπv0 + 2ωd
tan θ1 = − . (18.64)
4v0 + nπωd − ( nπ ) v0
2
 Multipactor Discharges 725

Figure 18.17:  Hatch diagram showing the regions in which multipactor is theoretically

possible.

Thus, the minimum value of θ1 is the greater of the two values determined by the
condition that the electron can escape from the surface, and by the second condi-
tion for phase focusing.
If the RF voltage is increased, at a fixed value of ωd , from the minimum for
which a discharge can occur, then θ1 decreases until it reaches the minimum value
described above. The lower and upper limits of the RF voltage between which a
discharge is possible are obtained by substituting the two values of θ1 obtained
from (18.57) and (18.64) into (18.56). We may therefore plot V1 against fd = ωd 2 π
to show the boundaries of possible regions of multipactor for each mode, as shown
in Figure 18.17. The lines representing the limiting phase conditions for the lowest
mode are AB and DC.
In order that an electron avalanche can develop the product of the secondary elec-
tron emission coefficients of the two plates must be greater than unity. The impact
velocity is found by substitution of the resonant condition into (18.52) to give

2eV1
vi = cos θ1 + v0 (18.65)
m0 ωd
which may be rearranged, eliminating θ1 using (18.56), to give

2
m0 ωd  nπ
V1 = ωd − (vi + v0 ) + (vi − v0 )2 . (18.66)
2e  2 
The limits set by this condition are found by inserting the two values of vi for which
the secondary electron emission coefficient is unity into (18.66). These boundaries
are shown as DA and CD in Figure 18.18. This plot is known as a Hatch diagram
[100, 101].
726 Emission and Breakdown Phenomena

The boundaries of the three lowest multipactor regions in Figure  18.17 were
computed using the theory developed above for comparison with the experimental
results in Figure 3 of [101] (see Worksheet 18.1). The lowest experimental break-
down voltages were obtained when the aluminium electrodes had been left at a gas
pressure of 2 microns Hg for 24 hours. It is probable that the surfaces were then
fully oxidised. In the calculations it was assumed that δ m = 3.5 and E pm = 400 V
[23]. The electron emission energy was fixed at 2 eV and the secondary electron
emission curve was defined by (18.19) with s = 1.65. The calculations predict that
multipactor can exist in well defined regions with clear gaps between them. In the
experiments it was found that discharges existed at all points above the dashed line
in Figure 18.17. This discrepancy can be explained by a number of factors which
have been ignored in the simple theory: the spread in the secondary electron ener-
gies; the effect of elastically and inelastically reflected electrons; and the possibility
of asymmetric modes in which the value of n differs for the outward and backward
motion [102–​106]. All these effects serve to broaden the bands so that they tend to
merge. The boundary below which multipactor cannot occur is then a line which
is tangential to the bottoms of the theoretical regions shown in Figure 18.17. The
points on that line can be found by eliminating V1 between (18.56) and (18.65) and
using θ1 from (18.57) to give

nπ 2
ωd ( n ) = (vi + v0 ) + (vi − v0 ) (18.67)
2 nπ
where vi is the minimum impact velocity for which δ = 1. The corresponding values
of V1 ( n ) are found by substitution into (18.65).
In the experiments described in [101] the voltage at which the discharge com-
menced increased when the electrodes were first subjected to a discharge at a higher
pressure for half an hour. This presumably reduced the thickness of the oxide layer,
and increased the minimum impact voltage for which the secondary electron emis-
sion coefficient was unity. The effect can be modelled by reducing the value of δ m
used in the calculations. This raises the boundary line described above. There is
then a distinction between those multipactor discharges which can be eliminated by
conditioning, and those which cannot. The properties of two-​surface multipactor
discharges between parallel plates can be explored using Worksheet 18.1.
The assumption of a uniform RF electric field between plane electrodes is neces-
sarily idealised. The experiments of Hatch and Williams used a pair of circular
plates without any kind of compensation for the field concentration at the edges.
The assumption of a uniform field is a useful approximation when the discharge is
in a rectangular waveguide or in a re-​entrant cavity resonator [92, 107, 108]. It can
be seen from Figure 18.17 that, theoretically, the discharge cannot occur if the RF
voltage is great enough. That limit is represented by a line which is tangential to
the tops of the multipactor regions. However, even if the maximum RF voltage is
greater than the upper limit, its spatial variation ensures there is always some point
in the waveguide or cavity where multipactor can occur. Thus, in a non-​uniform
field the discharge is located at a position where the conditions are favourable.
 Multipactor Discharges 727

18.8.2 Multipactor Discharges between Coaxial Cylinders

The other simple geometry in which multipactor can be studied is the vacuum-​
spaced coaxial line [94, 109, 110]. This case has practical importance in the coaxial
waveguides and couplers used with many vacuum tubes. The symmetry of the field
allows the problem to be treated as one-​dimensional, but it is not possible to find
a simple analytical solution. An important difference from the parallel plate case
is that the electron transit times are different for inward and outward motion. It
is therefore necessary to find trajectories for which the sum of the transit times is
an integral number of RF periods. Both classical, and hybrid, two-​surface modes
are possible, in a manner very similar to those for parallel plates. But, it is found
that the modes gradually disappear, as the ratio of the radius of the outer con-
ductor to that of the inner conductor increases, until two-​surface multipactor is no
longer possible. However, the velocity of an electron starting from the outer elec-
trode can be reversed before it reaches the inner electrode. If it returns to the outer
electrode in an integral number of RF periods then a single-​surface discharge is
possible. Hatch diagrams similar to Figure 18.17 are obtained for both two-​surface
and single surface discharges [110]. Further information is given in the references
cited above.

18.8.3 Multipactor Discharges in Crossed Fields

Static magnetic fields are used to control the motion of the electrons in most vacuum
tubes. It is important to consider how multipactor discharges are affected by them.
The simplest case is the parallel plate geometry considered above, with the addition
of a uniform static magnetic field parallel to the planes of the plates [111]. The deri-
vations of the equations are lengthy and will not be reproduced here. It turns out
that both single surface, and double surface, discharges are possible. Figure 18.18
shows examples of single and double-​surface trajectories starting from x = 0. The
impact of the single-​surface trajectory in Figure 18.18(a) on the plane at x = 0 is
normal so that the energy involved is similar to that in a two-​surface multipactor
without a magnetic field. On the other hand, the trajectories in Figure  18.18(b)
make oblique impacts so that the electron energy can be much higher, while the sec-
ondary electron yield remains greater than unity (see Figure 18.8). In consequence
the energy dissipated can be much greater, and more damaging, than in the other
kinds of multipactor [86]. The properties of crossed-​field discharges of both kinds
can be explored using Worksheet 18.2.

18.8.4 Modelling Multipactor Discharges

In general, the possibility that a multipactor discharge will occur can be explored
for any geometry for which the RF electric field can be computed. If the satu-
rated discharge is to be computed then it is necessary to use a full particle in
cell (PIC) code so that the modification of the RF fields by the electronic space
728 Emission and Breakdown Phenomena

Figure 18.18:  Examples of multipactor trajectories in crossed fields: (a) single-​surface, and

(b) double-​surface.

charge can be included. However, if the objective is only to determine whether

multipactor is possible, then space-​charge effects can be ignored. The electron
trajectories can be computed using RF electric field components which have
been calculated in advance and stored. The conditions under which the discharge
can happen may be quite specific so it is necessary to track sample electrons from
a large number of points on the surfaces with a range of starting phases and
energies. The spatial and energy distributions of the secondary electrons, and
the time delay following the impact of the primary electron, must be represented
correctly. Elastically and inelastically reflected primary electrons must also be
included. It is evident that a large number of particles are needed at the start
of the simulation, and that the number will grow very rapidly with successive
impacts. The growth in the number of electrons can be used as an indication
of the existence of a discharge. Examples of the simulation of multipactor dis-
charges are given in [94, 108, 112–​115].

References

[1]‌ A. J. Dekker, Solid State Physics. London: Macmillan, 1963.

[2]‌ G. Gaertner and W. W. P. Koops, ‘Vacuum electron sources and their materials and tech-
nologies’, in J. A. Eichmeier and M. K. Thumm, eds, Vacuum Electronics: Components
and Devices. Berlin: Springer, pp. 429–​481, 2008.
[3]‌ G. W. C. Kaye and T. H. Laby, Tables of Physical and Chemical Constants. Available
at: www.kayelaby.npl.co.uk/​(accessed 5 October 2017), 2014.
[4]‌ A. Sharma et  al., ‘Emission poisoning studies on impregnated tungsten dis-
penser cathode under CO2 and O2 environment’, Applied Surface Science, vol. 40,
pp. 97–​101, 1989.
 References 729

[5]‌ W. H. Kohl, Handbook of Materials and Techniques for Vacuum Devices. New  York:
American Institute of Physics, 1995.
[6]‌ J. L. Cronin, ‘Modern dispenser cathodes’, IEE Proceedings I: Solid-​State and Electron
Devices, vol. 128, pp. 19–​32, 1981.
[7]‌ R. H. Fowler and L. Nordheim, ‘Electron emission in intense electric fields’, Proceedings
of the Royal Society of London. Series A, Containing Papers of a Mathematical and
Physical Character, vol. 119, pp. 173–​181, 1928.
[8]‌ I. Brodie and P. R. Schwoebel, ‘Vacuum microelectronic devices’, Proceedings of the
IEEE, vol. 82, pp. 1006–​1034, 1994.
[9]‌ N. M. Jordan et al., ‘Electric field and electron orbits near a triple point’, Journal of
Applied Physics, vol. 102, p. 033301, 2007.
[10] J. A. Eichmeier, ‘Radiation sensitive vacuum electronic components and devices’, in
J. A. Eichmeier and M. Thumm, eds, Vacuum Electronics:  Components and Devices.
Berlin: Springer, pp. 127–​154, 2008.
[11] J. W. Gewartowski and H. A. Watson, Principles of Electron Tubes. Princeton, NJ: D. van
Nostrand, 1965.
[12] J. A. Nation et al., ‘Advances in cold cathode physics and technology’, Proceedings of
the IEEE, vol. 87, pp. 865–​889, 1999.
[13] R. J. Umstattd, ‘Advanced electron-​beam sources’, in R. J. Barker et al., eds, Modern
Microwave and Millimetre-​ Wave Power Electronics. Piscataway, NJ:  IEEE Press,
pp. 393–​443, 2005.
[14] E. Sternglass, ‘Theory of secondary electron emission by high-​speed ions’, Physical
Review, vol. 108, pp. 1–12, 1957.
[15] J. J. Scholtz et al., ‘Secondary electron emission properties’, Philips Journal of Research,
vol. 50, pp. 375–​389, 1996.
[16] E. Rudberg, ‘Inelastic scattering of electrons from solids’, Physical Review, vol. 50,
pp. 138–​150, 1936.
[17] A. Shih et al., ‘Secondary electron emission properties of oxidized beryllium CFA cath-
odes’, IEEE Transactions on Electron Devices, vol. 41, pp. 2448–​2454, 1994.
[18] M. A. Furman and M. T.  F. Pivi, ‘Probabilistic model for the simulation of second-
ary electron emission’, Physical Review Special Topics –​Accelerators and Beams, vol. 5,
p. 124404, 2002.
[19] D. Chernin et al., ‘A model of secondary emission for use in computer simulation of vac-
uum electronic devices’, in International Electron Devices Meeting, pp. 773–​776, 1993.
[20] H. Seiler, ‘Secondary electron emission in the scanning electron microscope’, Journal of
Applied Physics, vol. 54, pp. R1–​R18, 1983.
[21] V. Baglin et  al., ‘A summary of main experimental results concerning the secondary
electron emission of copper’, CERN, Geneva, 2001.
[22] B. Henrist et al., ‘Secondary electron emission data for the simulation of electron cloud’,
in Proc. Mini Workshop on Electron Cloud Simulations for Proton and Positron Beams
(ECLOUD’02), pp. 75–​78, 2002.
[23] Y. Lin and D. C. Joy, ‘A new examination of secondary electron yield data’, Surface and
Interface Analysis, vol. 37, pp. 895–​900, 2005.
[24] P. Tolias, ‘On secondary electron emission and its semi-​empirical description’, Plasma
Physics and Controlled Fusion, vol. 56, p. 123002, 2014.
[25] V. Baglin et al., ‘The secondary electron yield of technical materials and its variation
with surface treatments’, in EPAC 2000, Vienna, Austria, pp. 217–​221, 2000.
730 Emission and Breakdown Phenomena

[26] C. Walker et al., ‘The secondary electron emission yield for 24 solid elements excited
by primary electrons in the range 250–​5000 eV:  a theory/​experiment comparison’,
Scanning, vol. 30, pp. 365–​380, 2008.
[27] C. Walker et al., ‘The secondary electron emission yield for 24 solid elements excited
by primary electrons in the range 250–​5000 eV:  a theory/​experiment comparison’,
Scanning, vol. 30, pp. 365–​380, 2008.
[28] K. Kanaya et  al., ‘Secondary electron emission from insulators’, Journal of Physics
D: Applied Physics, vol. 11, p. 2425, 1978.
[29] A. Dallos et al., ‘Effects of ionized oxygen on primary and secondary emission, and total
current of a CFA’, IEEE Transactions on Electron Devices, vol. 34, pp. 1201–​1208, 1987.
[30] J. Lorkiewicz et al., ‘Surface TiN coating of TESLA couplers at DESY as an antimulti-
pactor remedy’, in The 10th Workshop on RF Superconductivity, 2001.
[31] S. Michizono et  al., ‘Secondary electron emission of sapphire and anti-​multipactor
coatings at high temperature’, Applied Surface Science, vol. 235, pp. 227–​230, 2004.
[32] G. Fuentes et al., ‘Spectroscopic investigations of Cr, CrN and TiCr anti-​multipactor
coatings grown by cathodic-​arc reactive evaporation’, Applied Surface Science, vol. 253,
pp. 7627–​7631, 2007.
[33] M. Pivi et al., ‘Sharp reduction of the secondary electron emission yield from grooved
surfaces’, Journal of Applied Physics, vol. 104, p. 104904, 2008.
[34] A. N. Curren and K. A. Jensen, ‘Secondary electron emission characteristics of ion-
textured copper and high-​purity isotropic graphite surfaces’, NASA Lewis Research
Center, Cleveland OH, July 1984.
[35] M. Ye et al., ‘Suppression of secondary electron yield by micro-​porous array structure’,
Journal of Applied Physics, vol. 113, p. 074904, 2013.
[36] V. Nistor et al., ‘Multipactor suppression by micro-​structured gold/​silver coatings for
space applications’, Applied Surface Science, vol. 315, pp. 445–​453, 2014.
[37] A.-​G. Xie et al., ‘Maximum secondary electron yield and parameters of secondary elec-
tron yield of metals’, Surface Review and Letters, vol. 23, p. 1650039, 2016.
[38] B. K. Agarwal, ‘Variation of secondary emission with primary electron energy’,
Proceedings of the Physical Society, vol. 71, p. 851, 1958.
[39] J. R. M. Vaughan, ‘A new formula for secondary emission yield’, IEEE Transactions on
Electron Devices, vol. 36, pp. 1963–​1967, 1989.
[40] G. F. Dionne, ‘Effects of secondary electron scattering on secondary emission yield
curves’, Journal of Applied Physics, vol. 44, pp. 5361–​5364, 1973.
[41] G. F. Dionne, ‘Origin of secondary-​electron-​emission yield-​curve parameters’, Journal
of Applied Physics, vol. 46, pp. 3347–​3351, 1975.
[42] M. Salehi and E. Flinn, ‘An experimental assessment of proposed universal yield
curves for secondary electron emission’, Journal of Physics D: Applied Physics, vol. 13,
p. 281, 1980.
[43] Z. Insepov et  al., ‘Comparison of candidate secondary electron emission materials’,
Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with
Materials and Atoms, vol. 268, pp. 3315–​3320, 2010.
[44] R. Vaughan, ‘Secondary emission formulas’, IEEE Transactions on Electron Devices,
vol. 40, p. 830, 1993.
[45] A. Shih and C. Hor, ‘Secondary emission properties as a function of the electron inci-
dence angle’, IEEE Transactions on Electron Devices, vol. 40, pp. 824–​829, 1993.
 References 731

[46] S. Yu et  al., ‘Secondary electron emission for layered structures’, Journal of Vacuum
Science & Technology A, vol. 20, pp. 950–​952, 2002.
[47] R. Kitchen, RF and Microwave Radiation Safety Handbook. Newnes, 2001.
[48] AMPTEK Inc., Mini-​X X-​Ray Tube System for XRF, Available at: http://​amptek.com/​
products/​mini-​x-​ray-​tube/​#5 (accessed 5 October 2017), 25 July 2017.
[49] J. H. Hubbell and S. M. Seltzer, Tables of X-​Ray Mass Attenuation Coefficients and
Mass Energy-​Absorption Coefficients (version 1.4), 2 March. Available at: http://​phys-
ics.nist.gov/​xaamdi (accessed 5 October 2017), 2004.
[50] J. K. Shultis and R. E. Faw, ‘Radiation shielding technology’, Health Physics, vol. 88,
pp. 587–​612, 2005.
[51] R. Forman, ‘Update of thermionic cathode progress’, in International Electron Devices
Meeting, pp. 387–​390, 1991.
[52] J. Hasker et al., ‘Properties and manufacture of top-​layer scandate cathodes’, Applied
Surface Science, vol. 26, pp. 173–​195, 1986.
[53] G. Gaertner and D. Barratt, ‘Life-​limiting mechanisms in Ba-​oxide, Ba-​dispenser and
Ba-​Scandate cathodes’, in The 5th International Vacuum Electron Sources Conference,
pp. 59–​61, 2004.
[54] R. Tuck, ‘Thermionic cathode surfaces: the state-​of-​the-​art and outstanding problems’,
Vacuum, vol. 33, pp. 715–​721, 1983.
[55] L. R. Falce, ‘Dispenser cathodes: The current state of the technology’, in International
Electron Devices Meeting, pp. 448–​451, 1983.
[56] A. Shroff, ‘Review of dispenser cathodes’, Revue technique-​Thomson-​CSF, vol. 23,
pp. 947–​1026, 1991.
[57] W. E. Harbaugh, ‘Tungsten, thoriated tungsten and thoria emitters’, in Electron Tube
Design. Harrison, NJ: Radio Corporation of America, pp. 90–98, 1962.
[58] T. E. Yingst et al., ‘High-​power gridded tubes -​1972’, Proceedings of the IEEE, vol. 61,
pp. 357–​381, 1973.
[59] M. Zhang et al., ‘Influence of plasma spraying on the performance of oxide cathodes’,
IEEE Transactions on Electron Devices, vol. 58, pp. 2143–​2148, 2011.
[60] R. T. Longo et al., ‘Dispenser cathode life prediction model’, in International Electron
Devices Meeting, pp. 318–​321, 1984.
[61] R. Longo, ‘Physics of thermionic dispenser cathode aging’, Journal of Applied Physics,
vol. 94, pp. 6966–​6975, 2003.
[62] R. S. Raju, ‘Studies on W-​Re mixed-​matrix cathodes’, IEEE Transactions on Electron
Devices, vol. 56, pp. 786–​793, 2009.
[63] C. A. Spindt et al., ‘Field-​emitter arrays for vacuum microelectronics’, IEEE Transactions
on Electron Devices, vol. 38, pp. 2355–​2363, 1991.
[64] C. Spindt, ‘A brief history vacuum nanoelectronics, the IVNC, and the present sta-
tus of the Spindt cathode’, in 25th International Vacuum Nanoelectronics Conference,
pp. 1–​2, 2012.
[65] D. R. Whaley et al., ‘Application of field emitter arrays to microwave power amplifiers’,
IEEE Transactions on Plasma Science, vol. 28, pp. 727–​747, 2000.
[66] D. R. Whaley et al., ‘Experimental demonstration of an emission-​gated traveling-​wave
tube amplifier’, IEEE Transactions on Plasma Science, vol. 30, pp. 998–​1008, 2002.
[67] D. R. Whaley et al., ‘100 W operation of a cold cathode TWT’, IEEE Transactions on
Electron Devices, vol. 56, pp. 896–​905, 2009.
732 Emission and Breakdown Phenomena

[68] D. Whaley et al., ‘High average power field emitter cathode and testbed for X/​Ku-​band
cold cathode TWT’, in IEEE International Vacuum Electronics Conference (IVEC),
pp. 1–​2, 2013.
[69] C. Spindt et al., ‘11.1: A reliable improved Spindt cathode design for high currents’, in
IEEE International Vacuum Electronics Conference (IVEC), pp. 201–​202, 2010.
[70] W. D. Kilpatrick, ‘Criterion for vacuum sparking designed to include both RF and DC’,
Rev. Sci. Instrum., vol. 28, pp. 824–​826, 1957.
[71] W. Peter et al., ‘Criteria for vacuum breakdown in RF cavities’, IEEE Transactions on
Nuclear Science, vol. 30, pp. 3454–​3456, 1983.
[72] A. S. Gilmour, Jr., Microwave Tubes. Dedham, MA: Artech House, 1986.
[73] G. Faillon, ‘Technical and industrial overview of RF and microwave tubes for fusion’,
Fusion Engineering and Design, vol. 46, pp. 371–​381, 1999.
[74] S. Döbert, ‘Gradient limitations for high-​ frequency accelerators’, presented at the
LINAC 2004, Lübeck, Germany, 2004.
[75] H. H. Braun et al., ‘Frequency and temperature dependence of electrical breakdown at
21, 30, and 39 GHz’, Physical Review Letters, vol. 90, p. 224801, 2003.
[76] G. A. Farrall, ‘Vacuum arcs and switching’, Proceedings of the IEEE, vol. 61,
pp. 1113–​1136, 1973.
[77] I. I. Beilis, ‘State of the theory of vacuum arcs’, IEEE Transactions on Plasma Science,
vol. 29, pp. 657–​670, 2001.
[78] J. D. Cobine, Gaseous Conductors. New York: Dover, 1941.
[79] A. S. Denholm, ‘High voltage technology’, IEEE Transactions on Nuclear Science,
vol. 12, pp. 780–​791, June 1965.
[80] D. Xiao, ‘Fundamental Theory of Townsend Discharge’, in Gas Discharge and Gas
Insulation. Shanghai: Shanghai Jiao Tong University Press, pp. 47–​88, 2016.
[81] J. Townsend, Electrons in Gases. London; New  York:  Hutchinson’s Scientific and
Technical Publications, 1948.
[82] E. Husain and R. S. Nema, ‘Analysis of Paschen curves for air, N2 and SF6 using the
Townsend breakdown equation’, IEEE Transactions on Electrical Insulation, vol. EI-​17,
pp. 350–​353, 1982.
[83] J. Kuffel et al., High Voltage Engineering Fundamentals. Amsterdam: Newnes, 2000.
[84] R. Arora and W. Mosch, High Voltage and Electrical Insulation Engineering, vol. 69.
John Wiley & Sons, 2011.
[85] H. C. Miller, ‘Surface flashover of insulators’, IEEE Transactions on Electrical
Insulation, vol. 24, pp. 765–​786, 1989.
[86] J. R.  M. Vaughan, ‘Multipactor’, IEEE Transactions on Electron Devices, vol. 35,
pp. 1172–​1180, 1988.
[87] R. Kishek et al., ‘Multipactor discharge on metals and dielectrics: historical review and
recent theories’, Physics of Plasmas (1994 –​present), vol. 5, pp. 2120–​2126, 1998.
[88] S. Riyopoulos, ‘Multipactor saturation due to space-​ charge-​induced debunching’,
Physics of Plasmas (1994–​present), vol. 4, pp. 1448–​1462, 1997.
[89] C. Hill and R. G. Carter, ‘Investigation of possible multipactor discharge in a
klystron input cavity’, in 2006 IEEE International Vacuum Electronics Conference
Held Jointly with 2006 IEEE International Vacuum Electron Sources, Monterey, CA,
pp. 81–​82, 2006.
[90] J. R. M. Vaughan, ‘Observations of multipactor in magnetrons’, IEEE Transactions on
Electron Devices, vol. 15, pp. 883–​889, 1968.
 References 733

[91] R. L. Geng et  al., ‘Suppression of multipacting in rectangular coupler waveguides’,

Nuclear Instruments & Methods in Physics Research Section A:​Accelerators Spectrometers
Detectors and Associated Equipment, vol. 508, pp. 227–​238, 11 August 2003.
[92] R. L. Geng et  al., ‘Dynamical aspects of multipacting induced discharge in a rec-
tangular waveguide’, Nuclear Instruments & Methods in Physics Research Section A:
Accelerators Spectrometers Detectors and Associated Equipment, vol. 538, pp. 189–​205,
11 February 2005.
[93] P. Ylä-​Oijala, ‘Analysis of electron multipacting in coaxial lines with traveling and
mixed waves’, Deutsche Elektronen-​Synchrotron DESY, MHF-​SL Group, 1997.
[94] E. Somersalo et al., ‘Analysis of multipacting in coaxial lines’, in Proceedings of the
1995 Particle Accelerator Conference, pp. 1500–​1502, 1995.
[95] J. R. M. Vaughan, ‘Some high-​power window failures’, IRE Transactions on Electron
Devices, vol. 8, pp. 302–​308, 1961.
[96] S. Yamaguchi et al., ‘Trajectory simulation of multipactoring electrons in an S-​band
pillbox RF window’, IEEE Transactions on Nuclear Science, vol. 39, pp. 278–​282, 1992.
[97] A. Woode and J. Petit, ‘Investigations into multipactor breakdown in satellite micro-
wave payloads’, ESA Journal, vol. 14, pp. 467–​478, 1990.
[98] C. Chang et al., ‘Review of recent theories and experiments for improving high-​power
microwave window breakdown thresholds’, Physics of Plasmas, vol. 18, p. 055702, 2011.
[99] V. Shemelin, ‘Generalized phase stability in multipacting’, Physical Review Special
Topics-​Accelerators and Beams, vol. 14, p. 092002, 2011.
[100] A. J. Hatch and H. B. Williams, ‘The secondary electron resonance mechanism of
low-​pressure high-​frequency gas breakdown’, Journal of Applied Physics, vol. 25,
pp. 417–​423, 1954.
[101] A. J. Hatch and H. B. Williams, ‘Multipacting modes of high-​frequency gaseous
breakdown’, Physical Review, vol. 112, pp. 681–​685, 1958.
[102] S. Riyopoulos et  al., ‘Effect of random secondary delay times and emission vel-
ocities in electron multipactors’, IEEE Transactions on Electron Devices, vol. 44,
pp. 489–​497, 1997.
[103] A. Dexter and R. Seviour, ‘Rapid generation of multipactor charts by numer-
ical solution of the phase equation’, Journal of Physics D: Applied Physics, vol. 38,
p. 1383, 2005.
[104] R. Seviour, ‘The role of elastic and inelastic electron reflection in multipactor dis-
charges’, IEEE Transactions on Electron Devices, vol. 52, pp. 1927–​1930, 2005.
[105] S. Riyopoulos, ‘Higher-​order, asymmetric orbit multipactors’, Physics of Plasmas
(1994–​present), vol. 14, p. 112101, 2007.
[106] V. E. Semenov et al., ‘Importance of reflection of low-​energy electrons on multipactor
susceptibility diagrams for narrow gaps’, IEEE Transactions on Plasma Science, vol.
37, pp. 1774–​1781, 2009.
[107] D. Proch et al., ‘Measurement of multipacting currents of metal surfaces in RF fields’,
in Proceedings of the 1995 Particle Accelerator Conference, pp. 1776–​1778, 1995.
[108] V. Semenov et al., ‘Multipactor in rectangular waveguides’, Physics of Plasmas (1994–​
present), vol. 14, p. 033501, 2007.
[109] R. Woo, ‘Multipacting discharges between coaxial electrodes’, Journal of Applied
Physics, vol. 39, pp. 1528–​1533, 1968.
[110] R. Udiljak et  al., ‘Multipactor in a coaxial transmission line. I.  Analytical study’,
Physics of Plasmas (1994–​present), vol. 14, p. 033508, 2007.
734 Emission and Breakdown Phenomena

[111] S. Riyopoulos et  al., ‘Theory of electron multipactor in crossed fields’, Physics of
Plasmas (1994–​present), vol. 2, pp. 3194–​3213, 1995.
[112] V. Semenov et al., ‘Multipactor in a coaxial transmission line. II. Particle-​in-​cell simu-
lations’, Physics of Plasmas (1994–​present), vol. 14, p. 033509, 2007.
[113] G. Burt et al., ‘Benchmarking simulations of multipactor in rectangular waveguides
using CST-​particle studio’, in SRF 2009, Berlin, pp. 321–​325, 2009.
[114] C. Lingwood et  al., ‘Phase space analysis of multipactor saturation in rectangular
waveguide’, Physics of Plasmas (1994–​present), vol. 19, p. 032106, 2012.
[115] J. W. You et  al., ‘Highly efficient and adaptive numerical scheme to analyze mul-
tipactor in waveguide devices’, IEEE Transactions on Electron Devices, vol. 62,
pp. 1327–​1333, 2015.
19 Magnets

19.1 Introduction

Virtually all microwave tubes employ a magnetic field to control the flow of elec-
trons. This chapter reviews the theory and design of both electromagnets and per-
manent magnets for this purpose. The objective is to create a specified magnetic
field in the volume of space in which the electron interaction takes place.
In linear beam tubes, gyrotrons, and long anode magnetrons, the volume is a
long thin cylinder. The magnetic field may vary in the axial direction, for example
to provide the correct entry conditions at the electron gun, to increase the beam
stiffness near the output gap of a klystron, or to allow beam expansion into the col-
lector of a gyrotron. It is important that the field should be cylindrically symmet-
rical about the axis of the beam. Multiple-​beam tubes require a very uniform axial
field to minimise the radial field components which can cause the beams to drift
in the radial direction. The field is provided either by an electro-​magnet, or by one
or more permanent magnets [1–​3]. Electro-​magnets may be normal conducting, or
superconducting, and operated pulsed or continuously. They have the advantage
that the strength of the magnetic field can be adjusted, as required, for the opti-
mum operation of the tube; however, the power required to energise and cool an
electromagnet reduces the overall efficiency of the tube. Permanent magnets do not
have these disadvantages, but heavy and expensive magnets are required to produce
uniform fields. The size and weight of the magnet system can be reduced substan-
tially by using periodic permanent magnet (PPM) focusing (see Section 7.6). This is
common for TWTs and has also been used for a few klystrons.
In crossed field tubes, other than long anode magnetrons, the interaction space is
a relatively short annular cylinder. The field, which is normally supplied by a per-
manent magnet, should be uniform to within 5% to 15%, and be symmetrical about
the axis to better than 3% [2].
The chapter begins with a review of the theory of magneto-​statics, including the
properties of iron and other ferromagnetic materials. This leads to a discussion of
the use of magnetic circuits for analysis and design. The properties of common
ferromagnetic materials are reviewed in Section 19.4. The remainder of the chapter
considers the properties, and design, of three types of magnet:
• In coil dominated magnets (Section 19.5) the magnetic field and its distribution in
space are determined by the positions of conducting coils, and the currents in them.
736 Magnets

• In iron dominated magnets (Section 19.6) the magnetic field is produced by cur-
rents in conducting coils but its distribution in space is determined by the shape
of iron yokes and pole-​pieces.
• In permanent magnet systems (Section 19.7) the magnetic field is provided by
one, or more, permanent magnets, while iron yokes and pole-​pieces are used to
determine its distribution in space.

19.2 Review of Theory

The theory of electro-​magnets and permanent magnets is discussed in textbooks

on electromagnetism [4–​6]. The main results are reviewed below for convenience.
Paramagnetic and diamagnetic effects are too small to be of importance for the
applications considered in this book. Therefore the magnetic properties of all
materials, except ferromagnetic materials (iron, cobalt, nickel and their alloys), will
be assumed to be those of free space. Two vectors are used to represent magnetic
fields: the magnetic flux density B and the magnetic field H. In a linear magnetic
material they are related by

B = µH = µ 0 µ r H, (19.1)

where µ is the permeability of the material, µ 0 = 4 π × 107 H m −1 is the primary mag-

netic constant (or permeability of free space), and µ r is the relative permeability of
the material. The normal component of B and the tangential component of H are
continuous at boundaries between regions occupied by different materials.
The magnetic flux density generated by a current element I dl at a point P, whose
distance from the element is r, is given by

µ0 I
dB = (dl ∧ rˆ ) , (19.2)
4 πr 2

where r̂ is a unit vector directed from the current element to P. The vector product
ensures that the vector dB is normal to the plane containing dl and r̂. A steady elec-
tric current must flow in a closed circuit and, taking the integral of (19.2) around
the circuit, we obtain the Biot-​Savart Law

µ 0 I ⌠ dl ∧ rˆ
B=
4π ⌡  r 2 . (19.3)


This can be used to compute the flux density, at any point, generated by an arrange-
ment of current-​carrying conductors in free space.
The magnetic flux and field vectors obey Maxwell’s integral equations

∫∫ B ⋅ dS = 0, (19.4)

 Review of Theory 737

where the integral is taken over a closed surface, and

⌠⌠  ∂D 
∫ H ⋅ dl = ⌡⌡  J +  ⋅ dS, (19.5)
∂t 
where J is the current density, and D = εE is the electric flux density. The integral
on the left-​hand side of (19.5) is taken around a closed path, and the integral on the
right-​hand side is taken over a surface which spans the path. For steady currents
flowing in conductors the right-​hand side is the total current enclosed by the path
of integration. The equation can then be written
N
∫ H ⋅ dl = ∑ I
n =1
n (19.6)

where N is the number of conductors and the current in the nth conductor is I n .
This form is known as the Magnetic Circuit Law. The application of this law to the
analysis and design of magnets is discussed in Section 19.3.
The differential forms of equations (19.4) and (19.5) are

∇ ⋅ B = 0 (19.7)
and
∂D
∇×H = J+ . (19.8)
∂t
It is convenient to define a magnetic scalar potential U by analogy with the electro-
static potential so that

H = −∇U. (19.9)
The magnetic scalar potential difference between two points is
2
U 2 − U1 = − ∫ H ⋅ dl, (19.10)
1

where the integral is taken along any convenient path joining them. The magnetic
scalar potential difference only has a unique value if the path does not encircle any
current. In that case, substituting into (19.7) from (19.9) we obtain
∇ ⋅ µ∇U = 0. (19.11)
For a homogeneous material µ is constant and U satisfies Laplace’s equation

∇2U = 0. (19.12)
Then the magnetic field can be determined from the gradient of the scalar potential.

19.2.1 Ferromagnetism
Iron, cobalt, and nickel, and alloys based on these metals, are known as ferromag-
netic materials. They have magnetic properties which make them important for
738 Magnets

engineering applications. The flux density in a ferromagnetic material depends not

only on the magnetising field to which it is subjected, but also on the history of
magnetisation. This is displayed in a B − H curve, as shown in Figure 19.1(a). If
the specimen is initially un-​magnetised, then its state is represented by the point a.
When the magnetising field Hi is gradually increased the curve a-​b, known as initial
magnetisation curve, is followed. The flux density can be written
Bi = J + µ 0 Hi , (19.13)
where J is the magnetic polarisation of the material. At high values of Hi the
polarisation reaches saturation ( J s ) and thereafter an increase in Hi only serves
to increase Bi by the change in µ 0 Hi. If Hi is then reduced to zero, the flux density
does not retrace the curve a-​b. Instead, it follows a curve such as b-​c, so that the flux
density in the circuit does not fall to zero when the magnetising field is zero. The
value of Bi at c is known as the remanence ( Br ) of the material. In order to reduce
the flux density to zero it is necessary to reverse the direction of Hi. The value of Hi
at which Bi becomes zero (point d) is known as the coercive force ( H c ). Increasing
the magnitude of Hi beyond the value needed to reach d eventually produces sat-
uration of the material at e with the direction of the flux opposite to that at b.
Finally, reducing the magnitude of Hi to zero and then increasing it with positive
polarity produces the curve e-​f-​g-​b. This behaviour, in which the magnetisation lags
behind the magnetising field, is known as hysteresis. When the magnetising field is
repeatedly taken through the same cycle, the hysteresis loop b-​c-​d-​e-​f-​g-​b is traversed
repeatedly in a stable manner, provided only that the maximum magnitude of Hi is
the same for both polarities. If, on the other hand, the magnitude of the magnetis-
ing field is smaller, it is found that the behaviour of the material is described by
a smaller loop, such as the one shown dotted in Figure 19.1(a). From a practical
point of view the main division is between ‘hard’ and ‘soft’ magnetic materials, as
shown in Figure 19.1(b). Soft magnetic materials are easy to magnetise and demag-
netise, whereas both processes are more difficult for hard magnetic materials. Soft
magnetic materials are used as conductors of magnetic flux, as described in the next
section. Hard magnetic materials are used as permanent magnets. The properties of
some common ferromagnetic materials are reviewed in Section 19.4.

19.2.2 Conduction of Magnetic Flux by Soft Magnetic Materials

Figure 19.2 shows a magnetic flux line as it passes from air into a material such
as pure iron which has a high relative permeability. Applying the boundary
conditions gives
B1 cos α1 = B2 cos α 2 (19.14)
and
B1 B
sin α1 = 2 sin α 2 . (19.15)
µ0 µ0 µr
 Magnetic Circuits 739

Figure 19.1:  (a) A typical hysteresis loop showing the initial magnetisation curve and
(b) typical hysteresis loops for hard and soft magnetic materials.

Figure 19.2:  Change in the direction of a magnetic flux line at a boundary between air
and iron.

Dividing (19.15) by (19.14) we obtain

1
tan α1 = tan α 2 . (19.16)
µr
For soft magnetic materials µ r is large so that α1  α 2 . For example, if µ r = 5000
and α 2 = 89° then α1 = 0.7°. Thus the flux lines in air are very nearly normal to the
magnetic surface. It is therefore possible to treat soft magnetic materials as perfect
magnetic conductors, to a good approximation. Laplace’s equation can then be
used to find the distribution of a magnetic field in the air spaces between sections
of soft magnetic material.

19.3 Magnetic Circuits

The analogy between the conduction of electricity by conductors, and the conduc-
tion of magnetic flux by iron, leads to the useful concept of the magnetic circuit.
740 Magnets

Figure 19.3:  A simple magnetic circuit made up of an iron core with an air gap in it. The flux
is supplied by a winding of N turns of wire.

Figure 19.3 shows a simple magnetic circuit formed by a square iron core with a
narrow air gap in it. A coil of N turns of wire is wound on the core and carries cur-
rent I. To make the problem easy to handle we make some simplifying assumptions.
Let us suppose that the magnetic field strength Hi is the same everywhere within the
iron, and the magnetic field in the air gap is Ha. Now consider a closed path around
the circuit which follows the centre line (shown dotted). Applying the magnetic
circuit law (19.6) to this path gives
LHi + gH a = NI (19.17)
where L is the length of the path in iron. If, for the moment, we assume that the
cross-​sectional area of the air gap is equal to that of the iron then
Φ
Bi = Ba = , (19.18)
A
where Φ is the total flux circulating, and A is the cross-​sectional area of the iron
at right angles to the direction of the magnetic field. We have therefore assumed
that all the flux due to the coil is contained within the bounds of the iron core and
its projection across the air gap. Making use of (19.1) we can write Bi = µ Hi and
Ba = µ 0 H a. Substituting for Hi and Ha in Equation (19.17) in terms of Φ and the
various constants gives

1 L g 
NI = + Φ, (19.19)
A  µ µ 0 
which can be written

M = RΦ, (19.20)
where M is known as the magneto-​motive force and R as the reluctance. Equation
(19.20) is analogous to Ohm’s law for electric circuits, but the analogy must not be
pressed too far. Unlike an electric current, the magnetic flux in the circuit dissipates
no energy, and there are no circulating magnetic charges. Moreover, it cannot be
 Magnetic Circuits 741

Figure 19.4:  Initial magnetisation curves of Armco iron and mild steel showing a typical
load line (dashed).

emphasised too often that magnetic materials do not normally behave in a linear
fashion, so the reluctance of a circuit is only approximately constant, and then only
under a limited range of conditions.
If the permeability of the iron is not constant, the application of the magnetic
circuit law to the circuit in Figure 19.3 gives
gBi
NI = LHi + gH a = LHi + (19.21)
µ0
or
µ 0 L  NI 
Bi =  − Hi  . (19.22)
g  L 
But we know that Bi and Hi must be related to each other by the initial magnetisa-
tion curve of the iron

Bi = f ( Hi ) . (19.23)
The non-​linear simultaneous equations (19.22) and (19.23) can be solved graphi-
cally, or by using an analytical approximation to the equation of the hysteresis loop.
The graphical solution is shown in Figure 19.4. The working point is at the intersec-
tion of the straight line represented by (19.22) with the initial magnetisation curve.
The intercept of this load line on the horizontal axis is the magneto-​motive force in
the circuit divided by the path length, while its slope is determined by the relative
sizes of the iron path and the air gap.
The preceding discussion of magnetic circuits made some rather crude assump-
tions about the distribution of the magnetic flux. A more realistic view is shown in
Figure 19.5(a). The flux lines in the air gap spread out into a fringing field as shown.
Because the magnetic flux in the circuit is constant, the fringing reduces the flux
742 Magnets

density in the air gap and, hence, the reluctance of the gap. A second difficulty in
calculating the flux in the air gap is that some of the flux simply takes a short cut
such as A-​A´ and never reaches the gap at all. This flux is known as leakage flux.
It exists because, although iron conducts flux better than air, the air is a poor mag-
netic insulator. If we wish to pursue the analogy with electric circuits it is neces-
sary to think of the material surrounding the conductor as being a poor conductor
rather than an insulator. Fortunately, in most cases, a very crude estimate of the
leakage flux is adequate to provide a satisfactory value for the flux in the air gap.
The flux leakage can be minimised by placing the source of the magneto-​motive
force as close to the air gap as possible. Thus, the magnetising coil can be replaced
by a pair of coils close to the air gap, as shown in Figure 19.5(b). It can be seen
that the leakage flux is then greatly reduced. Estimates of the fringing and leak-
age fluxes can be made using formulae to be found in the literature [7–​9] (see also
Section 19.7.1). These methods are only suitable for the creating an initial design
and computational magnetics software must be used to obtain accurate results [10].
Figure 19.5 and similar images in this chapter were produced using Gemini 1.2 [11].

19.3.1 Circuits Including Permanent Magnets

We have already noted that hard magnetic materials can be magnetised so that they
produce a substantial magnetic flux even when the magnetising field is removed.
These materials are used for making permanent magnets which are used especially
in magnetrons, CFAs, and PPM focused TWTs. Figure 19.6 shows a portion of a
typical B–​H plot for a permanent magnet material. This curve, which is the part of
the hysteresis loop lying in the second quadrant, is known as the demagnetisation
curve of the material.
Suppose that the circuit shown in Figure 19.5 is made of this material and mag-
netised to saturation by passing a current through the coil when the air gap is
bridged by a piece of soft iron. When the current in the coil is reduced to zero the
working point will be at P in Figure 19.6. If the soft iron is removed from the gap, I
is set to zero in (19.22), and fringing and leakage fluxes are neglected, we have

 µ L
Bi = −  0  Hi . (19.24)
 g 
This load line passes through the origin and has negative slope. It is plotted in
Figure 19.6 as the line OQ. The effect of opening the gap is therefore to demagnet-
ise the magnet to some extent. For this reason it is common to speak of the demag-
netising field of the air gap.
If the piece of soft iron is reinserted in the air gap, the operating point of the
magnet moves along a minor loop (or recoil line) to the point P’. In this state the
magnet is largely immune to the effects of external fields. For this reason it is usual
to store a permanent magnet with the air gaps bridged by a piece of soft iron which
is known as a ‘keeper’. If the keeper is absent, and the magnet is exposed to an
external demagnetising field, then the working point might move to R. When the
 Magnetic Circuits 743

Figure 19.5:  (a) An electromagnet showing the fringing field around the air gap and a typical
flux leakage path A − A′, and (b) an alternative arrangement of the coil to minimise the
leakage flux.

Figure 19.6:  The working point for a permanent magnet circuit lies in the second quadrant
of the B–​H plot at a point such as Q. A permanent magnet can be stabilised against
demagnetisation by external fields by operating it at a point Q’ on a minor loop such
as R–​S.

external field is removed the working point now lies at Q’ on the minor loop RS.
This illustration shows that it is necessary to treat permanent magnets carefully if
their properties are to be preserved. If the demagnetising field is strong enough for
the point R to move into the third quadrant then the magnet may be completely
744 Magnets

demagnetised. It is undesirable for the working point of the magnet to be sensitive

to external influences in this way and permanent magnets are often stabilised by
deliberately demagnetising them beyond their working points on the main hystere-
sis loop. The operation of a magnet which has been stabilised is then along a minor
loop such as RS unless the external demagnetising field is very strong. Smaller
external fields produce a temporary shift in the working point, but it returns to Q’
when the perturbing field is removed.
In a circuit involving a permanent magnet, the cost of the magnet is usually a
considerable part of the cost of the whole circuit. It is, therefore, desirable to use the
magnet material as efficiently as possible. A formula for this purpose can be derived
by considering a uniform magnetic circuit with an air gap. Let the lengths of the
magnet and the air gap be li and la and let their cross-​sectional areas be Ai and Aa.
The magnetic circuit law gives
li Hi + la H a = 0 (19.25)
and the conservation of flux gives
Ai Bi = Aa Ba (19.26)
so that
Vi Bi Hi = −Va Ba H a , (19.27)
where Vi = li Ai and Va = la Aa are, respectively, the volumes of the iron and the air
gap. Now the density of the stored energy in a magnetic field is BH 2 so (19.27)
shows that magnitude of the stored energy in the magnet must equal that in the air
gap. Thus the minimum magnet volume to provide a stored energy in a given air
gap is achieved by making the energy product of the magnet ( Bi Hi ) as great as pos-
sible. This is known as Evershed’s criterion. Figure 19.7 shows curves of Hi and BiHi
against Bi for a typical permanent magnet material. The optimum working point of
the material is then Qopt where the value of BiHi is greatest. It is usual to design the
magnetic circuit so that the magnet operates near this point.

19.4 Magnetic Materials

The properties of magnetic materials can be varied by making alloys with different
proportions. They may be isotropic, or anisotropic, and are strongly influenced by
the ways in which the finished material is prepared, especially any heat treatments
used. A very great variety of materials now exists whose properties can be looked
up in the literature [7, 12, 13] and in manufacturers’ catalogues.

19.4.1 Soft Magnetic Materials

Soft magnetic materials are characterised by narrow hysteresis loops, low rema-
nence and small coercive force. The area of the hysteresis loop is proportional to
 Magnetic Materials 745

Figure 19.7:  Typical curves of Hi and BiHi versus Bi for a permanent magnet material
showing the choice of working point at the maximum value of BiHi to make the most
efficient use of the material.

the work required to move the working point around it. Thus, narrow loops are
needed to minimise the generation of heat through hysteresis loss in transformer
cores and other AC applications [14]. Microwave tubes generally use static magnetic
fields and hysteresis loss does not occur. The properties of soft magnetic materials
can be approximated by their initial magnetisation curves. Figure 19.4 shows two
examples. The curves are nearly straight for fields well below saturation, so it is pos-
sible to make the approximation that the relative permeability is constant. The rela-
tive permeability decreases with increasing temperature, falling to zero at the Curie
temperature (Tc ). Soft magnetic materials are used in vacuum tubes for pole-​pieces,
for shaping the magnetic field for electron guns, for yokes for electromagnets, and
for magnetic shielding. It is useful to remember that materials such as nickel alloys,
which are sometimes used for other purposes in tube construction, are also ferro-
magnetic and may affect the magnetic field close to them. Typical properties of soft
magnetic materials are given in Table 19.1.

19.4.2 Permanent Magnet Materials

The choice of permanent magnets for use in microwave tubes is based on magnetic
strength, operating temperature, corrosion resistance, stability, and cost. The prop-
erties of these materials can be changed irreversibly by vibration and shock, and by
exposure to high temperatures. The available materials fall into four main classes as
shown in Table 19.2 [3, 16]. Typical demagnetisation curves are shown in Figures.
19.8 and 19.9.
Ferrite magnets are made from mixed oxides of iron, and certain other metals,
sintered at a high temperature to form an insulating ferromagnetic compound. They
are inexpensive and have good stability but the energy product and the maximum
working temperature are low. Forcing two magnets together in repulsion can cause
demagnetisation of up to 10%. Below  –​40 °C they suffer from large irreversible
746 Magnets

Table 19.1:  Typical properties of soft magnetic materials used in microwave tubes [15]

Material µr ( max ) J s (T ) H c ( A m −1 ) Br ( T ) Tc (°C)

Armco iron (99.85% Fe) 7000 2.16 80 1.3 770

Mild steel 2000 2.15 150 –​ –​
Nickel 600 0.615 400 –​ 358
Mumetal 240,000 0.77 1.0 0.45 350

Table 19.2:  Typical properties of permanent magnet materials [17]

Ferrite Alnico Samarium Neodynium

cobalt iron boron

BH max (kJ m −3 ) 26 40 210 280

Bopt (T ) 0.14 0.13 0.35 0.45
Br (T ) 0.22–​0.40 0.80–​1.26 0.85–​1.0 1.2
H c (kA m −1 ) 110–​240 40–​155 700–​800 820–​1040
Density (g cm ) −3
4.9 7.3 8.4 7.5
Working temperature (°C) 250 550 300 120
Reversible change in Br (C−1) 0.19% 0.02% 0.03% 0.12%
Corrosion resistance Excellent Fair Excellent Poor
Ease of demagnetisation Low High Very low Very low
Cost comparison 1 5 20 10

Figure 19.8:  Typical demagnetisation curves of permanent magnet materials

(copyright 2010 Delta Magnets, reproduced, with permission, from [16]).
 Coil Dominated Magnets 747

Figure 19.9:  Typical demagnetisation curves of Alnico magnets

(copyright 2010 Delta Magnets, reproduced, with permission, from [16]).

losses in magnetisation. These materials are brittle and must be shaped by grinding,
which is an expensive process.
Alnico materials are a group of cast alloys of iron with aluminium, cobalt, nickel
and other metals. The preferred magnetic axis is determined during manufacture.
They are relatively cheap, can be machined and have a high working temperature.
Because the coercive force is low they are easily demagnetised. Forcing two magnets
together in repulsion can cause demagnetisation by up to 50%. For this reason it
is usual to stabilise them. They should be stored with a keeper in place to avoid
demagnetisation, and are best magnetised after assembly [18].
Samarium cobalt, and neodymium iron, magnets are known collectively as rare
earth magnets. Samarium cobalt magnets have a high energy product, excellent sta-
bility, and a moderate working temperature. They are hard, brittle, materials which
are manufactured by sintering and must be shaped by grinding or abrasive cutting.
Their corrosion resistance is excellent but they will break down if exposed to hydro-
gen. The cost of these magnets is high but they are very useful when it is important
to minimise the volume of the magnet. Neodymium iron boron magnets have the
highest energy product of all and their cost is lower than that of samarium cobalt
magnets. However, the maximum working temperature is too low for most vacuum
tube applications and they break down when exposed to hydrogen.

19.5 Coil Dominated Magnets

The magnetic field generated by a system of coils without any iron pole-​pieces is
sometimes used for experimental purposes [19]. It is also relevant to the design of
superconducting magnets where iron cannot be used because the high flux densities
would saturate it [20, 21]. The simplest example is a circular current loop of radius
748 Magnets

Figure 19.10:  Calculation of the flux density on the axis of a circular wire loop carrying
current I.

Figure 19.11:  Calculation of the flux density on the axis of a single-​layer solenoid having n
turns per unit length carrying current I.

a carrying current I, as shown in Figure 19.10. It can be shown that the flux density
at a point P on the axis distant z from the plane of the loop is given by
µ0 I 3 µ I 1
Bz = sin ψ = 0 ⋅ 3
. (19.28)
2a 2a
(1 + z 2 a2 )2
The calculation of the flux density at points which do not lie on the axis is more
difficult because it requires the use of elliptic integrals [22].
A second simple case is the single-​layer solenoid of finite length shown in
Figure 19.11. The flux density on the axis is given by
µ 0 nI
Bz = (cos ψ 2 − cos ψ1 ) , (19.29)
2
where n is the number of turns per unit length. Note that if P is inside the solenoid
then the sign of cos ψ1 is reversed. The flux density at the centre of a solenoid of
length L is then
µ 0 nIL
Bz = . (19.30)
L2 + ( 2a )
2

If L → 0 then Bz tends to the value given by (19.28) at z = 0. When L → ∞ then

Bz → µ 0 nI which can also be derived by the use of the magnetic circuit law (19.5).
 Coil Dominated Magnets 749

Figure 19.12:  Arrangement of a coil having finite cross-​sectional area.

Figure 19.13:  Comparison between the normalised axial flux densities of coils having square
cross-​sections of different sizes (a = 1).

Practical coils differ from the idealised current loop and solenoid described
above because the windings have finite length and thickness in all cases, as shown
in Figure 19.12. The flux density of a practical coil having N turns is calculated by
summing the contributions of all the turns to the total flux density (see Worksheet
19.1). To find the field on the axis of this coil it is convenient to normalise all dimen-
sions to the mean radius ( a ). Then the flux density on the axis, normalised to that
at the centre of a current loop, is given by
L
⌠ 2  1+ 2t 3 
⌠ 2 −
1    ( z − x )  2 
Bz ( z ) =   1 +  dr  dx. (19.31)
Lt    r2  
 L ⌡1− t
⌡−  2 
2
750 Magnets

Figure 19.14:  Comparison between the normalised axial flux densities of solenoids having
different thicknesses (a = 1, L = 8) .

Figure 19.13 shows a comparison of the flux density on the axis of coils having
square cross-​section. It can be seen that the flux density can be approximated quite
closely by that of a current loop even when the cross-​sectional dimensions of the
coil are comparable with the mean radius. Figure 19.14 shows similar results for a
solenoid for which L = 8a. All the cases plotted can be modelled quite accurately by
a single-​layer solenoid

19.5.1 Arrays of Coils

The flux density produced by two or more coils can be calculated using the principle
of superposition. The flux density on the axis of a circular current loop located at
z = d is obtained from (19.28) as
µ0 I a2
Bz = ⋅ . (19.32)
(a )
3
2
+ (z − d )
2 2 2

This expression can be expanded as a Taylor series about the origin as

Bz ( z ) = b0 + b1 z + b2 z 2 + etc., (19.33)
where

1  d n Bz 
bn =   . (19.34)
n !  dz n  z = 0
When a pair of coils is arranged symmetrically with respect to the origin the odd
order terms cancel. Using (19.34) it is found that

3a 2 (a 2 − 4d 2 )
b2 = . (19.35)
(a 2 + d 2 )7 2
 Coil Dominated Magnets 751

Figure 19.15:  Normalised, maximally flat, axial flux density distributions for two, three and
four coils when the total current is unity (a = 1).

Figure 19.16:  Arrangement of a solenoid for a microwave tube.

Thus setting d = a 2 produces a maximally flat field at the origin giving the well-​
known Helmholtz coils. The region of uniform field can be extended by the addi-
tion of more coils and by using equal ripple solutions [23–​25]. Figure 19.15 shows
the maximally flat results for two, three and four coils when the total current is
unity (see Worksheet 19.1). The method has also been extended to coils of finite
cross-​section [26, 27]. The design of superconducting magnets to produce a long
length of uniform axial magnetic field is described in [20, 21]. Detailed information
about the design of coil dominated magnets can be found in [28, 29].

19.5.2 Solenoids
Figure 19.16 shows a typical arrangement of a cylindrically symmetrical solenoid
for a microwave tube. The air space is bounded at each end by a soft iron pole-​piece
and the return path for the flux is provided by an external yoke. This may take the
752 Magnets

form of an iron cylinder or of a number of iron bars. The magneto-​motive force is

provided by a winding made of copper or aluminium wire or tape [1, 3, 28, 30–​32].
It is possible to wind the coils directly onto the body of the tube but this is a dif-
ficult process which requires the complete tube to be spun in a lathe and can cause
distortion of the tube [30]. A wound-​on solenoid limits the temperature at which
the tube can be baked during evacuation and increases the cost of replacing a failed
tube. Thus, it is much more common to wind the solenoid on a separate former (as
shown in Figure 19.16) and arrange for the tube to be inserted into it. Part of each
pole-​piece is incorporated in the tube body and care must be taken to ensure that
there is a good magnetic joint between the sections of each pole-​piece when the
tube is installed. If the tube fails it can be replaced without changing the solenoid
which remains a permanent part of the equipment in which it is installed.
The basic design of a solenoid as shown in Figure 19.16 is very simple. For the
time being we ignore the holes in the pole-​pieces which are necessary in a linear
beam tube for the passage of the electron beam. The effect of these holes is dis-
cussed in Section 9.3.1. The inner faces of the pole-​pieces can be regarded as perfect
magnetic conductors to a good approximation, and the problem is solved using the
method of images. The region between the pole-​pieces is regarded as a section of
an infinitely long uniform solenoid which therefore has a uniform flux density ( B0 )
within it. The application of the magnetic circuit law (19.6) then gives
µ 0 H 0 L = NI , (19.36)
where the solenoid winding has N turns, each carrying current I. The shape and cross-​
sectional area of the conductors are chosen to give a convenient combination of volt-
age and current for the solenoid, bearing in mind the space needed for insulation.
The current density in the winding can be chosen independently depending on the
trade-​off between size and cooling requirements. The current density is typically 2
to 3 A  cm−​2. It is straightforward to calculate the resistance of the solenoid and,
hence, the power dissipated in it. For pulsed operation it is also necessary to calculate
the self-​inductance. The design of the solenoid must include a thermal analysis, and
arrangements for cooling, to ensure that the local temperature is never high enough
to damage the insulation between the turns of the winding. The design of the pole-​
pieces and the yoke should include checks to ensure that the flux density is never suf-
ficient to cause local magnetic saturation because that would cause the flux inside the
solenoid to be non-​uniform, and would increase the leakage flux outside it.
In practice a solenoid usually comprises a number of separate coils as shown in
Figure 19.16 allowing the sides of the coils to be cooled. Alternatively there may
be gaps between the coils to allow access the tube body. The effect of this arrange-
ment of the winding is to cause a periodic variation in the axial magnetic field.
The magnitude of the ripple can be estimated by assuming that each coil can be
represented by a single current loop located at the centre of the winding and that
the effect of the pole pieces can be represented by the mirror images of the coils in
them. The final design of the solenoid is checked by numerical modelling using one
of the many computer programs which are commercially available. The design of
solenoids is discussed in [3, 33, 34].
 Iron Dominated Magnets 753

Figure 19.17:  Electromagnet for a small microwave tube: (a) general arrangement, and

(b) flux distribution.

19.6 Iron Dominated Magnets

A few tubes, such as small klystrons, have dipole electromagnets arranged as

shown in Figure 19.17(a). It can be seen from Figure 19.17(b) that the flux den-
sity in the gap between the pole-​pieces is not uniform. Also, there are two regions
where the flux concentration may be great enough to saturate the iron. The flux
concentration at A can be reduced by rounding the corners of the pole-​pieces. The
concentration at B is caused by the increase in flux towards the roots of the pole-​
pieces caused by the fringing field. This effect can be reduced by tapering the pole-​
pieces so that the cross-​sectional area increases as the flux increases. It has been
shown that the saturation of the pole-​piece tips is almost uniform if they have
Rogowksi profiles [29]. This profile was originally devised to produce a uniform
electric field between electrodes for studies of voltage breakdown. The profile is
such that the electric field on the surface of an electrode is greatest on the axis
and decreases monotonically with increasing distance from it [35]. In practice
this profile can be approximated quite well by a circular arc and a taper, as shown
in Figure 19.18(a) [29]. If saturation of the pole-​piece does not occur, then the
region where the field is uniform can be increased by shaping the surface of the
pole-​piece, as shown in Figure 19.18(b). If the reluctance of the iron is neglected
then the number of ampere-​turns required to produce a given flux density in the
gap can be calculated directly. The flux density in the iron can be obtained from
the flux density in the gap multiplied by a factor to allow for the fringing field.
Then the additional ampere-​turns needed to overcome the reluctance of the iron
can be calculated if necessary. Further information on iron dominated magnets
can be found in [29, 36].
754 Magnets

Figure 19.18:  Pole-​piece design: (a) for uniform field in the pole-​piece, and (b) for uniform
field in the air gap.

19.7 Permanent Magnet Design

A permanent magnet system comprises one, or more, permanent magnets and may
include a soft iron yoke and pole-​pieces. The pole-​pieces may concentrate the flux
so that the flux density in the air gap is greater than the maximum which can be
achieved in the magnets. For initial design purposes it is possible to neglect the
reluctance of the soft iron parts of the circuit. The permanent magnets are then
designed to provide the magneto-​motive force required to produce the desired flux
in the air space, as described in Section 19.3.1. If the air space is between soft iron
pole-​pieces, which are not saturated, then its reluctance can be calculated using a
numerical solution of Laplace’s equation. It is usually important to minimise some
combination of the size, cost, and weight of the magnet.

19.7.1 Permanent Magnets for Magnetrons and CFAs

Early conventional magnetrons and CFAs employed Alnico magnets. The rela-
tively low energy products and low remanence of these materials required the
use of large and heavy magnets, as shown in Figure  19.19(a) [2]. The higher
energy products of ferrite, and rare earth, materials allow the use of smaller
magnets, as shown in Figure  19.19(b). In some cases there may be iron pole-​
pieces between the permanent magnets and the gap. Alternative versions of
both types of magnet exist with rotational symmetry about the vertical axis,
and the geometry may be annular instead of cylindrical. The general principles
of design are that the permanent magnet material should be located as close
to the air gap as possible, and that it should be operated close to its maximum
energy product.
To illustrate the design procedure for a magnet of this kind let us suppose that we
wish to generate a uniform flux density Bg between cylindrical magnets of radius R
separated by a gap g. The flux density in the magnets must be greater than that in
the air gap because of the fringing field. Figure 19.20 shows one quadrant of the
 Permanent Magnet Design 755

Figure 19.19:  Arrangement of permanent magnets for magnetrons and CFAs using

(a) Alnico, and (b) ferrite and rare earth magnets.

Figure 19.20:  Geometry of a dipole magnet showing the form of an approximate flux line.

system with dimensions normalised to the radius of the magnets. The fringing is
represented by approximate flux lines each comprising a circular arc and a straight
line as shown [9]. We will assume that the magneto-​motive force for a flux line is
given by

 h−z
M=  M (19.37)
 h − d  0
so that it is constant on the flat surface of the magnet ( z = d ) and varies linearly
along the curved surface ( r = 1). In the absence of fringing the flux generated by the
magnet is

µ0 π
Φ= M0 . (19.38)
d
The contribution to the fringing flux from an element of the surface dz is

 h−z 1
d Φ = 8µ 0M0  dz, (19.39)
 h − d  z + c1 g
756 Magnets

where c1 = ( 4 π ) − 1 and h is an initial estimate. Then the additional flux is

h
8µ M ⌠ h − z
∆Φ = 0 0  dz, (19.40)
h − d ⌡ z + c1 g
d

which can be integrated to give

 h + c1d   h + c1d  
∆Φ = 8µ 0M0   ln   − 1 . (19.41)
 h − d   d + c1d  
Dividing by Φ gives

∆Φ 8d  h + c1d   h + c1d  
=   ln − 1 . (19.42)
Φ π  h − d   d + c1d  
Since the flux is continuous the flux density in the magnet must be
 ∆Φ 
Bm = 1 +  Bg . (19.43)
 Φ 
We suppose that the operating point lies on a linear demagnetisation curve (or
recoil loop) defined by the intersections on the axes ( 0, Br ) and ( − H c ,0 ). Then
 B 
H m = 1 − m  H c (19.44)
 Br 
and, from the magnetic circuit law, we find that
 Bg 
h = 1 + d . (19.45)
 µ 0 H m 
The final value of h is found after a few iterations. Comparison between the results of
this calculation and those from finite element modelling shows good qualitative agree-
ment. Figure 19.21 shows, as an example, the flux plot and the variation of Bz with
radius at the mid-​plane of the gap for a typical case, using a samarium cobalt magnet,
for a target flux density in the gap of 0.4 T (pole radius 15 mm; gap 10 mm; magnet
thickness 4.2 mm). These calculations are implemented in Worksheet 19.2. It should be
noted that the theory given above assumes that the leakage flux is negligible. It is, in fact,
rather difficult to find approximate formulae for the leakage flux in practical geometries
and it is best to design the circuit so that the return flux path is kept well away from the
permanent magnets. The method of estimating the fringing flux can be adapted for
annular magnets, and for those fitted with cylindrical or conical pole-​pieces.

19.7.2 Permanent Magnets for Linear Beam Tubes

Magnet structures similar to those in Figure 19.19(b) have been used for small klys-
trons and TWTs but they are not well-​adapted to producing a long region of uni-
form field. An alternative arrangement is shown in Figure 19.22. This is suitable
for use with Alnico magnets where a long magnet is required to generate sufficient
 Permanent Magnet Design 757

Figure 19.21:  Results of finite element calculations for a dipole magnet (a) flux plot, and
(b) variation of Bz with r when z = 0.

Figure 19.22:  Alnico magnet for a linear-​beam tube: (a) general arrangement, and

(b) flux plot.

magneto-​motive force. The length of the magnet relative to the air space can be
increased, if necessary, by making it barrel-​shaped and by making the pole-​pieces
re-​entrant. If rare earth magnets are used then it may be necessary to create magnet
stacks with alternate layers of permanent magnet and soft iron to obtain the cor-
rect combination of flux and magneto-​motive force. The flux density on the axis is
758 Magnets

Figure 19.23:  Rare earth magnet for a linear-​beam tube: (a) general arrangement, and
(b) flux plot.

approximately uniform within the magnet but the sign is reversed outside the mag-
net. There is also a substantial leakage flux outside the magnet [1, 3]. An example
of a klystron focused in this way is given in [37].
Figure 19.23 shows an alternative arrangement using radially magnetised rings,
which is suitable for use with rare earth magnets [3, 38]. This suffers from the same
problems of flux uniformity and leakage as the previous example. The reversal of
the direction of the axial flux at the collector end of the magnet can be reduced by
using a single ring magnet at the gun end. The non-​uniformity of the axial flux, in
this and other magnets, can be reduced by including a number of transverse soft
iron plates as flux straighteners.
Further details of magnet design can be found in [1–​3]. The problems of flux
leakage and non-​uniform flux density can be overcome by cladding the magnet
system with additional magnets having appropriate directions of magnetisation
[39–​44].

19.7.3 Periodic Permanent Magnet (PPM) Systems

The theory of electron beam focusing by a periodic magnetic field is discussed in
Section 7.6. In this section we consider the problem designing the periodic perma-
nent magnet stack to produce the field required. This subject was considered by a
number of authors who developed graphical methods [45–​49]. The availability of
powerful personal computers means that the formulae used in the graphs can be
 Permanent Magnet Design 759

Figure 19.24:  Arrangement of a periodic permanent magnet stack

(copyright 2009, IEEE, reproduced, with permission, from [50]).

computed readily, so that graphical methods are no longer required. The discussion
here follows that of Santra et al. which has the widest applicability and has been
shown to give good results [50].
Figure 19.24 shows the arrangement and dimensions of a PPM structure. This
allows a number of common variations to be considered:
• The relationship between the outer radius of the pole-​pieces ( r3 ) and the outer
radius of the magnets ( r5 ). If r3 < r5 then the external leakage flux is reduced,
and the peak flux density on the axis is increased, for a given magnet. However,
it is then necessary to introduce some non-​magnetic material to increase the
overall radius of the pole-​piece if it is to be cooled by conduction. This prob-
lem is avoided if r3 > r5 , although the peak magnetic flux density on the axis
is reduced, and the leakage flux is increased. This arrangement is also better
adapted to the addition of small pieces of soft iron (shunts) to the outside
of the stack to adjust its properties when hot-​testing. The third possibility
(r3 = r5 ) is said to keep the variations in the field at the ends of the stack to a
minimum [48].
• The relationship between the inner radius of the magnets ( r4 ) and the outer radius
of the ferrule ( r2 ). In helix TWTs it is possible to make the inner radius of the mag-
nets only slightly greater than the outer radius of the ferrule. This corresponds to
the good design practice of placing the permanent magnet as close to the gap as
possible. In coupled-​cavity TWTs the ferrule forms part of the RF structure (see
Figure 4.29) while the magnets must be placed outside the structure.
• Some PPM structures have ferrules, whereas others do not. The choice is deter-
mined partly by the ratio of the gap length to the pitch of the structure which
gives the greatest fundamental component of the flux density on the axis.
However, it may be constrained by compatibility with the dimensions of the RF
structure.
The model used here allows for all these possible variations.
760 Magnets

Figure 19.25:  Flux paths in a periodic permanent magnet stack

(copyright 2009, IEEE. reproduced, with permission, from [50]).

The possible flux paths in air around the structure are illustrated in Figure 19.25.
Because these paths are mostly in parallel with one another it is convenient to
work in terms of permeance ( P ) which is the reciprocal of the reluctance defined
in (19.20). The calculation of the permeances is based on treating the surface of
the pole-​piece as a magnetic scalar equipotential so that the same magneto-​motive
force acts along each path. The case shown in Figure 19.25(a) was discussed in [46]
and only the results are given here. It is assumed that the flux density is uniform in
the gap between the ferrules ( g ). We have seen, in discussions of the RF electric field
in a gap, that this is quite a good approximation if the ends of the ferrules are blunt
(see Section 3.5.3). Then, in SI units,

µ 0 π ( r22 − r12 )
P1 = , (19.46)
g
∞
4µ 0 Lr1 I1 ( 2 nπ r1 L )  nπ g 
P2 = ∑ ⋅
I 0 ( 2 nπ r1 L )
⋅ sin 
 L 
, (19.47)
n = 1,3,5 π n g
2

and
∞
4µ 0 Lr3 K1 ( 2 nπ r3 L )  n πT 
P3 = ∑ ⋅
K 0 ( 2 nπ r3 L )
⋅ sin 
 L 
, (19.48)
n = 1,3,5 π n T
2

where I 0 , I1 , K 0 and K1 are modified Bessel functions and the dimensions of the struc-
ture are defined in Figure 19.24. The additional permeances in Figure 19.25(b) are

µ 0 π ( r42 − r22 )
P4 = (19.49)
T
 Permanent Magnet Design 761

and

µ 0 π ( r32 − r52 )
P5 = . (19.50)
T
Finally, the indented pole pieces in Figure 19.25(c) add permeances given by
2µ 0 πt p
P6 =
ln ( r3 r5 ) (19.51)
in series with P3 so that the modified permeance of this flux path is
P3 P6
P3 m = . (19.52)
2 P3 + P6
Note that this equation is given incorrectly in [50]. The total permeance Pt is found
by summing the appropriate permeances in each case. Then the slope of the load
line is
T
K= Pt . (19.53)
π ( r − r42 )
5
2

If rare-​earth magnets are used then the demagnetisation curve is linear and the
operating point ( H d , Bd ) is given by
Br
Hd = , (19.54)
Br
K+
Hc
where Br and H c are the remanence and coercive force of the material. Ferrite and
Alnico magnets must be stabilised so that the operating point is on a minor loop
[48]. We have seen that minor loops are effectively linear so that (19.54) can be used
in this case also if Br and H c are taken to be the intersections of the projections of
the minor loop on the two axes. Finally, the axial component of the magnetic flux
density in the beam hole of an infinite PPM stack is given by
∞
4 Bg sin ( nπ g L ) I 0 ( 2 nπr L )  2 nπ z 
Bz ( r, z ) = ∑ ⋅ ⋅ cos  ,
n = 1,3,5 nπ I 0 ( 2 nπr1 L )  L  (19.55)

where Bg = µ 0 H d T g . Equations (19.46) to (19.55) are the working equations in

Worksheet 19.3.
The design of a PPM stack starts from the period of the magnetic field ( L ) and
the peak flux density on the axis ( Bz ( 0, 0 )). The inner radius of the ferrule ( r1 ) is
normally fixed by the outer radius of the shield of a helix slow-​wave structure, or
by the beam hole radius of a coupled-​cavity slow-​wave structure. The flux density
in the gap ( Bg ) can be determined from (19.55) as a function of the gap length ( g )
which can be chosen for the best results if it is not constrained by the dimensions
of the RF structure. The outer radius of the ferrule ( r2 ) needs to be great enough to
avoid saturation of the ferrule. In practice this means that the flux density must be
less than 1T (see Figure 19.4). The choice must also be compatible with the design
762 Magnets

of the RF structure. The inner radius of the magnet ( r4 ) will normally be taken to be
as small as possible. There then remain only three dimensions to be chosen namely
the outer radius of the magnet ( r5 ), the outer radius of the pole-​piece ( r3 ), and the
thickness of the pole-​pieces (t p ). If the two radii are set to the same value then the
number of free parameters is reduced to two. The thickness of the pole-​piece must
be great enough to avoid saturation. Subject to these restrictions the free param-
eters can be chosen to minimise the volume of the magnets. Final optimisation
of the design requires the use of computational electromagnetics software. That
analysis can also investigate the fields at the ends of the stack, and the effects of any
gaps which are required to allow for the passage of the RF input and output con-
nections [51]. Methods similar to those described above can be used to analyse and
design long-​period PPM structures (see Section 7.6.1) and those where additional
permanent magnets have been added to reduce the leakage flux [44, 52–​54].

References

[1]‌ J. F. Gittins, Power Travelling-​Wave Tubes. English Universities Press, 1965.

[2]‌ W. Harrold and W. Reid, ‘Permanent magnets for microwave devices’, IEEE Transactions
on Magnetics, vol. 4, pp. 229–​239, 1968.
[3]‌ M. J. Smith and G. Phillips, Power Klystrons Today. Taunton, UK: Research Studies
Press, 1995.
[4]‌ R. G. Carter, Electromagnetism for Electronic Engineers. Available at:  http://​book-
boon.com/​en/​electromagnetism-​for-​electronic-​engineers-​ebook (accessed 5 October
2017), 2010.
[5]‌ G. W. Carter, The Electromagnetic Field in its Engineering Aspects, 2nd ed. London:
Longman, 1967.
[6]‌ B. I. Bleaney and B. Bleaney, Electricity and Magnetism. London: Oxford University
Press, 1957.
[7]‌ M. MacCaig, Permanent Magnets in Theory and Practice. London: Pentech Press, 1987.
[8]‌ A. Edwards, ‘Magnet design and selection of material’, in D. Hadfield, ed., Permanent
Magnets and Magnetism. London: Iliffe, pp. 191–​296, 1962.
[9]‌ W. L. Upson and E. L. Furth, ‘Abridgment of magnetic leakage and fringing flux calcu-
lations’, Journal of the A.I.E.E., vol. 47, pp. 340–​344, 1928.
[10] J. Sykulski, Computational Magnetics. Springer Science & Business Media, 2012.
[11] Infolytica Corp., ‘Gemini’, Montreal, Canada, 1966.
[12] P. Campbell, Permanent Magnet Materials and Their Application. Cambridge University
Press, 1996.
[13] H. R. Kirchmayr, ‘Permanent magnets and hard magnetic materials’, Journal of Physics
D: Applied Physics, vol. 29, p. 2763, 1996.
[14] R. G. Carter, Electromagnetism for Electronic Engineers, 2nd ed. London: Chapman &
Hall, 1992.
[15] G. W. C. Kaye and T. H. Laby, Tables of Physical and Chemical Constants. Available
at: www.kayelaby.npl.co.uk/​(accessed 5 October 2017), 2014.
[16] Magdev Ltd, Permanent Magnets. Available at: www.magdev.co.uk/​permanent-​magnets
(accessed 5 October 2017), 8 August 2017.
 References 763

[17] Delta Magnets, Permanent Magnets. Available at:  www.magdev.co.uk/​ permanent-​

magnets (accessed 5 October 2017), 2011.
[18] A. J. Bamford et al., ‘A 1-​MW L-​band coaxial magnetron with separate cavity’, IEEE
Transactions on Electron Devices, vol. 14, pp. 844–​851, 1967.
[19] D. Bhasavanich and H. G. Hammon, III, ‘Pulsed magnet system considerations for
repetitive operation of high-​power magnetrons’, IEEE Transactions on Electron Devices,
vol. 38, pp. 791–​795, 1991.
[20] Y. Dai et al., ‘Design study on A 9.2 T NbTi superconducting magnet with long-​length
uniform axial field’, IEEE Transactions on Applied Superconductivity, vol. 25 (3), June,
paper no. 4101204, 2015.
[21] R. Hirose et al., ‘Development of 7 T cryogen-​free superconducting magnet for gyro-
tron’, IEEE Transactions on Applied Superconductivity, vol. 18, pp. 920–​923, 2008.
[22] W. R. Smythe, Static and Dynamic Electricity, 3rd ed. New York: McGraw-​Hill, 1950.
[23] L. McKeehan, ‘Combinations of circular currents for producing uniform magnetic
fields’, Review of Scientific Instruments, vol. 7, pp. 150–​153, 1936.
[24] M. W. Garrett, ‘Axially symmetric systems for generating and measuring magnetic
fields. Part I’, Journal of Applied Physics, vol. 22, pp. 1091–​1107, 1951.
[25] R. G. Carter, ‘Coil-​system design for production of uniform magnetic-​fields’, Proceedings
of the Institution of Electrical Engineers-​London, vol. 123, pp. 1279–​1283, 1976.
[26] W. Franzen, ‘Generation of uniform magnetic fields by means of air-​core coils’, Review
of Scientific Instruments, vol. 33, pp. 933–​938, 1962.
[27] W. Gosling and M. J. Cunningham, ‘Gapped solenoid as a means of producing a highly
uniform magnetic field over an extended volume’, Proceedings of the Institution of
Electrical Engineers, vol. 121, pp. 1589–​1593, 1974.
[28] D. B. Montgomery, Solenoid Magnet Design. New York: Wiley-​Interscience, 1969.
[29] S. Russenschuck, Field Computation for Accelerator Magnets: Analytical and Numerical
Methods for Electromagnetic Design and Optimization. John Wiley & Sons, 2011.
[30] A. Staprans et al., ‘High-​power linear-​beam tubes’, Proceedings of the IEEE, vol. 61, pp.
299–​330, 1973.
[31] W. G. Worcester et al., ‘Light-​weight aluminum foil solenoids for traveling-​wave tubes’,
IRE Transactions on Electron Devices, vol. 3, pp. 70–​74, 1956.
[32] R. C. Glass and P. Gottfeldt, ‘Aluminum foil solenoids for traveling-​wave tubes’, IRE
Transactions on Electron Devices, vol. 4, p. 186, 1957.
[33] E. A. Perigo et  al., ‘Design of a multi-​sections solenoid for power microwave tubes’,
in IEEE MTT-​ S International Conference on Microwave and Optoelectronics,
pp. 233–​236, 2005.
[34] D. Borodin and M. Einat, ‘Copper solenoid design for the continuous operation of a
second harmonic 95-​GHz gyrotron’, IEEE Transactions on Electron Devices, vol. 61,
pp. 3309–​3316, 2014.
[35] J. Harrison, ‘A computer study of uniform-​field electrodes’, British Journal of Applied
Physics, vol. 18, p. 1617, 1967.
[36] J. T. Tanabe, Iron Dominated Electromagnets:  Design, Fabrication, Assembly and
Measurements. Singapore: World Scientific, 2005.
[37] J. Lebacqz, ‘High power klystrons’, IEEE Transactions on Nuclear Science, vol. 12,
pp. 86–​95, 1965.
[38] C. Fu et al., ‘A type of tube-​shaped permanent magnet focusing system of high-​power
klystron’, in Asia-Pacific Microwave Conference Proceedings, vol. 1, pp. 333–​335, 1997.
764 Magnets

[39] J. P. Clarke and H. A. Leupold, ‘Shaping of cylindrically symmetric magnetic fields with
permanent magnets’, IEEE Transactions on Magnetics, vol. 22, pp. 1063–​1065, 1986.
[40] Y. Fuwa et  al., ‘Focusing system with permanent magnets for klystrons’, IEEE
Transactions on Applied Superconductivity, vol. 24, pp. 1–​5, 2014.
[41] D. K. Abe and J. P. Calame, ‘Advanced materials technologies’, in R. J. Barker et al., eds,
Modern Microwave and Millimeter-​Wave Power Electronics. Piscataway, NJ: IEEE Press,
pp. 649–​689, 2005.
[42] H. Leupold et  al., ‘Permanent magnet sources for extended interaction klystrons’,
Journal of Applied Physics, vol. 70, pp. 6624–​6626, 1991.
[43] H. Leupold and E. Potenziani II, ‘Augmentation of field uniformity and strength in
spherical and cylindrical magnetic field sources’, Journal of Applied Physics, vol. 70,
pp. 6621–​6623, 1991.
[44] Q. L. Peng et  al., ‘Axial magnetic field produced by axially and radially magnet-
ized permanent rings’, Journal of Magnetism and Magnetic Materials, vol. 268,
pp. 165–​169, 2004.
[45] F. Sterzer and W. Siekanowicz, ‘The design of periodic permanent magnets for focusing
of electron beams’, RCA Review, vol. 18, pp. 39–​59, 1957.
[46] J. E. Sterrett and H. Heffner, ‘The design of periodic magnetic focusing structures’, IRE
Transactions on Electron Devices, vol. 5, pp. 35–​42, 1958.
[47] M. Schindler, ‘The magnetic field and flux distributions in a periodic focusing stack for
traveling-​wave tubes’, RCA Review, vol. 21, pp. 414–​436, 1960.
[48] M. Schindler, ‘An improved procedure for the design of periodic-​ permanent-​
magnet assemblies for traveling-​wave tubes’, IEEE Transactions on Electron Devices,
pp. 942–​949, 1966.
[49] M. Corsi et al., ‘Graphical method for optimum design of periodic magnetic focusing
structures’, IEEE Transactions on Electron Devices, vol. 21, pp. 2–​11, 1974.
[50] M. Santra et al., ‘An improved analysis of PPM focusing structures including the effect
of magnetic saturation in the iron pole pieces’, IEEE Transactions on Electron Devices,
vol. 56, pp. 974–​980, May 2009.
[51] C. L. Kory, ‘Effect of geometric azimuthal asymmetries of PPM stack on elec-
tron beam characteristics [TWTs]’, IEEE Transactions on Electron Devices, vol. 48,
pp. 38–​44, 2001.
[52] H. A. Leupold and J. P. Clarke, ‘Bulk reduction and field enhancement in peri-
odic permanent-​magnet structures’, IEEE Transactions on Electron Devices, vol. 34,
pp. 1868–​1872, August 1987.
[53] H. A. Leupold et al., ‘Lightweight permanent magnet stack for traveling-​wave tubes’,
Journal of Applied Physics, vol. 67, pp. 4656–​4658, 1990.
[54] M. Santra et al., ‘Analysis of long-​periodic permanent magnet structures for electron
beam focusing’, IEEE Transactions on Electron Devices, vol. 60, pp. 1776–​1781, 2013.
20 System Integration

20.1 Introduction

The previous chapters of this book have provided a detailed discussion of the prin-
ciples of vacuum tube amplifiers and oscillators. The purpose of this chapter is to
review the factors affecting the satisfactory integration of a tube into the system of
which it is a part. The principal elements of a high power RF system employing a
vacuum tube amplifier are shown in Figure 20.1. The arrangement is similar if the
tube is an oscillator, except that there is no low-​level RF system.
The key to satisfactory operation of any system is good integration of all the
parts. Usually the different elements are obtained from a number of sources and
successful system integration requires good communication between all the engin-
eers concerned. This is particularly important where vacuum tubes are concerned
because the knowledge bases of the tube and the system engineers are not usu-
ally the same (see Section 1.5) [1]. Some of the interface problems can be avoided
if a single organisation takes responsibility for the design and manufacture of a
packaged amplifier incorporating the tube and its subsystems. A good example is
provided by the microwave power module in which a solid-​state driver amplifier, a
miniature TWT and its power supply are combined into a single unit [2].
The sections that follow consider how each of the blocks in Figure 20.1 contrib-
utes to the successful operation of the system. A detailed analysis of the nature and
causes of reliability problems in tube-​based systems can be found in [3]. Although
great advances have since been made in tube manufacturers’ quality assurance it is
to be suspected that many of the other problems identified still remain. An over-
view of the factors affecting the reliability of systems incorporating vacuum tubes
is in given in Section 20.9. More detailed information can be found in [4–​6] and in
other sources cited in this chapter.

20.2 DC Power Supplies

The high-​voltage supply is the source of DC power to the tube, and any instability
or noise in it is likely to lead to degradation of the tube performance. There may be
combinations of voltage and current which must be avoided because the tube does
not work correctly (see Sections 14.3.8 and 15.4.1) [7]. Some tubes have regions
with negative resistance characteristics, and the impedance of the power supply
766 System Integration

Figure 20.1:  Block diagram of a typical vacuum tube amplifier.

must then be chosen correctly if oscillations are not to occur. When the tube is
pulsed that may be achieved by applying a pulsed cathode voltage (see Section 20.3)
or by using a DC high-​voltage supply and applying a pulsed voltage to a control
electrode (see Section 9.5). The design of the high-​voltage supply is more compli-
cated if the tube has a multi-​element depressed collector. The collector electrodes
are connected to taps on the high-​voltage supply, and these intermediate voltages
must be kept steady when the currents drawn change with the operating conditions
of the tube. The possible voltages at the taps of the high-​voltage supply may be
determined by its design, and that places constraints on the design of the collector.
In addition to the main high-​voltage supply there may be DC supplies at lower
voltages for the cathode heater, electromagnets, control electrodes, and ion pumps.
If these supplies are not accurately calibrated, and well-​regulated, then malfunc-
tion, and possibly damage, of the tube may result. The heater power is important
in ensuring the correct emission from the cathode. Noise in the heater power supply
can cause additional noise in the output from the tube. Incorrect magnet currents
and control voltages may lead to imperfect control of the electrons leading to add-
itional power dissipation within the tube, and possible damage to internal struc-
tures. All the power supplies draw their power from an external AC or DC source
which may not be well-​regulated, and may exhibit voltage transients.

20.2.1 High-​Voltage Switches

All power supplies rely on electronic high-​voltage switches for their operation.
These devices can be grouped into three classes:
• switched on and off by the applied voltage;
• switched on by a control signal, and off by the applied voltage;
• switched on and off by a control signal.
 DC Power Supplies 767

Table 20.1:  Classification of electronic high-​voltage switches

Switch on Switch off Semiconductor Vacuum Gas discharge

Applied voltage Applied voltage Diodes Diodes

Control signal Applied voltage Thyristors Thyratrons
Spark gaps
Ignitrons
Control signal Control signal MOSFETS Triodes
IGBTs Tetrodes
GTOs Hobetron [15]
IGCTs [14]

Semiconductor devices switching up to 8 kV are now used in all DC supplies, but

switches using vacuum or gas discharge technology are to be found in older systems
[8]. Thyratrons (triggered gas discharge tubes) which can work at voltages up to 200
kV and peak currents of 100 kA are still used in many pulsed supplies [9, 10], but
semiconductor-​based alternatives are being developed [11–​13]. Table 20.1 shows a
classification of electronic high-​voltage switches.

20.2.2 Load Impedance

The impedance presented to the high voltage power supply by the tube depends
upon the type of tube and the operating conditions. The DC current drawn by a
gridded tube operated in class B or C, or by an IOT, depends upon the RF input
power (see Figure 12.21(b)). The examples in Chapter 12 show that typical values
of the DC load impedance are 1 kΩ for a tetrode, and greater than 25 kΩ for an
IOT. The load impedance presented by a linear-​beam tube with a space-​charge lim-
ited gun depends upon the perveance ( K ) and the anode voltage (Va ) as

1
RL = . (20.1)
K Va
Figure  20.2 shows the impedance for typical values of voltage and perveance.
If the current is reduced from the space-​charge limited value by the voltage on
a control grid, or by temperature limited operation (as in a gyrotron), then the
impedance is increased. The operating points of both gridded tubes and linear-​
beam tubes are determined by the anode voltage so the power supply should
approximate to an ideal voltage source. In models of the tube with its power
supply, a linear-​beam tube may be represented by a fixed resistor in series with a
perfect diode.
Crossed-​field tubes draw no current until the threshold voltage (VT ) is exceeded.
Thereafter the voltage varies slowly for a large variation in current (see Figures 15.27
and 16.8) so that the incremental impedance is of the order of 100 Ω. The operating
point is determined by the current so the power supply should approximate to an
768 System Integration

Figure 20.2:  D.C. input resistance of tubes with space-​charge limited cathodes.

ideal current source. A crossed-​field tube can be modelled by a diode and a fixed
resistance in series with a DC voltage source [16, 17]. The current drawn is then
given by

Va − VT
I0 = , (20.2)
RL
where Va is the anode voltage, VT is the threshold voltage and RL is the incremental
resistance.
For all tubes the regulation of the power supply must be able to handle the vari-
ation in the impedance during start-​up as well as in normal operation. Figure 20.3
compares the static characteristics of typical linear-​beam, and crossed-​field, tubes
including a linear approximation for a linear-​beam tube. In practice this idealised
behaviour is modified by the inter-​electrode capacitance of the tube and by the
parasitic inductance of the cables connecting it to the power supply.

20.2.3 Electric Power Converters

The stable high voltage supply is provided by an electric power converter [14, 18].
Converters typically have the bridge configuration shown in Figure 20.4 where the
switches represent semiconductor power devices.
According to the type of switches, and the way in which they are controlled, the
basic converter can be used to convert power from DC or AC sources to supply DC,
or variable frequency AC, power to a load. The output voltages of circuits using
devices other than diodes can be controlled by the timing of the switching. The
simplest example of an AC to DC converter is the single-​phase bridge rectifier in
which the switches are diodes as shown in Figure 20.5(a). The output waveform is
shown in Figure 20.5(b). Referring to Figure 20.4 we see that when the input volt-
age is positive switches A and D are closed, and when the input voltage is negative
switches B and C are closed. Thus the voltage applied to the load always has the
 DC Power Supplies 769

Figure 20.3:  Comparison between the static characteristics of typical tubes: linear-​beam

( K = 1 µperv, RL ≈ 4 kΩ) and crossed-​field (VT = 20 kV, RL = 100 Ω).

Figure 20.4:  Generic single phase bridge converter.

Figure 20.5:  Single-​phase diode bridge rectifier: (a) circuit diagram, (b) unsmoothed output

waveform showing the mean voltage.
770 System Integration

Figure 20.6:  Three-​phase bridge rectifier: (a) circuit diagram and (b) output waveform

showing the mean voltage.

same polarity. The working voltages of individual switching devices are typically
1–​5 kV depending upon the type. Higher voltages are obtained by stacking con-
verter units in series, or by using voltage multiplier circuits. The properties of single-​
phase bridge converters of different kinds can be explored using Worksheet 20.1.
The output voltage provided by a power converter is modulated by the switch-
ing frequency and its harmonics. The ripple on the output of the rectifier can be
reduced by employing a three-​phase bridge circuit, as shown in Figure 20.6(a). This
reduces the amplitude of the ripple, and increases the ripple frequency, as shown
in Figure 20.6(b). In order to obtain a nearly constant voltage the output must be
passed through a low pass filter, comprising a series inductor and shunt capacitor,
as shown in Figure  20.6(a). Provided that the reactance of the capacitor (1 ωC )
is small compared with the load resistance, the ripple in the voltage at the load is
related to the ripple in the voltage at the output of the rectifier by

1
V1,L = V1,R , (20.3)
1 − ω 2 ω 20

where ω 0 = 2 π f0 = 1 LC and V1,R and V1,L are the amplitudes of the first harmonic
at the output of the rectifier, and the output of the filter, respectively.
A typical three-​phase power supply of this kind, capable of delivering 16 A at
52 kV with 0.7% ripple at 300 Hz, is described in [19]. The smoothing circuit comprises
a 2 H inductor and a 24 µF capacitor so that f0 = 23 Hz. Substitution of these
numbers into (20.3) shows that the filter reduces the amplitude of the ripple at the
output of the rectifier by a factor of 0.006. The energy stored in the capacitor is 32
kJ which is sufficient to cause damage to the tube if an arc occurs. It is therefore
necessary to provide an additional fast acting switch (known as a crowbar), in par-
allel with the capacitor, to divert the energy so that it can be dissipated harmlessly
under fault conditions. Capacitors are available with capacitances up to 2 μF and
working voltages of 100 kV [12, 20].
 Pulse Modulators 771

The sizes of the smoothing inductor and capacitor and, therefore, the magni-
tude of the energy stored in the capacitor, can be further reduced by converting
the AC input to the rectifier to single phase AC at a higher frequency [21]. This AC
to AC conversion is accomplished using a bridge converter connected to a circuit
which is resonant at the higher AC frequency. An alternative topology employs a
number of converters connected in such a way that their collective output has a
high-​frequency ripple so that only a small smoothing capacitor is required [11].
Further information on electronic power converters for vacuum tubes can be
found in [5, 22].
The current drawn from the power supply may vary according to the operating
conditions of the tube, causing variations in the voltage as a result of the internal
impedance of the supply. This is particularly true when tubes with a DC high-​
voltage supply are pulsed by a voltage applied to a control electrode. The variation
in voltage can be controlled by a post regulator which is a gridded tube, or a tran-
sistor, in series with the tube. The voltage drop across this component can be varied
rapidly to compensate for changes in the voltage as the current changes [5].

20.3 Pulse Modulators

Many tubes, especially those for radar and for particle accelerators, are pulsed by
applying voltage pulses to the cathode. The pulse duration can be from a few micro-
seconds up to several milliseconds, and the pulse repetition frequency from a few
pulses per second up to a few kilohertz. The product of the pulse duration and
the pulse repetition frequency, known as the duty cycle, is typically of the order of
0.001. Pulse modulators work by storing energy over a period of time and then dis-
charging it into the load [17]. The mean power delivered by the DC power supply is
then the product of the peak power and the duty cycle. The energy may be stored in
either a capacitance or an inductance. Comparison between capacitive and induct-
ive energy storage shows that the energy density in the latter can be greater by a
factor of about one hundred. However the specification for the switches for an
inductive system is considerably more demanding, and modulators of this type are
not widely used with vacuum tubes [10]. Further information about magnetic pulse
modulators can be found in [23, 24]. The discussion below is restricted to capacitive
energy storage.
Figure 20.7 shows two basic arrangements for a pulsed modulator with a capaci-
tive energy store. It is usual for the anode, and the tube body, to be earthed because
that makes it easier to make RF connections and provide cooling. During the stor-
age period the switch is open and the stored energy is replenished from the DC sup-
ply. Then, while the switch is closed, all or part of the stored energy is transferred to
the tube. It is not essential for the pulses to occur at regular intervals, provided that
the minimum inter-​pulse period is sufficient for the energy store to be recharged
completely. The purpose of the isolating element is to restrict the current drawn
from the power supply during the pulse.
772 System Integration

Figure 20.7:  Pulsed modulators with a capacitive energy store: (a) direct coupled, and
(b) capacitor coupled.

The shape of the voltage pulse can have important effects on the performance of
the tube. Excessive voltages of either polarity can cause voltage breakdown. Ripple
on the pulse top can produce temporal variations in power output, in the gain and
phase of an amplifier, and in the frequency of an oscillator. Unwanted oscillations
may occur if the rise and fall times of the pulse are too slow or if the over-​swing
following the rise of the pulse is too great (see Figure 20.13).

20.3.1 Active-​Switch Modulator with a Resistive Load

When the energy store is a capacitor the voltage across it rises slowly during the
inter-​pulse period, and decays exponentially during the pulse, at a rate set by the
storage capacitance and the resistance of the load. Since the voltage across the load
should be nearly constant during the pulse, the switch must be opened again after a
time which is short compared with the decay constant of the circuit. Therefore the
switch must be of a type which can be both opened and closed by a control signal
(see Table 20.1). The pulse duration is variable because the switch can be opened at
any time. Modulators of this type are known as active-​switch or hard-​tube modula-
tors. As only a small fraction of the energy stored in the capacitor is delivered to
the load during the pulse it is necessary to provide a crowbar circuit to discharge the
capacitor safely under fault conditions. The isolating element can be either a high
resistance or an inductor.
The pulse shape delivered to a resistive load can be found by considering the
equivalent circuit shown in Figure  20.8(a). The resistance of the load is RL , the
residual resistance of the switch in the on state is RS , and the capacitor CS represents
 Pulse Modulators 773

the stray capacitance of the load, the switch, and the connecting cables. During the
pulse the voltage across the storage capacitor varies as

VC (t ) = V0 exp ( −αt ) , (20.4)

where V0 is the initial voltage,

1
α= , (20.5)
(RL + RS )C
and the capacitance of the storage capacitor is C. If the switch is closed at t = 0 the
voltage across the load, during the rise of the pulse, is given by

RL
VL (t ) = V0 exp ( −αt ) ⋅ (1 − exp ( −α1t )) , (20.6)
RL + RS
where

RL + RS
α1 = (20.7)
RL RSCS
and it has been assumed that the rise-​time is small compared with the duration of
the pulse. During the fall of the pulse the stray capacitance is discharged through
the load and the voltage across the load is

RL
VL (t ) = V0 exp ( −αt1 ) ⋅ exp ( −α 2 (t − t1 )) , (20.8)
RL + RS
where

1
α2 = , (20.9)
RLCS

and t1 is the time at which the switch is opened. Note that the fall of the pulse is
much slower than the rise (see Figure 20.8(b)). The long fall time can be reduced
by adding a ‘tail-​biter’ switch, in parallel with the load, which is closed at the same
moment that the main switch is opened [5]. It is customary to define the rise time
of a pulse as the time taken for the voltage to change from 10% to 90% of its final
value. The fall time is defined in the same way. The peak voltage is less than the
initial voltage on the storage capacitor because of the voltage drop in the switch. It
follows that some of the power input to the modulator is dissipated as heat in the
switch so that the efficiency of the modulator is less than 100%. If the switching
device is a vacuum triode, or a transistor, it is possible to operate in its active region
to control the current drawn from the storage capacitor. But the penalty for this is
greater power dissipation in the active device and, therefore, reduced modulator
efficiency. The voltage on the pulse top falls slowly during the pulse as the capacitor
is discharged. The extent of the fall is known as the droop of the pulse. The droop
can be reduced by increasing the capacitance of the storage capacitor. Alternatively
774 System Integration

Figure 20.8:  Pulse generation by capacitor discharge: (a) equivalent circuit diagram for the
rise and fall of the pulse, and (b) a typical pulse shape plotted with the voltage normalised
to the initial voltage on the storage capacitor and time normalised to the pulse duration.

a bouncer circuit which provides a compensating ramped voltage can be inserted

between the load and earth [17]. This approach has been used to compensate for a
20% droop in a 2 msec pulse, leading to a considerable reduction in the energy to be
stored in the capacitor [25–​27].

20.3.2 Active-​Switch Modulator with a Biased Diode Load

When the load is a crossed-​field tube, modelled by a biased diode, the equivalent cir-
cuit for the rise and fall of the pulse is as shown in Figure 20.9(a). The switch S1 con-
trols the pulse, and the biased diode model of the load is represented by S2 together
with the voltage VT . The additional impedance ZR is necessary to allow the stray cap-
acitance to be discharged at the end of the pulse because the tube stops conducting
before the voltage has fallen to zero. In a capacitor-​coupled modulator it also provides
a recharging path for the storage capacitor. In the simplest case this component is a
resistor RR. For purposes of analysis the circuit can be redrawn in terms of current
sources as in Figure 20.9(b). Switch S2 is closed when VL > VT and open otherwise.
The voltage across the storage capacitor can be approximated by

VC (t ) = V0 exp ( −α ′t ) , (20.10)

where

1
α′ = , (20.11)
( 1 RS )C
R +
and

RL RR
R1 = . (20.12)
RL + RR
 Pulse Modulators 775

Figure 20.9:  Equivalent circuits for an active-​switch modulator with a biased diode load:
(a) with voltage sources, and (b) with equivalent current sources.

The rise of the pulse is analysed in two steps [17]:

i) From t = 0 until t = t1 when VL = VT

RR
VL (t ) = V0 exp ( −α ′t ) ⋅ (1 − exp ( −α1t )) , (20.13)
RR + RS

where

RR + RS
α1 = , (20.14)
RR RSCS

and it is assumed that the rise time is short compared with the duration of the pulse.
ii) From t = t1 until S1 is opened when t = t2

V V 
VL (t ) =  0 exp ( −α ′t ) + T  R2 1 − exp ( − α 2 (t − t1 )) + VT exp ( −α 2 (t − t1 )) ,
 RS RL 
(20.15)

where

1 1 1 1
= + + , (20.16)
R2 RS RR RL

and

1
α2 = . (20.17)
R2CS
776 System Integration

If the droop is small then the voltage on the pulse top is

V V 
VL (t ) =  0 + T  R2 . (20.18)
 S RL 
R

The fall of the pulse is analysed in a similarly manner:

iii) From t = t2 until t = t3 when VL = VT

RR
VL (t ) = VT 1 − exp ( −α 3 (t − t2 )) + VL (t2 ) exp ( −α 3 (t − t2 )) , (20.19)
RL + RR 
where

1
α3 = . (20.20)
R1CS
iv) From t = t3 onwards

VL (t ) = VT exp ( −α 4 (t − t3 )) , (20.21)

where

1
α4 = . (20.22)
RRCS

Figure 20.10 shows a typical waveform calculated in this way. The pulse top lies
only just above the threshold voltage and the capacitance of the storage capacitor
must be large enough so that the droop is small and the voltage does not fall below
VT before the end of the pulse. The losses incurred by the use of the resistor RR can
be reduced by replacing it by an inductor [17]. However, the storage of energy in
the inductor during the fall of the pulse can cause the decaying voltage to become
oscillatory so that its polarity is periodically reversed. It is undesirable for a large
reverse voltage to be imposed on the tube, and an ‘over-​swing’ diode may be added
in parallel with the tube to provide a path for the reverse current.
During the pulse an active switch modulator can be regarded as a voltage source
whose source resistance is the forward resistance of the switch. The current through
the switch is approximately equal to that through the load. Thus, to minimise the
power dissipated in the switch, it is desirable for the switch resistance to be small
compared with the load resistance. The usual principle of matching the load to the
source for maximum energy transfer is not relevant in this case. However, there are
circumstances in which it is useful to place a pulse transformer between the modu-
lator and the load. Chief among these is the use of a transformer to provide a high
voltage output pulse from a lower voltage modulator. This provides one way of
obtaining high voltage pulses from a modulator employing solid state switches. The
disadvantages of using a transformer are: the increased current and switch losses in
the modulator; losses in the transformer itself; and degradation of the pulse shape
 Pulse Modulators 777

Figure 20.10:  Typical voltage pulse for an active-​switch modulator connected to a biased

diode load.

caused by reactances in the transformer. An alternative approach is the Marx bank

modulator in which a number of capacitors are charged in parallel and then dis-
charged in series [28, 29]. A good comparison between different modulator topolo-
gies is given in [27] and examples of modulator design in [25, 30]. For discussion of
the design of pulse transformers see [16, 17, 31–​33].

20.3.3 Line-​Type Modulators

An alternative way of generating a flat-​topped pulse is to charge a length of TEM
transmission line and then discharge it by connecting it to the load, as shown in
Figure 20.11(a). Initially both switches are open and the line in uncharged. Switch
S1 is closed for long enough for the line to charge to the source voltage V0. When
S2 is closed the line appears to the load as a voltage source V0 with an impedance
equal to the characteristic impedance of the line Zc . The load voltage rises instant-
aneously to a value given by the potential divider rule. If the load is matched to
the line ( RL = Zc ) then the voltage across the load is 0.5V0. The change in voltage
propagates along the line towards the source. If the line is dispersionless then the
change in voltage propagates as a step having amplitude −0.5V0 and phase velocity
v p . The voltage on the line is the sum of the initial voltage and this step wave, as
shown in Figure 20.11(b). When the wave reaches the start of the line it encoun-
ters an open circuit and is reflected as a wave having amplitude −0.5V0 , as shown
in Figure 20.11(c). The time taken for the reflected step wave to reach the load is
τ = 2l v p where l is the length of the line. At that moment the wave is absorbed by
the load and the voltage falls to zero. Thus, the load experiences a voltage pulse of
amplitude 0.5V0 and a duration τ which is fixed by the electrical length of the line.
If the load is not matched to the line then the reflected wave is reflected by the load
with an amplitude which satisfies the boundary conditions. As a result the pulse
778 System Integration

Figure 20.11:  Principle of operation of a line-​type modulator: (a) circuit diagram, (b) voltage

distribution on the line during the first half of the pulse, and (c) voltage distribution on the
line during the second half of the pulse.

does not end cleanly and the subsequent voltage may be positive, or oscillatory,
depending upon the boundary conditions [17, 34].
The simple arrangement described is not practical, in most cases, because it
requires an inconveniently long transmission line. For example, a line with a phase
velocity half the velocity of light designed to produce a 1 µs pulse would be 75 m
long. Thus a practical line-​ type modulator employs a pulse-​ forming network
having a number of sections with series inductance and shunt capacitance, as shown
in Figure 20.12. The switch S1 is a diode, while S2 is a thyratron or a thyristor which
is switched on by a control signal and switches off automatically when the voltage
across it falls to zero. The inductor in series with the DC source, and the capaci-
tance of the pulse-​forming network, form a resonant circuit. As the line is charged
the voltage rises sinusoidally until it reaches a maximum of 2V0 , at which value it
is held by the diode. Thus the amplitude of the pulse is equal to the DC supply
voltage. The impedance of the load is not normally matched to a practical value of
characteristic impedance and modulators of this type require a pulse transformer
to match the load to the line.
A lumped-​element network cannot reproduce the behaviour of a uniform trans-
mission line exactly. Figure 20.13 shows the pulse shape when a uniform five-​section
network is terminated by a matched resistance. It can be seen that there is an over-
shoot at the start of the pulse, ripples on the pulse top, and an overshoot at the
end of the pulse. The properties of a uniform transmission line can be approached
more exactly by increasing the number of sections in the line but the pulse remains
imperfect to some extent. A detailed discussion of theoretical methods for design-
ing pulse-​forming networks is given in [17]. In one useful form of pulse-​forming
 RF Systems 779

Figure 20.12:  Arrangement of a line-​type modulator.

Figure 20.13:  Pulse generated by a uniform five-​section network terminated by a matched

resistance (reproduced, with permission, from [17]).

network the series inductors are wound on a common former so that there is mutual
inductance between the sections. It is then possible to use capacitors which all have
the same capacitance by suitable choice of the tapping points at which they are con-
nected to the inductors. It is important to recognise that the actual performance of a
line-​type modulator depends upon the impedance presented by the load, including
the effects of reactances in the pulse transformer, and of parasitic reactances. The
performance of modulators can be optimised by making empirical adjustments to
the components in the pulse-​forming network. It has been shown that the ripple in
the pulse top can be reduced by using a pair of pulse-​forming networks arranged so
that their contributions to the ripple are in anti-​phase [35]. The design of a simple
line-​type modulator with a pulse transformer is described in [36].

20.4 RF Systems

The RF system into which the tube is incorporated has an influence on its perform-
ance. The construction and calibration of the system are important, as also are
780 System Integration

differences between the system, and the test stand used by the tube manufacturer.
When the tube is an amplifier the correct drive level is important. Overdriving a
tube increases its non-​linearity and can cause oscillations and overheating. Because
some RF power may emerge from the input port of the tube it is necessary to ensure
that a good match is presented by the input system. Under fault conditions the
returned power may be high, and at frequencies other than the normal operating
frequency, so that it is necessary to protect the input system by a circulator.
When a tube is tested by the manufacturer the output is usually directed to a high
power load which has a good match over a wide range of frequencies. The effects of
an incorrect match may be variations in the power output, in the gain and phase of
an amplifier, and in the frequency and spectrum of an oscillator. In extreme cases
it may lead to RF voltage breakdown within the tube. The effects of the RF match
are indicated on a Rieke diagram which may show regions where operation is for-
bidden (see Figures 13.12 and 15.29). The RF match presented by the system may
be very different from that of the test stand and may affect the correct operation of
the tube. For this reason the output of the tube is normally directed through a high-​
power circulator. However, the performance of a circulator is usually only specified
over the working frequency band of the system. Thus, its properties at harmonic
and other out-​of band frequencies may not be defined. If a filter is included in the
system to block power at these frequencies, the effect of reflection into the output
port of the tube must be considered. The construction of the RF output system
must conform to good high power practice and particular care must be taken to
avoid resonances. The output power of the tube and the power reflected by the
system are measured using directional couplers whose calibration is important. The
measurement of power at harmonic frequencies is not straightforward because the
power may be divided between several waveguide modes [37]. Reflection of output
power into the tube may cause overheating within it which may be indicated by
increased gas pressure. It can also cause failure of the output window if a peak of
the standing wave occurs there. Abnormal causes of power reflection include arcs
and multipactor discharges in the output waveguide. Reflected power, and wave-
guide arc, detectors connected to fast trips are used to protect the tube (see Section
20.6.2). The design and maintenance of high power RF systems is discussed in
[4, 38].

20.5 Cooling System

The methods used for cooling the collectors of IOTs, linear-​beam tubes and gyro-
trons are discussed in Section 10.5. Similar methods are used for cooling the anodes
of gridded tubes, magnetrons and CFAs. The greatest part of the heat is dissipated
on these electrodes. However, it is usually necessary to cool the body of the tube,
and cooling may also be required for electromagnets and output windows. The sur-
faces to be cooled may not be at earth potential. The density of the power to be
dissipated is frequently great enough for the tube to be permanently damaged if the
 Control System 781

cooling is inadequate. Overheating can cause the release of adsorbed gases lead-
ing to an increase in the pressure within the tube. If the background gas pressure
rises then poisoning of the cathode may lead to reduced emission, and ion oscilla-
tions may add noise to the tube output. Large tubes are usually fitted with at least
one ion pump whose function is to monitor, and maintain, the vacuum within the
tube. Permanent damage is revealed by changes in the RF performance of the tube.
Under normal operation the internal dimensions of the tube change because of the
thermal expansion, leading to changes in performance with temperature. Vibration
caused by cooling fans, pumps, and turbulence in the cooling system can cause
internal parts of a tube to vibrate, imposing low frequency modulation on the RF
output.
If the tube is to be cooled correctly it is important that air and liquid flow rates
should be adequate, that the incoming coolant temperature is correct, and that the
surfaces to be cooled are clean and not damaged by corrosion. In air-​cooled sys-
tems corrosion can be caused by excessive humidity. In water-​cooled systems gal-
vanic corrosion may occur, leading to shortening of the life of the tube [39]. To
prevent this it is important to consider the metals used in the construction of the
cooling system, the purity and pH value of the water used, and the nature of any
additives such as anti-​freeze. Further information on the design and maintenance
of cooling systems is given in [4, 5, 38, 40].

20.6 Control System

The purpose of the control system is to ensure correct and safe operation of the
tube, and to protect it under fault conditions. The details of the system depend
upon the tube type, and the power level at which it is operating, but the general
principles are the same for all.

20.6.1 Interlocks
The parts of the system shown in Figure 20.1 must be interlocked in such a way that
the different parts are turned on and off in the correct order. In addition, the inter-
locks provide protection if any of the correct operating conditions, including the
maximum ratings specified by the tube manufacturer, are violated. The interlocks may
also include conditions relating to other parts of the complete system (for example
the cooling of a high-​power circulator). Examples of switch-​on sequences are given
in [5–​7]. At each stage the operating conditions are monitored so that the next stage
cannot commence unless they are within prescribed limits. If any quantity lies out-
side the permitted limits then the system is shut down. A typical start-​up sequence is:

i) The cooling and auxiliary supplies (electromagnets, ion pump) are turned on.
ii) The cathode heater is turned on with a time delay which inhibits the next step
until the full heater temperature has been reached.
782 System Integration

iii) The control grid bias voltage is applied.

iv) The anode voltage is applied. In some cases the voltage is increased gradually
until the working voltage is reached.
v) The collector voltages are applied to a tube with depressed collector electrodes.
The screen grid voltage is applied to a tetrode.
vi) The RF drive is applied. Note that in some linear-​beam tubes the collector may
not be designed to dissipate the full beam power in the absence of RF drive. It
is then necessary for the anode voltage and the RF drive to be applied simul-
taneously and for the beam power to be reduced if the RF drive fails [41].
vii) The heater power of a magnetron or CFA is reduced to compensate for heating
by back-​bombardment of the cathode.

The same steps are followed in the reverse order to turn the tube off. Because parts
of the tube may still be quite hot after it has been turned off it may be necessary to
keep the cooling systems in operation for some time.

20.6.2 Tube Protection

The interlocks described above cannot normally act fast enough to protect the tube
under fault conditions. For this purpose, trips are required which can remove power
from the tube under fault conditions within tens of microseconds, before damage
can occur. For tubes with DC supplies this is achieved by firing the crowbar switch
of the power supply. For cathode-​pulsed tubes, the triggering of the next pulse must
be suppressed. Faults to be detected include:
• DC arcs inside the tube indicated by excessive cathode, or grid, current;
• excessive reflected power and RF arcs in the output waveguide;
• excessive current interception on the tube body;
• grid bias failure;
• depressed collector bias failure;
• excessive gas pressure

20.7 Care of Tubes

The performance of a tube can be affected by its environment during storage,

transportation, installation, and operation. Because tubes appear to be mechan-
ically robust they may be treated carelessly, with the result that the performance is
degraded and the life of the tube reduced. Some common problems are:
• The copper body of a tube is very soft as a result of annealing during manufac-
ture so that the precision of the alignment of internal parts may be disturbed by
shocks.
• Ceramic insulators are brittle and can be fractured so that the vacuum is lost.
 Safety 783

• The cathode assembly of a tube is often cantilevered, making it liable to be dis-

placed by transverse shocks. Internal spacings, especially those of grids, can be
very small and critical to the operation of the tube.
• The magnetisation of permanent magnets can be altered by exposure to external
magnetic fields or ferromagnetic materials. The presence of the magnet of a high-​
power circulator close to the tube may cause problems with focusing.
• The external surfaces of insulators can become dirty, and metal surfaces can be
corroded.
For all these reasons it is important that a tube should always be transported in
the packing provided by the manufacturer, and the handling procedures specified
should be followed with care. If ion pumps are fitted they can be used to monitor
and maintain the vacuum during storage. When the manufacturer specifies a gas
test this should be carried out before the tube is installed. Some tubes are very heavy
and should only be handled and installed with appropriate equipment. After a tube
has been installed it is usually necessary to follow a conditioning procedure before
it is ready for normal use. That can include gradually raising the operating voltage
over an extended period of time (see Section 18.7.1). A detailed discussion of these
issues, and of the correct procedures is given in [4, 38].

20.8 Safety

Vacuum tubes are high power devices whose safe operation requires careful atten-
tion to the construction and use of the systems in which they are employed.
A detailed discussion of the issues involved is given in [4, 38, 42].
Tubes operate at voltages which are potentially lethal and proper high-​voltage
precautions must be taken. All high-​voltage connections must be enclosed, or
adequately insulated, and interlocks provided so that high-​voltage terminals are
inaccessible when they are live. The design of the equipment must ensure that all
high voltage terminals are earthed before any access to the enclosure is possible.
High voltage hazard warning notices should be clearly displayed. Great care must
be taken to keep earthing straps short and of very low inductance. Under fault
conditions the rate of change of current with time can be very high so that high-​
induced voltages may exist on exposed metalwork (see Section 18.7.1).
RF power is dangerous to biological tissue especially the eyes. For this reason
one should never look into the end of a waveguide carrying microwave power at
any level. All joints in the high-​power waveguide system must be correctly made
and checked for leakage. It is important to be aware that RF power may be emit-
ted from parts of the tube such as the collector insulator of a linear-​beam tube, or
the cathode insulator of a crossed-​field tube. There should, therefore, be adequate
screening against such emissions, including those which only exist under fault con-
ditions. National regulations for safe levels of exposure to radio-​frequency radia-
tion should be observed.
784 System Integration

X-​rays are produced by the impact of electrons on solid surfaces at voltages

of 5 kV or more, as described in Section 18.4. Low-​energy X-​rays are screened
out by the metal body of a tube but may still escape through ceramic insulators.
The energies of electron impacts in crossed-​field tubes are smaller than those
in other tubes, but X-​rays can still be emitted under fault conditions. If X-​rays
are likely to be generated then the tube must only be operated in a properly
screened enclosure and national regulations for the protection of personnel must
be observed [43, 44].
Some tubes contain beryllium oxide ceramics or beryllium metal parts. Beryllium
oxide dust is highly toxic, and great care must be taken in handling a tube in which
a ceramic has been broken. Otherwise a tube containing these materials is safe to
handle unless an attempt is made to dismantle it. The tube should be clearly marked
with a hazard label.
Sulphur hexafluoride (SF6) and Freon gases are sometimes used to prevent break-
down in high-​voltage systems and high-​power waveguides (see Section 18.7.2). If
a breakdown does occur the gas may decompose, and combine with air or water
vapour, to produce highly toxic and corrosive compounds. When systems use these
gases it is essential to establish safe procedures for handling the consequences of a
breakdown.
If an insulator is broken by a mechanical shock, the vacuum within a tube can
cause fragments to fly outwards at high velocity, leading to injury. Correct handling
procedures should be used to avoid this. Especial care is required if beryllium oxide
ceramics are present.
Exposed surfaces on an air-​cooled tube may be at temperatures of 200 °C to 300
°C during operation, and they may remain dangerously hot for some time after
the tube has been turned off. The cooling fans should continue to operate until
the temperature has been reduced to a safe level. Liquid coolants can also become
dangerously hot, and care is needed to avoid breakage of cooling pipes, especially
if they are under pressure.

20.9 Reliability

The reliability of a vacuum tube can impinge on the reliability of the system, in
which it is installed, in two ways. First, there are short-​term system outages caused
by trips when the tube is otherwise performing correctly. These can be common
when a tube is first installed, after a period of storage, and their frequency decreases
if the tube is conditioned by gradually raising the anode voltage. Sometimes the fre-
quency of tripping increases after a long period of operation, but it can be reduced
by repeating the conditioning process. The duration of each outage depends upon
the length of time required to restart the tube after a trip. The effect of the outages
may be only a minor irritation, or it may be serious where continuous availability of
 Conclusion 785

the system is important. The second way in which the reliability of the tube affects
that of the system is the point at which the tube no longer delivers the performance
required so that it has to be replaced. This can be a major operation during which
the system is not available for hours, or even days. Both kinds of outage affect the
cost of ownership of the system which needs to be minimised by careful design and
operation.
The normal end of life of a tube is determined by reduction in emission of elec-
trons from the cathode, caused by evaporation of barium. If the gas pressure in a
tube is too high, then the ions generated may bombard the cathode surface. This is
more likely to occur in CW tubes than in pulsed tubes. The result may be damage
by sputtering of material from the surface, or by ‘poisoning’, leading to reduced
emission. A third cause of low emission is low cathode heater power, which may be
caused by an internal short circuit, or by incorrect setting of the heater supply. If
the cathode becomes temperature-​limited, because the cathode work function has
increased, or the heater power is too low, then the cathode current varies with the
heater current. In addition, the reduction in the current drawn from the high volt-
age supply may cause its voltage to rise and voltage breakdown may occur. In mag-
netrons reduced emission leads to a poor RF spectrum. The lifetime of a tube can
sometimes be extended by de-​rating it to extend the cathode life, or by increasing
the heater current when the emission starts to fall. The tube manufacturer should
be consulted before either of these courses of action is pursued.
It is found that many tube failures are caused by incorrect handling, or by sys-
tem faults. The ways in which these affect the performance of the tube have been
discussed in the preceding sections. The tendency to assume that the tube is faulty,
and to attempt to cure system faults by replacing it, should be resisted. The tube is
an expensive component and the cost of replacing it is not trivial. Therefore proper
diagnostic procedures should be followed to ensure that the fault does not lie else-
where, before the tube is replaced [4]. Table 20.2 shows some typical symptoms and
causes of tube failure. Further information is to be found in [4, 5].

20.10 Conclusion

Vacuum tubes continue to be important components in systems of many kinds.

Progress in design, and in manufacturing technology, continue to yield improve-
ments in performance, reliability, and lifetime, as noted in Chapter 1 of this book.
As a result, the performance of systems incorporating tubes is also improving,
and new applications are becoming possible. However, if the full benefits are to
be achieved, it is essential that the tube engineers, and the systems engineers, have
a good understanding of each others’ technologies. It is hoped that this book will
contribute to that understanding by providing a detailed, and authoritative, account
of the principles of vacuum electron devices.
786 System Integration

Table 20.2:  Symptoms and causes of tube failure

Symptom Internal causes External causes

Low cathode current Cathode end of life Low anode voltage

Cathode poisoning Low heater power
Cathode erosion
Heater short circuit
Voltage breakdown Contamination of insulators Contamination of insulators
Increased gas pressure High anode voltage
High body current Mechanical damage
Demagnetised magnets
Overheating Mechanical damage Wrong electromagnet current
Demagnetised magnets High RF drive
Inadequate cooling
High ambient temperature
High ion pump current Mechanical damage Wrong electromagnet current
Overheating High RF drive
Inadequate cooling
High ambient temperature
High grid current Short-​circuit Wrong bias voltage
Wrong RF power Mechanical damage Wrong RF match
Demagnetised magnets Wrong grid bias
Wrong electromagnet current
Wrong anode voltage
Wrong RF drive
Oscillation Mechanical damage Wrong RF match
Demagnetised magnets Wrong electromagnet current
Wrong anode voltage
Wrong RF drive
Poor spectrum Reduced cathode emission Wrong anode voltage
(magnetrons) Demagnetised magnet

References

[1]‌ J. W. Hansen et al., Applications Note: System Aspects of Communications TWTAs, or

How to Deal with the Tube Manufacturer to Your Best Advantage. Torrance, CA: Hughes
Aircraft Company, Electron Dynamics Division, 1982.
[2]‌ C. R. Smith et  al., ‘The microwave power module:  a versatile RF building block for
high-​power transmitters’, Proceedings of the IEEE, vol. 87, pp. 717–​737, 1999.
[3]‌ M. Katzman and R. Wills, Microwave Tube Reliability. San Diego, CA: Naval Electronics
Lab Center, CA, 1976.
[4]‌ Varian, Technical Manual:  Installation, Operation, Maintenance, Care and Handling
Instructions, General:  Microwave Tubes, Magnetron Tubes, Electron Tubes. 1
October 1979.
[5]‌ L. Sivan, Microwave Tube Transmitters. London: Chapman and Hall, 1994.
 References 787

[6]‌ M. J. Smith and G. Phillips, Power Klystrons Today. Taunton, UK: Research Studies
Press, 1995.
[7]‌ EEV Ltd, A User’s Guide to the Coupled-​Cavity TWT. Chelmsford, UK: EEV Ltd, 1987.
[8]‌ E. Carroll, ‘High-​power active devices’, in D. Brandt, ed., Specialized Course on Power
Converters. Geneva: CERN, pp. 43–​56, 2004.
[9]‌ EEV Ltd, Hydrogen Thyratron Preamble. Chelmsford, UK: EEV Ltd., 2002.
[10] P. D. Pearce, Pulsed Power for Future Linear Accelerators. Geneva: CERN, 1999.
[11] D. Poole et al., ‘A crowbarless high voltage power converter for RF klystrons’, in EPAC
96, pp. 2326–​2328, 1996.
[12] D. E. Anderson, ‘Recent developments in pulsed high power systems’, presented at
LINAC 2006, Knoxville, Tennessee, 2006.
[13] C. Burkhart et al., ‘Development of a solid state thyratron replacement for the LCLS
klystron modulator’, in 19th IEEE Pulsed Power Conference, pp. 1–​4, 2013.
[14] N. Mohan et  al., Power Electronics:  Converters, Applications and Design, 3rd ed.
Hoboken, NJ: Wiley & Sons Inc., 2003.
[15] R. B. True et  al., ‘The Hobetron  –​a high power vacuum electronic switch’, IEEE
Transactions on Electron Devices, vol. 48, pp. 122–​128, 2001.
[16] H. W. Lord, ‘Pulse transformers’, IEEE Transactions on Magnetics, vol. 7, pp. 17–​28, 1971.
[17] G. N. Glasoe and J. V. Lebacz, eds, Pulse Generators (MIT Radiation Lab Series).
New York: McGraw-​Hill, 1948.
[18] D. Brandt, ed., CAS Specialized Course on Power Convertors. Geneva: CERN, 2004.
[19] J. E. Theed et al., ‘52 kV power supply for energy recovery linac prototype RF’ in EPAC
2004, Lucerne, pp. 1762–​1764, 2004.
[20] Sorrento Electronics Inc., High Energy Capacitors Overview. San Diego, Sorrento
Electronics, Inc, 2002.
[21] D. J. Cook et al., ‘Design of a direct converter for high power RF applications’, in PAC
07, Albuquerque, NM, pp. 2221–​2223, 2007.
[22] I. Barbi and R. Gules, ‘Isolated DC-​DC converters with high-​output voltage for TWTA
telecommunication satellite applications’, IEEE Transactions on Power Electronics, vol.
18, pp. 975–​984, 2003.
[23] E. W. Manteuffel and R. E. Cooper, ‘D.C.  charged magnetic pulse modulator’,
Transactions of the American Institute of Electrical Engineers, Part I: Communication
and Electronics, vol. 78, pp. 843–​850, 1960.
[24] J. Mankowski and M. Kristiansen, ‘A review of short pulse generator technology’, IEEE
Transactions on Plasma Science, vol. 28, pp. 102–​108, 2000.
[25] H. Pfeffer et al., ‘A long pulse modulator for reduced size and cost’, in 21st International
Power Modulator Symposium, Costa Mesa, CA, 1994.
[26] H. Pfeffer et  al., ‘A second long pulse modulator for TESLA using IGBTs’, in Fifth
European Particle Accelerator Conference, pp. 2585–​2587, 1996.
[27] C. A. Martins and K. Rathsman, ‘An assessment on klystron modulator topologies for
the ESS project’, in Pulsed Power Conference, Chicago, IL, 2011.
[28] J. Casey et  al., ‘Marx bank technology for accelerators and colliders’, in LINAC08,
Victoria, BC, Canada, pp. 1002–​1004, 2008.
[29] R. L. Cassel, ‘An all solid state pulsed Marx type modulator for magnetrons and klys-
trons’, in IEEE Pulsed Power Conference, pp. 836–​838, 2005.
[30] R. Richardson et al., ‘High energy pulse modulator for microwave tube applications’, in
25th International Power Modulator Symposium, pp. 240–​243, 2002.
788 System Integration

[31] M. Akemoto et al., ‘Pulse transformer R & D for NLC klystron pulse modulator’, in
11th IEEE International Pulsed Power Conference, vol. 1, pp. 724–​729, 1997.
[32] M. Akemoto and A. Tokuchi, ‘A 600 kV, 150 ns rise-​time pulse transformer for klys-
tron pulse modulator’, in IEEE International Power Modulators and High Voltage
Conference, p. 396, 2008.
[33] D. Bortis et  al., ‘Design procedure for compact pulse transformers with rectangular
pulse shape and fast rise times’, in IEEE International Power Modulator and High
Voltage Conference, pp. 298–​302, 2010.
[34] R. G. Carter, Electromagnetism for Electronic Engineers. Available at:  http://​book-
boon.com/​en/​electromagnetism-​for-​electronic-​engineers-​ebook (accessed 5 October
2017) 2010.
[35] C. B. Figley, ‘Results of a modulator pulse top ripple reduction study’, IEEE Transactions
on Nuclear Science, vol. 40, pp. 114–​119, 1993.
[36] N. Carleto and C. C. Motta, ‘Design, construction and characterization of a line-​type
pulse modulator for driving high power magnetron’, in IEEE MTT-​S International
Conference on Microwave and Optoelectronics, pp. 330–​333, 2005.
[37] V. Price, ‘Measurement of harmonic power generated by microwave transmitters’, IRE
Transactions on Microwave Theory and Techniques, vol. 7, pp. 116–​120, 1959.
[38] J. Whitaker, Power Vacuum Tubes Handbook. CRC Press, 2012.
[39] R. Heppinstall and G. Clayworth, ‘The importance of water purity in the successful
operation of vapour-​cooled television klystrons’, Radio and Electronic Engineer, vol. 45,
pp. 391–​399, 1975.
[40] EIMAC, Care and Feeding of Power Grid Tubes, 5th ed. San Carlos, CA: CPI Inc. Eimac
Division, 2003.
[41] H. Frischholz, ‘Generation and distribution of radio-​frequency power in LEP’, IEEE
Transactions on Nuclear Science, vol. NS-​32, pp. 2791–​2793, 1985.
[42] R. Kitchen, RF and Microwave Radiation Safety Handbook. Oxford: Newnes, 2001.
[43] H. E. Johns and R. J. Cunningham, The Physics of Radiation Protection, 3rd ed.
Springfield, IL: Charles C. Thomas, 1974.
[44] W. Meredith and J. B. Massey, Fundamental Physics of Radiology, 2nd ed.
Baltimore: Williams and Wilkins, 1972.
Appendix: Mathcad Worksheets

Available online at www.cambridge.org/9780521198622

WS 1.1 Coupled Modes
WS 1.2 Modulation
WS 1.3 Linearity
WS 2.1 Ridged Waveguides
WS 2.2 Waveguide Capacitive Iris
WS 2.3 Waveguide Inductive Iris
WS 3.1 Resonant Circuits
WS 3.2 Cavity Resonators
WS 3.3 Re-​entrant with Beam Hole
WS 3.4 Fujisawa’s model
WS 3.5 Gap Field
WS 3.6 Iris-​Coupled Cavity
WS 4.1 Capacitance of a Planar Grid
WS 4.2 Meander Line
WS 4.3 Sheath Helix
WS 4.4 Capacitance of a Cylindrical Grid
WS 4.5 Tape Helix Model
WS 4.6 Ring-​Bar Structure
WS 4.7 Folded Waveguide
WS 4.8 Space-​Harmonic Structure
WS 5.1 Planar Diode
WS 5.2 Cylindrical Diode, Relativistic Effects
WS 5.3 Spherical Diode, Relativistic Effects
WS 5.4 Planar Diode with Initial Velocity
WS 6.1 Electrostatic Triode with Island Formation
WS 6.2 Effective Diode Spacing
WS 6.3 Tetrode Fields
WS 7.1 Calculation of Beam Parameters
WS 7.2 Solenoid Focusing
WS 7.3 Beam Spreading Curves
WS 7.4 PPM Focusing (Mean Radius)
WS 7.5 Periodic Electrostatic Focusing
790 Appendix: Mathcad Worksheets

WS 8.1 Electron Trajectories

WS 8.2 Planar Crossed-​Field Diode
WS 8.3 Cylindrical Crossed-​Field Diode
WS 9.1 Pierce Electron Gun
WS 9.2 PPM Stack Entry
WS 10.1 Collector Design
WS 10.2 Collector Depression (Klystron)
WS 10.3 Collector Depression (IOT)
WS 10.4 Collector Depression (TWT)
WS 11.1 Gap Coupling
WS 11.2 Reduced Plasma Frequency
WS 11.3 Beam Gap Interaction
WS 11.4 Continuous TWT
WS 11.5 Discrete TWT
WS 11.6 Large Signal Gap Model
WS 12.1 Amplifier Classes
WS 12.2 ML-​5681 Triode Characteristics
WS 12.3 RS 2058 Tetrode
WS 12.4 Tetrode Design
WS 12.5 IOT Model
WS 12.6 IOT Collector Depression
WS 13.1 Klystron Small Signal
WS 13.2 Klystron Simplified Small Signal
WS 13.3 Klystron Large Signal
WS 13.4 Klystron Efficiency
WS 13.5 Klystron Spent-​Beam Characteristics
WS 13.6 Klystron Design
WS 14.1 Helix TWT Small Signal
WS 14.2 Helix TWT Large Signal
WS 14.3 Helix TWT Design
WS 15.1 Magnetron Anode Worksheet 4J50
WS 15.2 Rising Sun Anode Worksheet
WS 15.3 4J50 Magnetron
WS 15.4 Magnetron Design Sheet
WS 16.1 SFD-​262 Model
WS 17.1 Gyrotron Model
WS 18.1 Two Surface Multipactor Model
WS 18.2 Crossed Field Multipactor Model
WS 19.1 Coil Field
WS 19.2 Permanent Magnet Design
WS 19.3 PPM Stack Design
WS 20.1 Power Converters
WS 20.2 Pulse Shape
Index

absolute instability, 16, 417 oscillation, 16, 416

aliasing, 144 growth rate, 416
AM/​AM characteristics, 8, 11, 29 in higher-​order mode, 544
crossed-​field amplifier, 643, 651 start current, 416
travelling-​wave tube, 8, 539 suppression in travelling-​wave tube, 543
AM/​AM conversion, 29 in travelling-​wave tube, 543
travelling-​wave tube, 541, 543 oscillator, 668
AM/​PM characteristics, 8, 11, 29 band edge oscillation
travelling-​wave tube, 540 in coupled-​cavity travelling-​wave tubes, 544
AM/​PM conversion in crossed-​field amplifier, 641
travelling-​wave tube, 541, 552 bandwidth
Amplitron, 631 definition, 11
analogue modulation, 20, 23–6 efficiency, 23, 26
amplitude, 24 beam filling factor, 265, 530, 533, 545, 660
double sideband, 24 beam spreading, 266–9
double sideband suppressed universal beam spreading curve, 267
carrier, 24 with angular velocity, 268
frequency, 26 without angular velocity, 266
phase, 24 beam stiffness, 263–6
anisotropic boron nitride, 508 beam-​gap interaction
Anti-​Karp structure, 148 ballistic theory, 376–85
Applegate diagram, 419, 456 beam loading, 380–2, 400–2
modified, 419, 524 conductance, 381, 400, 423
applications of vacuum tubes, 17–18 susceptance, 381, 400
communications systems, 508, 510 coupling factor, 378
electronic counter-​measures, 508, 633 axial, 384, 400
instrumentation, 508 radial, 384
medical linear accelerators, 565 electric field of gap, 383
microwave ovens, 565 equivalent circuit, 402
nuclear fusion, 468 induced current, 377, 399
particle accelerators, 456, 468 large-​signal, 419–29
plasma heating, 468 coupling factor, 428
radar, 468, 510, 565, 631, 684 induced current, 423
satellite communications, 28, 468, 550 induced gap voltage, 425
tropospheric scatter, 468 maximum power transfer, 427, 429
UHF television transmitters, 456, 468 power transfer, 426–9
velocity modulation, 422
Babinet’s principle, 152 matrix representation, 402–4
backward-​wave space-​charge wave amplitude, 398
amplification, 416 space-​charge wave theory, 397–404
interaction, 415–17 transit angle, 378
two-​wave approximation, 415 velocity modulation, 376, 378
792 Index

beam-​plasma amplifier, 375 R/​Q, 130

beam-​wave interaction, 375–6, see also backward-​ resonant frequency, 130
wave interaction; beam-​gap interaction; unloaded Q, 131
continuous beam-​structure interaction; centipede structure, 179–81
discrete beam-​structure interaction; large-​ equivalent circuit, 179
signal beam-​wave interaction; space-​charge higher-​order modes, 180
wave theory total impedance, 180
beryllia, 508 Cerenkov interaction, 138
Bessel function, 55 CFA, see crossed-​field amplifier
I0, 135 characteristic field
Bessel’s equation, 55 magnetron, 574
Biot-​Savart law, 736 characteristic voltage
bit error rate, 23, 26 magnetron, 574
Boltzmann’s constant, 22, 695 Child-​Langmuir current, 609
Brillouin field, 260 Child-​Langmuir law, 195, 607
Brillouin zone, 144, 583, 629, 637 circuit efficiency, 11
Buckingham’s theorem, 34 circular waveguide, 64–7
Busch’s theorem, 256, 259 modes in gyrotron oscillator, 670
Butterworth, see matching: maximally flat polarisation, 65
TE mode field patterns, 65
capture ratio, 550 TM mode field patterns, 65
Carcinotron, see crossed-​field backward-​wave whispering gallery modes, 670
oscillator circulator, 102, 483, 598, 602, 722, 780, 783
care of tubes classification of vacuum tubes, 17–18
conditioning, 783 cloverleaf structure, 176–8
installation, 782 dispersion, 178
operation, 782 equivalent circuit, 178
storage, 782 higher-​order modes, 178
transportation, 782 total impedance, 178
CARM, see cyclotron auto-​resonance maser coaxial line, 45, 52–7
Carson’s rule, 26 characteristic impedance, 54
cathode loss parameter, 54
cold maximum power, 54
in crossed-​field amplifier, 634 multipactor discharge in, 55
field emission, 712 TE and TM modes, 55
Spindt, 712 TE mode field patterns, 58
in travelling-​wave tube, 558 TEM mode, 53
oxide TM mode field patterns, 58
in crossed-​field amplifier, 635 co-​channel interference, 29, 31
photo-​cathode, 700 collector, 352–70
platinum depressed, 353, 357–66
in crossed-​field amplifer, 635 efficiency, 359–​60
thermionic, see thermionic cathode energy recovered, 358–​9, 361
cavity resonator, 93, see also coupling to cavity in gyrotron oscillator, 682
resonator probability of collection, 363
equivalent circuit, 93 multi-​element, 360–70, 490, 508
extended interaction, 483 asymmetric, 369
gyrotron oscillator, 670 DC efficiency, 365
klystron, 466 design, 366–9
magnetron, 584 dispersive-​lens, 367
pill-​box, 93 hyperbolic asymmetric, 368
rectangular, 110 individual-​lens, 368
surface roughness, 107–9 inductive output tube, 459
cavity resonator measurement, 129–32 klystron, 488
higher-​order modes, 131 non-​ideal, 363
loaded Q, 131 parabolic symmetric, 369
 Index 793

RF efficiency, 365 crossed-​field amplifier, 629–38; see also crossed-​

symmetric, 367 field amplifier design; crossed-​field amplifier
travelling-​wave tube, 525, 552 model; crossed-​field amplifier performance
particle in cell simulation, 357 anode current, 638
power density, 355, see cooling arrangement, 629
secondary electrons, 352–​3, 366 backward-​wave, 629
suppression, 369 cathode-​driven, 635
space-​charge, 357 control electrode, 641
potential depression, 357 emitting cathode, 629, 631–3
virtual cathode, 357 forward-​wave, 629
spent beam injected beam, 629, 633
distribution, 354, 366 basic principles, 636–8
reconditioning, 370 cathode, 629, 634
virtual cathode in, 481, 487 back-​bombardment, 634
X-​ray generation, 352 cold, 634, 704
conservation of energy, 5, 9, 194, 291, 302, 394, 574–​5 thermionic, 634
continuity equation, 294, 385 efficiency, 632
continuous beam-​structure interaction, 404–11 electron transit time, 638
backward wave, 408 inter-​modulation products, 631
bandwidth, 411 line, see crossed-​field amplifier: slow-​wave
coupled-​mode equation, 406 structure
decaying wave, 408 magnetic field, 639
dispersion equation, 407, 527 noise, 632
electric field of structure, 405 operating point, 636
fast space-charge wave, 408 oxygen dispenser, 635
gain per circuit wavelength, 410 secondary electrons in, 634
growing wave, 408 slow-​wave structure, 629, 634
Pierce parameters, 409 heat dissipation, 634
slow space-charge wave, 408 helix-​coupled vane, 634
synchronous point, 407 inter-​digital line, 634
two-​wave approximation, 409 meander line, 634
convective instability, 15, 667 split folded waveguide, 634
cooling, 371–2 strapped ladder line, 634
air, 371 strapped-​vane, 634
conduction, 371 stub-​mounted helix, 634
hypervapotron, 372 sole, see crossed-​field amplifier: cathode
pumped liquid, 372 space-​charge hub, 629, 634, 636
vapour phase, 372 spokes, 636, 638
coupled-​cavity slow-​wave structure, 173–81; angular velocity, 636
see also centipede structure; cloverleaf synchronism, 636
structure; space-​harmonic structure tuning, 629
coupling to, 180 crossed-​field amplifier design, 654–5
forward fundamental structure, 173 circuit efficiency, 655
sever, 181 electronic efficiency, 654
in travelling-​wave tube, 509, 556 equivalent magnetron, 654
coupled-​mode theory, 12–17, 408 modified Slater factor, 655
coupling to cavity resonator, 120–9 power added efficiency, 654
coupling factor, 122, 124–​5, 129 crossed-​field amplifier model, 645–54
electric antenna, 120 Brillouin hub, 646
equivalent circuit, 122, 129 case study, 649
external Q, 124 cold loss, 648
input impedance, 128 comparison of models, 651
iris coupling, 122, 124 efficiency, 647
loop coupling, 122 electron transit time, 646, 653
magnetic antenna, 120 equivalent magnetron, 648
voltage reflection coefficient, 123 gain, 647
794 Index

crossed-field amplifier model (cont.) space-​charge

Gauss lines, 649 density, 304, 313
guiding centre theory, 646 hub, 301, 307
induced current, 647 modes, 315
non-​linear fluid mechanics, 647 oscillations, 312, 314
output power, 649 cylindrical thermionic diode, 206–9
particle in cell, 645 space-​charge limited, 207
rigid spoke, 647 with relativistic velocity, 208
secondary electron emission, 645
soliton theory, 646 DC power supply, 765–71
spoke formation, 645 bridge converter
threshold voltage, 648 3-​phase, 770
crossed-​field amplifier performance, 639–45 single-​phase, 768
band-​edge oscillations, 641 crowbar, 770, 782
bandwidth, 643 electric power converter, 768–71
gain, 642 high-​voltage, 765
heat dissipation, 645 high-​voltage switch
load line, 639, 643 semiconductor, 767
matches, 644 thyratron, 767
performance chart, 639 high-​voltage switches, 766
phase, 643 impedance, 765
pulse nesting, 641 load impedance, 767
pulsed, 641 crossed-​field tube, 767
signal growth, 645 gridded tube, 767
transfer characteristics, 642 linear-​beam tube, 767
crossed-​field backward-​wave oscillator, 633 low-​voltage, 766
crossed-​field electron flow, 287–8; see also for multi-​element depressed collector, 766
cylindrical magnetron diode; planar noise, 765–​6
magnetron diode post regulator, 771
categories, 287 smoothing capacitor, 770
equation of motion, 287 stored energy, 771
crossed-​field electron gun, 344–8 voltage ripple, 770
Kino gun, 345 DEMATRON, 632
magnetron injection gun, 346 dielectric material
design, 346 alumina, 84
CVD diamond window, 679 anisotropic boron nitride, 508
cyclotron auto-​resonance maser, 663 beryllia, 84, 508
cyclotron frequency, 289, 592, 601, 606, 610, CVD diamond, 84, 679
624, 670 digital modulation, 26–8
relativistic, 661 amplitude shift keyed (ASK), 26
cylindrical magnetron diode, 300–15 binary phase-​shift keyed (BPSK), 26
back-​bombardment, 310 constellation diagram, 26
Brillouin flow, 303 frequency shift keyed (FSK), 26
cathode current density, 306 phase shift keyed (PSK), 26
current when conducting, 307 quadrature amplitude (QAM), 26
current when cut off, 309 quadrature phase-​shift keyed
electron flow in, 308 (QPSK), 26
electron trajectories digital signals, 20
with space-​charge, 303, 305 dimensionless parameters
without space-​charge, 302 magnetron, 619
equations of motion, 301 travelling-​wave tube, 526
experimental results, 308–14 diocotron instability, 281, 293
Hull cut-​off voltage, 302 discrete beam-​structure interaction, 411–15
the magnetron problem, 314 dispersion equation, 413
potential in, 304 space-​harmonic waves, 414
secondary electron emission, 311 dispersion diagram, 13
Slater trajectories, 307 dispersion equation, 12
 Index 795

for coupled modes, 13 in klystron, 477

double beam amplifier, 375 RF convection current, 379, 399
RF current, 398
E modes, see TM modes in travelling-​wave tube, 512
ECM, see electron cyclotron maser velocity modulation, 378–80, 398
effective plasma frequency, 389, 411 electron current modulation
efficiency by deflection, 7
circuit, 11 by emission density, 7
electronic, 10, 359, 482 by velocity, 7
gridded tube amplifier, 437 electron cyclotron maser, 660–8
gyrotron oscillator, 674, 678 convective instability, 667
inductive output tube, 458 cyclotron modes, 667
klystron, 481 dispersion diagram
magnetron oscillator, 580 coupled, 667
travelling-​wave tube, 528–34 uncoupled, 663
overall, 10 dispersion equation, 664
power added, 10 Doppler shift in, 663
of crossed field amplifier, 654 electron bunches, 661
radiofrequency, 10 electronic efficiency, 662
with depressed collector, 358 feedback oscillations, 668
electromagnetic spectrum Larmor radius, 665
regulations, 8 non-​convective instability, 668
electromagnetic structure pitch factor, 662
fast-​wave,  11 relativistic cyclotron frequency, 661
resonant, 12 RF space-​charge, 666
slow-​wave,  12 small-​signal theory, 664–8
electron synchronism, 661–​2
relativistic threshold current, 668
kinetic energy, 5 waveguide modes, 664
longitudinal mass, 6 electron emission, 694–700, see also
mass, 5 free electron model of metal;
momentum, 6, 671 secondary electron emission;
transverse mass, 6 thermionic cathode
velocity, 5 field emission, 698
rest energy, 5, 204 current density, 699
electron beam, 250, see also beam spreading; Fowler-​Nordheim equation, 699
electron optics; periodic focusing; solenoid triple point, 698, 713, 720
focusing field-​enhanced emission, 697
annular, 281 photo-​electric emission, 699
cylindrical, 250 quantum efficiency, 700
diocotron instability, 281 thermionic emission, 696–7
electronic admittance, 395 current density, 697
relativistic, 395 Richardson-Dushman equation, 696
harmonic currents, 420 saturation current, 698
potential depression, 251, 680 electron gun, 317, see also crossed-​field electron
pre-​bunching, 557–​8 gun; Pierce electron gun
sheet, 281, 502 area convergence, 317
thermal velocities, 282 Kino gun, 345, 633
transmission, 330 perveance, 317
trapped ions, 282 electron optics, 252–8
electron bunching, 7, 420 aperture lens, 256
beam kinetic voltage, 399 Busch’s theorem, 256
current harmonics, 7, 379 einzel lens, 255
current modulation, 398 electrostatic lens, 254–6
drift length, 380 electrostatic paraxial ray equation, 254
idealised, 426, 479 equations of motion, 252
in inductive output tube, 456 focal length
796 Index

electron optics (cont.) remanence, 738

aperture lens, 256 saturation, 738
electrostatic lens, 255 magnetic polarisation, 738
magnetic lens, 258 permanent magnet, 738, 745
magnetic lens, 258 Alnico, 747
magnetostatic paraxial ray equation, 258 ferrite, 745
electronic efficiency, 10 rare earth, 747
electronic propagation constant, 378 samarium cobalt, 747
energy per bit, 23 soft, 738, 744
engineering design, 32–3, see also statement of stabilisation, 744
requirements Floquet’s Theorem, 143
conceptual design, 32 Fourier series, 7, 143
design consolidation, 33 Fowler-​Nordheim equation, 699
detailed design, 33 free electron laser, 689, see also fast-​wave devices:
dimensional analysis, 34 ubitron
dimensionless groups, 34 free electron model of metal, 694–6
dynamic similarity, 35 Fermi energy, 694
geometrical similarity, 34 Fermi-Dirac distribution, 695
modelling, 36 work function, 695
scaling, 33 frequency domain multiplexing, 28
EPC, see DC power supply: electric power frequency-​multiplier tube, 380
converter
Eulerian analysis, 417 gain, 8
evanescent wave, 14 gated field emission arrays, 457
extended interaction oscillator, 494 gridded tube amplifier, 433–9
classes of amplification, 436
fast-​wave devices, 659–68, see also electron common cathode, 433
cyclotron maser; gyrotron oscillator common grid, 433
cyclotron auto-​resonance maser, 663, 683 conduction angle, 436
free electron laser, 689 design, 451–5
gyro-​BWO, 663 electronic efficiency, 437
gyro-​klystron, 663, 684 gain, 438, 453
gyrotron oscillator, 663 input impedance, 453
gyro-​TWT, 663, 685 load line, 436, 452
peniotron, 662, 685 output impedance, 453
cavity, 686 practical details, 454
dispersion equation, 686 push-​pull amplifier, 439
resonant condition, 685 Q point, 434
periodic electron beam, 660 group velocity, 13
synchronism, 660 guiding centre theory
ubitron, 686–9 crossed-​field amplifier, 646
dispersion equation, 687 magnetron, 609
undulator, 687 gyrotron design, 679–83
wiggler, 687 cavity length parameter, 679
feedback oscillations, 668 competing modes, 680
in travelling-​wave tube, 544 cyclotron harmonics, 682
FEL, see free electron laser detuning parameter, 679
Fermi energy, 694 high-​order cavity modes, 680
Fermi-Dirac distribution, 695 optimum beam radius, 680
ferromagnetic materials, 737 output coupling, 681
energy product, 744 stored energy in cavity, 680
hard, 738 velocity spread, 681
hysteresis, 738 gyrotron oscillator, 16, 668–84, see also gyrotron
coercive force, 738 design, large-​signal gyrotron theory
demagnetisation curve, 742 accumulation of ions, 677
initial magnetisation curve, 738 arrangement, 670
minor loop, 742 case study, 677–8
 Index 797

cavity resonator, 670 inductive output tube, 433, 455–62

coaxial cavity, 682 annular beam, 458
collector depression, 682 Applegate diagram, 456
cyclotron frequency, 670 arrangement, 455
diffractive Q, 674 bandwidth, 456
effective beam voltage, 677 case study, 460–2
electron bunching, 678 constant efficiency, 460
electronic efficiency, 678 deflection modulation, 457
hard-​excitation region, 677 electron bunch formation, 455–​6
heating of cavity walls, 679 electronic efficiency, 458
large orbit, 682 maximum output power, 458
magnetic field, 670 multi-​element depressed collector, 459
magnetron injection gun, 670 multiple beam, 458
ohmic Q, 674 radial beam, 458
output window, 679 space-​charge debunching, 457
potential depression, 677, 680 spent beam distribution, 459
quasi-​optical, 683 inductive wall amplifier, 375
quasi-​optical mode converter, 682 inselbildung, see triode: island
RF electric field, 670 formation
soft-​excitation region, 677 instability
stability, 679 absolute, 16
state of the art, 669 convective, 15, 667
step-​tuned, 683 non-​convective, 16, 668
total Q, 674 inter-​modulation distortion, 32
transverse efficiency, 678 inter-​modulation products, 31
tuneable, 683 fifth order, 31
waveguide modes in, 670 in klystron, 491
waveguide window, 670 third order, 31
whispering gallery modes, 670 IOT, see inductive output tube

H modes, see TE modes Karp structure, 148

Hartley’s law, 21 Kilpatrick’s criterion, 482, 713
Hartree voltage, see magnetron oscillator: threshold Kino electron gun, 345, 633
voltage klystrode, see inductive output tube
helix slow-​wave structure, 152–66 klystron, 466–8, see also klystron design;
attenuator, 166, 509 large-​signal klystron theory; small-​signal
characteristic impedance, 164 klystron theory
cold loss, 550, 553 clustered-​cavity,  494
coupler, 165 electro-​statically focused, 499
dielectric loading, 164 extended interaction, 493
dielectric support rods, 152, 156 external cavity, 120, 467
dispersion shaping, 158 high frequency limits, 659
equivalent circuit model, 162–6 integral cavity, 467
Pierce impedance, 164 monotron oscillations, 494
sever, 166 multi-​cavity, 466
sheath helix, 154 multiple-​beam, 500
dielectric loading, 156 reflex oscillator, 499
dispersion, 155, 157 relativistic, 500
equivalent circuit, 155 super-​power, 495
Pierce impedance, 158 klystron design, 491–9
tape helix, 159–62 broad-​band, 479, 493
in travelling-​wave tube, 508 tuning, 493
vane loading, 158, 554 case study, 497–9
Helmholtz coils, 751 high efficiency, 494
Hull cut-​off long drift, 494
curve, 574 optimisation, 499
voltage, 567 second harmonic cavity, 495
798 Index

Lagrangian analysis, 417 output gap

Laplace’s equation, 45–​6, 572, 688, 737, 739, 754 effective coupling factor, 481
large-​signal beam-​wave interaction, 417–19 extended interaction, 483
disc (1D) models, 417 re-​accelerated electrons, 481
distance step models, 418 voltage breakdown, 482
energy balance, 419 Rieke diagram, 485
induced current, 418 sections, 477
ring (2.5D) models, 417 final bunching, 479
self-​consistent solution, 419 initial bunching, 478
space-​charge field, 418 output, 481–3
time step models, 418 spent beam distribution, 488
large-​signal gyrotron theory, 671–7 terminal characteristics, 490
beam current, 674 phase variation with beam voltage, 490
contra-​rotating modes, 676 large-​signal travelling-​wave tube theory, 521–44
co-​rotating modes, 676 AM/​AM
cyclotron harmonic, 674 characteristics, 539
detuning parameter, 672 conversion, 541, 543
electron energy parameter, 672 AM/​PM
electronic efficiency, 674 characteristics, 540
energy balance, 674 conversion, 541
equations of motion, 671 capture ratio, 529
interaction length, 672 dimensionless parameters, 526
loaded Q, 674 efficiency
normalised current parameter, 674 dependence on operating point, 531–4
overall efficiency, 674 effects of sever, 534–6
particle in cell code, 671 electronic, 528–34
reference current, 674 maximum, 532–​3, 555
RF electric field, 673 at synchronous point, 528
rotating mode, 673 forward wave growth, 521
soft-​excitation region, 677 harmonics, 537–8
starting current, 675 in broad-​band tubes, 538
transverse efficiency, 674 maximum gain, 521, 534, 555
large-​signal klystron theory, 475–91 mean electron velocity, 525
bunch model
compression, 479 Eulerian, 521
potential energy, 479, 482 Lagrangian, 521
velocity spread, 482 particle in cell, 521
bunching simple disc (1D), 521
figure of merit, 477, 482 modified Applegate diagram, 524
collector phase space diagrams, 524
depression, 488 RF beam current
virtual cathode, 481, 487 harmonics, 523
efficiency maximum, 532–​3
electronic, 481–​2 phase, 523
theoretical limit, 485 scaling, 528
variation with frequency, 483 space-​charge, 524
variation with perveance, 487 spent beam distribution, 525
model stability, 543–4
distance-​step, 475 taper, 542–3
particle in cell, 475 design, 542
time-​step, 475 dynamic velocity, 543
non-​linearity, 476 positive, 542
gain compression, 476 transfer characteristics, 538–42
harmonic output, 483 gain compression, 540
inter-​modulation products, 491 saturation, 525
saturation, 476 trapped electrons, 522, 524
output coupling, 483 Larmor frequency, 258, 260, 305
 Index 799

Larmor radius, 662, 665 strap break, 590

lasers, 659 strapped, 587–90
Llewellyn-​Peterson equations, 211 dispersion curve, 589
loaded Q, 674 equivalent circuit, 587
long anode magnetron, 594 tuning, 586
vane-​type, 580
magnet, 735–54, see also ferromagnetic materials; capacitance, 585
permanent magnet design dispersion curve, 582
coil-​dominated, 747–52 equivalent circuit, 581
arrays of coils, 750 fringing capacitance, 585
circular loop, 747 inductance, 585
Helmholtz coils, 751 resonant frequency, 584
practical, 749 π –​1 mode, 583, 590
solenoid, 748, 751 π + 1 mode, 583
superconducting, 751 π mode, 583, 590
computational magnetics, 742, 762 magnetron design, 619–24
for crossed-​field tubes, 735 anode cooling, 623
iron-​dominated, 753 case study, 620
pole-​piece design, 753 cathode current, 623
for linear-​beam tubes, 735 DC impedance estimate, 620
periodic permanent magnet, 759 dimensionless parameters, 619
theory, 736 magnetic flux density, 623
Biot-​Savart law, 736 modified Slater factor, 622
demagnetising field, 742 number of vanes, 623
Evershed’s criterion, 744 overall efficiency estimate, 620
flux conduction, 738 practical problems, 624
fringing field, 741, 754 rising sun
leakage flux, 742 cyclotron frequency, 592
load line, 741 scaling, 619
magnetic circuit, 739–45 Slater factor, 622
magnetic circuit law, 737, 740 threshold voltage, 624
magnetic scalar potential, 737 vane tip temperature, 623
magneto-​motive force, 740 magnetron injection gun, 346, 662, 681
permanent magnet, 742 in gyrotron oscillator, 670
permeance, 760 magnetron model, 603–19
reluctance, 740 comparison of models, 604, 612, 614
magnetic DC anode current, 607–​9, 612
boundary conditions, 736 DC cathode current, 606–​7
circuit law, 737 back bombardment, 608
field, 736 primary emission dominated, 603
flux density, 736 primary emission limited, 604
scalar potential, 737 secondary emission dominated, 603
magnetron anode, 580–95 dissipation
coaxial, 593–4 anode, 614
tuning, 593 cathode, 614
degenerate modes, 583 particle in cell, 603
long, 593–5 properties
modes, 594 frequency pushing, 616–19
output coupler, 595 operating point, 614
output coupler, 586 output power, 614
rising sun, 590–3 performance chart, 614
circuit ratio, 590, 592 Rieke diagram, 616
dispersion curves, 591 RF anode current, 618
equivalent circuit, 590 conduction, 617
pass-​bands, 591 induced, 617
stop band, 591 simplifying assumptions, 605
zero mode, 592 space-​charge hub, 603, 606–8
800 Index

magnetron model (cont.) magnetron properties, 596–602

electron trajectories, 606 efficiency, 596
first order (Slater), 607 external match, 598
secondary emission current, 606 frequency, 597
Slater trajectories, 603 pulling, 593, 599
space-​charge density, 606 pushing, 593, 598
zero order (Brillouin), 606 Gauss line discontinuities, 596,
spoke, 603 600–​1
conduction angle, 609, 612 Gauss lines, 596
electron trajectory model, 612 harmonics, 599
equations of motion, 609 long line effect, 599
guiding centre, 609–11 mode selection, 601–2
limiting trajectory, 612 azimuthal periodicity, 586
phase, 609, 612, 617 phase locking, 602
rigid, 608 priming, 602
Welch criterion, 608, 615 moding, 596
synchronous velocity, 611 performance chart, 596
threshold voltage, 608 magnetic field, 596
magnetron oscillator, 565–81, see also operating point, 596, 601
magnetron anode; magnetron design; pulse waveforms, 601
magnetron model; magnetron Rieke diagram, 598
properties spectrum, 599–601
arrangement, 565–​6 chaotic behaviour, 604
anode, 565 dependence on current, 600
cathode, 566 deterioration with age, 600
end hats, 566 twinning, 600
magnetic field, 565 MAGY, 671
output coupler, 566 manufacturing data package, 19
cathode manufacturing specification, 19
back-​bombardment, 568 Marx bank, 777
secondary electron emission, 568 masers, 659
thermionic emission, 568 matching, 75–9
characteristic field, 574 broad-​band, 76
characteristic voltage, 574 equal ripple, 78
circular maximally flat, 77
interaction field, 572–3 stepped-​impedance transformer, 79
space-​harmonic waves, 573 using stubs, 75
standing wave, 573 Maxwell’s equations, 44
electron Maxwell-​Boltzmann energy spectrum, 196
angular velocity, 569, 578 mean free path, 718
equations of motion, 575 monotron oscillator, 494
motion, 567, 569 M-​type Carcinotron, see crossed-​field
spokes, 568 backward-​wave oscillator
electronic efficiency, 580 multipactor, 52, 721–8
mode competition, 567 asymmetric (hybrid) modes, 726–​7
operating point, 577 basic principles, 721
planar on ceramic surface, 722
interaction field, 569–72 coaxial cylinders, 727
space-​harmonic waves, 570 in coaxial lines, 722
standing wave, 571 in crossed fields, 722, 727
relativistic, 565 double-​surface, 727
space-​charge hub, 566, 569 single-​surface, 727
synchronous in klystron cavities, 478
condition, 574, 579 in klystrons, 722
velocity, 575, 578 in magnetrons, 722
threshold voltage, 568, 573–9 modelling, 727
space-​charge wave theory, 578–9 particle in cell code, 727
 Index 801

parallel plates minimum ripple, 279

Hatch diagram, 725 periodic permanent magnet focusing, 270–7
impact velocity, 725 cathode flux, 271
lower boundary, 726 design, 276
phase focusing, 724 double period, 275
resonant condition, 723 equilibrium condition, 274
RF voltage limits, 725 equilibrium radius, 275
theory, 723–6 field variations, 276
upper boundary, 726 long period focusing, 275
phase focusing, 721 magnetic field parameter, 270
saturation, 722 minimum ripple, 271
signs of, 722 random field variations, 276
suppression, 722 ripple, 271
in waveguides, 722 scalloping, 275
in windows, 722 space-​charge parameter, 271
multiple beam stiffness, 275
inductive output tube, 458 permanent magnet design, 754–62
klystron, 500 for crossed-​field tubes, 754–7
multiplexing, 28 for linear-​beam tubes, 756
frequency domain, 28 periodic, 758–62
adjustment, 759
Newton’s second law of motion, 385 permeability, 736
noise, 22 of free space, 736
effective noise temperature, 23 relative, 736
noise figure, 23 phase velocity, 13
noise power ratio (NPR), 32 plane waves, 42
non-​convective instability, 16 phased-​array radar, 3
non-​linear fluid mechanics photo-​cathodes, 457
theory of crossed-​field amplifier, 647 PIC code, see particle in cell code
Pierce electron gun, 318–44
oscillation anode
backward-​wave, 543 aperture, 319
in higher-order mode, 544 effective potential, 325
band-​edge, 544 nose, 321–4
suppression, 544 nose design, 331
feedback, 544 anode lens, 320, 322
monotron, 494 improved model, 324–7
out-​of-​band emissions, 11 spherical aberration, 326
output back-​off, 31 beam
compression, 337
parallel operation control electrodes, 341
of transistors, 3 scraper, 343
of vacuum tubes, 3 waist, 321
particle in cell code, 37, 417 case study, 332
model of collector, 357 computer modelling, 331, 336
model of crossed-​field amplifier, 645 control focus electrode, 341
model of gyrotron, 671 control grid
model of inductive output tube, 458 intercepting, 342
model of klystron, 475 non-​intercepting, 343–4
model of magnetron, 603 transparency, 342
model of multipactor, 727 electrostatic
model of travelling-​wave tube, 521 design, 331–2
Paschen curve theory, 318–22
theory, 717 focus electrode, 321–4
Paschen’s Law, 716, 719 design, 331
Pauli exclusion principle, 694 for hollow beam, 340
periodic electrostatic focusing, 277–80 magnetic design, 333–40
802 Index

Pierce electron gun (cont.) with initial velocity, 196

bucking coil, 337 with injected current, 214
periodic permanent magnet, 338–40 with thermal velocities, 196–203
pole-​piece aperture, 333 Planck’s constant, 697, 707
solenoid, 333–8 plasma frequency, 259, 387, 578
modulating anode, 341 effective, 389
shadow grid, 343–​4 reduction factor, 389
for sheet beam, 340 relativistic correction, 387
thermal velocities, 326–31 Poisson’s equation, 194, 385
Pierce parameters of travelling-​wave tube, 409 cylindrical geometry, 206
pill-​box cavity, 104–10 spherical geometry, 209
modes, 104, 109 Poynting vector, 44, 48, 58
R/​Q, 105 PPM, see periodic permanent magnet
resonant frequency, 105 primary magnetic constant, 736
stored energy, 105 propagation constant, 13
unloaded Q, 106 pulse modulator, 771–9
planar magnetron diode, 287–98 active switch, 639
Brillouin flow, 292, 294 biased-​diode load, 774
charge density, 292 resistive load, 772
conducting bouncer circuit, 774
current, 296 duty cycle, 771
electron trajectory, 297 energy storage in, 771
cut-​off, 288, 298–301 hard tube, see pulse modulator: active switch
electron layer, 298 line type, 639, 777
injected beam, 293 pulse-​forming network, 778
electron trajectories Marx bank, 777
with space-​charge, 292 over-​swing diode, 776
without space-​charge, 290 pulse transformer, 776, 778
electrostatic potential in, 293 voltage
emitting cathode, 294 droop, 773
equations of motion, 289 fall time, 772–​3
guiding centre, 289 over-​swing, 772
Hull cut-​off voltage, 291 ripple, 772
magnetically insulated, 288 rise time, 772–3
Slater two-​stream flow, 299
cathode current, 300 Q factor
charge density, 300 diffractive, see external Q
charge striations, 300 external Q, 98
electron trajectories, 299 loaded Q, 98
planar slow-​wave structure, 145–52 measurement, 128, 131
equivalent circuit, 145 ohmic, see unloaded Q
inter-​digital line, 152, 550 total, see loaded Q
in crossed-​field amplifier, 634 unloaded Q, 95
ladder line, 148 rectangular cavity, 111
meander line, 149
dispersion, 150 Rabbits’ Ears, see band-​edge oscillation
in crossed-​field amplifier, 634 rectangular cavity, 110–11
surface impedance, 151 R/​Q,  111
in travelling-​wave tube, 509 resonance, 110
in travelling-​wave tube, 556 stored energy, 111
planar thermionic diode, 194–206 unloaded Q, 111
dimensionless groups, 195 rectangular waveguide, 58–60
pulsed, 212 attenuation constant, 59
relativistic velocity, 204–6 characteristic impedance, 59
space-​charge limited, 195 frequency band, 60
transit time, 195, 211 higher-order modes, 60
virtual cathode, 196–​7, 218 maximum power, 59
 Index 803

TE mode safety
cut-​off frequencies, 58 high temperatures, 784
equivalent circuit, 59 high-​voltage, 783
field patterns, 60 implosion, 784
TM mode RF power, 783
cut-​off frequencies, 60 toxic materials, 784
field patterns, 60 X-​ray, 784
re-​entrant cavity, 111–21 saturation current density
design, 120 cathode, 196
external, 120 saturation of iron, 670
Fujisawa’s model, 115 scaling
interaction field, 118 travelling-​wave tube, 528
method of moments model, 112 Schottky current, 193, see also saturation current
tuner, 120 density
relativistic magnetron, 565 Schottky effect, see field-​enhanced emission
reliability, 784 second-order intercept point, 8
symptoms and causes, 785 secondary electron emission, 700–6
requirements specification, see statement of coefficient, 701
requirements angle of incidence, 702
resistive wall amplifier, 375 of ceramic, 721
resonant circuit, 94–104 of insulators, 703
bandwidth, 96 of metals, 703
coupled resonators, 102 universal curve, 702
coupling factor, 99 elastically reflected primaries, 701
excitation, 99 electron bombardment, 700
critically coupled, 100 energy distribution, 701
over-​coupled, 101 inelastically reflected primaries, 701
under-​coupled, 101 ion bombardment, 700
external Q, 98 model, 704–6
frequency, 95 angle of incidence, 705
impedance, 96 Furman-​Pivi, 705
loaded Q, 98 surface layers, 706
loss in, 97 surface
parallel, 94–​5 conditioning, 703
phase, 97 contamination, 703
R/​Q, 95 layers, 703
shunt resistance, 96 preparation, 703
stored energy, 96 texture, 704
time constant, 97 true secondary electrons, 701
unloaded Q, 95 in vacuum tubes, 700
voltage reflection coefficient, 101 yield, 701
Richardson-Dushman equation, 696 secondary electrons
ridged waveguide, 60–4 in crossed-​field amplifier, 634
bandwidth, 61 in electron collector, 352–​3, 366
characteristic impedance, 61, 63 in gridded electron gun, 342
TE10 mode cut-​off frequency, 62 in magnetron, 568
Rieke diagram, 780 in multipactor, 721–8
klystron, 485 in tetrode, 246, 444
magnetron, 598, 616 in triode, 440
ring-​bar slow-​wave structure, 166–9 Shannon’s theorem, 22
dispersion, 167 sheet beam
Pierce impedance, 168 klystron, 502
theoretical models, 167 signal to noise ratio, 22
in travelling-​wave tube, 509 signals, 20–33, see also analogue modulation;
ring-​loop slow-​wave structure, 168 digital modulation; multiplexing
in travelling-​wave tube, 509 analogue, 20
Rogowksi profile, 753 bit error rate, 23
804 Index

signals (cont.) space-​charge wave amplitudes, 470

constant envelope, 28 transfer function, 472
digital, 20 transient behaviour, 491
energy per bit, 23 transmission line model, 471
Shannon’s theorem, 22 small-​signal travelling-​wave tube theory, 513–21
signal to noise ratio, 22 coupled-​cavity, 520–1
skin depth, 106 transfer matrix, 520
Slater factor helix, 514–20
modified boundary conditions, 517
crossed-​field amplifier, 655 coupled modes, 515
cylindrical magnetron, 622 gain, 518
planar magnetron, 622 launching loss, 518
slow-​wave structure, 134–44, see also coupled-​ sever, 519
cavity slow-​wave structure; helix slow-​wave sever loss, 520
structure; planar slow-​wave structure; SNR, see signal to noise ratio
ring-​bar slow-​wave structure; ring-​loop solenoid, 670, 751
slow-​wave structure; waveguide slow-​wave current density in, 752
structure design, 752
anti-​Karp structure, 148 for linear-​beam tube, 751
characteristic impedance, 136 superconducting, 670
coupling impedance, 405 solenoid focusing, 259–66
dielectric loaded waveguide, 138 angular velocity of electrons, 261
electric field, 137 beam stiffness, 261, 263–6
equivalent circuit, 136, 139 Brillouin field, 260
forbidden region, 144 Brillouin flow, 260
group velocity, 137, 143 bulk current, 265
Karp structure, 148 cathode flux, 261
period, 139 confined flow, 261
periodic, 139 electron velocity, 261
phase velocity, 136, 143 equation of motion, 259
photonic band-​gap structure, 138 equilibrium beam radius, 261, 396
Pierce impedance, 138, 405 scalloping, 262
space harmonics, 141 space-​charge balanced flow, 261
stop band, 141, 144 surface current, 265
surface impedance, 138 thermal velocities, 550
total impedance, 141 solid-​state devices, see transistors
uniform structures, 134 compared with vacuum tubes, 2–4
slow-​wave structure measurement, 181–3 solid-​state power amplifiers, 3
coupled-​cavity structure soliton theory of crossed-​field
dispersion, 181 amplifier, 646
impedance, 182 space harmonic structure, 174–7
helix structure Chodorow-​Nalos structure, 176
dispersion, 183 dispersion, 175
impedance, 183 equivalent circuit, 174
small-​signal klystron theory, 468–75 parameter calculation, 176
current modulation at output gap, 471 Hughes structure, 174
equivalent circuit, 469 total impedance, 175
gain, 471 space-​charge wave theory, 385–97
gain vs. frequency, 475 beam kinetic voltage, 394
idler cavities, 470 current
idler cavity transfer matrix, 470 body, 390
input cavity, 469 convection, 388, 392, 395
input gap voltage, 470 displacement, 388
output cavity, 471 induced, 390–4
output power, 471 surface, 390
poles and zeroes, 472 cyclotron wave, 397
simplified small-​signal model, 474 dispersion diagram, 387, 390
 Index 805

displacement current, 392 propagation constant, 45

effective tunnel radius, 394 test specification, 19
electromagnetic power, 396 tetrode, 242–8, 433
electronic admittance, 395 case study, 444–8
relativistic, 395 design, 448–51
fast space-​charge wave, 387 electrostatic models, 242
higher-order modes, 396 equivalent triode, 245–​6
kinetic power flow, 395 grid currents, 246
Landau damping, 396 high-​power, 443
on non-​ideal electron beam, 396 island formation, 246
plasma frequency penetration factors, 243
effective, 389 screen grid, 225
reduction factor, 389 space-​charge in, 246
relativistic correction, 387 static characteristics, 245
propagation constant, 390 thermionic cathode, 708–12
radial coupling factor, 393 brightness temperature, 708
slow space-​charge wave, 387 current density, 709
transmission line representation, 394–6 dispenser, 710
spent beam distribution ion bombardment, 709
inductive output tube, 459 life, 708
klystron, 488 measurement, 711
travelling-​wave tube, 525 metal, 709
spherical thermionic diode, 209–11 mixed-​matrix, 712
space-​charge limited, 210 oxide, 710
with relativistic velocity, 210 poisoning, 709
statement of requirements, 19–21 reservoir, 710
statement of work, see statement of saturated current, 711
requirements scandate, 711
sulphur hexafluoride, 716–​17 temperature dependence of emission, 711
surface resistance, 54, 106 temperature measurement, 708
synthetic diamond, 508 thoriated tungsten, 709
system integration, 765–86, see also DC power type M, 551, 710
supply; pulse modulator thermionic diode, 190–4, see also
block diagram, 765 cylindrical thermionic diode; planar
care of tubes, 782 thermionic diode; spherical
control system, 781 thermionic diode
interlocks, 781 3/​2 power law, 192
tube protection, 782 dimensional analysis, 191
cooling system, 780 perveance, 193
reliability, 765, 784 space-​charge limited, 190
RF system, 779 temperature limited, 193
circulator, 780 two-​dimensional flow, 221
input match, 780 thermionic emission, 696
output match, 780 in a magnetron, 568
safety, 783 thyratron, 767
start-​up sequence, 781 TM modes, 49–50
turn-​off sequence, 782 equivalent circuit, 50
field patterns, 50
Tchebychev, see matching: equal ripple wave impedance, 50
TE modes, 46–9 Townsend discharge, 715
characteristic impedance, 48 breakdown voltage, 715
equivalent circuit, 49 Townsend ionisation coefficients, 718
field patterns, 46 transistors, 3
wave impedance, 48 diamond, 3
TEM mode, 45–6 state of the art, 4
characteristic impedance, 46 transmission line
equivalent circuit, 45 TEM mode, 45
806 Index

travelling-​wave tube, 15, 507–14, see also perveance, 224

large-​signal travelling-​wave tube theory; static characteristics, 234–8
small-​signal travelling-​wave tube theory; tunnel effect, 698
travelling-​wave tube design TWT, see travelling-​wave tube
continuous interaction, 507 Twystron, 558
coupled-​cavity, 509
power flow, 510 ubitron, 686–9
discrete interaction, 507 universal beam spreading curve, 321
electron bunching in, 512 unloaded Q
energy conversion in, 510 pill-​box cavity, 106
feedback oscillations, 507
folded waveguide, 509 vacuum tube amplifier, 7–9
gain ripple, 508 bandwidth, 8, 11
helix, 508 efficiency, 9–11
arrangement, 508 electromagnetic structure, 11
mean power dissipation, 509 gain, 8
power holes, 509 gain compression, 8
high frequency limits, 659 harmonics, 7
linearity, 552 linearity, 8
meander-​line, 509 output power, 10
phase space diagram, 512 saturation, 8
ring-​bar, 509 second-order intercept point, 8
ring-​loop, 509 transfer characteristic, 8
sever, 508–​9 vacuum tubes, 3–​12
synchronous point, 510 classification of, 17–18
termination matches, 507 compared with solid-​state devices, 2–4
trapped electrons, 513 design constraints, 20
travelling-​wave tube design, 544–58 efficiency, 3
case study, 545–9 geometry, 5
collector depression, 547 performance requirements, 20
coupled-​cavity, 555 reliability, 3
high efficiency, 550–3 state of the art, 4
hybrid tubes, 556–8 type M, 6
millimetre wave tube, 549 type O, 6
taper, 555 velocity jump amplifier, 375
differential, 556 virtual cathode oscillator, 219
double, 552, 554 Vlasov equation, 666
positive phase velocity, 554 voltage breakdown, 52, 713–21
ultra broad-​band, 553–4 in gas, 715–20
triode, 224–42, 433 ionising collisions, 717
amplification factor μ, 224 mean free path, 718
anode (plate), 224 Paschen’s Law, 716
anode current, 224 RF, 716
case study, 441–3 Townsend discharge, 715
control grid, 224 Townsend ionisation coefficients, 718
design, 448–51 on insulators, 720
dynamic anode (anode slope) resistance, 225 in vacuum, 713–15
electrostatic models, 225–8 contamination, 714
equivalent diode, 234 field enhancement factor, 713
geometrical screening factor, 226 frequency dependence, 714
grid current, 238–40 Kilpatrick’s criterion, 713
high-​power, 439 pulsed DC, 714
inter-​electrode capacitances, 227 RF, 714
island formation, 240–2 surface finish, 714
mututal conductance (transconductance), 224 vacuum arc, 714
penetration factor, 224, 226–​34 lifetime, 715
 Index 807

wave equations, 41 evanescent mode attenuation, 44

wave impedance group velocity, 43
of free space, 54 guide wavelength, 48
waveguide, 41, 52–68, see also circular waveguide; phase velocity, 43
coaxial line; rectangular waveguide; ridged summary of impedances, 65
waveguide waveguide window, 52, 83–8
filled with sulphur hexafluoride, 52 broadband, 87
frequency band, 52 in coaxial line, 86
maximum power, 52 dielectric materials, 84
overmoded, 52 failure mechanisms, 83
pressurisation, 52 in gyrotron oscillator, 670
waveguide coupling half-​wavelength,  84
with change of mode, 81–3 in magnetron, 586
without change of mode, 79–81 multipactor discharge in, 83
waveguide discontinuities, 65–75 pill-​box, 88
capacitive iris in rectangular waveguide, 72 resonant in rectangular
height step in rectangular waveguide, 70 waveguide, 87
inductive iris in rectangular waveguide, 74 stresses in, 84
waveguide slow-​wave structure, 169–73 trapped modes in, 86
folded waveguide, 169 work function, 695, 700
coalesced mode, 173 apparent, 697
dispersion, 170, 172 work-​energy theorem, 575
impedance, 172
inverted, 173 X-​ray absorption, 707
normal, 173 attenuation coefficient, 707
split in crossed-​field amplifier, 634 edges, 707
in travelling-​wave tube, 509, 556 mass attenuation coefficient, 708
helical waveguide, 173 X-​ray emission, 706–8
waveguide theory, 41–52, see also TEM mode; brake radiation, 706
TE modes; TM modes brehmsstrahlung, 706
cut-​off frequency, 43 spectrum, 706
dispersion diagram, 43 X-​ray safety, 784
evanescent mode, 44 X-​ray scattering, 707