LATEST ADVANCES IN VNA

ACCURACY ENHANCEMENTS
A vector network analyzer (VNA) measurement accuracy depends on the accuracy of the calibration
standards, the calibration method employed and the instrumentation accuracy. Advances in signal
processing technology, computation technology, manufacturing technology and computer modeling
have pushed instrumentation and calibration standards accuracies to their limit. To improve accuracy,
new calibration methodologies are being explored. This article presents some of the new developments
in network analyzer accuracy enhancements, including data-based calibration standard models,
weighted least squares calibration and “unknown through” calibration. Comparisons are made
between these new methods and the current widely used traditional methods.

C
alibration standards such as opens and duced by the fitting process, as illustrated in
shorts have traditionally been modeled Figure 2.
by fitting the actual response of the Often, more calibration standards than the
standard to a third-order polynomial. 1 This minimum required are measured. This may
approach has limitations as the frequency be necessary to maintain distinction between
range of the standards increases. Multiple- calibration standards over a very wide fre-
band models for the same calibration standard
were created in an effort to minimize the fit-
ting errors. This required the user to measure Fit Nominal
Response to
the same standard several times — once for Polynomial
Fig. 1 Comparison of 1.85 each model. Model
mm short 1 (5.4 mm offset) Figure 1 shows a comparison of the accu-
polynomial models racy of three different polynomial models for
accuracy. ▼ Compute Nominal
an offset short. Model accuracy Nominal Polynomial
is the difference between the Response from Model (Includes
Low band Nominal Fitting Errors)
High band polynomial model and the data Dimensions
Broadband used to fit the polynomial. The
0.05 low band and high band models
Nominal
optimize the accuracy over por-
MAGNITUDE DEVIATION

Data-based
0.04 tions of the frequency band Model (Elimi-
LINEAR VECTOR

nates Fitting
while the broadband model op- Errors)
0.03
timizes the accuracy over the
0.02 full frequency band. A new cali-
bration standard model type, ▲ Fig. 2 Data-based and polynomial models data
0.01 the data-based standard model, flow.
has been introduced to over-
0
0 10 20 30 40 50 60 70 come this limitation and to re- DAVE BLACKHAM AND KEN WONG
FREQUENCY (GHz) duce the errors in calibration Agilent Technologies Inc.
that had previously been intro- Santa Rosa, CA

Reprinted with permission of MICROWAVE JOURNAL® from the July 2005 issue.
©
2005 Horizon House Publications, Inc.
 T ECHNICAL F EATURE
Equation 1 can be derived from the
signal flow graph:
Γm = e00 – (e11e00 – e10e01) Γa
+ e11ΓmΓa
Compute Nominal e10

Γ Γ
Nominal Data-based (1)
Response from Model (Valid
Nominal for a Particular
e00 e11
To determine the systematic errors
Part Number) m a
Dimensions at least three distinct and known de-
vices must be measured. The general
Characterize e01 solution may be formulated in matrix
Actual Response Data-based form:
1 Γ m1 Γ a1 – Γ a1 
for a Particular Model (Valid

 
Standard (For for a Particu-

1 Γ m2 Γ a 2 –Γa2 
Example, a lar Standard)
Fixed Load)

1 –Γa3 
▲ Fig. 4
Γ m3 Γ a 3
Flow graph representation of a

 
measured reflection coefficient.

M M 
1 – Γ an 
response of calibration standards to a M
 Γ mn Γ an
▲ Fig. 3 Data-based models used as a set of predefined calibration models,

 Γ m1 
generic nominal model or as a model for a
such as coaxial or waveguide. This in-
 
specific device.

   Γ m2 
creased flexibility will also enable

  
quency range. When the accuracy of users to define more easily their cali- e00

×  =  Γ m3 
a single standard is not adequate over bration standards and virtually elimi-
e11 (2)
 e11e00 – e10 e01   M 
the full frequency range, more stan- nates any error that previously would

Γ 
dards are required, such as a fixed have been introduced by the curve
 mn 
load and sliding load combination. fitting process. An example of the
Traditionally, in each of these cases, data-based standard model data file,
all of the standards are measured based on the CITIFILE format, is Better accuracy can be achieved
over the entire frequency range, al- shown in Appendix A. The data may by measuring more than three stan-
though some of the data is simply ig- be obtained by device modeling dards. Electronic calibration, (ECal),
nored. This can lead to discontinu- based on physical dimensions and introduces a least squares solution to
ities in subsequent error corrected properties or from accurate measure- increase the accuracy obtainable with
measurements. Weighted least ments, as illustrated in Figure 3. Un- the over-determined system.7 Equa-
squares (WLS) calibration uses all of certainties in the data are included. tion 2 can be rewritten as a matrix
the data measured on multiple stan- The generic VNA model for a equation:
dards, while seamlessly weighting fixed load is that its reflection coeffi- A×x=b (3)
each standard relative to its accuracy. cient is equal to zero. The actual re-
The data-based standard model in- flection coefficient of the load is the The optimal least squares solution to
cludes uncertainties as part of the de- dominant factor in the directivity and Equation 3 is given as
finition. For non-data-based calibra- source match errors. If a fixed load x = (AH × A)–1 × AH × b (4)
tion standard models, default uncer- can be characterized using a more ac-
tainties are assigned to these models, curate calibration, the resulting char- where
based on the specifications of the cal- acterization data, with uncertainties, AH = conjugate transpose of A
ibration standard. is used as the data-based standard de-
The “unknown through” calibra- finition. Calibrations using the fixed The least squares solution works
tion method was introduced by An- load and its associated data-based well when all observations of calibra-
drea Ferrero and Umberto Pisani2,3 model will have an accuracy ap- tion standards are known with the
in 1992. It was based on the general proaching the accuracy of the system same accuracy. This is a reasonable
theory of VNA calibration that that characterized the fixed load. assumption for ECal, but may not be
brought about the TRL family of cali- Residual calibration errors now de- valid when it is applied to calibrations
brations in the mid-1970s.4,5 This full pend on the uncertainty of the char- using other calibration kits.
two-port calibration technique is well acterization rather than the specifica- A weighted least squares solution
suited for calibrating a vector net- tion of the load. approach provides a simple solution
work analyzer with immovable test to handle the case where the calibra-
ports, to measure non-insertables, WEIGHTED LEAST SQUARED tion standards do not have the same
odd shape and multi-port devices. (WLS) CALIBRATION accuracy.8 If the uncertainties of the
The through “standard” does not Basic Theory6 standards are uncorrelated, an opti-
need to be known and does not need Figure 4 is a signal flow graph mal solution is best obtained by mul-
to be perfect. representation of the relationship be- tiplying each equation by a weighting
tween the measured reflection coeffi- factor that includes both the accuracy
DATA-BASED CALIBRATION cient (Γm) and the actual reflection of the standard’s model and the prox-
STANDARDS coefficient (Γa) plus systematic errors imity of the standard’s response to the
Data-based calibration standards of directivity (e00), reflection tracking other measured calibration standards.
eliminate the requirement to fit the (e 10 e 01 ) and source match (e 11 ). Equation 2 becomes
 T ECHNICAL F EATURE

 1 Γ m1 Γ a1 – Γ a1  tions from those ideal values indicate

 
 σ1 σ1 σ1 
DB WLS Cal
errors in the measurement. The DB DB Standard Cal

 1 Γ m2 Γ a 2 –Γa2 
WLS (data-based standard, weighted Poly Standard Cal

 
least squares) calibration results are
 σ2 σ2 σ2 
0.15
much closer to the ideal values.
 1 Γ m3 Γ a 3 –Γa3 
A precision bead-less coaxial air- 0.10

 
line introduces very low reflections.
 σ3 σ3 σ3 
0.05

|S | (dB)
−0.05
 
When terminated with a short, the
0

 M M 
magnitude of the measured reflection

−1.10

11
 1 – Γ an 
M
Γ mn Γ an
at the input of the coaxial line should
 
−0.15 0
be similar to its insertion loss. Ripples
 σ n σn σ n  in the measured data indicate mea-

 Γ m1 
surement errors caused by residual 10 20 30 40 50 60

 σ 
directivity, source match and reflec- FREQUENCY (GHz)

 1 
tion tracking errors. Figure 7 shows ▲ Fig. 5
 Γ m2 
Comparison of the measured S11
how the three calibration methods
 
magnitude of a flush short.

   σ2 
compared.

   
e00 The benefits of both weighted least
×  =  Γ m3 
DB WLS Cal
squares calibration and data-based cali-

−178
e11 (5)
e11e00 – e10 e01   σ 3 
DB Standard Cal
bration standard models are clearly Poly Standard Cal

 
−179
demonstrated by comparing them to a
 M 
commercially available sliding load kit.
 Γ mn 
−180
The sliding load calibration kit uses the
 
 σn 
traditional modeling for the coaxial

−181

∠S (°)
open, short, load and sliding load. A

11
−182
customized set of definitions for each
where calibration device was provided. A stan-
σi
−1830
dard open, short, load (OSL) sliding
= weighting factor for the ith
load calibration was performed using
equation.
this kit. The Agilent 85058B kit, with 10 20 30 40 50 60

MEASUREMENT RESULTS data-based calibration standard models, FREQUENCY (GHz)

used the WLS calibration. After cali- ▲ Fig. 6 Comparison of the measured S11
The benefits of the weighted least bration, two different devices were phase of a flush short.
squares calibration in conjunction measured, a flush short and the flush
with the data-based model are evalu- short at the end of a 5 cm airline. Each
ated using the Agilent 85058B cali- DB WLS Cal
device was connected and then mea- DB Standard Cal
bration kit. This kit has similar cali- sured with each calibration activated. Poly Standard Cal

−0.2
bration standards for each sex, con- None of the comparison devices were 0
sisting of a low band load, an open, a

−0.4
used as calibration standards during ei-
5.4 mm offset short, a 6.3 mm offset ther calibration. Figures 8, 9 and 10

−0.6
short, a 7.12 mm offset short and a compare the measurements based on
|S | (dB)

−0.8
7.6 mm offset short. the two calibrations.
The improvements due to the

−1.0
11

weighted least squares calibration and UNKNOWN THROUGH

−1.2 0
the data-based model become readily CALIBRATION
apparent when comparing measure- The theoretical base for the un-
ments made with this technique to known through calibration is provid- 10 20 30 40 50 60
measurements made after a traditional ed in the literature.8 This method has FREQUENCY (GHz)
frequency limited, three standard, one- the following requirements: ▲ Fig. 7 Comparison of the measured S11
port calibration and to the traditional • The systematic errors, directivity, magnitude of an airline terminated in a short.
polynomial model-based weighted source match and reflection tracking
least square calibration. of each test port can be completely APPLICATIONS
Measurements of devices that characterized. Immobile Test Ports or Physically

reciprocal, Sij = Sji.

were not used in the calibration • The unknown through must be Long Devices
process, a flush short and the flush Quite often, a direct port 1 to port
short connected to a 5 cm airline, • The phase response of the un- 2 through connection cannot be made
were performed. Figures 5 and 6 known through must be known to because of physical restrictions, such
compare the measured S 11 magni- within a quarter of a wavelength. as wafer probe stations and some test
tude and phase of a flush short. The • The VNA signal path switch errors fixtures. To calibrate such a system,
S11 parameter of a perfect flush short can be quantified. A more detail the traditional SOLT calibration re-
should have a magnitude of 1 (or 0 treatment of this topic was presented quires a known through adapter or ca-
dB) and a phase angle of 180°. Devia- in References 9 and 10. ble assembly. These known devices
 T ECHNICAL F EATURE

DB WLS Cal Good Cable Unknown Through

OSL Sliding Load Cal Flush Through

−0.20
Bad Cable

−0.04
0.2 0

−0.40
0

−0.2 −0.08 −0.60

0

−0.4 −0.12 −0.80

|S | (dB)
|S | (dB)

|S | (dB)
−0.10
−0.6 −0.16

21
11

−0.12

11
−0.8 0 −0.200 −0.14
10 20 30 40 50 60 70 5 10 15 20 25 30 0 20 40 60 80 100
FREQUENCY (GHz) FREQUENCY (GHz) FREQUENCY (GHz)
▲ Fig. 8 Measured S11 magnitude of a flush ▲ Fig. 12 Comparison of S21 measurements
short. ▲ Fig. 11 Errors caused by cable bending. of a 3.5" long cable.

physically long de-

−170
DB WLS Cal
OSL Sliding Load Cal vices, the test ports

−172
must be moved sub-
stantially from the

−174
calibration through

−176
position. By using VNA
One-port
∠S (°)

the unknown

−178
Calibrations
11

through calibration

−180
Cable Movement
method, the test

−1820
ports can be placed
and calibrated at the Full Two-port
Calibration
10 20 30 40 50 60 70
connection point Remove
Adapter
FREQUENCY (GHz)
and thus cable
▲ Fig. 9 Measured S11 phase of a flush
movements are re- One-port Insert Adapter
duced. Figure 12 Calibration (Unknown Through)
short.
shows the difference Adapter Remover Cal Unknown Through Cal
DB WLS Cal between a tradition- ▲ Fig. 13 Adapter removal calibration sequence versus unknown
OSL Sliding Load Cal al flush through cali- through calibration.
0.4 bration and the un-
known through calibration. Notice accommodate their respective orienta-

−0.4
0 that the unknown through results are tion. This is especially true for multi-
less noisy and less spurious. The total port devices such as power splitters

−0.8
|S | (dB)

movement is only 3.5 inches. and directional couplers.

Many characterization methods

−1.2
11

Non-insertable Devices have been proposed to extract three-

−1.6 0
To test non-insertable devices, the port S-parameters from two-port
adapter removal calibration has been measurements.11 Because most three-
10 20 30 40 50 60 70 used. It usually requires two full two- port devices are non-insertable de-
FREQUENCY (GHz) port calibrations or one full two-port vices, an adapter removal calibration
▲ Fig. 10 Measured S11 magnitude of a 5
calibration plus a one-port calibration.
cm long airline terminated in a short. In contrast, the unknown through cali-
1.85 Adapter Removal Cal
bration is as simple as a standard 1.85 Unknown Through Cal
require periodic re-characterization to SOLT calibration, as illustrated in Fig-

−0.05
0
maintain calibration accuracy. In addi- ure 13. Figure 14 compares the mea-

−0.10
tion, uncertainties of the known surement results of the two calibration
through characterization are propa- methods. A 1.85 mm female-to-female

−0.15
|S | (dB)

gated to the uncertainties of the trans- adapter was measured using the two
mission tracking terms and the load calibration methods. Again, the un-

−0.20
21

match terms. No characterization is known through calibration provided a

−0.250
needed using the unknown through cleaner measurement.
calibration method.
Test port cables, even the best Not In-line Test Ports
10 20 30 40 50 60 70
ones, degrade calibration accuracy and Multi-port Devices
FREQUENCY (GHz)
when moved. Figure 11 shows how When the test ports of the device ▲ Fig. 14 Comparison of S21 magnitude
much transmission error some cables under test are not in-line, the test port measurements of a 1.85 mm female-to-female
can cause with a 90° bending. For cable(s) must be bent and moved to adapter.
 T ECHNICAL F EATURE
method must be used. To measure a
three-port device with a two-port net-
work analyzer, the device must be mea-
sured three times, as illustrated in Fig-
ure 15. In addition, most three-port de-
vices do not have in-line connectors. Test
VNA
port cables need to be moved and bent
Full Two-port Changed Cable Position Changed Cable Position to accommodate the device’s connector
Calibration Plane
orientations. These cable movements, af-
ter calibration, cause degradation to the
Changed measurement accuracy. The unknown
Cable
Position through calibration method minimizes
these cable bending and movements, as
Three-port
Device
OR illustrated in Figure 16.
In conjunction with the unknown
through calibration, a two-port to three-
port reconstruction method is proposed
Change Terminations Rotate Port Connections
to characterize the S-parameters of
▲ Fig. 15 Measuring a three-port device with a two-port VNA using the conventional method. three-port devices.12 This makes recip-
rocal three-port measurements almost as
simple as two-port measurements.
If a multi-port VNA is available, the multi-port calibra-
tion is greatly simplified by using the unknown through
method.13 Each test port is calibrated first using the ap-
propriate one-port calibration standards, mechanical or
VNA electronic. Any connector combination is allowed. A mini-
mum set of adapters, cables or even the device under test
No Change to can be used as the unknown through to finish the calibra-
Cable Position tion. For example, a power splitter or a directional cou-
pler can be the unknown through since it meets the reci-
procity requirement, as shown in Figure 17.

One-port Any Through Device

CONCLUSION
OR
Calibrations Device Under Test The enhancements to vector error correction, provided
in the new generation of VNAs, open the possibility to
▲ Fig. 16 Measuring a three-port device with a two-port VNA using more accurate measurements and greater flexibility in the
the unknown through method. calibration process. Data-based standards eliminate the
standard modeling challenges previously faced. Now users
Multiport VNA have a convenient way to model accurately a variety of
standards, including on-wafer standards, stripline stan-
dards, microstrip standards, etc. This article also demon-
strates the unprecedented accuracy that weighted least
squares calibration has achieved. The unknown through
calibration makes non-insertable calibration as simple as
insertable calibration and at the same time minimizes er-
rors caused by test port cable movements. It also elimi-
nates the need of a “perfect through” to calibrate a multi-
port network analyzer. ■
No Change to Cable Position
References
1. Specifying Calibration Standards for the Agilent 8510 Network Ana-
lyzer, Agilent Technologies Inc., Product Note 8510-5B, 2001.
2. A. Ferrero and U. Pisani, “Two-port Network Analyzer Calibration
One-port
Any Reciprocal Using an Unknown ‘Thru,’” IEEE Microwave and Guided Waves
Calibrations Letters, Vol. 2, No. 12, December 1992, pp. 505–507.
Through Device
3. A. Ferrero, U. Pisani and F. Sanpietro, “Save the ‘Thru’ in the ANA
Calibration,” 40th ARFTG Conference Digest, December 1992, pp.
128–135.
4. B. Bianco, M. Parodi, S. Ridella and F. Selvaggi, “Launcher and Mi-
crostrip Characterization,” IEEE Transactions on Instrumentation
and Measurement, Vol. 25, December 1976, pp. 320–323.
5. R.A. Speciale, “A Generalization of the TSD Network Analyzer Cali-
▲ Fig. 17 Multi-port VNA using the unknown through calibration bration Procedure, Covering N-port Scattering Parameter Measure-
method. ments, Affected by Leakage Errors,” IEEE Transactions on Mi-
 T ECHNICAL F EATURE
crowave Theory and Techniques, Vol. 25, David Blackham Ken Wong received
No. 12, December 1977, pp. 1100–1115. received his BSEE his BSEE degree from
6. D.V. Blackham, “Application of Weighted degree from Brigham Cal Poly, SLO, in 1972,
Least Squares to OSL Vector Correction,” Young University, and has been with
61st ARFTG Conference Digest, Philade- Provo, UT, his MSEM Hewlett-Packard and
phia, PA, June 13, 2003. degree from Stanford now Agilent
7. K.H. Wong and R.S. Grewal, “Microwave University, Stanford, Technologies Inc. since
Electronic Calibration: Transferring Stan- CA, his MSEE degree then. His experience at
dards Lab Accuracy to the Production from National HP/Agilent includes
Floor,” Microwave Journal, Vol. 37, No. 9, Technological product design,
September 1994, pp. 94–105. University, Fort manufacturing process
8. G. Strand, Linear Algebra and Its Applica- Collins, CO, and his development, and test
tions, Second Edition, Academic Press PhD degree from the University of Leeds, process development of microwave hybrid
Inc., New York, NY, 1980. Leeds, UK. He has been an R&D engineer with microcircuits and instruments. He is currently
9. D. Rytting, “Appendix to an Analysis of Hewlett-Packard and now Agilent Technologies a senior engineer responsible for the
Vector Measurement Accuracy Enhance- Inc. since 1979. He has worked on the development, design, modeling and
ment Techniques,” Hewlett-Packard RF & HP8340A Synthesized Sweeper, scalar measurement of precision microwave electronic
Microwave Symposium, March 1982. detectors and bridges, RF and microwave and mechanical calibration standards,
10. R.B. Marks, “Formulation of the Basic Vec- vector network analyzers, and microwave verification standards, and calibration
tor Network Analyzer Error Model Includ- characterization of materials. His current methodology of vector network analyzers. He is
ing Switch Terms,” 50th ARFTG Conference responsibilities include vector error correction an officer of the ARFTG (Automatic Radio
Digest, December 1997, pp. 115–126. and measurement accuracy associated with Frequency Techniques Group) Executive
11. S.B. Goldberg, M.B. Steer and P.D. Fran- vector network analyzers. Committee and a senior member of the IEEE.
zon, “Accurate Experimental Characteriza-
tion of Three Ports,” IEEE International
Microwave Symposium Digest, 1991,
pp. 241–244.
12. M. Davidovitz, “Reconstruction of the S-
matrix for a 3-port Using Measurements at
Only Two Ports,” IEEE Microwave and
Guided Wave Letters, Vol. 5, No. 10, Octo-
ber 1995, pp. 349–350.
13. A. Ferrero, F. Sanpietro and U. Pisani,
“Multiport Vector Analyzer Calibration: A
General Formulation,” IEEE Transactions
on Microwave Theory and Techniques, Vol.
42, No. 12, December 1994, pp. 2455–2461.

APPENDIX A VAR Freq MAG 509

Data-based standards are described by a data list. A data-based DATA S[1,1] RI
standard should include a list of frequencies, the actual response for DATA U[1,1] MAG
each S-parameter at each frequency and an estimate of the accuracy of VAR_LIST_BEGIN
the actual response to be used in determining the weighting factor. The 0
example file shows the most basic file format required to define a data- 10000000
based calibration standard. The file must be saved as a text file with a 15000000
“.dat” suffix. …
70000000000
EXAMPLE FILE VAR_LIST_END
CITIFILE A.01.01 BEGIN
COMMENT MODEL: 85058-60101 –1,0
COMMENT SERIAL NUMBER: NOMINAL –0.99976354927,0.00249289367
NAME DATA –0.99970950119,0.00367766097
COMMENT This section describes the s parameter data and …
weighting 0.9772034982,–0.14575300423
COMMENT factor for the calibration standard END
COMMENT COVERAGEFACTOR is used to scale the weighting BEGIN
factor 0.00028
COMMENT S[i,j] is Sij for the standard. Supported formats: RI 0.00028
COMMENT U[i,j] is the weighting factor for Sij 0.00028
COMMENT Supported U[i,j] formats: RI, MAG …
#PNA COVERAGEFACTOR 2 0.005
COMMENT note number of points is 509 below END