DIGITAL

SIGNAL
PROCESSING
lowers Signals and Systems

(ee MATLAB Programs

en S Salivahanan
=) A Vallavaraj
C Gnanapriya
 Information contained in this work has been obtained by
Tata McGraw-Hill, from sources believed to be reliable.
However, neither Tata McGraw-Hill nor its authors
guarantee the accuracy or completeness of any information
published herein, and neither Tata McGraw-Hill nor its
authors shall be responsible for any errors, omissions, or
damages arising out of use of this information. This work
is published with the understanding that Tata McGraw-Hill
and its authors are supplying information but are not
attempting to render engineering or other professional
services. If such services are required, the assistance of an
appropriate professional should be sought.

Tata McGraw-Hill

© 2000, Tata McGraw-Hill Publishing Company Limited

21“ reprint 2007

DZLCRRXYRCYZR

No part of this publication may be reproduced in any form or by any

means without the prior written permission of the publishers

This edition can be exported from India by the publishers,

Tata McGraw-Hill Publishing Company Limited

ISBN 0-07-463996-X

Published by Tata McGraw-Hill Publishing Company Limited,

7 West Patel Nagar, New Delhi 110 008, typeset at
Anvi Composers, A1/33 Pashchim Vihar, New Delhi 110 063 and printed at
A P Offset Pvt. Ltd., Naveen Shahdara, Delhi 110 032

The McGraw-Hill Companies ae

 Contents

Foreword v
Preface vii

1. Classification of Signals and Systems 1

1.2 Classification of Signals 3

1.3 Singularity Functions 9
1.4 Amplitude and Phase Spectra 15
1.5 Classification of Systems 17
1.6 Simple Manipulations of Discrete-time Signals 21
1.7___ Representations of Systems 23
18 Analog-to-Digital Conversion of Signals 28
Review Questions 37
2. Fourier Analysis of Periodic and
Aperiodic Continuous-Time Signals and Systems 40

2.2 Trigonometric Fourier Series 41

2.3 Complex or Exponential form of Fourier Series 52
2.4 Parseval’s Identity for Fourier Series 58
2.5 Power Spectrum of a Periodic Function 59

2.7 __ Properties of Fourier Transform 64

2.8 Fourier Transform of Some Important Signals 75
2.9 Fourier Transform of Power and Energy Signals 103
Review Questions 119

3. Applications of Laplace Transform to System Analysis 127

1 _tutwaiantt 127
3.2 Definition 128
3.3 Region of Convergence (ROC) 128
3.4 Laplace Transforms of Some Important Functions 129
3.5 Initial and Final Value Theore: 137
3.6 Convolution Integral 138
3.7 _ Table of Laplace Transforms 142
3.8 Partial Fraction Expansions 144
Copyrighted material
 3.9 Network Transfer Function 146
3.10 s-plane Poles and Zeros 147
3.11 Laplace Transform of Periodic Functions 154
3.12 Application of Laplace Transformation in
Analysing Networks 157
Review Questions 183

4 l In troid uctiion$ 193

4.3 Properties of z-transform 203

J i ]
Review Questions 228

5. Linear Time Invariant Systems 236

5.2 Properties of a DSP System 238

5.3 Difference Equation and its Relationship with
System Function, Impulse Response and
Frequency Response 256
5.4 Frequency Response 260
Review Questions 272

6.4 Fast Fourier Peepe (FFT) 319

6.5 _Computing an Inverse DFT by Doing a Direct DFT 344
6.6 Composite-radix FFT_352
6.7 Fast (Sectioned) Convolution 368

Review Questions 376

7. Finite Impulse Response (FIR) Filters 380

L1 Introduction 380
7.2 Magnitude Response and Phase Response of
Digital Filters 381
7.3 Frequency Response of Linear Phase FIR Filters 384
7.4 Design Techniques for FIR Filters 385
7.5 Design of Optimal Linear Phase FIR Filters 409
Review Questions 414
8._Infinite Impulse Response (IIR) Filters 417
8.2 IIR Filter Design by Approximation of Derivatives 418
Copyrighted material
 Contents xi

8.3 IIR Filter Design by Impulse Invariant Method 423

8.4 IIR Filter Design by the Bilinear Transformation 427
8.5 Butterworth Filters 432
8.6 Chebyshev Filters 439
8.7 Inverse Chebyshev Filters 444
8.8 Elliptic Filters 445
8.9 Frequency Transformation 446
Review Questions 450

9. Realisation of Digital Linear Systems 453

9.1
Introduction 453
9.2 Basic Realisation Block Diagram and the
Signal-flow Graph 453
9.3 Basic Structures for IIR Systems 455
9.4 Basic Structures for FIR Systems 482
Review Questions 489

10. Effects of Finite Word Length in Digital Filters 496

10.1
Introduction 496
10.2 Rounding and Truncation Errors 496
10.3 Quantisation Effects in Analog-to-Digital
Conversion of Signals 499
10.4 Output Noise Power from a Digital System 502
10.5 Coefficient Quantisation Effects in Direct Form
Realisation
of IIR filters 505
10.6 Coefficient Quantisation in Direct Form

10.7 Limit Cycle Oscillations 510

10.8 Product Quantisation 513
10.9 Scaling 518
10.10 Quantisation Errors in the Computation of DFT 519
Review Questions 521

11. Multirate Digital Signal Processing 523

111
Introduction
523
11.2 Sampling 524
11.3 Sampling Rate Conversion 525
11.4 Signal Flow Graphs 535
115
Structures
Filter 539
11.6 Polyphase Decomposition 541
11.7 Digital Filter Design 551
11.8 Multistage Decimators and Interpolators 555
11.9 Digital Filter Banks 565
11.10 Two-channel Quadrature Mirror Filter Bank 572
11.11
Multilevel Filter Banks 578
Review Questions 581
xii Contents

12. Spectral Estimation 584

12,1
Introduction 584
12.2 Energy Density Spectrum 584
12.3 Estimation of the Autocorrelation and Power Spectrum |
of Random Signals 586
12.4 DFT in Spectral Estimation 591
12.5 Power Spectrum Estimation: Non-Parametric
Methods 593
12.6 Power Spectrum Estimation: Parametric methods 606
Review Questions 628

13. Adaptive Filters 631

13.1 Introduction 631
13.2 Examples of Adaptive filtering 631
13.3 The Minimum Mean Square Error Criterion 643
13.4 The Widrow LMS Algorithm 645
13.5 Recursive Least Square Algorithm 647
13.6 The Forward—Backward Lattice Method 650
13.7 Gradient Adaptive Lattice Method 654
Review Questions 655

14. Applications of Digital Signal Processing 658

14.1 658
Introduction
14.2 Voice Processing 658
14.3 Applications to Radar 671
14.4 Applications to Image Processing 673
14.5
Introduction to Wavelets 675
Review Questions 686

15. MATLAB Programs 688

15.1 Introduction 688
15.2 Representation ofBasic Signals 688
1
15.4 Discrete Correlation 693
15.5 Stability Test 695
15.6 Sampling Theorem 696
15.7 Fast Fourier Transform 699
15.8 Butterworth Analog Filters 700
15.9 Chebyshev Type-1 Analog Filters 706
15.10 Chebyshev Type-2 Analog Filters 712
15.11 Butterworth Digital IIR Filters 718
15.12 Chebyshev Type-1 Digital Filters 724
15.13 Chebyshev Type-2 Digital Filters 729
15.14 FIR Filter Design Using Window Techniques 735
15.15 Upsampling a Sinusoidal Signal 750
15.16 Down Sampling a Sinusoidal Sequence 750
 Contents xiii

15.17 Decimator 751

15.18 Estimation of Power Spectral Density (PSD) 751
15.19
Estimator
PSD 752
15.20 Periodogram Estimation 753
15.21 State-space Representation 753
15.22 Partial Fraction Decomposition 753
15.23
z-transform
Inverse 754
15.24 Group Delay 754
15.25 Overlap-add Method 755
15.26 IIR Filter Design-impulse Invariant Method 756
15.27 IIR Filter Design-bilinear Transformation 756
15.28 Direct Realisation of IIR Digital Filters 756
15.29 Parallel Realisation of IIR Digital Filters 757
15.30 Cascade Realisation of Digital IIR Filters 757
15.31 Decimation by Polyphase Decomposition 758
15.32 Multiband FIR Filter Design 758
15.33 Analysis Filter Bank 759
15.34 Synthesis Filter Bank 759
15.35 Levinson-Durbin Algorithm 759
15.36 Wiener Equation’s Solution 760
15.37 Short-time Spectral Analysis 760
15.38 Cancellation of Echo produced on the
Telephone—Base Band Channel 761
15.39 Cancellation of Echo Produced on the
Telephone—Pass Band Channel 763
Review Questions 765
Appendix A 773
Appendix B 774
Appendix C 782
Index 802
 Chapter 1

Classification of Signals
and Systems

1.1 INTRODUCTION
Signals play a major role in our life. In general, a signal can be a
function of time, distance, position, temperature, pressure, etc., and it
represents some variable of interest associated with a system. For
example, in an electrical system the associated signals are electric
current and voltage. In a mechanical system, the associated signals may
be force, speed, torque, etc. In addition to these, some examples of
signals that we encounter in our daily life are speech, music, picture
and video signals. A signal can be represented in a number of ways.
Most of the signals that we come across are generated naturally.
However, there are some signals that are generated synthetically. In
general, a signal carries information, and the objective of signal
processing is to extract this information.
Signal processing is a method of extracting information from the
signal which in turn depends on the type of signal and the nature of
information it carries. Thus signal processing is concerned with
representing signals in mathematical terms and extracting the
information by carrying out algorithmic operations on the signal.
Mathematically, a signal can be represented in terms of basic functions
in the domain of the original independent variable or it can be
represented in terms of basic functions in a transformed domain.
Similarly, the information contained in the signal can also be extracted
either in the original domain or in the transformed domain.
A system may be defined as an integrated unit composed of diverse,
interacting structures to perform a desired task. The task may vary such
as filtering of noise in a communication receiver, detection of range of a
target in a radar system, or monitoring steam pressure in a boiler. The
function of a system is to process a given input sequence to generate an
output sequence.
2 Digital Signal Processing

It is said that digital signal processing techniques origin in the

seventeenth century when finite difference methods, numerical
integration methods, and numerical interpolation methods were
developed to solve physical problems involving continuous variables and
functions. There has been a tremendous growth since then and today
digital signal processing techniques are applied in almost every field.
The main reasons for such wide applications are due to the numerous
advantages of digital signal processing techniques. Some of these
advantages are discussed subsequently.
Digital circuits do not depend on precise values of digital signals for
their operation. Digital circuits are less sensitive to changes in
component values. They are also less sensitive to variations in
temperature, ageing and other external parameters.
In a digital processor, the signals and system coefficients are
represented as binary words. This enables one to choose any accuracy
by increasing or decreasing the number of bits in the binary word.
Digital processing of a signal facilitates the sharing of a single
processor among a number of signals by time-sharing. This reduces the
processing cost per signal.
Digital implementation of a system allows easy adjustment of the
processor characteristics during processing. Adjustments in the
processor characteristics can be easily done by periodically changing
the coefficients of the algorithm representing the processor
characteristics. Such adjustments are often needed in adaptive filters.
Digital processing of signals also has a major advantage which is not
possible with the analog techniques. With digital filters, linear phase
characteristics can be achieved. Also multirate processing is possible
only in the digital domain. Digital circuits can be connected in cascade
without any loading problems, whereas this cannot be easily done with
analog circuits.
Storage of digital data is very easy. Signals can be stored on various
storage media such as magnetic tapes, disks and optical disks without
any loss. On the other hand, stored analog signals deteriorate rapidly as
time progresses and cannot be recovered in their original form.
For processing very low frequency signals like seismic signals, analog
circuits require inductors and capacitors of a very large size whereas,
digital processing is more suited for such applications.
Though the advantages are many, there are some drawbacks
associated with processing a signal in the digital domain. Digital
processing needs ‘pre’ and ‘post’ processing devices like analog-to-digital
and digital-to-analog converters and associated reconstruction filters.
This increases the complexity of the digital system. Also, digital
techniques suffer from frequency limitations. For reconstructing a
signal from its sample, the sampling frequency must be atleast twice the
highest frequency component present in that signal. The available
frequency range of operation of a digital signal processor is primarily
 Classification of Signals and Sypeme 3

determined by the sample-and-hold circuit and the analog-to-digital

converter, and as a result is limited by the technology available at that
time. The highest sampling frequency is presently around 1GHz
reported by K.Poulton, etal., in 1987. However, such high sampling
frequencies are not used since the resolution of the A/D converter
decreases with an increase in the speed of the converter. But the
advantages of digital processing techniques outweigh the disadvantages
in many applications. Also, the cost of DSP hardware is decreasing
continuously. Consequently, the applications of digital signal processing
are increasing rapidly.

1.2 CLASSIFICATION OF SIGNALS

Signals can be classified based on their nature and characteristics in the

time domain. They are broadly classified as (i) continuous-time signals
and (ii) discrete-time signals. A continuous-time signal is a mathemati-
cally continuous function and the function is defined continuously in
the time domain. On the other hand, a discrete-time signal is specified
only at certain time instants. The amplitude of the discrete-time signal
between two time instants is just not defined. Figure 1.1 shows typical
continuous-time and discrete-time signals.
x(t)

= t
0
(a) Continuous-time signal

xr (n)

E UE 2T
EE n oe
(b) Discrete-time signal
Fig. 1.1 Continuous-Time and Discrete-Time Signals
4 Digital Signal Processing

Both continuous-time and discrete-time signals are further classified

as
(i) Deterministic and non-deterministic signals
(ii) Periodic and aperiodic signals
(iii) Even and odd signals, and
(iv) Energy and power signals.
1.2.1 Deterministic and Non-deterministic Signals
Deterministic signals are functions that are completely specified in time.
The nature and amplitude of such a signal at any time can be predicted.
The pattern of the signal is regular and can be characterised
mathematically. Examples of deterministic signals are
(i) x(t)=at This is aramp whose amplitude increases linearly with
time and slope is a.
(ii) x(t) = A sin wt. The amplitude of this signal varies sinusoidally
with time and its maximum amplitude is A.
1 n20
(iti) x(n) = { This is a discrete-time signal whose
0 otherwise
amplitude is 1 for the sampling instants n 2 0 and for all other
samples, the amplitude is zero.
For all the signals given above, the amplitude at any time instant can
be predicted in advance. Contrary to this, a non-deterministic signal is
one whose occurrence is random in nature and its pattern is quite
irregular. A typical example of a non-deterministic signal is thermal
noise in an electrical circuit. The behaviour of such a signal is
probabilistic in nature and can be analysed only stochastically. Another
example which can be easily understood is the number of accidents in
an year. One cannot exactly predict what would be the figure in a
particular year and this varies randomly. Non-deterministic signals are
also called random signals.
1.2.2 Periodic and Aperiodic Signals
A continuous-time signal is said to be periodic if it exhibits periodicity,
i.e.
x(t+T)=x(t), -o<t<c (1.1)
where T is the period of the signal. The smallest value of T that satisfies
Eq. 1.1 is called the fundamental period, T,, of the signal. A periodic
signal has a definite pattern that repeats over and over, with a
repetition period of T,. For a discrete-time signal, the condition for
periodicity can be written as,
x(n + N,) = x(n), — œ <n < æ ( 1.2)
where N, is the sampling period measured in units of number of sample
spacings. Periodic signals can be in general, expressed as
 Classification of Signals and Systems 5

(i) Continuous-Time Periodic Signals

x,t) = E Xt-iT,) (1.3)

iy
where
x(t), t <t<t n
= 1.
XD {0, elsewhere a4)

(ii) Discrete-Time Periodic Signals

x(n) = [YX(n-iN,), r| (1.5)

where
X(n)= ES nysn<(n, + Nol (1.6)
0, elsewhere
and T is the sampling period in seconds.
A signal which does not satisfy either Eq. 1.1 or 1.2 is called an
aperiodic signal. Some examples of periodic signals are shown in
Fig. 1.2. Some periodic signals can be simply modelled using a single
equation. For example,

xt) =A sin(2249) (1.7)

T
is a continuous-time sinusoidal signal which is valid for all t. The
constant A represents the maximum amplitude and 9 represents the
phase shift of the sinusoidal signal. Similarly, a periodic discrete-time
sinusoidal signal is represented as

x(n) = [asin(222= B)r| (1.8)

No
The term ß represents the delay and T represents the sampling period.
The sum of two or more periodic continuous-time signals need not be
periodic. They“ will be periodic if and only if the ratio of their
fundamental periods is rational. In order to determine whether the sum
of two or more periodic signals is!periodic or not, the following steps may
be used.
(i) Determine the fundamental period of the individual signals in the
sum signal.
(ii) Find the ratio of the fundamental period of the first signal with
the fundamental periods of every other signal.
(iii) If all these ratios are rational, then the sum signal is also periodic.
In the case of discrete-time signals, the sum of a number of periodic
signals is always periodic because the ratio of individual periods is
always the ratio of integers, which is rational.
6 Digital Signal Processing

(a) Continuous-time

x(n)

Ei: -0.6 -0.4 -0.2 0 02 04 06 08 10 12 14

A
soe |È xm-imar
inne

Where
-n, -2sn<0
x[nj=4 n Osn<2
0, otherwise
No=4 and T=0.25
(b) Discrete-time

Fig. 1.2 Some Examples of Periodic Signals

|Example 1.1|Determine which of the following signals are periodic.

(a) x,(t)=sin15 nt (b) xo(t)=sin20nt (c) x,(t)=sin V2nt

(d) x,(t) = sin 5xt (e) x5(t) = x(t) + x(t)
(f) xg(t) = x(t) + x(t)

Solution
(a) x,(t) = sin 15 rt is periodic.
ee P 2n_ 27
The fundamental period is T, = eT
—— = —

= 0.1333333333... seconds
(b) x(t) = sin 20 nt is periodic.
2n_ 20
The fundamental period is T, = w ae 0.1 seconds
 Classification of Signals and Systems 7

(c) x,(t) = sin V2nt is periodic.

ee 2n_ 20
fundamentaltal period dis is T,T, =
The e fund: = ~>— =——
Tr

= 1.41421356... seconds
(d) x4(t) = sin 5rt is periodic.
The fundamental period is T, = =— = 0.4 seconds

(e) x5 (t) = x,(t) + x2 (t)

The fundamental period ofx, (t) = T,, = 0.13333333... seconds and
the fundamental period of x, (t) = Tpz = 0.1 seconds. The ratio of
Ty _ 0.13333333333...
fundamental frequencies, , cannot be
Toa 0.1
expressed as a ratio of integers. Hence, x,(t) is not periodic.
(£) xg (t) = xo (t) + x4 (t)
The fundamental period of x, (t) = T, = 0.1 seconds and the
fundamental period of x, (t) = T,4 = 0.4 seconds. The ratio of

fundamental frequencies, Toa _ 0.1 _ 1 can be expressed as a

Ta 04 4
ratio of integers. Hence, x, (t) is periodic.

1.2.3 Even and Odd Signals

If a signal exhibits symmetry in the time domain, it is called an even
signal. The signal must be identical to its reflection about the origin.
Mathematically, an even signal satisfies the following relation.
For a continuous-time signal, x(t) = x(-t) (1.9a)
For a discrete-time signal, x(n) = x(-n) (1.9b)
An odd signal exhibits anti-symmetry. The signal is not identical to
its reflection about the origin, but to its negative. An odd signal satisfies
the following relation.
For a continuous-time signal, x(t) = —x (—t) (1.10a)
For a discrete-time signal, x(n) = —x (—n) (1.10b)
x(t) = sin wt and x(t) = cos wt are good examples of odd and even
signals, respectively. Figure 1.3 shows the typical odd and even signals.
An even signal which often occurs in the analysis of signals is the sinc
function. The sinc function may be expressed in the following two ways
sin x
according to our convenience: (i) sine (x) = and (ii) sinc (x) =

mati In chapter 2, the first expression is used.

The area under the sinc function is unity. The sinc function is shown
in Fig. 1.3c. The positive portions of the sinc function have angles of
8 Digital Signal Processing

+ nr where n is an even integer, and the negative portions of the sinc

function have angles of + mn where m is odd. It can be seen from Fig.
1.3c that the sinc function exhibits symmetry about x = 0.
x(t) x(n)
A A

(a) Odd signals

x(t) x(n)

(b) Even signals

Sinc (x)

(c) Sinc (x) function

Fig. 1.3 Typical Examples for (a) Odd Signal and (b) Even Signal
(c) the Sinc (x) Function

A signal can be expressed as a sum of two components, namely, the

even component of the signal and the odd component of the signal. The
even and odd components can be obtained from the signal itself, as given
below.
X(t) = Xeyen (É) + Xoga lO) (1.11)
where

Xeven (t) = aa) + x(-t)] and x,a) = iko ~x(-t)]

 Classification of Signals and Systems 9

1.2.4 Energy and Power Signals

Signals can also be classified as those having finite energy or finite
average power. However, there are some signals which can neither be
classified ‘as energy signals nor power signals. Consider a voltage source
v(t), across a unit resistance R, conducting a current i(t). The
instantaneous power dissipated by the resistor is
2
p(t) = v(t) i(t) = Ge = POR
Since R = 1 ohm, we have
p(t) = v%t) = 7(t) (1.12)
The total energy and the average power are defined as the limits
£
E= lim [i?(t)de, joules (1.13)
aay
and
1 T
P= Sng [OG mote (1.14)
The total energy and the average power normalised to unit resistance
of any arbitrary signal x(t) can be defined as
K .
E= lim Jixtt)?? dt, Joules (1.15)
i
asec
and
1 T
P= lim =~ [|x(t)|? de, watts (1.16)
T>e 2T y

The energy signal is one which has finite energy and zero average
power, i.e. x(t) is an energy signal if 0 < E < œ, and P = 0. The power
signal is one which has finite average power and infinite energy, i.e.
0< P < œ, and E = æ. If the signal does not satisfy any of these two
conditions, then it is neither an energy nor a power signal.

1.3 SINGULARITY FUNCTIONS

Singularity functions are an important classification of non-periodic
signals. They can be used to represent more complicated signals. The
unit-impulse function, sometimes referred to as delta function, is the
basic singularity function and all other singularity functions can be
derived by repeated integration or differentiation of the delta function.
The other commonly used singularity functions are the unit-step and
unit-ramp functions.
10 Digital Signal Processing

1.3.1 Unit-Impulse Function

The unit-impulse function is defined as
&(t) = 0,¢ #0 (1.17)
and

J Sode=1 (1.18)
The Equations 1.17 and 1.18 indicate that the area of the impulse
function is unity and this area is confined to an infinitesimal interval on
the t-axis and concentrated at t = 0. The unit impulse function is very
useful in continuous-time system analysis. It is used to generate the
system response providing fundamental information about the system
characteristics. In discrete-time domain, the unit-impulse signal is
called a unit-sample signal. It is defined as
1 n=0
8(n)(n) = {
6, n20 (1.19) J

1.3.2 Unit-step Function

The integral of the impulse function &(¢) gives,
t
1 t>0
J:a(t)t)dt = i Sen (1.20)
1.20

Since, the area of the impulse function is all concentrated at t = 0, for

any value of t < 0 the integral becomes zero and for t > 0, from Eq.1.18,
the value of the integral is unity. The integral of the impulse function is
also a singularity function and called the unit-step function and is
represented as
0, t<0
EER fSi ve
u(t) (1.21 )

The value at ¢ = 0 is taken to be finite and in most cases it is

unspecified. The discrete-time unit-step signal is defined as
0, n<O
um =f sa (1.22)

1.3.3 Unit-ramp Function

The unit-ramp function, r(t) can be obtained by integrating the unit-
impulse function twice or integrating the unit-step function once, i.e.
t @

r(t)= f f&(t)dtdo (1.23)

= fore
 Classification of Signals and Systems U1

That is,
24
t, t>0 “ee
A ramp signal starts at t = 0 and increases linearly with time, t. In
discrete-time domain, the unit-ramp signal is defined as

rin) = {> nso

n, n20
(1.25)
1.3.4 Unit-pulse Function
An unit-pulse function, M(t), is obtained from unit-step signals as shown
below.

M(t)t) = u(t+2) uf 2)
1
ult+—)-uft-= 1
(1.26)
1.26

The signals u(t+ 3)and u(t- 3)are the unit-step signals shifted by i
units in the time axis towards the left and right, respectively.
Figure 1.4 shows some of the singularity functions. The advantage of
the singularity function is that any arbitrary signal that is made up of
straight line segments can be represented in terms of step and ramp
functions.
1.3.5 Properties of ô (t)
1. f 5) dt=1

2. f x(t) 8(t) dt = x(0)

3. f x(t) S(t - to) dt= x(t)

4. | MS-A dà = xlt)

5. Slat) = + 8t)
lal
6. x(t) (t — to) = x(ty)
A x(to) clt - to) = x(to)

t2
8. J x(t) &" (t — to) dt = (-1)" x(t)
ti
12 Digital Signal Processing

öt) 5(n)

:
3
2
1

— ke t -3 -2 -1 0 1 2 3 4 an
í.
A
(a) Unit-impulse function

r(t)

Oo 4 2 3 t -3 -2 -1 0 1 2 3 4n

nie
(c) Unit-ramp function

oT -05 0
(d) knee

Fig. 1.4 Singularity Functions (a) Unit-Impulse Function (b) Unit-Step Function
(c) Unit-Ramp Function (d) Unit-Pulse Function
 Classification of Signals and Systems 13

Proof

4 [x(t) Slt — to] = x(t) S(t — to) + å (t) 8(t — to)

= x(t) 5(t — to) + X (ty) Ölt — to), ty < tọ < tz
Integrating, we get
t d ty J t
[ž(to)(t — to)) dt
lt - ty)] dt + Íi
f Le@se-o) dt = J (x(t)
ti ti ti
ty

[ x ( t )
8(¢- t)]? = jx(t) 5(t — to) dt + å (to)
ti
LHS = 0.
t

Therefore, jix(t) 5(t — ty)dt+ žlto) =0

ti
ty

i.e. j x(t) Èl — to) dt = — x (ty)

t
Similarly,

J x(t) Èl-to)dt =#(¢,)

Hence, Í x(t) 8" (t — to) dt = (-1)" x”(to)

1.3.6 Representation of Signals

In the signal given by x (at + b), i.e., x(a (t + b/a )), a is a scaling factor
and b/a is a pure shift version in the time domain.
If b/a is positive, then the signal x(t) is shifted to left.
If b/a is negative, then the signal x(t) is shifted to right.
If a is positive, then the signal x(t) will have positive slope.
If a is negative, then the signal x(¢) will have negative slope.
Ifa is less than 0, then the signal x(t) is reflected or reversed through
the origin.
If |a| < 1, x(t) is expanded, and if |a| > 1, x(t) is compressed.

Sketch the following signals

(a) x(t) = TI(2ż + 3) (b) x(t) = 2M(t — 1/4)
(c) x(t) = cos(20 nt — 5r) and (d) x(t) = r (— 0.5¢ + 2)
Solution
(a) MM(2t + 3) = TI(2(t + 3/2))
Here the signal shown in Fig. E1.2(a) is shifted to left, with centre at
-3 /2. Since a = 2, i.e. ja | > 1, the signal is compressed. The signal
width becomes 1/2 with unity amplitude.
14 Digital Signal Processing

x(t)

t t
-3⁄2 0 -144 01/4 3/4
Fig. E1.2 (a) Fig. El.2 (b)
(b) x(t) = 2M(¢ — 1/4)
Here the signal shown in Fig. E1.2(b) is shifted to the right, with
centre at 1/4. Since a = 1, the signal width is 1 and amplitude is 2.
(c) x(t) = cos(20 nt- 52)

Here the signal x(t) shown in Fig. E1.2(c) is shifted by quarter cycle
to the right.

Fig. El.2(c)
(d) x(t) =r(—0.5¢ + 2) x

=r (-05 (:a &))

0.5
=r(-0.5 (t -—4)) 2
The given ramp signalis
reflected through the origin and 0 4 r
shifted to right at ¢ = 4. Fig. E1.2 (d)
The signal is expanded by 5 = 2. When ¢ = 0, the magnitude of
the signal x(t) = 2, shown in Fig. E1.2(d).
 Classification of Signals and Systems 15

|Example 1.3|Write down the corresponding equation for the given

signal.
x(t)

Fig. E1.3
Solution
Representation through addition of two unit step functions
The signal x (t) can be obtained by adding both the pulses, i.e.
x(t) = 2[u(t) — u (t — 2)]+[u(t — 3) - u (t - 5))
Representation through multiplication of two unit step functions
x(t) = 2[u (t) u(-t + 2)) + [u (t — 3) u(t + 5))
= 2(u (t) u(2 — t) + u (t — 3) u(5
- t))

1.4 AMPLITUDE AND PHASE SPECTRA

Let us consider a cosine signal of peak amplitude A, frequency f and
phase shift ®, in order to introduce the concept of amplitude and phase
spectra, i.e.,
x(t) = A cos (2nft + ) (1.27)
The amplitude and phase of this signal can be plotted as a function of
frequency. The amplitude of the signal as a function of frequency is
referred to as amplitude spectrum and the phase of the signal as a
function of frequency is referred to as phase spectrum of the signal. The
amplitude and phase spectra together is called the frequency spectrum
of the signal. The units of the amplitude spectrum depends on the
signal. For example, the unit of the amplitude spectrum of a voltage
signal is measured in volts, and the unit of the amplitude spectrum of a
current signal is measured in amperes. The unit of the phase spectrum
is usually radians. The frequency spectrum drawn only for positive
values of frequencies alone is called a single-sided spectrum.
The cosine signal can also be expressed in phasor form as the sum of
the two counter rotating phasors with complex-conjugate magnitudes,
i.e.
16 Digital Signal Processing

ei(2aft+o)
x(t) = Se
2
From this the amplitude spectrum for the signal x(t) consists of two
components of amplitude, viz. A/2 at frequency ‘f’ and A/2 at frequency
‘f’. Similarly, the phase spectrum also consists of two phase
components one at ‘f’ and the other at ‘-/’. The frequency spectrum of
the signal, in this case, is called a double-sided spectrum. The following
example illustrates the single-sided and double-sided frequency spectra
of a signal.

| Example 1.4|Sketch the single-sided and double-sided amplitude

and phase spectra of the signal

x(t) =8 sin (20nt-2),-w <tc

Solution: The single-sided spectra is plotted by expressing x(t) as
the real part of the rotating phasor. Using the trigonometric identity,
cos (u- =). sin u, the given signal is converted into a form as in
Eq.1.27, i.e.
Amplitude Phase shift (radians)
A
84

$3 |
q
i
g 10

Amplitude

]
‘J|
$ 2 +

|
1=
-10 te)

Fig. E.1.4 Amplitude and Phase Spectra (a) Single-Sided and (b) Double-Sided
 Classification of Signals and Systems \7

x(t) =8 sin (20n¢ a z)= 8c0s (20x -ž- z)

= 8 cos (zone - 22)

The single-sided amplitude and phase spectra are shown in

Fig. E.1.4a. The signal has an amplitude of 8 units at f= 10 Hz anda
phase angle of “x radians at f = 10 Hz. To plot the double-sided

spectrum, the signal is converted into the form as in Eq.1.28. Therefore,

j(2one- =) glare - 35)
x(t) = 4e +4
The double-sided amplitude and phase spectra are shown in
Fig. E.1.4b. The signal has two components at f = 10 Hz and f = -10 Hz.
The amplitude of these components are 4 units each and the phase of
27
these components are -—— and 2 radians, respectively.

1.5 CLASSIFICATION OF SYSTEMS

As with signals, systems are also broadly classified into continuous-time
and discrete-time systems. In a continuous-time system, the associated
signals are also continuous, i.e. the input and output of the system are
both continuous-time signals. On the other hand, a discrete-time system
handles discrete-time signals. Here, both the input and output signals
are discrete-time signals.
Both continuous and discrete-time systems are further classified into
the following types.
(i) Static and dynamic systems
(ii) Linear and non-linear systems
(iii) Time-variant and time-invariant systems
(iv) Causal and non-causal systems, and
(v) Stable and unstable systems.
1.5.1 Static and Dynamic Systems
The output of a static system at any specific time depends on the input
at that particular time. It does not depend on past or future values of
the input. Hence, a static system can be considered as a system with no
memory or energy storage elements. A simple resistive network is an
example of a static system. The input/output relation of such systems
does not involve integrals or derivatives.
The output of a dynamic system, on the other hand at any specified
time depends on the inputs at that specific time and at other times.
Such systems have memory or energy storage elements. The equation
characterising a dynamic system will always be a differential equation
18 Digital Signal Processing

for continuous-time system or a difference equation for a discrete-time

system. Any electrical circuit consisting of a capacitor or an inductor is
an example of a dynamic system. The following equations characterise
the dynamic systems.
2
(i) D2. xo,+ Sy(t)=
wa x(t)
Gi) y(n- 1) + ide= 4x(n)- x(n — 1)
1.5.2 Linear and Non-linear Systems
A linear system is one in which the principle of superposition holds.
See Fig. 1.5. For a system with two inputs x,(¢) and x(t), the super-
position is defined as follows.
Hla,x,(t) + agxo(t)] = a Hix (t) + a pH [x,(t)] (1.28)
where, a, and a, are the weights added to the inputs, and H[x(t)} = y(t)
is the response of the continuous-time system to the input x(t). Thus, a
linear system is defined as one whose response to the sum of the
weighted inputs is same as the sum of the weighted responses. If a
system does not satisfy Eq.1.28, then the system is non-linear. For a
discrete-time system, the condition for linearity is given by Eq.1.29.
H{a,x,(n) + agx (n)] = a,H [x,(n)] + aH [x,(n)] (1.29)
where H [x(n)] = y(n) is the response of the discrete-time system to the
input x(n).

x A
Hlaz
(t)+azx(t)]

x(t)

x(t)

aHix(t)] + a2gH(x20)]

x2(h

Fig. 1.5 Illustration of the Superposition Principle

|Example 1.5 Determine whether the system described by the

differential equation oe + 2y(t) = x(t) is linear.

 Classification of Signals and Systems 19

Solution Let the response of the system to x,(t) be y,(t) and the
response of the system to x(t) be y(t) . Thus, for the input x,(¢), the
describing equation is
dy, (t)
“= + 2y; (t) = x,(t)

and for the input x(t),

(t)
2e. + 2yg (©) = x(t)
Multiplying these equations by a, and ag, respectively, and adding
yields,

a oO +a BO + 2a,yı (t) + 2ag

ya (t) = ay x(t) + agxo (t)

i.e.

4 (a,yı (t) + ag yo (t)) + 2a,y; (£) + Ag yo (t)) = a,x, (t) +a 9 XQ (t)

The response ofthe system to the input a,x, (t) + agx; (t) is a; y, (t)
+ Q@2Yq(t). Thus, the superposition condition is satisfied and hence the
system is linear.

|Example 1.6|Determine whether the system described by the

differential equation dy (t) + y (t) + 4 = x(t) is linear.

dt
Solution Let the response of the system to x,(t) be y,(t) and the
response of the system to xz (t) be y(t). Thus, for input x, (t), the
describing equation is

oth
(9 +y (t)+4= x(t)
and for input x, (t),

be + Yq (t) + 4 = x3 (t)

Multiplying these equations by a, and ag, respectively, and adding

yields,

4 (a,yy (t) + a272 (t)) + (ayyı (£) + Gg Hq (t)) + 4a; + ag)

= a,x, (t) + agx (t)
This equation cannot be put into the same form as the original
differential equation describing the system. Hence, the system is non-
linear.
20 Digital Signal Processing

1.5.3 Time-variant and Time-invariant Systems

A time-invariant system is one whose input-output relationship does
not vary with time. A time-invariant system is also called a fixed system.
The condition for a system to be fixed is
H {x(t - 0d) =y (t- 1) (1.30)
A time-invariant system satisfies Eq.1.30 for any x (t) and any value
of t. Equation 1.30 states that if y (t) is the response of the system to
any input x(t), then the response of the system to the time-
shifted input is the response of the system tox (t) time shifted by
the same amount. In discrete time, this property is also referred to as
shift-invariance. For a discrete-time system, the condition for shift-
invariance is given by
H [x (n-k)] =y (n-k) (1.31)
where k is an integer. A system not satisfying either Eq.1.30 or 1.31 is
said to be time-variant. The systems satisfying both linearity and time-
invariant conditions are called linear, time-invariant systems, or
simply LTI systems.
1.5.4 Causal and Non-causal Systems
Acausal system is non-anticipatory. The response of the causal system to
an input does not depend on future values of that input, but depends only
on the present and/or past values of the input. If the response of the
system to an input depends on the future values of that input, then
the system is non-causal or anticipatory. Non-causal systems are
unrealisable. The following difference equations describe causal systems.
(i) y(n) = 0.5 x (n)—x (n - 2)
(ii) y(n) = x (n)
(iii) y(n — 2) + y (n) = x(n) + 0.98 x (n - 1)
The following equations describe non-causal systems.
Gi) y(n- 1 =x(n)
(ii) y(n) = 0.11x(n-1) + x(n) - 0.8 x(n + 1)
1.5.5 Stable and Unstable Systems
A system is said to be bounded-input, bounded-output (BIBO) stable, if
every bounded input produces a bounded output. A bounded signal has
an amplitude that remains finite. Thus, a BIBO stable system will have
a bounded output for any bounded input so that its output does not grow
unreasonably large. The conditions for a system to be BIBO stable are
given below.
(i) If the system transfer function is a rational function, the degree of
the numerator must be no larger than the degree of the
denominator.
(ii) The poles of the system must lie in the left half of the s-plane or
within the unit circle in the z-plane.
 Classification of Signals and Systems 21

(Gii) If a pole lies on the imaginary axis, it must be a single-order one,

i.e. no repeated poles must lie on the imaginary axis.
The systems not satisfying the above conditions are unstable.

1.6 SIMPLE MANIPULATIONS OF DISCRETE-TIME SIGNALS

When a signal is processed, the signal undergoes many manipulations
involving the independent variable and the dependent variable. Some of
these manipulations include (i) shifting the signal in the time domain,
(ii) folding the signal and (iii) scaling in the time-domain. A brief
introduction of these manipulations here, will help the reader in the
following chapters.
1.6.1 Transformation of the Independent Variable

Shifting
In the case of discrete-time signals, the independent variable is the time,
n. A signal x(n) may be shifted in time, i.e. the signal can be either
advanced in the time axis or delayed in the time axis. The shifted signal
is represented by x(n — k), where k is an integer. If ‘k’ is positive, the
signal is delayed by k units of time and if k is negative, the time shift
results in an advance of signal by k units of time. However, advancing
the signal in the time axis is not possible always. If the signal is available
in a magnetic disk or other storage units, then the signal can be delayed
or advanced as one wishes. But in real time, advancing a signal is not
possible since such an operation involves samples that have not been
generated. As a result, in real-time signal processing applications, the
operation of advancing the time base of the signal is physically
unrealizable.

Folding
This operation is done by replacing the independent variable n by —n.
This results in folding of the signal about the origin, i.e. n = 0. Folding is
also known as the reflection of the signal about the time origin n = 0.
Folding of a signal is done while convoluting the signal with another.

Time scaling
This involves replacing the independent variable n by kn, where k is an
integer. This process is also called as down sampling. If x(n) is the
discrete-time signal obtained by sampling the analog signal, x(t), then
x(n) = x(nT), where T is the sampling period. If time-scaling is done,
then the time-scaled signal, yin] =x (kn) =x(knT). This implies that the
sampling rate is changed from 1/T to 1/kT. This decreases the sampling
rate by a factor of k. Down-sampling operations are discussed in detail
in Chapter 11 of this book. The folding and time scaling operations are
shown in Fig. 1.7(a) and (b).
22 Digital Signal Processing

-8 -6 -4 -2 (0) 2 4 6 8
(a) Unit-step signal
u(n- 4)

TA . = . = .—ei

(b) Delaying the unit-step signal by 4 units

u(n+ 4)

eer =8 -4
TE LLTI
-2 0 2 4 6 8
L,
(c) Advancing
the unit-step signal by 4 units

Fig. 1.6 Graphical Representation of a Signal, and

Its Delayed and Advanced Versions

x(n)
Ma HOOKS S81 1.08

tet,
-7 -6 -6 -4 -3 -2

-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8H
(a) Folding

Fig. 1.7 (Contd.)

 Classification of Signals and Systems 23

x(n)
x(n) ={1, 1,2,1, = 2,0, 1, 3, 1}

x(2=n{1), 2,1,2, 1, 1}

a a

eee
-7 -6 -5 -4 -3 -2 -1 0
|
1 N a a a a s aol ay

(b) Time scaling

Fig. 1.7 Illustration of Folding and Time Scaling Operations

1.7 REPRESENTATIONS OF SYSTEMS

The representation of a system helps in visualising the system with its
components and their interconnections. A system can be represented
using a diagram featuring various components of the system. These
components are represented by symbols. An electrical system is thus
represented by a diagram consisting of different symbols representing
resistors, capacitors or any other device. A mechanical system can be
represented using symbols for different elements like damper,
acceleration, zero friction, etc. Thus, a system can be visualised when
represented in the form of a diagram.
A more convenient form of representing a system is the block-diagram
representation. In this form of representation, each box is an operator
on the input signal and the operation is shown on the box itself. Some
typical operations include integration, differentiation, scalar
multiplication, delay, etc. The lines connecting the individual boxes are
directional and these lines show the direction of the signal flow. A
number of lines may terminate at a node. The node may be either an
accumulator or a multiplier. These nodes are represented by circles with
the symbols ‘+’ or ‘x’ marked on them. Figure 1.8 shows the block
diagram representation of continuous-time and discrete-time systems,
A system can also be represented using mathematical models. The
analysis of system characteristics and performance can be carried out
using the mathematical model of a system. The mathematical model of a
24 Digital Signal Processing

(a) Continuous-time

: lon | y(n)
=Hy(x(n))
Jes

—
y(n)
— T7

ys(n) = Ha[ya(n)]

yaln) =Haly(n)]

(b) Discrete-time

Fig. 1.8 Block Diagram Representation of (a) Continuous-Time System

(b) Discrete-Time System

system consists of equations relating to the signals of interest. Both

discrete-time and continuous-time signals can be modelled using the
following three methods.
Gi) A linear difference/differential equation
(ii) The impulse-response sequence
(iii) A state-variable or matrix description.
All the above three methods help in determining the output of the
system from the knowledge of the input to the system. A system can be
interpreted in different ways depending on the model since each model
emphasise certain aspects of the system. Together, these models provide
a very good understanding of the system and how the system works. In
the following sections, these models are discussed further.
1.7.1 Linear Difference/Differential Equations
A discrete-time system is modelled by a difference equation, whereas a
continuous-time system is modelled by a differential equation. In a linear
discrete-time system, the input sequence {x,} is transformed into an
output sequence {y,,} according to some difference equation. For example,
y(n) = x(n) + 3x(n — 1) + 2x(n — 2) (1.32)
is a linear difference equation which tells that the nth member of the
output sequence y(n) is obtained by accumulating (adding) the input at
the present moment, x(n), with thrice the previous input, x(n — 1) and
 Classification of Signals and Systems 25

twice the input delayed twice, x(n — 2). Let the input sequence be x(n) =
{0,1, 1, 2, 0, 0, 0, ...}. The output sequence for the system as described by
Eq. 1.32 is y(n) = {0, 1, 4, 7, 8, 4, 0, 0, ...}. The block diagram
representation of the system described by Eq.1.32 is shown in Fig. 1.9.
x(n) y(n)
Asn

y(n) =x [n] +3x[n - 1] +2x[n- 2]

Fig. 1.9 Discrete-Time System Corresponding to Eq. 1.32

In digital signal processing applications, our prime concern is of

linear, time-invariant discrete-time systems. Such systems are modelled
using linear difference equations with constant coefficients. The block
diagram representation of these systems contain only unit delays,
constant multipliers and adders.
A continuous-time system is modelled by a linear differential
equation. An ordinary linear differential equation with constant
coefficients characterises linear, constant parameter systems. For
example, an nth order system is represented by
n n-1
a, 2 or ea cores we
+ y(t)=x(t) (1.33)

The general solution of the above equation consists of two

components, namely, the homogeneous solution and the particular
solution. The homogeneous solution is the sourcefree, natural solution
of the system, whereas the particular solution is the component due to
the source x(t).
1.7.2 Impulse Response of a System
The impulse response of a system is another method for modelling a
system. The impulse response of a linear, time-invariant system is the
response of the system when the input signal is an unit-impulse
function. The system is assumed to be initially relaxed, i.e. the system
has zero initial conditions. The impulse response of a system is
represented by the notation A(t) (continuous-time) or h(n) (discrete-
26 Digital Signal Processing

time). If y(t) is the system response for an input x(t), then the response
of the system when x(t) = &(t) is y(t) = A(t).
The impulse response of a system can be directly obtained from the
solution of the differential or difference equation characterising the
system. The impulse response is also determined by finding out the

output of the system to the rectangular pulse input x(¢) = zIl ($) and
then taking the limit of the resulting system response, y(t) as € > 0. The
unit-impulse function is nothing but the derivative of the unit-step
signal. Therefore, the impulse response of the system can also be
obtained by computing the derivative of the step response of the system.
1.7.3 State-Variable Technique
The state-variable technique provides a convenient formulation
procedure for modelling a multi-input, multi-output system. This
technique also facilitates the determination of the internal behaviour of
the system very easily. The state of a system at time tọ is the minimum
information necessary to completely specify the condition of the system
at time fy and it allows determination of the system outputs at any time
t > to, when inputs upto time ¢ are specified. The state of a system at
time fy is a set of values, at time tọ, of a set of variables. These variables
are called the state variables. The number of state variables is equal to
the order of the system. The state variables are chosen such that they
correspond to physically measurable quantities. It is also convenient to
consider an n-dimensional space in which each coordinate is defined by
one of the state variables
x,, Xo, ...,X,, where n is the order of the system.
This n-dimensional space is called the state space. The state vector is
an n-vector x whose elements are the state variables. The state vector
defines a point in the state space at any time t. As the time changes, the
system state changes and a set of points, which is nothing but the locus
of the tip of the state vector as time progresses, is called a trajectory of
the system.
A linear system of order n with m inputs and k outputs can be
represented by n first-order differential equations and & output
equations as shown below.
dx,
ry = 04, X1 + AygQXqt ... + AyyXp_ + Oy) Uy + Ogu g+... + bim Um

dxs
E = 1 Xy + Ogg Xo + ... + Aon Xn + O91 Uy + Ogg g+... + DamUm

` (1.84)

we
d a= an1 ž1 + Ang Xz+... t brougt
+ Ann Xn + b On)Uy+b ... + Onm Um
+...+6,,,u
 Classification of Signals and Systems 27

and
Yı = Cy Xy + Cig Xot... + Cy_ Xp, + yy Uy t+diglot ... + dy, Up
Yq = C21Xy + Cog Xo+ ... + ConXn+ dgy Uy + doggt ... + dom Um

. (1.35)

Yp = CyyXy
+Cho Xot ... + Can Xn +tAyy Uy + Ayg llgt ... + dpmUm
where u;, i = 1, 2, ..., m are the system inputs, x;, i = 1, 2,3, ..., n are
called the state variables and y; i = 1, 2, 3, ..., k are the system outputs.
Equations 1.34 are called the state equations, and Eqs 1.35 are the
output equations. Equations 1.34 and 1.35 together constitute the state-
equation model of the system. Generally, the a’s, b’s, c’s and d’s may be
functions of time. The solution of such a set of time-varying state
equations is very difficult. If the system is assumed to be time-invariant,
then the solution of the state equations can be obtained without much
difficulty.
The state variable representation of a system offers a number of
advantages. The most obvious advantage of this representation is that
multiple-input, multiple-output systems can be easily represented and
analysed. The model is in the time-domain, and one can obtain the
simulation diagram for the equations directly. This is of much use when
computer simulation methods are used to analyse the system. Also, a
compact matrix notation can be used for the state model and using the
laws of linear algebra the state equations can be very easily
manipulated. For example, Eqs 1.34 and 1.35 expressed in a compact
matrix form is shown below. Let us define vectors
xy uy yı
x u
x= tg , L= $ > ya sa (1.36)

Xn Um Yk
and matrices
Oy Gg - + + Om by biz ~~ + bim
G2, An > - + Ban ba ba . . . bom
A=]. Pia sae Sh. BS s E r -& (1.37)

ani an2 ann bn 1 bnz bam

Cu Ci Cin dı d dim
C21 C22 Con dı dz dom
Ce , D=

Cer Ceo + + + Chn dya Gyo => e dpm

28 Digital Signal Processing

Now, Eqs 1.34 and 1.35 can be compactly written as

x = Ax + Bu (1.38a)
y = Cx + Du (1.38b)
where x = dv/dt. Equations 1.38 may be illustrated schematically as
shown in Fig.1.10. The double lines indicate a multiple-variable signal
flow path. The blocks represent matrix multiplication of the vectors and
matrices. The integrator block consists ofn integrators with appropriate
connections specified by the A and B matrices.
k

Fig. 1.10 Block Diagram of the State-Variable Model of Eq. 1.38

State Equations for Discrete-time Systems

For a discrete-time system, the state equations form a set of first-order
difference equations constituting a recursion relation. This recursion
relation allows determination of the state of a system at the sampling
time kT from the state of the system and the input at the sampling time
(k — 1)T, where k is an integer. The state-equations for a discrete-time
system can be modelled as shown below.
Xho = Fx + Gu, (1.39a)
Yk = Hx, + Ju, (1.39b)

The dependence of these parameters on T is suppressed for simplicity.

For a single input, single output system u, and y, are scalars and G and
H become vectors g and h, and J is a null in most cases. The state-
variable modelling of a discrete-time system finds application in the
digital simulation of a continuous time systems.

1.8 ANALOG-TO-DIGITAL CONVERSION OF SIGNALS

A discrete-time signal is defined by specifying its value only at discrete
times, called sampling instants. When the sampled values are quantised
and encoded, a digital signal is obtained. A digital signal can be obtained
from the analog signal by using an analog-to-digital converter. In the
following sections the process of analog-to-digital conversion is
 Classification of Signals and Systems 29

discussed in some detail and this enables one to understand the

relationship between the digital signals and discrete-time signals.
Figure 1.11 shows the block diagram of an analog-to-digital
converter. The sampler extracts the sample values of the input signal at
the sampling instants. The output of the sampler is the discrete-time
signal with continuous amplitude. This signal is applied to a quantiser
which converts this continuous amplitude into a finite number of sample
values. Each sample value can be represented by a digital word of finite
word length. The final stage of analog-to-digital conversion is encoding.
The encoder assigns a digital word to each quantised sample. Sampling,
quantizing and encoding are discussed in the following sections.

Ls|Sampler jra
cee BEBie eal
Continuous-time Discretoiüme Discrete-time poni output
continuous-amplitude Continuous-amplitude discrete-amplitude
input signal signal signal

Fig. 1.11 Analog-to-Digital Converter

1.8.1 Sampling of Continuous-time Signals

Sampling is a process by which a continuous-time signal is converted
into a discrete-time signal. This can be accomplished by representing
the continuous-time signal x(t), at a discrete number of points. These
discrete number of points are determined by the sampling period, T, i.e.
the samples of x(t) can be obtained at discrete points t = nT, where n is
an integer. The process of sampling is illustrated in Fig.1.12. The
sampling unit can be thought of as a switch, where, to one of its inputs
the continuous-time signal is applied. The signal is available at the
output only during the instants the switch is closed. Thus, the signal at
the output end is not a continuous function of time but only discrete
samples. In order to extract samples of x(t), the switch closes briefly
every T seconds. Thus, the output signal has the same amplitude as x(t)
when the switch is closed and a value of zero when the switch is open.
The switch can be any high speed switching device.
The continuous-time signal x(t) must be sampled in such a way that
the original signal can be reconstructed from these samples. Otherwise,
the sampling process is useless. Let us obtain the condition necessary to
faithfully reconstruct the original signal from the samples of that signal.
The condition can be easily obtained if the signals are analysed in the
frequency domain. Let the sampled signal be represented by x,(¢). Then,
x, (t) = x(t) g(t) (1.40)
where g(t) is the sampling function. The sampling function is a
continuous train of pulses with a period of T seconds between the pulses,
and it models the action of the sampling switch. The sampling function
is shown in Fig. 1.12(c) and (d). The frequency spectrum of the sampled
30 Digital Signal Processing

Samples of x(t)
x(t)

0 T 2T 3T 4T 5T 6er t
(a) Samples of x (4)

|Switch (9 ai
x(t) x (i) e
= a g(t)
(b) Modelling a sampler as a switch (c) Mode! of a sampler

>t H-

| |
a ee khae A eee ki eer
T 2T 3T 4T 5T 6T

(d) Sampling function

Fig. 1.12 The Sampling Process

signal x,(t) helps in determining the appropriate values of T for

reconstructing the original signal. The sampling function g(t) is periodic
and can be represented by a Fourier series (Fourier Series and
transforms are discussed in Chapter six), i.e.

g= SC, ent (1.41)

na-

where
r

C= goetan: (1.42)
B
2

is the nth Fourier coefficient of g(t), and f, = 7 is the fundamental

frequency of g(t). The fundamental frequency, f, is also called the
sampling frequency. From Eqs.1.40, and 1.41, we have

nÒ FC = Eoen (143)
na-am Nao

The spectrum of x,(¢), denoted by X, (f), can be determined by taking

the Fourier transform of Eq. 1.43, i.e.
 Classification of Signals and Systems 31

X,(f)= |x) e7?" dt (1.44)

Using Eq.1.43 in the above equation,

X= | E Crt) e”? eiat (1.45)

=% n=-

Interchanging the order of integration and summation,

X= EC, fae ae (1.46)

n=- -o

But from the definition of the Fourier transform

Jaee- dt = Xf- nf)

Thus,

XQ = 2 C,, Xf- nf,) (1.47)

From Eq. 1.47, it is understood that the spectrum of the sampled

continuous-time signal is composed of the spectrum of x(t) plus the
spectrum of x(t) translated to each harmonic of the sampling frequency.
The spectrum of the sampled signal is shown in Fig. 1.13. Each
frequency translated spectrum is multiplied by a constant. To
reconstruct the original signal, it is enough to just pass the spectrum of
x(t) and suppress the spectra of other translated frequencies. The
amplitude response of such a filter is also shown in Fig. 1.13. As this
filter is used to reconstruct the original signal, it is often referred to as a
reconstruction filter. The output of the reconstruction filter will be
CoX(f) in the frequency domain and x(t) in the time-domain.

xh

Fig. 1.13 Spectrum of Sampled Signal

32 Digital Signal Processing

The signal x(t), in this case, is assumed to have no frequency

components above f,, i.e. in the frequency domain, X(f) is zero for
If | 2 fp- Such a signal is said to be bandlimited. From Fig.1.13, it is
clear that in order to recover X(f) from X,(/), we must have
h-hh,
or equivalently,
fs 2 2f;,, hertz (1.48)
That is, in order to recover the original signal from the samples, the
sampling frequency must be greater than or equal to twice the
maximum frequency in x(t). The sampling theorem is thus derived,
which states that a bandlimited signal x(t) having no frequency
components above f, hertz, is completely specified by samples that are
taken at a uniform rate greater than 2f, hertz. The frequency equal to
twice the highest frequency in x(t) , i.e. 2f,, is called the Nyquist rate.
Sampling by Impulse Function
The sampling function g(t), discussed above, was periodic. The pulse
width of the sampling function must be very small compared to the
period, 7. The samples in digital systems are in the form of a number,
and the magnitude of these numbers represent the value of the signal
x(t) at the sampling instants. In this case, the pulse width of the
sampling function is infinitely small and an infinite train of impulse
functions of period T can be considered for the sampling function. That
is,

g(t)= }8t-nT) (1.49)

n=-«
The sampling function as given in Eq.1.49 is shown in Fig.1.14. When
this sampling function is used, the weight of the impulse carries the
sample value.
The sampling function g(t) is periodic and can be represented by a
Fourier series as in Eq.1.41, which is repeated here.

gi= E C,e"r4
i.
where
i"
Gal face etar (1.50)
T 7

Since ô(t) has its maximum energy concentrated at t = 0, a more

formal mathematical definition of the unit-impulse function may be
defined as a functional
 Classification of Signals and Systems 33

g()

1
A
t
-6T -5T -4T-3T -2T-T 0 T 27 3T 4T ST 6T

(a)
Ax)

Fig. 1.14 (a) Impulse Sampling Function (b) Spectrum of the Signal x(t)
(c) Spectrum of Impulse Sampled Signal

Í x(t) &(t) dt = x(0) (1.51)

where x(t) is continuous at ¢ = 0. Using Eq. 1.51 in Eq. 1.50, we have

Cc n =7E
eet o_1_ "T fe (1.52)
5

Thus C, is same as the sampling frequency f,, for all n. The spectrum
of the impulse sampled signal, x,(t) is given by

XP=f, YX -nf,) (1.53)

n=-o

The spectra of the signal x(t) and the impulse sampled signal X, (t)
are shown in Figs 1.14 (b) and (c). The effect of impulse sampling is
same as sampling with a train of pulses. However, all the frequency
translated spectra have the same amplitude. The original signal X(f
can be reconstructed from X,(f) using a low-pass filter. Figure 1.15
shows the effect of sampling at a rate lower than the Nyquist rate.
Consider a bandlimited signal x(t), with f, as its highest frequency
content, being sampled at a rate lower than the Nyquist rate, i.e.,
sampling frequency f, < 2f}. This results in overlapping of adjacent
34 Digital Signal Processing

x(f)

-h 0 fn

(a) Spectrum of the input signal

Xan)

f
-6 -h O h fs-ħfhs fth
(b) Spectrum
of the sampled signal's tor fs > 2fp

>» f

(c) Sampled signal's spectrum for f, < 2f,

Fig. 1.15 Illustration of Aliasing

spectra i.e., higher frequency components of X,(/f) get superimposed on
lower frequency components as shown in Fig.1.15. Here, faithful
reconstruction or recovery of the original continuous time signal from
its sampled discrete-time equivalent by filtering is very difficult because
portions of X(f - f,) and X(f + f,) overlap X(f), and thus add to X(f) in
producing X,(f). The original shape of the signal is lost due to
undersampling, i.e. down-sampling. This overlap is known as aliasing
or overlapping or fold over. Aliasing, as the name implies, means that a
signal can be impersonated by another signal. In practice, no signal is
strictly bandlimited but there will be some frequency beyond which the
energy is very small and negligible. This frequency is generally taken as
the highest frequency content of the signal.
To prevent aliasing, the sampling frequency f, should be greater than
two times the frequency f, of the sinusoidal signal being sampled. The
condition to be satisfied by the sampling frequency to prevent aliasing is
called the sampling theorem. In some applications, an analog anti-
aliasing filter is placed before sample/hold circuit in order to prevent the
aliasing effect.
 Classification of Signals and Systems 35

A useful application of aliasing due to undersampling arises in the

sampling oscilloscope, which is meant for observing very high frequency
waveforms.
1.8.2 Signal Reconstruction
Any signal x(t) can be faithfully reconstructed from its samples if these
samples are taken at a rate greater than or equal to the Nyquist rate. It
can be seen from the spectrum of the sampled signal, X,(t) that it
consists of the spectra of the signal and its frequency translated
harmonics. Thus, if the spectrum of the signal alone can be separated
from that of the harmonics then the original signal can be obtained.
This can be achieved by filtering the sampled signal using a low-pass
filter with a bandwidth greater than f, and less than f,- f,. hertz.
If the sampling function is an impulse sequence, we note from Eq.1.53
that the spectrum of the sampled signal has an amplitude equal to
f, = 1/T. Therefore, in order to remove this scaling constant, the low-pass
filter must have an amplitude response of 1/f, = T. Assuming that
sampling has been done at the Nyquist rate, i.e. f, = 2f,, the bandwidth

of the low-pass filter will bef, = a. Therefore, the unit impulse response
of an ideal filter for this bandwidth is
h12
ht)=T fe df (1.54)
~f,/2

That is

h(t)
=jax"
—P_ (eit lt _ gg
T
git jx fat -jR fat

The above expression can be alternatively written as

sin t f,t
A(t)=Tf, a =sinc f, t (1.55)
s

The ideal reconstruction filter is shown in Fig.1.16a. The input to this

filter is the sampled signal x(nT) and the output of the filter is the
reconstructed signal x(t). The output signal x(t) is given by

xt)= Y xnT)ht-nT)
Using Eq.1.55, we get

x(t) = £ x(nT) sine f,(t —- nT) (1.56)

The above expression is a convolution expression and the signal x(t)

is reconstructed by convoluting its samples with the unit-impulse
response of the filter. Eq. 1.56 can also be interpreted as follows. The
36 Digital Signal Processing

original signal can be reconstructed by weighting each sample by a sinc

function and adding them all. This process is shown in Fig. 1.16b.

> x(nT)8(t- AT) |

ee cs, _,, [deal reconstruction filter x(0)
| h(t) = sinc ft 5 g
L
(a)
x(t)
\ Samples of x(t)

(b)
Fig. 1.16 Signal Reconstruction (a) Reconstruction Filter
(b) Time Domain Representation

1.8.3 Signal Quantisation and Encoding

A discrete-time signal with continuous-valued amplitudes is called a
sampled data signal, whereas a continuous-time signal with discrete-
valued amplitudes is referred to as a quantised boxcar signal.
Quantisation is a process by which the amplitude of each sample of a
signal is rounded off to the nearest permissible level. That is,
quantisation is conversion of a discrete-time continuous-amplitude
signal into a discrete-time, discrete-valued signal. Then encoding is
done by representing each of these permissible levels by a digital word
of fixed wordlength.
The process of quantisation introduces an error called quantisation
error and it is simply the difference between the value of the analog
input and the analog equivalent of the digital representation. This error
will be small if there are more permissible levels and the width of these
quantization levels is very small. In the analog-to-digital conversion
process, the only source of error is the quantiser. Even if there are more
quantisation levels, error can occur if the signal is at its maximum or
minimum value for significant time intervals. Figure 1.17 shows how a
continuous-time signal is quantised in a quantiser that has 16
quantising levels.
 Classification of Signals and Systems 37

Quantisation level

[0] E 2T 3T 4T
Fig. 1.17 Quantizing and Encoding

REVIEW QUESTIONS
1.1 What are the major classifications of signals?
1.2 With suitable examples distinguish a deterministic signal from
a random signal.
1.3 What are periodic signals? Give examples.
1.4 Describe the procedure used to determine whether the sum of
two periodic signals is periodic or not.
1.5 Determine which of the following signals are periodic and
determine the fundamental period.
(a) x(t) = 10 sin 25 nt (b) x(t) = 10 sin V5 xt
38 Digital Signal Processing

(c) x(t) = cos10nt (d) x(t) = x,(t) + x(t)

(e) x(t) = x(t) + x3 (t) P xat) = xQ(t) + x,(t)
What are even signals? Give examples.
What are odd signals? Give examples.
What is energy signal?
What is power signal?
What are singularity functions?
Define unit-impulse function?
What is unit-step function? How it can be obtained from an
unit-impulse function?
What is unit-ramp function? How it can be obtained from an
unit-impulse function?
What is pulse function?
Evaluate

(a) fe” §(t-10)dt ©) fe? 5(t45)dt

(c) | 40e? 5(t-10)dt and(d) fe} &(t-10)dt

Ans (a)e"* @)0 (40e? (d)e™™
1.16 Explain the terms single-sided spectrum and double-sided
spectrum with respect to a signal.
1.17 Sketch the single-sided and double-sided frequency spectra of
the signals

(a) x(t) = 10 sin (107t - 22), =o < £ < oo

(b) x(t) = 25 cos (sxe - z),=o < È < co

< t< o0
(c) x(t) = 100 sin (107: - zz) +50 cos{25 xt - z), —20
How are systems classified?
Distinguish static systems from dynamic systems.
What is linear system ?
Determine whether the following systems are linear
(a) we n oun
+ 5y(t) + 2 = x(t) b) 5 —— + y(t) = 5x(t)

lo) ao
o+ y(t) + 5 = 10x(t)
1,22 What isym system?
1.23 What is a causal system? Why are non-causal systems
unrealisable?
1.24 What is BIBO stability?
 Classification of Signals and Systems 39

1.25 What are the conditions for BIBO stability?

1.26 With illustrations, explain shifting, folding and time scaling
operations on discrete-time signals.
1.27 What are the different ways of representing a system?
1.28 Explain how difference! differential equations are used to model
a system.
1.29 Explain how impulse response can model a system.
1.30 Discuss the state-variable modelling of a system.
1.31 Explain the terms. (i) state variable (ii) state space (iii) state
vector (iv) trajectory (v) state equations and (vi) output
equations.
1.32 With a block diagram explain the process of analog-to-digital
conversion.
1.33 What is meant by sampling? State the sampling theorem.
1.34 Explain how sampling can be done with an impulse function.
1.35 Draw the spectrum of a sampled signal and explain aliasing.
1.36 Explain the process of reconstruction of the signal from its
samples. Obtain the impulse response of an ideal reconstruction
filter.
1.37 What is meant by quantization and encoding?
1.38 What is a quantized boxcar signal?
 Chapter 2

Fourier Analysis of Periodic and

Apertodic Continuous-Time
Signals and Systems

2.1 INTRODUCTION
A signal which is repetitive is a periodic function of time. Any periodic
function of time f(t) can be represented by an infinite series called the
Fourier Series. A function of time f(t) is said to be periodic of period T
if f(t) = f(t + T) for all ¢. For example, the periodic waveforms of
sinusoidal and exponential forms are shown in Fig. 2.1.
4 f(t)

0 Ti2 T t
{c)
Fig. 2.1 Waveforms Representing Periodic Functions
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 41

Examples of periodic processes are the vibration of a tuning fork,

oscillations of a pendulum, conduction of heat, alternating current
passing through a circuit, propagation of sound in a medium, etc.
Fourier series may be used to represent either functions of time or
functions of space co-ordinates. In a similar manner, functions of two
and three variables may be represented as double and triple Fourier
series respectively. Periodic waveforms may be expressed in the form of
Fourier series. Non-periodic waveforms may be expressed by Fourier
transforms.

2.2 TRIGONOMETRIC FOURIER SERIES

A periodic function f(t) can be expressed in the form of trigonometric
series as

£ + a3 cos 3t + ...
fH = Jao+ a, cos ot+ a, COS 2M
+ b, Sin Wp t+ by sin 20t + bz sin 30t + ... (2.1)

where @) = 2nf= z, f is the frequency and a’s and b’s are the

coefficients. The Fourier series exists only when the function ft)
satisfies the following three conditions called Dirichlets conditions.
(i) f(t) is well defined and single-valued, except possibly at a finite
number of points, i.e.
f (t) has a finite average value over the period T.
(ii) f(t) must posses only a finite number of discontinuities in the
period T.
(iii) f(t) must have a finite number of positive and negative maxima in
the period T.
Equation 2.1 may be expressed by the Fourier series

f(t) = Fay + Ya, COS NW, t + Xo, sin n Wọ t (2.2)

n=1 n=1

where a,, and b, are the coefficients to be evaluated.

Integrating Eq. 2.2 for a full period, we get
T12 1 72 Ti e
[fae = = ao fat + Í ¥ (a, cos n wot + bn sin n wet)dt
-T12 2 fn -Tzn
Integration of cosine or sine function for a complete period is zero.

Therefore, ff(t) dt = i aT
-T/2 2
T12
Hence, ap= > ffOdt (2.3)
FT -Ti2
42 Digital Signal Processing

T
or, equivalently ag = żj f(t)dt
0
Multiplying both sides of Eq. 2.2 by cos m Wot and integrating, we have
T/2 T/2
J Feos moot dt= > Jao cos moot dt+
-T12 -T12
TIZ 9 T a
J $a, cos nwo t cos moot dt + Yb, sin nwo t cos m wot dt
-Tign=1 -Tign=1

172
Here, — fÍdy cos M Wot dt
=0
-T/2
T/2 a, 72
Jaa, cos n wot cos mog dt= —* J[cos (m + nwo t + cos(m ~ n) wot] dt
-T/2 2 te
|0, form#n
—a,, form=n
2
T/2 T/2
fon sin N@ot cos mogtdi = m fisin (m + n) Wot - sin (m — n)@ot]dt
-T12 -T/2
=0
T12 Ta
Therefore, fro COS NWot dt = —*, form=n
-T/2
T12
Hence, an = 2 fro COS NWy t dt (2.4)
TT
T
or, equivalently a, = Zjf(t) cos n og t dt
0

Similarly, multiplying both sides of Eq. 2.2 by sin m œt and

integrating, we get
Ti2 T/2
fro sin m Wọ tdt = A Jao sin M@ot dt
-T/2 2 in

T w TI2 6
+ Í J a, cos n Wot sin m Wo t dt + Í F bn sin n wot sin m wp t dt
-Tign=1 -Ti2n=1
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 43

T/2
Here, > Í dy sin MWg
£dt = 0
-T12
T12
Í a, COS N Wọ É sin MWg tdt = 0
-T12
T12 0, form#n
¢ sin mota=lT,
fon sin nọ ieman
-T/2 gn
T12
Therefore, fro sin MWg tdt = Zon form=n
-T12
T/2
Hence, b, = 2 JF@)sin nootdt (2.5)
-T12
T
or, equivalently, b„ = Zj f(t)sin n Wotdt
0
The number n = 1, 2, 3, ... gives the values of the harmonic
frequencies.
Symmetry Conditions
(i) If the function f(t) is even, then f-t) = At). For example, cos t, t,
t sin t, are all even. The cosine is an even function, since it may be
expressed as the power series

cos t= 1 — —

The waveforms representing the even functions of ¢ are shown in

Fig. 2.2. Geometrically, the graph of an even function will be
symmetrical with respect to the y-axis and only cosine terms are present
(d.c. term optional). When flt) is even,

frod = affit)at
A 4
f(t) f(t)

(a) (b)

Fig. 2.2 (Contd.)

44 Digital Signal Processing

f(t) f(t)

(c) (d)
Fig. 2.2 Waveforms Representing Even Functions

The sum or product of two or more even functions is an even function.

(ii) If the function f(t) is odd, then f(-t) = —f(¢) and only sine terms
are present (d.c. term optional). For example, sin ż, t*, t cost are all odd.
The waveforms shown in Fig. 2.3 represent odd functions of t. The graph

of an odd function is symmetrical about the origin. Iff (t) is odd, ff(t) dt

= 0. The sum of two or more odd functions is an odd function and the
product of two odd functions is an even function.

f(t) AMD f(t)

|
|

(a) (b)

Fig. 2.3 Waveforms Representing Odd Functions

(iii) If f(t + T/2) = f (t), only even harmonics are present.

(iv) Iff (t + T/2) = -f (t), only odd harmonics are present and hence the
waveform has half-wave symmetry.

Obtain the Fourier components of the periodic square

wave signal which is symmetrical with respect to the vertical axis at
time ż¢ = 0, as shown in Fig. E2.1.
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 45

fit)
A

Fig. E2.1
Solution Since the given waveform is symmetrical about the
horizontal axis, the average area is zero and hence the d.c. term
a, = 0. In addition, f(t) = f(-t) and so only cosine terms are present,
i.e., b, = 0.
T12
Now, a= 2 fro cos n Wot dt
-T/2
~A, from -T/2<t<-T/4
where f(t)=4+A, from -T/4<t<+T/4
-A, from +T7/4<t<+T/2
Therefore ,
2a T T/4 T/2
a, = Al fc COS NWot)dt + Joos NWotdt + J(-cos NW, t) a
~T/2 -T74 T/4
i -T14 i T14 y T/2
_2A [2] (Saree e [=ne] |
T noo Jar noo Jra no Jra
-24 -sin (= $27) + sin (2287), sin (2907)
NO oT 4 2 4
` jeer : (7) : (**)]
-sin |——— |-sin + sin |——
4 2 4
8A (=27) 4A (752)
= sin |—— |- sin |——
nT 4 NWT 2
When wọ T = 2n, the second term is zero for all integer values of n.
Hence,
8A . (=) 4A. (=)
a, = —— sin |— |= — sin |—
2nn 2 nt 2
ay = 0 (d.c. term)
46 Digital Signal Processing

ay = tA sin(n)=0

4A in(32) 4A

Substituting the values of the coefficients in Eq. 2.2, we get

f= 44 cos(Wot) - icos (3Wot) + =cos(50t) - “|

Obtain the Fourier Components of the periodic

rectangular waveform shown in Fig. E2.2.
f(t)
A

|
A

~ 72 - 7/4 0 7/4 T2

Fig. E2.2
Solution The given waveform for one period can be written as
0, for -T/2<t<-T/4
f(t)= 4A, for -T/4<t<T/4
0, for T/4<t<T/2
For the given waveform, f (—t) = f (t) and hence it is an even function
and has b, = 0.
The value of the d.c.term is

2 T/2
a, = — J £© cos n wot dt
-T/2
 Fourier Analvsis of Periodic and Aperiodic Continuous-Time Signals and Systems 47

7 z nof
T -T74

4A
= noT sin (n0 T/4)

When œT = 2r, we have

2
=0, forn=2, 4,6,

=^, forn= 1,5,9, 13,

=-24 forn =3,7, 11, 16, ...

nn

Substituting the values of the coefficients in Eq. 2.2, we obtain

A 2A 1 1
f(t)= 3¢ == (cosWot = 7 008Soot + = cosSwot --)

Obtain the trigonometric Fourier series for the half-

wave rectified sine wave shown in Fig. E2.3.
|Kt)

F 372
Fig. E2.3
Solution As the waveform shows no symmetry, the series may
contain both sine and cosine terms. Here, f(t) = A sin Wot
To evaluate ap:
T
=afA sin Wot dt
0
T/2
a [A sin Wot dt
0
48 Digital Signal Processing

art 2A 2A
cos wt]? =ort cos (9T'/2) + 1]

Substituting © T = 2n, we have ay = 24.

To evaluate a,:

2
a= žfro COS Notdt

at
p 4i Wo
£ cos n Wotdt

2A [=sin Wot sin Be soos

ne! cos
oe)"
-n° +1 o
Substituting @) T = 2n, we kave

a, = g leona N
n(l-n

Hence, an= oe for n even

m1-n*)
= 0, for n odd
For n = 1, this expression is infinite and hence we have to integrate
separately to evaluate a4.
T/2
Therefore, a,= T j sin Wot COS Wp t dt

Age
==r en 2@ot dt

A T72
= T 7l cos 209 t],

When œT = 21, we have a, = 0.

To find 8,,:

2
6, = 2 sinNO, tdt

T/2
=7 JAsin Wy tsin NW,
t dt
0
A à T12
2A | nsin Wgt cos NW, t — sin n Og £ COS Wp t í
T -n? +1 b
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 49

When @, T = 2r, we have b, = 0.

For n = 1, the expression is infinite and hence b, has to be calculated
separately.
T/2
2A f.
bı = T ps Oot dt

_ 2A [st- sinzeit)"
2 a4
When 0) T = 21, we have b, = A,
Substituting the values of the coefficients in Eq. 2.2, we get

fit)=gAEfit Rsinoat _230520 _2

Tp 08 Mot 254 }

Obtain the trigonometric Fourier series of the

triangular waveform shown in Fig. E2.4.

Fig. E2.4

Solution
(i) As the waveform has equal positive and negative area in one
cycle, the average value of aç = 0.
(ii) As f(t) = -f (t), it is an odd function and hence a, = 0
472
and b, = T [iain NWotdt

(iii) Here, f(t + 7/2) = -f (t). Hence it has half-wave odd symmetry
and a, = b, = 0 for n even.
(iv) To find f(t) for the given waveform

The equation of a straight line is Y-Mi_N-I2

x=% x%
50 Digital Signal Processing

For the region 0 <i < z

ft)-0 _ 0-A
t-0 0-T/4
4A
Therefore,
erefore, f(t) =—T

Pare T
7 <t< >
theregion —<t<—,
Foror th

fit)-A_ A-0
joe FT
4 4 2
4A T 4A
fit)-Az= aC *)- atta

Therefore, f(t) = - an +2A.

472
Now,6, = T Jr sin N Wotdt

T/4 T/2
_4 4A). 4 4A :
=T J(“A)esinnwg tdt + FJ (Fit 24) sinn oy a

T/4 T/2 T/2

= 8A Jtsinn wgetat-164 ftsinnagt +6 fsinnowot dt
0 TIA T14
_ 16A Pe t = _ "cos nat y
rT -n@ Jo 3 Tno
16 A peno,” f meno g
ria -n@o Jra ry -P2

+
8A [= naar)”
—— |M

T =n Oo Tl4

_ 16A|T cosnm@oT/4 | |sinn@ot ia

T? |4 (no) nos Jo
sa l z co s np T/ 2 7 20 s NM T/ 4 { = a
~ T° pe |2 (-n@) 4 nO 22
nO Jra
{ 8A| cos NW. T/2 ; COS NW) T'/4
T (nog) NWo
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 51

Substituting wp = =. we have

ba 16A 2sin nx/2 "i 16A T cosnn 8A cosnn

| T? n?4n°/T? T? 2n22wW/T T n2niT
Simplifying, we get

b, = SA. sin (nn/2)

nn

b, = 84 sin (x2) =84

T n?

b3 =
8A .
sin
3n =
8A

Substituting the values of the coefficients in Eq. 2.2, we get

f= $2 sinWot - a sin 3@ot + a sin 5@ot + |

Deduce the Fourier series for the waveform of a

positive going rectangular pulse train shown in Fig. E2.5.
f(t)
i

T- T T+d/2
Fig. E2.5

Solution The periodic function of the Fourier series for the given
pulse train is expressed by

f(t)= Fy + Sa cos norgt+ Sb, sin

n Wot
n=1 n=1
52 Digital Signal Processing

d/2
2Ad
=2 [ae 24
22 giz, = T
T an

Here, since the choice of ¢ = 0 is at the centre of a pulse, the b,

coefficients are zero.
2 T2 9 42
Therefore, a, = = fro COS
N Wy tdt = = Joos n Wot dt
Tn T än

2A/snneat)
= 22 2A [si(2904) : (=24)]
= sin -sin |———
T NW Jan NOT 2 2
4A . n@d
—————
81 Dr
NW 2
Ad , 2Ad © sin (nw, d/2)
Hence, f(t)= > +5 »y“aa oe
n=l

2.3 COMPLEX OR EXPONENTIAL FORM OF

FOURIER SERIES
From Eq. 2.2, the trigonometric form of the Fourier series is

f= $40 + F(a, cos nwt +b, sin nwo t)

n=1

An alternative but convenient way of writing the periodic function

fit) is in exponential form with complex quantities. Since
eire y eirag
cos n yt = 2

ej”
eot _ g7inoot
sin
n Mt =
2j
Substituting these quantities in the expression for the Fourier series
gives
jnwot — JN wot oo Jn@ot _ pinot
po= 4a, + È o,(oe $ [eee]
n=l n=1

1 a, — jb, ei" (a, + jbp)e 172o

=2%t -a 2 * -jb
Here, taking c, = z(a„-jb n)
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 53

cn = L(a, +jb,) (2.6)

Co = 4
Where c_, is the complex conjugate of c,. Substituting expressions for
the coefficients a, and 6, from Eqs 2.4 and 2.5 gives
172
°.= = J f@[cos noot- j sin noo t]dt
T -T/2

172
= [roete at (2.7)
T tye
172
and Cn? 5 Jro [cos n Wp t +j Sin n Wy t] dt
T -T/2

T/2
=2 frien! at (2.8)
T -T/2

- r =| i
with f= cot $ cepet Y cp ef"! (2.9)
nal n=-%

where the values of n are negative in the last term and are included
under the E sign. Also, co may be included under the = sign by using the
value of n = 0. Therefore,

f@= Fc, ea (2.10)

EER
It is clear from the result given in Eq. 2.10 that the periodic function
f(t) may be expressed mathematically by an infinite set of positive and
negative frequency components. The negative frequencies have not only
mathematical significance, but also physical significance, since a
positive frequency may be associated with an anti-clockwise rotation
and a negative frequency with a clockwise rotation.
The complex Fourier series furnishes a method of decomposing a
signal in terms of a sum of elementary signals of the form {e/atit |.This
representation may be used for signals f(t) that are
(i) Periodic, f(t) = f(t + T), in which case the representation is valid
on (— ©, ce)
(ii) Aperiodic, in which case the representation is valid on a finite
interval (¢,, t2). The periodic extension of f(t) is obtained outside of
(ty, ty).
Note that similar to the evaluation of integrals a„ and b,,, the limits of
integration in Eq. 2.7 may be the end points of any convenient full period
and not essentially 0 to T or 0 to 2n. For f(t) to be real, C_, = C,,, so that
only positive value of are considered in Eq. 2.7. Also, we have
54 Digital Signal Processing

a, = 2Re[c,] and b,„=- 2 Im [c,] (2.11)

For an even waveform, the trigonometric Fourier series has only cosine
terms and hence, by Eq. 2.6, the exponential Fourier series coefficients
will be pure real numbers. Similarly, for an odd waveform, the
trigonometric Fourier series contains only sine terms and hence the
exponential Fourier series coefficients will be pure imaginary.

Bg
(a) Find the trigonometric Fourier series of the waveform shown in
Fig. E2.6 and
(b) Determine the exponential Fourier series and hence find a, and b,,
of the trigonometric series and compare the results.

Fig. E2.6
Solution The function of the given waveform for one period can be
written as
fit)= a for -T/2 <t<0O
+A, for 0 < t < T/2
As the waveform is symmetrical about the origin, the function of the
waveform is odd and hence ay = a,, = 0, and
T/2
bes f f (t) sin nœtdt
-772
2 0 T12

-2l f (-A sin not) dt + f'Asinnostat

T lta 0
- 24[fosna] [ee]
T noo J-r72 nao 0
2A
= nT {[1 — cos (n wg T/2)] + [1— cos (n Wy T/2)}}
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 55

= 7- cos
(2 Wy T/2))
noT

When œ = ="
3 | we have

bn = “2-1 -cosn (22-772)

2A
=f
re g cos nt]

0 , ifniseven
4A
bn =) 44 |itn isodd
nn

Substituting the values of the coefficients in Eq. 2.2, we obtain

f= 44 [sinWy t+ $sin 3@ t+ $sin5wot+ ha where o = 2r

n T
(b) To determine exponential Fourier series

Here cp = |
$a =0
To evaluate c,,
Since the wave is odd, c, consists of pure imaginary coefficients. From
Eq. 2.7, we have

c, = 17
al f(t)e j2"! dt
0
1 0 ; T/2 f
-+ Í (- A) e270! dt + Í Ae Jnot J
T -T/2 0
o T/2
i A |-1) 1 ei af 1 e inet
E (- jno) -rz L(- Jn ) 0
„A,
s2. Lh +e Jno (712) en ina (T/2)
-
0
T (- cat i }

When œ = ZE we get

eA. {re +e AM@n/TITID 4 g-jn(2n/2(T/2) _ 20)

T -jn2n
= A {-e° +e/"™* +eI"* — ef = ja
A irn
(-j2rn) nt
Here, e?"* = + 1 for even n and e/"* = —1 for odd n

Therefore,c, = -j (24) te
for odd n only.
56 Digital Signal Processing

Hence, the exponential Fourier series is

— e Jot -j 2A „jot mid ,2A pj3oot

f=.. „+ JŽA erite +j2A
Tt 31
By using Eq. 2.11, the EIE ER Fourier series coefficients a„ and
b„ can be evaluated as

= 2Relc,] = 2|c,| =0 and b, =- 2 Im [e,] = 24 for odd n only.

n

These coefficients are the same as the coefficients obtained in the

trigonometric Fourier series.

a
Example 27
(a) Find the trigonometric Fourier series of the waveform shown in
Fig. E2.7 and
(b) Determine the exponential Fourier series and hence finda, and b,
of the trigonometric series and compare the results.
A A(t)

54

@gt
0 2n án °
Fig. E2.7
Solution (a) As the waveform is periodic with period 27 in œ t and
continuous for 0 < @t < 2n, with discontinuities at Mot = n (27),
where n = 0, 1, 2, ..., the Dirichlet conditions are satisfied.
To find f(t) for the given waveform of region 0 < Opt < 2n:
pee;aes es S
The equation of the straight line is———
~% %,—%
= (2n, 5), we get
Substituting (x, yı) =(0, 0) and (x, girs
ft)-0 _ 0-5
@)t-0 0-27

Therefore, f(t)= (%}e. t

erefore, f(t)

To find Fourier coefficients

Using Eq. 2.3, we obtain the average term,
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 57

T/2

sales
f fe ae
-T12
2r
2 5
On JOn Oo t d (wg t)

a t27?
“@ =
__10 (21)? _ 5
(Qn)? 2
T/2
Using Eq. 2.4, we obtain a, = Z Í f(t) cos nOg t dt
-T12
2z
2 5
=— — |@ t cos noy t d (0y £)
Qn J(4) i g $

= [2t sin NW t + | cos no J

Qn7L n ` n? 9 0

= 5 z (cos n2n - cos 0) = 0

2n°n
Hence, the series contains no cosine terms.
T/2
Using Eq. 2.5, we obtain b, = 2 J f@)sin nogt dt
-T/2

1 —
(5
==z tac t sin
si n W td
td (Wo t t)

2r
W o t y t ; non]
= z| - cos n a + -Fsin
2 n n 0

ss%
nt
Combining the average term and the sine-term coefficients, the series
becomes

+b, sin Wọ t + by sin 2Wyt +...

-Ž
= sin oot-
lo > sin 2a ¢ - > sin 3a ¢ -..
a
58 Digital Signal Processing

_5 < arot
rola r
(b) To determine exponential Fourier series

Here, co = la
3 2
To evaluate c,,:
From Eq. 2.7, we have
i?
7l fde d
= -jnwot t

2n
Cn l | (zante ac t)
“On
5 fein ir 5
= (-jnea,t-v| | =j
ean
(Qn)? Fax ee
Substituting the coefficients c, in Eq. 2.9, the exponential Fourier
series is
, 6 ~j2ogt . 6 ~j@ot 5 . 5 jOgt
(H) =... j —— eitt j eltt g Hy jel
f 4x Jon 2 7 on
. 5 -j2opt
+J — P ias
TIn
By using Eq. 2.11, the trigonometric Fourier series coefficients a, and
6, can be evaluated as

a, = 2Rele,] =2}c,|=0 and 6, =—2Im [c,} = -Š

nn

Hence, f(t) = 3-3 sinwt -Z sin2oyt—>nsin 38@_t-...

This result is the same as that of the trigonometric Fourier series
method.

2.4 PARSEVAL’S IDENTITY FOR FOURIER SERIES

A periodic function f(t) with a period T is expressed by the Fourier series
as

f(t)= EPA + ¥ (a, cos nw t +b, sin noot)

to n=l

Now, FOP = T f+ ¥ la, F(t) cosnot +b, f(t) sin not)

n=1

2 (ag/2) ey
Therefore, ae | [f(y dt= T f (f@jae
-T12 -T12
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 59

1< T12 T12

+T ¥ Ja, Í f(t) cos nog
t dt + b, Í f(t) sin nog
t dt
n=1 -T/2 -T/2
From Eqns 2.2, 2.3 and 2.4, we have
T/2

aE
te
= | f(t)dt

2 T/2
a,= J f(t) cos ng t dt
-T/2
T/2
bn 2 f f(t) sin ngtdt
T -T12
Therefore, substituting all these values, we get
T/2 à
1a I (peor? 2 aedt= =(S2.)
2) + 1+2 È (ito)2 (249
This is the Parseval’s identity.

2.5 POWER SPECTRUM OF A PERIODIC FUNCTION

The power of a periodic signal spectrum f(t) in the time domain is
defined as
1 T/2

P== pf iol 2 dt
The Fourier series for the signal f(t) is

f(t) = 5 Cp einot
n= -w

According to Parseval’s relation, we have

T12 7
Pw == | [0 dt
i T å
Ti2
f fO D c, e779 dt
T Tn neo
č T12
= te =f f(t) ef "0!dt
T iyo
60 Digital Signal Processing

= È lcn |?, watts

n=-=-%

From Eq. 2.12, the above equation becomes

Here, co = “| and c, = a2+b?, (n21 (2.13)

Thus the power in f(t) is

Ps=...+|e_,[? +...+[e yl? leol? +[e,[? +...4+]e,[? +... (2.14)

P = jel? + 21e? + 21c? +... +]e,|?+...
Hence, for a periodic function, the power in a time waveform f(t) can
be evaluated by adding together the powers contained in each harmonic,
i.e. frequency component of the signal f(t).
The power for the n“ harmonic component at n œ radians per sec is
|c, |? and that of -n wp is |c_, |”. For the single real harmonic, we have
to consider both the frequency components + n Wp.
Here, c, = c_, and hence |c,|? = |c_,|?. The power for the n” real
harmonic f(t) is
Pa - le, [? + je_.,]*= 2\c,/?

The effective or RMS value of f (t)

Using Eqns 2.12, 2.13 and 2.14, the RMS value of the function f(t)
expressed by Eq. 2.1 is
2 1 1 1 1):
Fms = ($) + yar + yan te +5 bt +5 bet...

= ob +S cf +3c5
+... (2.15)
| Example 2.8] The complex exponential Fourier representation of a
signal f(t) over the interval (0, T ) is

= — 3 jnnt
fo p 4+ (nn)? :
(a) What is the numerical value of T ?
(b) One of the components f(t) is A cos3nt. Determine the value of A.
(c) Determine the minimum number of terms which must be retained
in the representation off (t) in order to include 99.9% of the energy
in the interval.
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 61

Note: X AER = 0.669

neve |4+(nn)
Solution The complex exponential Fourier transform representa-
tion of a signal f(t) is

ft)= J, cpe" where wo -7

The given signal f(t) over the interval (0, T) is

f= 2 Tran
= < 3 jnrt

(a) Comparing the above two equations, we get

gece, ee
"" 4+(nn)?
e jn?žt
Fr ;
= ent

2n
Hence, T
— m =m ie
ie. T=2

(b) When n = 3, the component of f(t) will be

cg = — eism = —3__ [cos 3nt+jsin 3n¢]

4+(3n) 4+(3n)
Similarly, when n = — 3, the component will be
3 e f3nt = 3
c_3 = ———_— ——, [cos 37t - jsin 3nt}
43n? Tas A
Therefore, cz + c_3 = _ 6 O cos 3nt
44+(3n)?
Hence, when one of the components of f(t) is A cos 3 nt , the value of
Ais

ei Oe
4+(3n)?
a < 3
(c) Total (maximum) erP, = ——, | = 0.669
å ponen E 4+ (nn)?
The power in f(t) is

P= |c? +2[le)? +|co[? + les)? +|c4l?]

_=|=]13? +2 3 m
F Md 3
PEET
F A
a.
PET)
f 3a
alPa)
F
4 4+(n) 4+(2n) 4+ (37) 4+(4n)
62 Digital Signal Processing

= 0.5625 + 0.0935 + 9.52 x 10`? + 2.088 x 10-3 + 6.866 x 1074

= 0.66836
Therefore, energy contained in the four terms is

PP, 100 = 9:66836

0.669
| 100 = 99.9%
Hence, the first four terms include 99.9% of the total energy.

2.6 FOURIER TRANSFORM

The plot of amplitudes at different frequency components for a periodic
wave is known as discrete (line) frequency spectrum because amplitude
values have significance only at discrete values of n œ where Wy = 2n/T
is the separation between two adjacent (consecutive) harmonic
components. If the repetition period T increases, @ decreases .
Hence, when the repetition period T becomes infinity, i.e. T — œ, the
wave f(t) will become non-periodic, the separation between two adjacent
harmonic components will be zero, i.e. œ = 0. Therefore, the discrete
spectrum will become a continuous spectrum. When T —> æ, the adjacent
pulses virtually never occur and the pulse train reduces to a single
isolated pulse. The exponential form of the Fourier series given in
Eq. 2.10 can be extended to aperiodic waveforms such as single pulses or
single transients by making a few changes.
Assuming f(t) is initially periodic, from Eq. 2.10, we have,

f= Fc, e/ 7%!
T/2 l
where c, = VT f f(t) ereot at
-T/2
In the limit, for a single pulse, we have
T > =~, @ = 2n/T > dw (a small quantity)
or VT = W/2n > dw/2n
Furthermore, the n” harmonic in the Fourier series is n @ > nda.
Here n must tend to infinity as @) approaches zero, so that the product
is finite, i.e. n Wy > @.
In the limit, the £ sign leads to an integral and we have

_ do J
ae 7 fie -jot dt

and, fit)= jeli feio a

When evaluated, the quantity in bracket is a function of frequency
only and is denoted as F(j œ) where
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 63

F(jo)= | fei dt (2.16)

It is called the Fourier transform of f(t).

Substituting for f(¢) above, We obtain

a er do
f(t) = mad F(ja) e/
or, equivalently,

f= | Fjo) eih af (2.17)

which is called the inverse Fourier transform. Now the time function
f(t) represents the expression for a single pulse or transient only.
Equations 2.16 and 2.17 constitute a Fourier transform pair.
From Eqns 2.16 and 2.17, it is apparent that the Fourier transform
and inverse Fourier transform are similar, except for a sign change on
the exponential component.
2.6.1 Energy Spectrum for a Non-Periodic Function
For a non-periodic energy signal, such as a single pulse, the total energy
in (— , œ) is finite, whereas the average power, i.e. energy per unit

time, is zero because $ tends to zero as T tends to infinity. Hence, the

total energy associated with f(t) is given by

E= | fede

Since, f(t)== Í Fo) e/* do, we obtain

E==fJ ros,
L ftJ Fue)
riie)elmt
e/% dw dt

s>
= 1 n ruw]
; T
fOe jot
afao

= — | F(jo) F(-jo)do

is
“35 |FUer® (jo) do
64 Digital Signal Processing

ar e
oma! F(jo)|° do

J IE (PP af, joules

=%

E fir@Pat= f IFP af (2.18)

This result is called Rayleigh’s energy theorem or Parseval’s
theorem for Fourier transform. The quantity |F (f)|? is referred to as
the energy spectral density, S(f), which is equal to the energy per unit
frequency.
The integration in Eq. 2.18 is carried out over positive and negative
frequencies. If f(t) is real, then |FO w)| = |F(-j @)|, then the Eq. 2.18
becomes,
17 esd =-117
fjFf ii 2 r
E=} On Jirga| do | IF (jo) |?do JS(w) do
Here the integration is carried out over only positive frequencies. The
quantity S(w) = |F(j«)|?/n is called the energy spectral density.

2.7 PROPERTIES OF FOURIER TRANSFORM

Table 2.1 presents important properties of the Fourier transform.

Table 2.1 Important properties of the Fourier transform

Operation fW

Transform fit) Í fit) ei dt

Inverse transform x Í F(jo) ei% do F( jo)

Linearity af,(t) + bf,{t) aF ( jo) + bF jo)

Time-reversal f(t) F- jw) = F (jo), f(t) real
Time-shifting (Delay) fít-to) ji

Time-Scaling f (at)

Time-differentiation 2 fit) ( jo)" F(jo)

dF(j@)
Frequency-differentiation (— jt) f (t) da

(Contd.)
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 65

Operation
t

Time-integration Íf(t) dt an jo) + z F(0) ò (w)

Frequency-integration = f(t) Í f ( jo’) doy’

(- jt)
Time convolution fit) * folt) = F (jo) Fajo)

J A@he-vde
Frequency convolution

(Multiplication) Fat) - fle) EF

n
jo) * FA jo)
Frequency shifting
(Modulation) f (t) e12% F( jo- jm)
Symmetry FV jt) 2nf (w)
Real-time function f(t) F( jo) = F`- jo)
RelF( jw)] = RelF(-jo)]
Im[F( jo)] = -Iml Fjo)
|Ftjo)| = |Fjo)|
Dfi jo) = -Of jo)

Parseval’s theorem E= Í IFO? dt E= = Í IFG)? do

Duality fft) = g( jo),

then g(t) <> 2nf (- jw)

2.7.1 Linearity
The Fourier transform is a linear operation. Therefore, if
f(t) < Fy (j o)
h(t) e Fo (j o)
then, af; (t) + bf, (t) = aF, (j œ) + bF; (j w)
where a and b are arbitrary constants.
2.7.2 Symmetry
If f(t) e F (jo)
then, F( jt) <= 2nf (—)
Proof

Since ft)= + | F(ja)e!* do

2m *

anf(-t)= | FU) eie da’

where the dummy variable is replaced by w’.

66 Digital Signal Processing

Now if t is replaced by w, we have

2nf(-a)= | Fohe do’

Finally, w is replaced by¢to obtain a more recognisable form and we

have

2nf(-o)= | Flite i" dt =F (FO)

Therefore, F( jt) <= 2nf(- œ)
If f(t) is an even function, f (t) = f (- t).
Hence, FIF jt)] = 2nf(@)
2.7.3 Scaling
If fi) & F(jo),
then, flat)» Ł r(Z2)
laj a
Proof
If a > 0, then the transform of f (at) is

F [f (at)) = f flat) es dt

Putting x = at, we have dx = adt. Substituting in the above equation,

we get

FIFA FIANE | faye* 21 (le) a a a

Ifa < 0, thenF [f(at)] = sl r(Z2)

a a
Combining these two results, we get

fat) ka F(Z2)
ja} Xa
We conclude that larger the duration of the time function, smaller is
the bandwidth of its spectrum by the same scaling factor. Conversely,
smaller the duration of the time function, larger is the bandwidth of its
spectrum. This scaling property provides an inverse relationship
between time-duration and bandwidth of a signal i.e. the time-
bandwidth product of an energy signal is a constant.
2.7.4 Convolution
Convolution is a powerful way of characterising the input-output
relationship of time-invariant linear systems. There are two convolution
 Fourier Analysis of Periodic and Aperiodic Continuous-Time Signals and Systems 67

theorems, one for the time domain and another for the frequency
domain.
Time Convolution

If x(t) X(j@) and A(t) Hjo),

then y(t) = x(t)» A(t)

= Í x(t) h(t-1) dt ¥(jo) =X(jo) H(jo)

Proof

Fly = Yj o) = Í viel] x(t) A(t - vaydt

= jHof]h(t
-t)e1” alas

Putting a = ¢ — 1, then ¢ =a + t and da = dt

Therefore, Y( jw) = jxo]f haeie cola

= f x(t) eI dt Í h(a) e12? da

=X jo) H( jo)
Hence, the convolution of the signals in the time domain is equal to
the multiplication of their individual Fourier transforms in the
frequency domain.

| Example 2.9| In the system shown in Fig. E2.9 determine the

output response of the low-pass RC network for an input signal x(t)

x(t) = etPC

Fig. E2.9
68 Digital Signal Processing

Solution The input signal x(t) = e Re

Using the convolution theorem, we can find y(t)
y(t) = x(t) * A(t) = FX (jo) H (jo)

X (jo) =F [x(t)] = J e FC git dt

0
a fios
= j et jo} at
0

_aa1
-+—
10 RC
Similarly, the transfer function of the network is

H(jo)= _VjoC _ a
(z + een
1 ) (j@RC +1)
JoC

BSE N 1
~ RC (Jo+ 5)
Re

3 ; ‘ 1 1
Hence, Y¥(j@) = X(j@) H( jo) = RGT 1

(e+e)
ther = 1 [Y n = —1 te _RC
yt)=7 7 [YCj o) RG u(t)

example 210po the output response of the low-pass RC

network due to an input x(t; = te*"° by convolution.
Solution The transfer function of the network is
1
; joc 1
id Rl @+joRO)
H = SOF OC

joc
The given input time function is x(t) = te~”®
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
70 Digital Signal Processing

1
Asja +j) ljo- -a= Ga

B = Y(joXb + jo)| jo-

-b= (a 5

H . | Peat: a
(b-a) lz+jo3 )e i (b+ 1 ;
3 Yi
ence, Y(jo) = ———
jal
Taking inverse Fourier transform, we get

y(t)=gigle
phen | - D- -bt uw]
When b = a, the partial fraction expansion is invalid. Hence,
z 1
Y( jo) = ——
j (a+ jo)?
ig A 1
2/2]
Using dual of the differentiation property,

etulthe 1
+jo
- .d 1 1
te“ ult) j— ear
: iil (a+ jo)?
Therefore, y(t) = te~™ u(t)
2.7.5 Frequency Convolution
If f(t) <= F( j@) and g(t) @ Gj),
then f(t) g(t) = Z Foe G(jo)
Proof
The inverse transform of [F( jœ) * G( j @)]/2n is
2 = -
p-[Fue) on _(1
*«G(j@) |- t : ere
(4) le J FGwade ju) du dw

1
J FOW |GGo- ju) ei” dodu
(27)? a 00

Putting x = o — u, then
o = x + u and dx = do
Therefore
. . 2 œ Gd

2n
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
74 Digital Signal Processing

2.7.12 Time Reversal

If f(t) = F( jo), then f (-t) = F (jo)
It is clear that the time-reversal theorem is similar to the scaling
theorem with a = ~1. When the signal is real, time-reversal affects only
the phase spectrum because the amplitude spectrum is an even function
of frequency.
2.7.13 Complex Conjugation
If f(t) <> F( jo), then f(t) e Fjo)
2.7.14 Duality
If f(t) = g( jo), then g(t) = 2nf (jo)
2.7.15 Area Under f(t)

If f(t) > Ff), then f f(t) dt = F(0)

Thus, the area under a function f(t) is equal to the value of its Fourier
transform F (f) at f = 0.
The result can be obtained by substituting f = 0 in the formula
defining the Fourier transform of the function f(t).
2.7.16 Area Under F(f)

If f(t) > Fif), then | Fif) df= f0)

Thus, the value of a function f (t) at t= 0 is equal to the area under its
Fourier transform F(f ).The result can be obtained by substituting t = 0
in the formula defining the inverse Fourier transform of F(f).

| Example 2.14] A certain function of time f(t) has the following

Fourier transform

F(jo) =2
1 e-20 w+ D

+1
Using the properties of the Fourier transform, write the Fourier
transforms of
t

(a) f(2t), (b) flee, O4 fad d) ffar

In each case state clearly the properties you will use.
Solution
‘ 1 -20/4 D
F(jo)= BA nes:
g o? +1
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 awig, uipwop
J (1) Asuanbasgumwop
q (Ol)
78
Digital Signal Processing

eft-

l
+D

CPIu09)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
82 Digital Signal Processing

2.8.1 Gate Function

Let us consider the single gate function (rectangular pulse) shown in
Fig. 2.4. It has the analytic expression given by
1, for -T/2<t<T/2
t)=
f I otherwise

(t) A

Fig. 2.4 Single Gate Function

The Fourier transform of f(t) is

F(jo)=F(f@1 = | fe? at
T/2 ,
Í 1-e°/°! dt
-T/2
1 2, T/2
em dad
1 [e7402 - ej9712])
-jo

:=T sin (25)

2) «7 sine
(a7) =T sine ( z (22)
2
Hence, the amplitude spectrum is

IFG o)| = T |sine (F)

and the phase spectrum is LF (œ) =

The amplitude and phase spectra are shown in Fig. 2.5.

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
86 Digital Signal Processing

Comments
(1) The phase of the amplitude spectrum is exactly the same as that
in the previous case given in Fig. 2.7. :
(2) There is an additional uniform phase shift factor e/°”? which
changes the phase spectrum of the previous case.

(3) By using the time-shift theorem, [re - t\I- F(jo) e Jere

where F( jm) = T sinc (2), the above result can readily be

obtained.

|Example 2.16| Find the Fourier transform of a rectangular pulse 2

seconds long with a magnitude of 10 volts as shown in Fig. E2.16.
f(t) in (volts)

10

= > tin seconds

tt) 2
Fig. E2.16
Solution Fourier transform F( jœ) of the given pulse is given by

F(jo)= | fOe at
2 f -jot 2
=f 10677” dt = 10)£ |
0 “jo n

A 10(/ -jo

=10 E jeh -e*]

=20 eio fe -eit
(0) 2j
= 20 e1” sin@
(03

= 20e /° sine ©
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
90 Digital Signal Processing

2.8.4 Triangular Pulse

Consider the triangular pulse shown in Fig. 2.10.
f(t)

N
f
EIS
/
/
P,/ \ Ps ase
- Ti2 o 1/2
Fig. 2.10 Triangular Pulse

Equation of line P. A is
2
f(t)p,p, = a
——t+A= A(1+ T t)

Equation of line P,P, is

f tt PaPa =-4t+A=A(1-
T/2
A
21)
T
2

2 T
Therefore f(t)=A l+at for ~5<tso

= a(i- že) for osts?

Now,
T/2
F (jo)= f reese" dt = | f(t)e4 de
EA -T12
T/2
= ffiee dt + [|oase dt
-T/2
0 T72

f A(1+Ztlevmars f A(1-2t)e
-T/2 T 0 T
T42
=A fel dt+A Jeiet dt
-T?2

0 Ti2
2a jista a- 24 J tet” de
-T72 0
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
94 Digital Signal Processing

Therefore,
xf 2
F [e] = vx ete)
a

2.8.6 Impulse Function (Unit Impulse)

The impulse function is called the Dirac delta function ô(t) which has
an infinite amplitude and is infinitely narrow. This is defined as
(i) 5(t) = 0 for all values except at t = 0.

(ii) J 8(t) dt = 1, i.e. the area within the pulse is unity.

The Fourier transform of the impulse function (t) is obtained as

F( jo) = FÐ = f Se de=1
Hence, we have the pair 5(¢) = 1.
The frequency spectrum of the impulse function 6(t) shown in
Fig. 2.13 (a) has a constant amplitude and extends over positive and
negative frequencies.
4 A(t) F( ja)

8(t)

— f en ne
te) 0
Fig. 2.13 (a) Impulse Function and its Spectrum

Using the time-shift theorem, we get

5 (t — to) e 1%
The shifted impulse and its amplitude and phase spectra are shown
in Fig. 2.13 (b)
!f(t) |Fijo)| A 9(io)
ölt-t)
<

Slope =-b

> t — — > ©
0 b 0 0 g

Fig. 2.13 (b) Shifted Impulse and its Spectrum

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
98 Digital Signal Processing

F( jw) = F {sgn (8) f sgn (t) e/ dt

0 oo

Jc Deo dt + f (1) e747 dt

~-o% 0

- [eze] [=e]
© L-jo J. L -je Jo
1,1 _2

Therefore, F [sgn (t)} = 2

jo

Hence, we have the transform pair sgn (t) = 2.

jo
The amplitude and phase spectra of the signum function are shown
in Figs 2.18 (a) and (b) respectively.
A |F(jo)| (jo)

>
o

= = L Sa E
0 o
(a) (b)

Fig. 2.18 (a) Amplitude and (b) Phase Spectra of the Signum Function

Determine the Fourier transform of f(t) = e! I

sgn(t).

Solution f(t) = e~*'!! sgn (t)

ee for t<0
e“, for t>0

Therefore,F[f(¢)] = Í foei at

0 ‘a
=- Í e@ e J! at +f erg Ie" dt
- 0
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
102 Digital Signal Processing

Comparing the powers P and P,,,, we see that (0.181/0.2) x 100 =

90.5% of the total power in f(t) is contained within the first zero
crossing of the spectrum for f(t).

—32n-24n-16n-8n 0 lón
Ea |
24x 327
40n
eY

Fig. E2.20(b) Spectrum of the Function

2.8.11 Unit Impulse Train

Since f (t) = Föt — kT) is a periodic function, as shown in Fig. 2.20,

kz-%

the Fourier series representation of this unit impulse train is

fit) = px einwot
acess
17?
where, Ch= > ree dt
-T12
T/2
atT faire -jnoot qt = T
~T/2

tf(t)

t
-37 -27 -T 0 T 2T 3T

Fig. 2.20 Unit Impulse Train

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
106 Digital Signal Processing

Solution Here, the signal period is T = Z.Since x(t) is a periodic

signal, it cannot be an energy signal. Therefore, the power signal is
evaluated as
eT
P,= Í jx(t)[? dt
t
1
asla)
Pao | [Aef a

=a
n(2)
ja dt
ti
tet r
=Q [4%], a =A?
Since the signal has finite power, it is a power signal and E, = œ.

Emad Determine the magnitude and phase spectrum of the

pulse shown in Fig. E2.24(a).
A f(t)

Fig. E2.24(a)

Solution
Here f(t)=A, for-Tst<0
=-A,for0<t<T
=0, otherwise
- 0 e
F(jo)= Jre dt= O dt+ OT dt
= 0
o y
= fro el dt+ fro ei% dt
-TF 0
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
110 Digital Signal Processing

Fig. E2.26 (b) Phase Spectrum

Solution The magnitude |F(jw)| = 2, for-T<sosT

Mo) = ZF(jo) = 2/2, for œ < 0
=-
n /2, forw>0O
=0, fro=0
For the limits
-T < @<0, F (j ©) =ne/™?
For the limits
0 < œ< T, F (j œ) = ne 1"?
E 1 r . jot
O= oe FUD e do

o T
= alfre? eft do + frei”? ejot 2o)
2n T o
= lei” [=] lei [e]
2 jt jy 2 Jt Jo
2 [se] 2 fest
J
1 : P pia xi Şi ái
=— [e72 Z eie jTt +e jni2 eiT! -e sat |

1 ei™!2 _ g-i™!2 1 einl2-Tt) _ 4-j(n/2-Th

j2 t j2

= +[sin(n/2)~ sin(n/2- Tt)

1 2sin?( =) T*t Tt
—{1-cosTt
[ ] =—_12/ = — sinc? |—
t t 2 2

ene Obtain the Fourier transform of the trapezoidal

pulse shown in Fig. E2.27.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
114 Digital Signal Processing

ae i [ea + geese
2

1 eres jb)t _ ee ii

2 ~(a+ jo- jb) tarjo jo),

1 m i |
(a+ jœ- jb) (a+jo+
jb)
a+ jo+jb+a+
jo- jb]
7
oOo,
eee
r
a J (a + jo)? +b?
a+ jo
(a+ jo)? +b?

|Example 2.30) Find the Fourier transform of f (t) = t cos at

Solution
jat -jat
F [tcos at] = F jet
2 J
-ft e™ tei e-iot dt
A 2

1f, g`i atot | e`icatot |

Ei Eri jeto Jy
+ ieK aina - T aar 2 |
2| (-s(e+0)} [-j(a+o)]? Jo

šH t
2| (- i)? Ca +o)
;
1 1 1
mi =t
2|(-a+0) (a+) z
1a? +@? +2aw +a? +@? -2a0
2 (a? - a2}
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
118 Digital Signal Processing

Ea Determine the Fourier transform of the sinusoidal

pulse shown in Fig. E2.33.

0 Te
Fig. £2.33
Solution

Fijo)= |fe de
Here f(t) = A sin Wt for 0 < t < = 7/2 where wọ = 2n/T.
= 0, otherwise
T/S
F(j@)= fAsinogt e” dt
0

2 a (ea
m je dt
0 2j
T/2
z A ffo — Hoot dt
2j 4

A Fer eg A(@o + @)t i

2j J(@p-@) ~J(@o +w) 0

afen -1 r e`}(o +9)T/2 |

(Wp - 9) (Oo + @)
A (ei -w)T/2 _ 1)(Wy +0)+ enna - 1)(© - 0)

2 (o-o
2
?)
-A 3 d -i ý
= ae} ara G (et )T/2 +e jlog ae)

+ wfe -97/2 i e`ileo +72) _ 20)|

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
122 Digital Signal Processing

Fig. Q2.12
2.13 Obtain the Fourier series for the full wave rectified sine waves
shown in Figs Q2.13 (a) and (b).
A f(t

Fig. Q2.13

Ans : (a) fity=74

= (1- 2= cos2 wt - = cosd wt -Z cos6 wt =»)

(b) f(t = 24 (2+Š cos? at -Z cos wt + = cos6 wt~--]

2.14 Obtain the Trigonometric Fourier series expansion of the

periodic signal
A for kT <ts(k+DT
x(t) e |
-A for (k+1)T <t<s-kT
with k taking the values 0, 2, 4, 6...
2.15 A periodic triangular waveform starts at the origin with zero
value and increases linearly with respect to time. After a time T,
it becomes zero. Obtain its Fourier series.
2.16 With regard to Fourier series representation, justify the
following statement :
(i) Odd functions have only sine terms
(ii) Even functions have no sine terms
(iti) Functions with half-wave symmetry have only odd
harmonics.
2.17 Obtain the exponential Fourier series for the waveforms shown
in Figs Q17 (a) and (b).
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
126 Digital Signal Processing

2.44 Defining x(t) and y(t) as

íf ,-t
x= 16 fort20
0 fort<0O

y(t) = [oe fortz0

(0 fort<0
Find f(t) = x(t)« y(t).
2.45 With suitable waveform, explain the convolution of x(t), the
input and h(t), the system transfer function to get y(t), the
output.
2.46 Evaluate the Fourier transform of a single unit pulse of 1 volt.
2.47 Obtain the Fourier transform of a single symmetrical triangular
pulse.
2.48 Show that a time shift in the time domain is equal to a phase
shift in the frequency domain.
2.49 Find the Fourier transform of x(t) = Ae™*'Tu(t) and sketch its
magnitude and phase as functions of frequency.
2.50 Determine the Fourier transform of a two-sided exponential
pulse x(t) = e~'*!,
2.51 Find the Fourier transform of x(t) = Acos(@,t + 8).
2.52 Find the Fourier transform of f(t) = e"™ sinbt
2.53 Find Fourier transform of the following functions
(a) 5 sin?3t (b) cos(8t + 0.17)
Ans: (a) 2.52{2 5(@) + (w + 6) + w- 6)
(b)18.85.218° Alw- 8) + 18.85 418° lw + 8)
2.54 Find the Fourier transform of the single triangular pulse with
period T = 8 sec and amplitude A = 10V.
Ans: 40 sinc? 2w
2.55 Determine the Fourier transform of a one-cycle sine wave.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
130 Digital Signal Processing

ee: S —(at i A
y E ae e" (s+a)

Hence, £{Ae™}= a (3.4)

(s+a)
3. Sine Function

f(t) = sin @gt

Using Euler’s identity, we have
; Wt =aj”
sin 1 j sg £ dat)

t]
Hence, L{sin wt} = A [ecet — Lle}Po
2j
ak es DA E
eee pe
2ils-jM st+jo} s*+03
Q
H ence, „Lisin % nt}
L{si% t}= Saal
a (3.5)

4. Cosine Function
F(t) = cos Wot
We know that cos Wot = A ia karm

L {cos Wot} = [aeh + Lert )|

=-
1 1= +
1 C
sEE i
2[s- joo s+j®o s? +0
Hence, Li{cos W t} = z (3.6)
2 s? +03
5. Hyperbolic Sine and Cosine Functions

sinh œt = eror =g %t]

cosh Wo t = afe" +e)

(sinh wt) = Zete") - cet]

oy ae ee ee E
2[s-0 s+] s-o?
L{sinh ot) = 2 (3.7)
s“ -00

Similarly, £ {cosh at) = S[ete**) + £(e™*)]

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
138 Digital Signal Processing

Lt f(t)= Lt sF(s)
t30* se
Proof We know that
LiF = sle fO — f0)
By taking the limit s — œ on both sides
Ltt O= Lt [sF(s) — f(0)]

So
Lt |fe“ dt= Lt [sF(s)-f(0)]
0 soe

As s—> %, the integration of LHS becomes zero

i.e. J soe
Lt (P@e“Ide =0
0

Lt sF(s)- f (0) =0
s%

Therefore, Lt sF(s) =f (0) = ‘ Lt f(t)

se

3.5.2 Final Value Theorem

If f (t) and f (t) are Laplace transformable, then
Lt fo = lt SFO) (3,15)

Proof We know that

AFD} = sF(s) — f (0)
Taking the limit s — 0 on both sides, we get
it, Lif *(t)) = RSA [sF(s) — f (0)]

Lt, JPe“ dt= LtsF(s)- f(O)-

Therefore, Jf@at = Lt [sFis) — f(0)]

"FOR = Lt f(t)- Lt f(t)= Lt sF (s)-f
(0)
t t-+0 30
Sincef(0) is not a function ofs, it gets cancelled from both sides of the
above equation.
Therefore, Lt f(t) = “Lt sFis)
to s>

3.6 CONVOLUTION INTEGRAL

If X(s) and Hís) are the Laplace transforms of x(t) and A(t), then the
product of X(s)H(s) = Y(s), where Y(s) is the Laplace transform of y(t)
given by
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
142 Digital Signal Processing

pr Sa
RC (s+ —
1 i
RC
Taking inverse Laplace transform, we obtain

3.7 TABLE OF LAPLACE TRANSFORMS

Table 3.1 presents some functions and their corresponding Laplace
transforms. Table 3.2 lists the properties of the Laplace transform.
Table 3.3 gives the elements needed to develop the s-domain image of a
given time domain circuit.

Table 3.1 Laplace transforms pairs

ölt)
ölt- a)

u(t)

u(t - a)

L u(t), n positive interger

n!

e“ ult)

n,-at
tec u(t)
n!

sin (Wt) u(t)

cos(wgt) u(t)

t cos(t) u(t)

t sin(@pt) u(t)

e™ sin(wpt) u(t)

(Contd.)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
$46 Digital Signal Processing

For evaluating the constants Ay, Aj, Ag,....4,_ p we have to multiply

both sides of the above equation by (s — p; )”.
Hence, (s- p;)" F(s) = F(s) = Ag + Ay(s — p;) + Aas — p; )? + ...
-l4 N, (s)
+A, 4(s - pj)" D, (s) (s —p; )”

Substituting s = p;, we get

Ao = (s -p;i " F(s) |, <p,
Differentiating F(s) with respect to s, we get

Eno =A, +2A,(s—p,)+...+A,_,(n-1)(s—p,)"?

N,(s) n

biralD, at ~ Pi) }
Substituting s = p; in the above equation, we get

A= oe d

Similarly , A=

1
Generally, A,= 1a F,(s)|,-p, Where
n = 0, 1, 2,...n-1.

3.9 NETWORK TRANSFER FUNCTION

The transfer function H(s) of the LTI system, as shown in Fig. 3.2, is
(s) of the output signal to
equal to the ratio of the Laplace transform Y
the Laplace transform X(s) of the input signal when initial conditions
are zero. Thus

Hís) = Y(s) _
= Laplace transform of output (3.18)
e X(s) Laplace transform of input |4) initial conditions are zero

Fig. 3.2 Transfer Function of a System

The transfer function of a system H(s) is the Laplace transform of the

impulse response A(t). The transfer function H(s) is strictly analogous to
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
150 Digital Signal Processing

The magnitude response of the pure delay is |H(jm)| = 1 and the

phase response of the pure delay is ®( jw) = — wt.
Therefore, the time delay t is equal to the minus the derivative of the
phase response, i.e.
__ dO(ja)
do

CA Draw the poles and zero for the current I(s) in a

network given by
3s
I(s) = —
i” (s + 2) (s + 4)
and hence obtain i(t).

Solution The zero occurs at s = 0 and the poles at s = -2 and s =-4

as shown in Fig. E3.10.

z
Ps m— f $a + = oO

Fig. E3.10 Pole Zero Plot of I(s) .

The given function J(s) can be expanded by partial fraction as

A
I(s) = ——— + ——Ay
ʻe) (s+2) (s+4)
The coefficients A, and A, may be evaluated from the pole-zero
diagram.
ail Magnitude and phase angle of phasor from zero at zy to pole at p;
A, = k —
l Magnitude and phase angle of phasor from poleat p, to pole at p,

= 3.21180" _ 3 | 180° = 3 - (cos180° + j sin180°) = 3 x -1 = -3

2Lo°
Similarly,

4,09 A og
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
154 Digital Signal Processing

and Phase ®(j2)= L9o° ____|90°

Ltan-?(1)tan-1(3) L45°L71L8°
= 90° — 45° — 71.8° = — 26.8°

3.11 LAPLACE TRANSFORM OF PERIODIC

FUNCTIONS
The time-shift theorem is useful in determining the transform of
periodic time functions. Let function f (t) be a causal periodic waveform
which satisfies the condition f(t) = f(t + nT ) for all t > 0 where T is the
period of the function and n = 0, 1, 2....

F(s)= [f(e~™ dt
0

27 n+ DT
z dt +
dt+--+ f fitje
= fr@e-* dt + froe
T nT
°
As f (t) is periodic, the above equation becomes
T T T
= froe dt+e7*? [fOe dt + mte T ffe dt +-
0 o 0
T
=[L+e-8 +078 nt erT +-]ff@e* dt
0
= [1+ eT} (e-*7)? tent (eT) + |e)
T
where F(s)= fre dt
0
Here, F,(s) = £{[u(t)— u (t - T )) f(t)}, which is the transform ofthe first
period of the time function, and {[u (t) — u (t — T )] f @)} has non-zero only
in the first period of f(t).
When we apply the binomial theorem to the bracketed expression, it
becomes 1/(1- e~°7)

Fl) = ——
— T
r froe ita SER

ve Laplace transform of the periodic

rectangular waveform shown in Fig. E3.13
Solution Here the period is 2T

ee
Therefore, Z {f(t)}= soeer Jro e`“ dt
l]
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
158 Digital Signal Processing

Therefore, I(s)[Ls + R]= Yo,

s

Hence, I(s)=en ee
7 (e+)

2
LR
a
s+R/L Ris
E
s+R/L
Taking inverse Laplace transform, we get
R
io- i-e d (3.27)

Impulse Response
For the impulse response, the input excitation is x(t) = 5(t). Hence, the
differential equation becomes

pao + Ri(t) =8(t)

L {sIK(s) — i (0*)} + RI (s)=1

Since i(o*)=0
I(s)= 1 1 1
R+Ls Ls+R/L
Taking inverse Laplace transform, we get
å 1 -(R/L}t
i(t)=—-e
L u(t(t)

3.12.2 Step and Impulse Responses of Series R-C Circuit

Step Response
For the step response, the input excitation is x(t) = Vy. u(t). In the series
RC circuit shown in Fig. 3.4, the integro-differential equation is
t

A fi(t)dt+ Ri(t) = Voule) (3.28)

c -o

3 R

t=0

x(ĝ On’ C

mm
Fig. 3.4 Series R-C circuit
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
166 Digital Signal Processing

Solution Applying Kirchoffs voltage law, the loop equation can be

written as

2i(t)+1 dit) = 10sin25t

dt
Taking Laplace transform, we get

21(s) + [sI(s) -i (0) = 10x 73,

s^ + (25)
where i(0) is the initial current passing through the circuit. As the
inductor does not allow sudden changes in currents, i (0) = 0.

Therefore, sI(s) + 2I(s) = -0x25

s“ + (25)

250
I(s)= in
(s? + 625) (s + 2)
Using partial fractions, the above equation can be expanded as
250
I(s)=
(s + 2)(s + j25) (s - 725)

A, A, A;
I(s) =|—>+ deem T
s+2 s+ j25 s-j25

where A, =(s + 2) Us)|,-_2

Ag =(s +j25) I(s)|, --j25

: 250
(s + 2) (s - f25)|, - _ jos

a...
eee nee. a
© (2- j25)(-
j50) (25+ j2)
Ag = (s -j 25) I(s)|, = j25

7 250
(s + 2)(s + J25)}, - jo5

a. mene eee
(2 + j25)(j50) (25- j2)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
170 Digital Signal Processing

1
Ao 0 =F(s)s7|,-9=—~|
s)s“|s=0 Fl =1

1
A. sh, 4 ee 1 eee
1 ds s+1\,-6 (s +1)”

dente DF()|4.-1= 5] ad
S"ls=-1

Therefore, 3 1 = A _1i 4 1
s“(s+1) s s (s+1

Ha 1 jet 1+e7'
s“(s+1)
x -5 -2s
Therefore, i(t)= L '{U(s)]= £72 aroei
2s“ (s+ 1)
1 z
= slt-1+e |[u(t) - 2u(¢ - 1) + u(t - 2))

For the circuit shown in Fig. E3.22, determine the

resultant current i(t) when the switch is moved from position 1 to
position 2 at ¢ = 0. Initially the switch has been at position 1 for a long
time to get the steady state values.

Fig. E3.22
Solution Let us consider the switch be at position 2. By applying
Kirchhoff ’s law, we have
di(t) 5
0.2 —— + 4i (t) = 40
a e0
Taking Laplace transform on both sides, we get

0.21s1(s) - i (0)] + 41 (8) = =

i(0) is the initial current passing through the circuit just after the
switch is at position 2. Since the inductor does not allow sudden
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
178 Digital Signal Processing

ie.,i(t)= awe coswt + oC?R sinwt— Cet RC

+O

A periodic waveform shown in Fig. E3.28(a) is

applied to the RC network of Fig. E3.28(b). Find the transient current
and periodic or steady-state current.

|v(t) 12

; C fa T c

_ — j
0 m2 T 372 2T
(a) (b)
Fig. E3.28
Solution The function for the first period of the given waveform is
v(it)=1 for O<tsT/2
=0 forT/2<t<T
1 T
V(s) (s) = ———
1 ent t) dt
jro

1 T12 T

= mlfre“ dt + foe a
l-e 0 T12
i ent?
TIt [-sS |,
1 1—e 87/2

“=| s |
Alternate method to find Laplace transform of the given
periodic waveform
The input periodic pulse train can be represented as
v(t)= u(t)—u(t-T/2) + u(t-T)-u(t-3T/2) + u(t- 2T)- u(t- 5T/2) +...
Its Laplace transform is
Vis)=24 fi e872 4 eT -3572 ge-BT _e-55T/2;.]
s
= -[1- e857? 40 87 (1-087) + oT (e872)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
182 Digital Signal Processing

Comparing the coefficients of s, we get

A,+AgtAz =0
1- 1+4; =0
A; =0

= +... 4
Therefore, His) =<——__+—__-
(s+1)(s?+s+1) s+1 s?+s+1
We know that Y(s) = H(s) X(s)
1 s 1
|] =
1 1
EEE
Y (8) = | —— =a
@) (4 zo) +)

~ s(s+1) (s+1)(s+3) (s+1)?

ENPE: IEDERE. DOE P s
(s?+s+1) (s+3)(s?+s+1) (s+1)(s?+s+1)
By: using partial fraction expansions, the above functions can be
expanded as :

(i) |een PE
s(s+1) s s+1

Gi) 4 -1 __1
(s+1)(s+3) 2(s+1) 2(s+3)
Git) s = Z3/7 ,3/7s+1/7
(s+3)(s?+s+1) s+3 s°+s+1
1 s+1
>
:
(iv) —— i = - —— +
(s+1)(s?+s+1) stl s*+54+1
Therefore,Y(s) = 1- — +—I— -—
s+1 2(s+1) 2(s+3)
1 1 3/7 3/7s+1/7
{s+ s?+s+1 s+3 s?+s+l
1 +s+ 1
s+1 s?+s+1
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
190 Digital Signal Processing

3.44 A sinusoidal voltage 25sint is applied at the instant t = 0 to an

RL circuit with R = 5Q and L = 1H. Determine i(t) by using
Laplace transform method.
3.45 In the circuit shown in Fig. Q3.45, the steady state condition
exists with the switch in position 1. The switch is moved to
position 2 at t = 0. Calculate the current through the coil at the
switching instant and current for all values t > 0.

25Q

Xo
d : WIA
|
| l }10H

a Šasa ;
} Zsa
Fig. Q3.45
3.46 In the circuit of Fig. Q 3.46, the switch S is closed and steady-state
conditions have been reached. At t = 0, the switch S is opened.
Obtain the expression for the current through the inductor.

20
oe)

Lee a
tov * a a1 pF

Fig. Q3.46
Ans : 5cos1000t
3.47 In the circuit of Fig. Q3.47, the switch S is closed at t = 0 after the
switch is kept open for a long time. Determine the voltage across
the capacitor.
T $A |

i(th=10A t m i| S Xt=0

ee | | +
Fig. Q3.47
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
194 Digital Signal Processing

The analysis of any sampled signal or sampled data system in the

frequency domain is extremely difficult using s-plane representation
because the signal or system equations will contain infinite long
polynomials due to the characteristic infinite number of poles and zeros.
Fortunately, this problem may be overcome by using the z-transform,
which reduces the poles and the zeros to a finite number in the z-plane.
The purpose of the z-transform is to map (transform) any point
s = +o+j in the s-plane to a corresponding point z (r LO) in the z-plane
by the relationship
z=e*’,T where T is the sampling period (seconds)

Table 4.1

jo 0 of of 30/8 w2 508 30/4 To o

Z=1LwT 110° 1145° 190° 11135° 1L180° 1l225° 11270° 1l315° 11360°

Under this mapping, the imaginary axis, o = 0 maps on to the unit

circle |z| = 1 in the z-plane. Also, the left hand half-plane o < 0
corresponds to the interior of the unit circle |z | = 1 in the z-plane. This
correspondence is shown in Fig. 4.1.

mlS) In(2)

Fig. 4.1 Mapping of s-plane to z-plane for z = e/””

Considering that the real part of x is zero, i.e. o= 0, we have z = efor
= 1 | +jwT7, which gives the values of z (in polar form) shown as in
Table 4.1.
We know that the Laplace transform gives

Lx"(t)] = X(s) = Y x(nT)e*?

n=0
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
198 Digital Signal Processing

(vi) If x(n) is two-sided, and if the circle |z| = rg is in the ROC, then
the ROC will consist of a ring in the z-plane that includes the circle
|z| =r. That is, the ROC includes the intersection of the ROC’s of
the components.
(vii) IfX(z) is rational, then the ROC extends to infinity, i.e. the ROC is
bounded by poles.
(viii) If x(n) is causal, then the ROC includes z = æ.
(ix) If x(n) is anti-causal, then the ROC includes z = 0.
To determine the ROC for the series expressed by the Eq. 4.2, which
is called a two-sided signal z-transform, this equation can be written as
æ -1 ~

Zanr” = Fan)” + Yxn)"

na-u n=-0 n=0

xn) 27" + YFx(n) 2"

n=1 n=0

The first series, a non-causal sequence, converges for |z| <r , and the
second series, a causal sequence, converges for |z| > rj, resulting
in an annular region of convergence. Then the Eq. 4.2 converges for
rı < |z| < rg. provided r, < rg. The causal, anti-causal and two-sided
signals with their corresponding ROCs are shown in Table 4.2. Some
important commonly used z-transform pairs are given in Table 4.3.

Table 4.2 The Causal, anti-causal and two-sided signals and their ROCs
Signals ROCs
(a) Finite duration signals
Causal
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
202 Digital Signal Processing

íd) x(n) = { 0, 0, 1, 2, 5, 4, 0, 1)
Taking z-transform, we get
X(z) = 27 + 2294 5244425427,
ROC: Entire z-plane except z = 0.
(e) x(n) = &(n), hence X(z) = 1, ROC: Entire z-plane.
(f) x(n) = &n — k), k > 0, hence X(z) = z*, ROC: Entire z-plane except
z=0
(g) x(n) = &(n + k), k > 0, hence X(z) = z*, ROC: Entire z-plane except
z=,

| Example 4.3] Determine the z-transform including the region of

convergence of

a”, n20
x(n)=
0, n<O
Solution The z-transform for the given x(n) is

X(z) = Zla”] = Fare = ¥ (az)

n=-æ% n=0

< 1
We e know know that
tha 2a n= T i jaļ< 1

Hence, X(z)= l == z
-az z-a
This converges when |az™| < 1 or |z| > |a]. Values of z for which
X(z) = 0 are called zeros of X(z), and values ofz for which X(z) > œ are
called poles of X(z).
Here the poles are at z = a and zeros at z = 0. The region of
convergence is shown in Fig. E 4.3.

Zero at
origin z= 0

Iz|= lal

Fig. E4.3 ROC for the z-transform of x(n) = a”.

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
206 Digital Signal Processing

The ROC ofz™* X(z) is the same as that of X(z) except forz = 0 if k > 0
and z = œ if k <0.

|Example 4.8|By applying the time shifting property, determine the

z-transform of the signal
zt
X(z)= [at

Solution

X(z) = aa =z? X,(z)

where X,(z) = <L

-3z
Here, from the time shifting property, we have k = 1 and
x(n) = (3)" u(n)
Hence x(n) = (3)""! u (n - 1)

|Example 4.9| Find x(n)

147
if X(@) = —4—
1~=2z7

14,7 1 1z
Solution Given X(z) = 2 = +—2
l-1
1-=z la
1~=z t
1-=z
2 2 2

1
Therefore, x(n) = Z7! ———_ + =1 z”
1-22 2 1-1,71

)"at + Ha i -)

= (2) fu(n) + n(n — 1)

) [u(n) — u(n — 1) + 2u(n - 1)]

=(3) [&(n) + 2u(n - 1))

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
210 Digital Signal Processing

Taking inverse z-transform, we get

x(n) =27 ey -(3) u(n)

x2(n) =2 (te) =( 2) u(n)

OHE
acy
T A s T 1 n+1

If x(n) 2 >X{(z) and x(n) —— X,(z), then,

Nex, (l )= XY x4(n)xo(n -1) 25 R -Pg (z) = X(2z)Xx2"") (4.11)

n=-«

Seti te Determine the cross-correlation sequence r,,,,(l) of

the sequences:
x(n) = (1, 2, 3, 4)
x(n) = (4, 3, 2, 1)
Solution Cross-correlation sequence can be obtained using the
correlation property of z-transform, given in Eq. 4.11. Hence, for the
given x,(n) and x(n),
X,(z)= 1422743274425
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
214 Digital Signal Processing

with the coefficients representing the sequence values in the time

domain, namely
N(z E
X(z) = 2- =a, z” =az? +az! +agz? 4 +e
Diz) A%
where the coefficients a, are the values of x(n).
From the above equation it is clear that expansion does not result in
a closed form solution. Hence, if X(z) can be expanded in a power series,
the coefficients Tepresent the inverse sequence values. Thus, the
coefficient of z* is the K** term in the sequence. The region of
convergence will determine whether the series has positive or negative
exponents. For right hand sequences, called causal sequences will have
primarily negative exponents, while left hand sequences the anti-causal
sequences will have positive exponents. For annular regions of
convergence, a Laurent expansion will give both the positive and
negative exponents.
This method is only useful for having a quick look at the first few
samples of the corresponding signals.

|Example 4.19|A system has an impulse response A(n) = {1, 2, 3} and

output response
y(n) = {1, 1, 2, -1, 3}. Determine the input sequence x(n).
Solution Performing the z-transform of A(n) and y(n), we have
H(z) = Z[h(n)] = Z[1, 2, 3) = 1 + 2274 +327
Yiz) = Z[y(n)] = ZIL, 1, 2, -1, 3] = 1 +274 + 2277-27 + 324

We know that H(z)= Y)

XC)
-1 2.8 4
Therefore, X(z)= Y(z) aate the E tu.
H) 1+ 2z° +327
l-zl4+27

1422744327 | 1427242272 34324

1422714327?
~zt_-27_23%
-271-2277-323
—4
24227432
-4
274223432
0
Therefore, X(z) = 1- z! +2?
Taking inverse z-transform, we get
-11
x(n) = b 4 }
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
218 Digital Signal Processing

|Example 4.22 |Determine the causal signal x(n) having the

z-transform
1
Xz) = ——__, Sa
(1+z7%)(1-z71)?
Solution Expanding the given X(z) in terms of the positive powers
of z.

X(z2):) ———— z?
(z+ 1) (z- 1?
i 2 ý Áz :
Hence F(z)= 2 -— 2 -A , Ag
z (z+D(z-1? (z+1) (z-1) (z-1

Here, A1 =(2+1 F)|, 2-4 =—2zlz

2
1
ere =(z+1) Gps 13

= 2 DE a Pg:i
Ag =(z-1 Fa) |201= Cope =9g

d=| z? D 2z- z?
(z + , 3
A, = — re N ==
dz((z+1)j,-4 (z+) emy 4

Therefore, F(z) = +1 1 +33 1 +11 1

Me el Th dab sae
1-2 3 z 1 -2
fore,
Therefore, X(z)
X(z) = =4G+) =46-0 =2e-
Taking inverse z-transform of X(z), we obtain

x(n) = JCD'uln) + Žun) + gnun)

= [0 + 34 $n |uin)
Alternate Method

X(z)= aeae a
(1+274)(1-274)
A A
"Thetis ‘ach
Equating the numerators, we get
1= A,(1—27!)? + Ag (1+272) (1-271)+ A, (1+ 271)
= A (1-227) +.27*) + Ay (1-277) + Ag (14 277)
Here, A, +A, +A; =1
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
222 Digital Signal Processing

2+3z)
X(z) =
a+27)(14+2
+2
ae
at 27 JQ-2Pid )
i 1 4 l_-ı

Also verify the results in each case for 0 < n < 3.

Solution
(i) Long Division Method
-1
Xa) 2,24
142214527 -=2%
4 8

1+ 5 z7! + 1 z7?
8

EOE 94.1,-1_
Therefore, X(z) = 2+ 37 87
1,-2, 41,-8
+ 32 z

T D i
Taking inverse z-transform, we get x(n)=| ° 2° 8° 32’
T
(ii) Partial Fraction Expansion Method
24327!
ear dete )(i-2e)
A A A.
= AL+
l+z 1+7 ei

A\= 2+3z7 __8

(1+=
+
a
1-=a*
0-7") Asai 5
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
226 Digital Signal Processing

=> -1(1)'a" 2", |z| > jal

n=1 n

Taking inverse z-transform, we get

x(n) =
(pt *tta" u(n — 1)

For a low-pass RC network (R = 1 MQ and C = 1pF)

shown in Fig. E4.29, determine the equivalent discrete time
expressions for the circuit output response y(n), when the input is
x(t) = e~ and the sampling frequency is f, = 50 Hz.

Fig. £4.29
Solution The transfer function of the given circuit in the s-domain
can be expressed as
VRC 1
©) = S URC 341
Hi = = -——

Taking inverse Laplace transform, we get

hít) = et
z-domain approach:
Using Table 4.3, the above transfer function may be expressed in
z-plane as

H(z) = —
z-e?
Also, the given input function x(¢) = e`”! may be expressed in the
z-plane as
z
X(z )= peer

We know that the output function Y(z) = H(z) X(z).

Therefore,

Y¥(z) = =Z Z
@-eT) (z-e?)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
230 Digital Signal Processing

| =2 for 6sns10
| =3 for n>10
4 5
Ans: X(z) =1 +z +z? +23 427447754
2ft 74 rhe 4%) 4
Be Ue Pe.)
4.24 Determine the z-transform of the following sequences
ot
(a) um-4) fae =z) >1

(b) etul) Anë aa

() ôn- 5) Ans: 2°

o (Gju
íd)
1\"
3 u(-n) Ans:
depts
1
3; lal<3
1

92°
(e) 3"u(n - 2) Ans: ———
oe 1-327
4.25 Find the z-transform of the sequence x(n) = na"u(n)

Ans; X(z) = e _ |z| > Ja]

' (1-az")”’
4.26 Find the two-sided z-transform of
x(n) = (1/3)!" n20
=(-2)" ns-1

Ans: X(z) = 5. ~—2

z-+ z+2
3
4.27 Use convolution to find x(n) if X(z) is given by
1
X(z)=
det rade
(2 2° Jeg? )
Ans: EA EAN u(n)+3(
x(n) = 2a) 1(_1y
2) u(n)

4.28 Find y(n) using the convolution property of z-transform when

x(n) = {1, 2, 3, 1, - 1, 1} and h(n) = {1, 1, 1}
6,6,3,1,0 }
1,3,6,6,3,1,0,1
Ans: in) = {2°

4.29 Convolve the sequences x(n) and h(n) where

x(n)=0,n<0
=a",n20
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
234 Digital Signal Processing

1
4.45 Finindd ththe causal l sisigngnalal x(n) for Xa)X(z) = 1-
l+z`
iL27 +0
r.5oz°
s
Ans: x(n) = 104) cos (= - 71.565°)u(n)

4.46 Find the inverse z-transform of

Oe P
f 32* -4z +1

where the ROC is (i) |z| > 1 and (ii) |z| < i using the long
division method.

Ans: (i) x(n) = be

7 Colm

++» 121, 40, 13, 4,1,0

(ii) x(n) = { 3 A
4.47 Using long division, determine the inverse z-transform of
Xe) = 1+2z -1
1-227 +3”
if (a) x(n) is causal and (b) x(n) is anti-causal

;
1,4,7, 10,13,
Ans: (a) x(n) = {

++ 14, 11,8, 5, gi
(b) x(n) = { +
4.48 Determine the causal signal x(n) having the z-transform

X(z) = for the region of convergence |z| > :

Ans: x(n) -{80(2) - 20n(2)" +6[n(n- u/2I{=)

- 80 (2) |u(n)

4.49 Using (i) the long division method, (ii) partial fraction method
and (iii) residue method, find x(n) and verify the results in each
case for n in the range 0 sn <3.
z+3
(a) Xz) = 7025
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
238 Digital Signal Processing

This expression gives the output response y (n) of the LTI system as a
function of the input signal x (7) and the unit impulse (sample) response
h (n) and is referred to as a convolution sum.
5.1.2 Unit Step Response [u (n)]
The unit step sequence u (n) is defined by
0, n<O
u(n)= P nsô (5.3)

The shifted unit step sequence u (n — k) is given by

0, n<k
u(n—k) = {
1 nèk
The graphical representations of u (n) and u (n — 2) are shown in
Fig. 5.5.

-2 -1 0 1 2 3 4 5 n
(a) (b)
e
Fig. 5.5 (a) The Unit-step Sequence u (n) and
(b) The Shifted Unit-step Sequence u (n - 2).

The step response can be obtained by exciting the input of the system
by a unit-step sequence, i.e, x (n) = u (n). Hence, the output response
y (n) is obtained by using the convolution formula as

yia)= Ð h(k)u(n-k)
k=-

Relation Between the Unit Sample and the Unit-step Sequences

The unit sample sequence 6 (n) and the unit-step sequence u (n) are
related as

u(n)= X &(m), 5 (n) = u (n)- u(n - 1) (5.4)

m=0

5.2 PROPERTIES OF A DSP SYSTEM

The properties of linearity, time invariance, causality and stability of
the difference equations are required for the DSP system to be
practically realisable.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
242 Digital Signal Processing

Therefore,

aF [x,(t)] + bF fx) (t) = at

a SnO,+bcr
d
2
yt) Ti
LAO

bya (t) a
22 + ay; (t) + bys (t)
2
F [a x, (t) + b x (t)) = a la y; (t)+ b yo(t)]
+ la yi (t) +b y0) [2(ayt)+b ya (t))+ 1
Here aF [x,(t)] + bF [xo(t)] # F [ax, (t) + bx,(t)] and hence the system is
non-linear.
5.2.2 Time- Invariance
A DSP system is said to be time-invariant if the relationship between
the input and output does not change with time. It is mathematically
defined as
ify (n) = F [x (n)], then y (n — k) = F [x (n—k)] = 27 F [x (n)] (5.6)
for all values of k. This is true for all possible excitations. The operator
z™* represents a signal delay of k samples.

| Example 5.8|Determine whether the DSP systems described by the

following equations are time invariant.
(a) y(n)=F [x (n) =a nx (n).
(b) y (n) =F [x (n) =a x (n — 1) +b x (n - 2)
Solution
(a) The response to a delayed excitation is
F [x (n —k)] = an [x (n - k)]
The delayed response is y (n — k) = a (n — k) [x (n — k))
Here F [x (n — k) # y (n — k) and hence the system is not time
invariant, i.e. the system is time dependent .
(b) Here, F [x (n —k)] = ax [(n —k) — 1] + bx [(n — k) - 2]
=y(n-k)
Hence the system is time invariant.

|Example 5.9| Check whether the following systems are linear and
time invariant.
(a) F [x (n)) = n [æ (n)?
(b) F [x (n))=a [x (n)? +b
x (n)
Solution
(a) (i) F [x (n)) = n kx (ny
Here, F {x, (n)) = n [xy (n)? and
F ix; (n))= n [x (n)?
Therefore, F [x, (n)] + F [xq (n)] =n Ilx; (n)}? + {x3 (n)}]
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
246 Digital Signal Processing

= 2 [(% — tan™ œ) — (0 — tan“ 0))

2 2 [oo - 1/2) = o

Since the Paley-Wiener criterion is not satisfied, the amplitude

function is not a suitable amplitude response for a causal LTI system.

|Example 5.13) Using Paley-Wiener criterion, determine whether

i å ‘ 1
the magnitude function |H G @)| = is realisable.
v¥1+?
Solution According to Paley-Wiener criterion, the magnitude
function |H (j w)| is realisable only when the following condition is
satisfied.
Í IIH Go| do <o
1+9

Here, the given magnitude function is |H (j œ)| =

1+@

in|sober|
-in 1- log (1 + œ?)
V¥1+
o?
log(1 + w)

mS
Therefore,
1 log
2
(1 + œ’)

in{ 1 )
= = —lio
log (1+*)2
fi
diys
| =e | Eg
1+0?
i
17 loøg(1+w?)
-a ee 0

> j log (1+ 09")rm

A l+@
Substituting = tan 0, we get
1+? = 1+ tan? 0 = sec? 6,
dw=sec?@d@ and @ becomes 0 to n/2
n/2 2
= Í log æ s2 . sec? @dO
ọ «sec @
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
250 Digital Signal Processing

Check the stability condition for the DSP systems

described by the following equations .
(a) y (n) =a" u (n)
(b) y (n)=x(n)+e*y (n-1)
Solution
(a) y (n) =a" u (n)
Taking z-transform, we have

Y =
1 =æ
z
as l-az? z-a
Here the pole is atz =a and hence for the system to be stable, ja |< 1.
(b) y(n) =x (n) +e" y(n- 1)
Taking z-transform, we have
Y (z)=X(z) +e" z! Y(z)
Y (2) [1 - e" 27] =X (2)
Y (2z) I 1 he
Therefore, H(z)=
X(z) 1-etz! z-e"
Here the pole is at z = e° and hence |e*| < 1, i.e. a < 0 for stability.
5.2.5 Bounded Input- Bounded Output (BIBO) Stability
Stability is of utmost importance in any system design. There are many
definitions for stability. One of them is BIBO stability. A sequence x(n)
is bounded if there exists a finite M such that |x(n)| < M for all n. Any
system is said to be BIBO stable if and only if every bounded input gives
a bounded output.
For any linear time invariant (LTI) system, the BIBO stability
depends on the impulse response of that system. To obtain the necessary
and sufficient condition for BIBO stability, consider the convolution
property which relates the input and the output of a LTI system as

y(n)= ¥ x(n-k)h
(k)
k=-%

It follows that

|x(n)| = DY, x(n-h ACR) |

k=-

< ¥ |x(n-k)| |h(k)|

k=-

<M Y, |h(k)| [since |x(n)| < M for all n]

k=-

where M is a finite constant.

Therefore, the output is bounded if and only if
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
254 Digital Signal Processing

The necessary and sufficient condition for BIBO stability is

Y |A(k)| <2
k=0

Here, > [A(k)| = |b]

k=0
Hence, the given system is BIBO stable if |b] < œ.
(e) y (n) = ax (n). x(n- 1)
If x (n) = 8 (n), then y(n) = h(n).
The above equation can be changed into h(n) = aô (n) . 8 (n-1)
When n = 0, A(0) = ad (0) . 8 (-1) = 0
Whenn = 1, A(1)
= aô (1). 8 (0)
=0
Whenn = 2, A(2) = aô (2). 5 (1) = 0
The necessary and sufficient condition for BIBO stability is

È [hlk)| <
k=0

Here, $, |A(k)| =0.

k=0
So the given system is BIBO stable .
(f) y (n) = max. of [x (n), x (n — 1), x (n — 2)) .
If x(n) = ô (n), then y(n) = h(n).
The above equation can be changed into
h(n) = max . of [85 (n), ô (n — 1), ô (n - 2)]
When n = 0, h (0) = max . of [5 (0), 5 (-1), 5 (-2)}=1
When n = 1, A (1) = max. of [ò (1), è (0), 8 (-1)] =1
When n = 2, h (2) = max . of [ò (2), (1), (0) = 1
When n = 3, A (3) = max . of [5 (3), 8(2),5(1)] = 0
In general,
h(n)=1, forn=0,1,2.
=0, otherwise (i.e., n > 2).
The necessary and sufficient condition for BIBO stability is

|h (R)| < œ.
k=0

Here, > JACk)| = |h (O)| + |A (D| + [A (2)] +-+ |h (k) +

k=0
=1+1+1+0+-=3.
So, the given system is BIBO stable.
(g) y (n) = average of [x (n + 1), x (n), x (n -1)).
If x (n) = 8 (n), theny (n) = h (n).
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
258 Digital Signal Processing

bilap , OSnsm
h (n)= { : (5.16)
0 , otherwise
As this is obviously of finite duration, it represents a FIR system.

|Example 5.20 |A DSP system is described by the linear difference

equation
y (n) = 0.2 x (n) — 0.5 x (n — 2) + 0.4 x (n - 3)
Given that the digital input sequence (-1, 1, 0, —1} is applied to this
DSP system, determine the corresponding digital output sequence.
Solution Taking z-transform of the given linear difference equation,
we get

Y (z) = 0.2 X (z) - 0.5 2° X (z) + 0.4 7° X (2)

Therefore,
Y (z)
H(z)= = 0.2-0.527+042°
X (z)
The given input sequence is x (n) = {- 1, 1, 0; —1} and its z-transform is
X(z)=-1+2'-2%
Therefore, Y (z) = H (z).X (z)
=-0.2 +0.22" +0.5z7°?-1.127° +0.424+0.52°-0.42%
Taking inverse z-transform, we get the digital output sequence
y (n) = {- 0.2, 0.2, 0.5, — 1.1, 0.4, 0.5, — 0.4}

|Example 5.21 |Determine H (2) and its poles and zeros if

y(n)+ 2y(n- V+ Fy (n= 2 = x(n) +z(n-D

Solution Given

yin)+ 2y(n-1)+ gy (n= 2)=x(n)+x(n-1)

Taking z-transform,we get

¥(z)+ ir ¥ (z)+ 2° Y (2) =X (z) +27 X (2)

H@=-Y@ 2.
_1+e* eet
X 443,-1,1,-2 2243241
4 4 8
z(z+
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
262 Digital Signal Processing

Using the convolution property of the z-transform, we get

H (z) = H, (z) H, (2) s H, (2) (5.18)
Hence, H (e/®) = H, (e*®) H, (e”) --- H, (e7)
Here we observe that the cascade connection involves convolution of
the impulse responses in the time domain and multiplication of the
frequency responses in the frequency domain.
5.4.3 Energy Density Spectrum
In the digital system, the spectrum of the signal at the output of the
system is
¥ (z) = H(z) X(@)|, =e/® (5.19)
Hence, Y (e”) = H (e/”) X (e/”). This is the desired input-output
relation in the frequency domain, which means that the spectrum of the
signal at the output of the system is equal to the frequency response of
the system multiplied by the spectrum of the signal at the input.
1Y (e)|? = |H (e/*)[? |X (e4*)|? (5.20)
As the energy density spectra ofx(n)is S, (e/”) = |X (e+) |? and the
energy density spectra of y(n) is S, (e/®) = |Y (e?®)| 2 we have
S,, (e/”) = |H (e|? S,, (e”) (5.21)
5.4.4 Magnitude and Phase Spectrum
The magnitude response is the absolute value of a filter’s complex
frequency response. The phase response is the angle component of a
filter’s frequency response. For a linear time invariant system with a
real-valued impulse response, the magnitude and phase functions
posses symmetry properties which are detailed below. From the
definition of z-transform, H (e/®), a complex function of the real variable
@ can be expressed as

He) = $, hn) ei
n=-s0

= 5 h (n) cos on -j 3 h (n) sinon

n=-% n=-
= Hp (e*”) + jH, (e*™)
= |H (et)|ei?
; r ; -1 jo jo
= JH (e19) + H2 (e/®) e1" Hy (e%)/Hpce
where H p (e jie) and H; (e 4®) denote the real and imaginary components
of H (e).

Therefore, |H (e) | = (H2(e!®) + H?(e!®)

Hı (e/®) |
® (w)
(@) == tan tan? [eee
:
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
266 Digital Signal Processing

Therefore, H (z)|,..=H (e/) = ; =e

Magnitude and phase responses:

Therefore |H (e®)| =
GEDI4 GER)4
Ow) == o + arg(ejo _1)_
1) arg (ejo _1)_
1) arg (ejo _33)

Determine the frequency response, magnitude

response, phase response and time delay of the system given by

y(n)+ ja- 1) =x (n)-x

(n - 1)

Solution
To find the frequency response H (e/“):
Given,y (n)+ iya- 1)=x(n)- x(n- 1)
Taking z-transform, we get

¥(2)+ 521¥@)=X@)-27X()
1o-1). z1
¥@)[1+32 |-x@u zy

H (z) =
Y(z) _ l-z?
a X(z) 441,71
2
l-e
Therefore, the frequency response is, H (e ™) = -jù
i+i¢
2
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
270 Digital Signal Processing

Points A and C in Fig. 5.8 correspond to frequencies 0 and ,/2, the

Nyquist frequency or folding frequency, which is equal to nf, or n/T,
where f, is the sampling frequency and T is the sampling period. One
complete revolution of the phase e*? about the origin corresponds to a

frequency increment of ©, = zr, Here, H (e *7) is a periodic function of

frequency with a period @, . The frequency response has the property
H (e?°") = H (e*"). Therefore, the magnitude function M (w) is an even
function of œ, and the phase function © (w) is an odd function of œ .
Vectors are drawn from each pole and zero to e“”” point on the unit
circle.
The transfer function of a digital system may be expressed in terms of
its poles and zeros as
p x
Hy M1 (eT —z;)
He) = M () e/%™ = —=* __ (5.23)
ALS (eT — p;)
By substituting, (eT _ 2;) = My et
and (eT - p;)= Mpi fn
We have, Magnitude as
M (o) = |H (e*)|
Ho fiM;i
Z i=1
~ q
Il M

p
Ho LL {Vector magnitude from the i“ zero to the frequency point
i=

on the circumference of the unit circle}

q
A {Vector magnitude from the i” pole to the frequency point

on the circumference of the unit circle }

That is, the magnitude of the system function M (œ) equals the
product of all zero vector lengths, divided by the product of all pole
vector lengths.
Phase Shift
® (w) = LH (e’*)
Pp q

=}, 0,- Pp,

i=1 i=l
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
274 Digital Signal Processing č č č o

nt
onw =z) , —-1<n<15
0 , otherwise
Ans: (a), (c) and (e) are causal and stable.
5.26 For each of the following discrete-time signals, determine
whether or not the system is linear, shift-invariant, causal and
stable.
(a)y (n)=x(n+7), @©)y(n)= (n), Cy (n)=nx (n)
4
(dy m=a+ 2 x(n - k), wis a non-zero constant.

(by (mM =at+ x x(n- k), ais a non-zero constant.

h=-4
Ans: (a) Linear, time-invariant, non-causal and stable.
(b) Non-linear, time-invariant, causal and BIBO stable.
(c) Linear, not time-invariant, causal and not BIBO stable.
(d) Non-linear, time-invariant, causal and BIBO stable.
(e) Non-linear, time-invariant, non-causal and BIBO stable.
5.27 Determine whether the following systems are linear or non-
linear, causal or non-causal, shift invariant or shift-variant.
WOynT=xnT+T)+x(nT-T).
(i y(n T) =2 (nT+ Te" sin onT
(iy (n) =a y (n - 1) + x (n)
5.28 Using the Paley-Wiener criterion, show that |H | = eP? is
not a suitable amplitude response for a causal LTI phot
5.29 Discuss the stability of the system described by
-1
H (z) = ——1—__;
a | ee ia
5.30 Find the stability region for the causal system

Hj = ——-1___,
1+a,z-° +d9z
by evaluating its poles and restricting them to be inside the unit
circle.
5.31 A causal LTI system is described by the difference equation
y(n)=y(n-1)+y(n-2)+x(n-1)
where x (n) is the input and y (n) is the output.

(i) Find the system function H (z) = za for this system, plot the

poles and zeros of H(z) and indicate the region of convergence.

(ii) Find the unit sample response of the system.
(iii) Is the system stable or not?
(Contd. )
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
282 Digital Signal Processing

In the first case the impulse response A (n ) is folded and shifted, and
x(n) is the excitation signal. In the second case the input signal x(n) is
folded and shifted. Here k(n) acts as the excitation signal.
Commutative Law
Convolution satisfies commutative law, i.e. x(n) » h(n) = h(n) * x(n)
x(n) ho) y(n) =x(n) *h(n) hn) x(n) h(n) = h(n) +x (n)

Fig. 6.3 Commutative Property

Associative Law
[x(n) » hi(n)] * han) = x(n) » [h;(n) * ho(n)]
Take LHS of the above equation.
Consider x(n) to be the input signal to the LTI system with impulse
response h ,(n). The output y,(n) is given by
y(n) = x(n) + hy(n)
This y,(n) signal now acts as the input signal to the second LTI
system with impulse response h.(n).
Therefore, y(n) = y(n) * han)
= [x(n) x hy(n)) + h(n)
Now consider the RHS of the equation, which indicates that the input
x(n) is applied to an equivalent system h(n) and is given by
h(n) = h(n) * h(n)
and the output of the equivalent system to the input x(n) is given by
y(n) = x(n) = h(n)
= x(n) + [hy(n) » ho(n)]
Since convolution satisfies commutative property, the cascading of
two systems can be interchanged as shown in Fig. 6.4.

araia. e]
y(n) yin)
y yin)
t

2) enen E atm) ad
Fig. 6.4 Associative Property

If N linear time invariant systems are in cascade with impulse

responses h(n), h2(n),...Ay(n), then the equivalent system impulse
response is given by
A(n) =h (n) *hg(n)«... #hy(n)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
290 Digital Signal Processing

=
17 [60+4+4)=17
1 M ERT
+4j e)? e/*]
ri -4jjee" +4j
x3 (2) = İf60
= £{60-4j(-1)+4j(-D]=15
x4(3) = 260+ (-4j) 2 + 4je792]
= F160 +(-4))(-) +450)
= <[60~ 4-4] =13
Therefore, x,(n) = [ 15, 17, 15, 13 ]
Note: From the above results, we find that the resulting sequences
obtained by both linear convolution and circular convolution have
different values and length. Linear convolution results in an aperiodic
sequence with a length of (2N- 1), i.e. seven in this case, whereas
circular convolution results in a periodic sequence with a length of N,
i.e. four in this case. Circular convolution will produce the same
sequence values as those produced by linear convolution if three zeros
are padded at the end of the two given sequences x ,(n)and x,(n).

|Example 6.2|Find the response of an FIR filter with impulse

response A(n) = {1, 2, 4} to the input sequence x(n) = {1, 2).
Solution
Linear Convolution
Given h(n) = { 1, 2,4} and x(n) = {1,2}
Here N, = 3 and N, = 2. Hence N=N,+N2-1=4

We know that y(n) = x(n) + h(n)= Y x(k)h(n- k)

k=-

Therefore,

y(0)= J x(k)h(- k)
k=-m

=... + x(0) (0) + x(1) A(-1) +...

=0+1+0..=1

y)= J x(k)h(1-k)
k=-œ
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
294 Digital Signal Processing

When n=l

y= J x(k) h1- k)
ka-a
=... + x(—1) A(2) + x(0) ACL +)x(1) A(0) +...
1 ) + ( 1 ) ( 1 ) + 0 . . . = 2
=0+(1)(
When na2

y(2)= $ x (k) h(2- k)

hk=-0

=... +x (-1) A(3) + x(0) A(2) + x(1) A(1) +...

=0+(1)(1)+0...=1
When n=3

y(3)= F xk) h(8-—k) =0

h=-oo

When n=-1

y-1)= $ x(k) h(-1-k)

koe

=... +x(-1)
A(1) + x(0) A(-1)
+ x(1) A(-2) +...
=0+(1)(1)+(1)()+0...=2
When n=-2

y(-2)= J x(k) h(-2-k)

k=-

=... + x(—1) A(1) + x(0) h(—2) + x(1) A(-3) +...

=0+(1)(1)+0...=1
When n=-3

y(-3) = J x (k) A(-3- k) = 0

June
The convolution signal y(n) is [See Fig. E6.3(b)]
y(n)=0,n
5-3 and n23
yín)=1,n=+2
yín)=2,n=+1
y(n)= 3,n=0
Note: For the convolved signal, the left extreme and the right
extreme can be found using the left and right extremes of the two
sequences to be convolved. That is,
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
298 Digital Signal Processing

y(n)= J h(k) x(n -k)

kz=-%

When n=0

yO) = J h(k)
xi- k)
j

=... + A(0)x(0)
+ A) x(-1) +...
=0+1+0+...=1
When n=1

y= SAR) x(1-k)
how

«+ A(O) x(1) + ACL) x(0) + A(2) x(-1) +...

0+(1)(1)+(a)(1)+0...=1lt+a
When n=2

y(2)= F h(k) x(2-k)

ka=

.-. + A(O) x(2) + A(1) x(1) + A(2) x(0)+A(3) x(-1) +...

+... =1+a +a?
0 + (1) (1) + (a) (1) + (a?°X1)+0
When n=3

x(3 - k)
y(3)= J h(k)
kee
=... + A(O) x(3) + ACL) x(2) + A(2) x(1) + A(3) x(0)
+Ah(4)x(-l +...
= 0 + (1) (1) + (a) (1) + (a2) (1) + (a3) (1)+0 +...
=lt+a+a?+a°

Fig. E6.4(b)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
302 Digital Signal Processing

When n=3

y(3)= J xk)h(3-k)=0
huee

=... + x(0) A(3)

+ x(1) A(2)
+ x(2) AC) +...
+ (1) (0) + (2) (-1) + 3 (-2)+0...=-8
When n oO

y(4)= J x(k)h(4-k)
ku-
=... + x(0) A(4) + x(1) A(3) + x(2) A(2) +...
= 0 + (1) (0) + (2) (0) + 3 (-1)+0....=-3
These sequence values are plotted in Fig. E 6.5(b).

|Example 6.6] Compute the convolution y(n) = x(n) » h(n) of the

signals
a ee
x(n) = land A(n)= {
T T
Solution The sequences of the given two signals are plotted in
Fig. E6.6(a).

x(n) A h(n)

Fig. E6.6 (a)

From the graph,
x, =-2, x,=2, h,=-3, h,=0

Hence the left and right extremes of the convoluted signal y(n) are
calculated as
yp =x, +h, =-2+(-3)=-5
y-=x, +h, =2+0=2
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
306 Digital Signal Processing

and the corresponding IDFT is

N-1
x(n) = + See, n=0, 10,0 -1
k=0

6.3.1 Relationship between the DFT and Other Transforms

Relationship to the Fourier Series Coefficients of a

Periodic Sequence
The Fourier Series of a periodic sequence x, (n) with fundamental period
N is given by
Nal <o
x(n) = dep eitenkIN en
k=0
where the Fourier series coefficients are given by
1%! i
1... Keb, N-1.
C= = Sx, (n) e Jenner .
N n=0

By comparing the above equations with that of DFT pair and defining
a sequence x(n) which is identical to x, (n) over a single period, we get
X(k) = Ne,
If a periodic sequence x, (n) is formed by periodically repeating x(n)
every N samples, i.e.

xpin)= $ x(n-1N)
=-~%

The discrete frequency-domain representation is given by

N-1 :
X(k)= Fapa PYN =NC,, k=0,1,...,.N-1.
n=0

and the IDFT is

1%: i
x(n) = m E Xk ™EN, locn <o
k=0

Relationship to the Spectrum of an Infinite Duration

(Aperiodic) Signal
Let x(n) be an aperiodic finite energy sequence. The Fourier transform
is given by

X(e/”) = ¥ x(n) eJan

N=-0co

If X(e/”) is sampled at N equally spaced frequencies,

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
310 Digital Signal Processing

Hence,
x’(-n,(mod N)) = x"(N - n) ae x(k)
8. Circular Convolution

If x(n) T, X (k) and x,(n) OFT +X 1k), then

x (n)an) PET SX (k) Xalk)
F where x(x 2(n) denotes the circular convolution of the sequence
x(n) and x(n) defined as
N-1
x3(n) = Ea (m) xin - m, (mod N))
m=0
N-1
= X x2(m) x(n -m, (modN))
m=0

9. Circular Correlation
For complex-valued sequences x(n) and y(n),

if x(n) FT,Xk) and y(n) T,

DFT
Y(k), then
rall) PET+ Rik) = XHY)
where r,,(/) is the (unnormalised) circular crogs-correlation sequence,
given as
N-1
r (l) = Exin) y*(n-—l, (mod N))
n=0

10. Multiplication of Two Sequences

If x(n) OFT 9X (hk) and x(n) 2T Xyk ), then

x(n) x(n) DIT xo, (k)
I 1. Parseval’s Theorem
For complex-valued sequences x(n) and y(n),

if x(n) PET X(h) and y(n) oT Y(k), then

N-1 : 1 N-1 .
Lany (m = 57 LXO W
a=0 k=0
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
314 Digital Signal Processing

= 1+ 0.5 —j0.866 + 2(—0.5 -j 0.866) + 2(- 1)

+ 3(-—0.5 + j 0.866) + 3(0.5 + j 0.866)
=- 1.5 + j 2.598
Fork=2
5 P
X(2) = SY x(n) e 1282016
n=O

m $ x(n) e7211™3

n=0

= 14 e778 4 20 43 5 Qe I 4 ZeJN y Ze i103

= 1+ (—0.5) - j0.866 + 2(—0.5 + j 0.866) + 2(1)
+ 3(-—0.5 -j 0.866) + 3(-0.5 + 7 0.866)
=-1.5 +j 0.866
Fork =3
5
X(3) = F, x(n) e/200"/6
n=Q

5
= Dy x(n) e717"
nae F A
= 1407" + QeF™ + 2e" 4 Betr 4 Be V5m
=1-—1+ 2(1) + 2(-1) + 3(1)+3-D=0
For
k =4
5

X(4) = J, x(n) eo /2MOn

n=0
5
= X x(n) e7145"/3

n=0
= 14 eF4™3 4 278m3 4 2e vAn 4 BoV EMS 4 3p V20K/3
= 1+ (-0.5 + j 0.866) + 2(-0.5 -j 0.866) + 2(1)
+ 3(—0.5 + 0.866) + 3(-0.5 -j 0.866)
= -1.5 -j 0.866
Fork=5

X(5)= $ x(n) e7725 e

n=0

= ¥ x(n) a
n=0
as ~j5x/3 -j10x/3 jn ~j20%/3 ~j25n/3
=l+e +2e +2e7™ + 3e +3e
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
318 Digital Signal Processing

(1+ 2¢/9*/? 4 3e 3% 4.40 /9n/2)

(1+ 2(-j)+3(-)+4))

(-2+2j)=-=+j

Cun
Aj
Dje
Sia

|Example 6.14 |Determine the IDFT of X(k) = {3, (2 + j), 1, (2-J)}.

Solution The IDFT is defined as
-1 s

x(n) = pa X (kjel? "HN O<n<N-1.

k=0

3
Given N= 4, x(n) = +>: X(k) ef ™*2 O<n<3
k=0
When
n =0
3
x0)= 1 F, Xhe
4 2

=213+(24 +1+(2- pl=2

When
n =1

is j
x(1) e x X (k)
ef*#/2

= HE + (2+j)ei? + eit 4(2—j)ej®™2]

=f 13+(24+j)j-1+(2--) Cj =0
When
n = 2

[3+ (2+ j) ett +e/?® +(2- j) e37]

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
326 Digital Signal Processing

Table 6.1

Index Binary representation Bit reversed binary Bit reversed index

0 000 000 0
1 001 100 4
2 010 010 2
3 011 110 6
4 100 001 1
5 101 101 5
6 110 011 3
7 111 111 7

The basic computation in the DIT FFT algorithm is illustrated in

Fig. 6.9, which is called a butterfly because the shape of its flow graph
resembles a butterfly.
The symmetry and periodicity of Wy can be exploited to obtain
further reductions in computation. The multiplications by Wy = 1,
A2 = -1, WN4 = j and Wi” = -j can be avoided in the DFT
computation process in order to save the computational complexity.
a ai S iai Pi A=
* A=a+W r
wb
eal

we ae Teg:
be = a > « Baa-Wnb
-1
Fig. 6.9 Basic Butterfly Flow Graph for the Computation in the DIT FFT Algorithm

In the 8-point DIT FFT flow graph shown in Fig. 6.8, W,°,W,* and W,®
are equal to 1, and hence these scale factors do not actually represent
complex multiplications. Also, since W, Wi, and W equal to —1, they
do not represent a complex multiplication, where there is just a change
in sign. Further, W,', Wit, Wê and W? are j or -j, they need only sign
changes and interchanges of real and imaginary parts, even though they
represent complex multiplications. When N = 24, the number of stages
of computations is L = log, N. Each stage has N complex multiplications
and N complex additions. Therefore, the total number of complex
multiplications and additions in computing all N-DFT samples is equal
to N log, N. Hence, the number of complex multiplications is reduced
from N? to N log 2N.
In the reduced 8-point DIT FFT flow graph shown in Fig. 6.10, there
are actually only four non-trivial complex multiplications corresponding
to these scale factors. When the size of the transform is increased, the
proportion of nontrivial complex multiplications is reduced and N log,N
approximation becomes a little closer.
The reduced flow-graph for 16-point decimation-in-time FFT
algorithm is shown in Fig. 6.11.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 330

S
OL
(0)¥p= —o `> Aa K] o 0z=(0)K

o—¢ pit'z/-8z8's-=(
gM b= l-
o v= (b)x eS

€=(2)x
—o
Digital Signal Processing

0=(2)x

z=(9)x o O!-z2t0=
pty ()x
1=8m
=98
aoo
——oz=(
0=(p)x ;)x

¢=(g)x#—° co -=(S)x2210 pivot

b=3M j=

=2@
o y=(€)x . 0=(9)x ©

4—o 1 =(2)* + 5 pipeze's

2!+ -= (2)X ©
t=8m is =M Sa
-
202°0/
£04"0-

‘314
s193
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
334 Digital Signal Processing

Using DIT FFT algorithm, we can find X(k) from the given
sequence x(n) as shown in Fig. E 6.17.
Therefore, X(k) = (255, 48.63 + 7166.05, —51 + 7102, -78.63 + j46.05,
—85, —78.63 - j 46.05, -51 —j102, 48.63 — 7166.05}

| Example 6.18 |Given x(n) = {0, 1, 2, 3}, find X(k) using DIT FFT
algorithm.
Solution Given N= 4
- (2%)
Wize tw)
Wf =1 and Wi =e7*? =-j
Using DIT FFT algorithm, we can find X(k) from the given
sequence x(n) as shown in Fig. E 6.18.
Therefore, X(k) = {6, -2 + j2, -2, -2-j 2}

6.4.3 Decimation-in-Frequency (DIF) Algorithms

The decimation-in-time FFT algorithm decomposes the DFT by
sequentially splitting input samples x(n) in the time domain into sets of
smaller and smaller subsequences and then forms a weighted
combination of the DFTs of these subsequences. Another algorithm
called decimation-in-frequency FFT decomposes the DFT by recursively
splitting the sequence elements X(k) in the frequency domain into sets
of smaller and smaller subsequences. To derive the decimation-in-
frequency FFT algorithm for N, a power of 2, the input sequence x(n) is
divided into the first half and the last half of the points as discussed
below.
(N12) -1 N-1
X= $, x( n) Wa h + D,
n=N/2
xn ) Wy
a=0
(N/2)-1 N
(N/2)-1
= J mW + È x(n+ 2) wyrenme
n=0 2
n=0

=1
) x(n) Wy" + Wy? +
NADY (N /2 )- 1
= SY x( n +. N/ 2) Wr * (6.24)
n=0
n=0

_ ,2n N
Since, WiN/2* =e 45 2" = cos (nk) —j sin nk = (-1)*, we obtain
(N/2)-1 (N/2)-1
X(k)= J, x(W +CD? J, x(n+.N/2)
wy"
n=0 n=0

(N/2)-1
= > [x(n + D" x (n+ Z)] wy (6.25)
n=0 j 2
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 338

(05 oy*

<
<
ox
(0)

(o)x
A
4-

La
wy qwiogz
Lag
(ixo ae a 7 gs — a (px

(8
Miik

=
Nm
(2)
6Oe
O—-—

(ax> X J l:
Digital Signal Processing

(Ne
()x N ra i

(px HP
X Qa e 05° E
cca
e e
L ee
(g)x° o
WaS Z Wied la o~ (3)x

(9)xO aa2
a a eane
WEA

ax
b= l-

€1°9 “Bid 8 = N 10} 144 Aouanbai4-uj-uonowizeg Jo a8pxg puoras ay3Joydo1y Mojy

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 342

4=(o)x
d-— esz = (0x O--—~¢-
x
gM
Z=(L)¥
= 0—4 —¢+ O— $8-=(r)x

0
89 am iS-
p=(z)x
d óe —— l Q -4 pee
o 1S-=(2)Xx
Z01/+
SO pare
Digital Signal Processing

(e)xB= b ft’ p à p 0—+—0 (9x is-= zo-

l-
t-
°
ae ee B
=(p)x
9b o
L-
as
reas o o D
+€9'E9
eat ni
XX Hg 6901 sm
Zeo
=(9)x Q
aS x -4 F- K O a Q o (9)X-= -£984 so'9r!
t- t-

Q Q O O O (Ex
= +6984- S0'9p/

t9
o us
=(9)x
(2)x
= 82k —p Q so'991 /-€9°8h =(L)x ©
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
346 Digital Signal Processing

Use the 4-point inverse FFT and verify the DFT

results (6, -2 + j2, -2, -2 —j2} obtained in Example 6.18 for the given
input sequence {0, 1, 2, 3}.
FELS
Solution We know that W $ = e Ay a Hence,
W Y= 1 and WY =e =j
Using IFFT algorithm, we can find the input sequence x (n) from the
given DFT sequence X(k) as shown in Fig. E6.23.

X(0) = 6 _ x(0) =0

XN =-2 + 20—
X(2)=-2

(3) =-2 nto

Fig. E6.23
Hence, x(n) = {0, 1, 2, 3)

Given X(k) = (20, -5.828 -j 2.414, 0,-0.172 -j 0.414,

0, — 0.172 + j 0.414,0, —5.828 + j 2.414 }, find x(n).
- (28)
Solution We know that W X = e Aw) . Given N = 8. Hence,
W,’=1
wz! = 0.707 + j 0.707
We =i
Wg? =-0.707 + j 0.707
Using IFFT algorithm, we can find x(n) from X(k) as shown in
Fig. E. 6.24.
Hence, x(n) = (1, 2, 3, 4, 4, 3, 2, 1)

GivenX(k) = {255, 48.63 +j166.05,—51 +j102,-78.63

+ j46.05, -85, -78.63 — j46.05, -51 -j 102, 48.63 -j 166.05 }, find x(n).
-j{(ž5)r
Solution We know that Wy = e ITE Given N = 8. Hence,
W,°=1
W, |= 0.707 + j0.707
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 350

(0)x
= 9€
(0)x
=1

oO
9s9'°6/+p-=(1)x g
(p)x= g
€=
(2)x

(6X
= +p- 9S
Digital Signal Processing

9 9'1/ L=(9)x

vl+y-=(2)x

v-=(b)x
z=(4)x

9s9'1/
o 9 = (g)x

y!-y-=(9)Xx
T
=(€)
y x
Z S J bi i
i= i ; A

O O Q F : c 8 =(Z)x
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
354 Digital Signal Processing

X(0) = X,(0) + W2X,(0) + W2X (0)

X(1) = X,(1) + W3X,(1) + W2X,(1)
X(2) = X,(2) + W3X,(2) + W4X (2)
X(3) = X,(0) + W3X,(0) + W8X,(0)
X(4) = X\(11) + W$X 1) + W8X,(1)
X(5) = X,(2) + W8X.(2) + W2°X,(2)
X(6) = X(0) + W§X,(0) + W}?
X (0)
X(7) =X) + W3XQ(1) + WHEX,(1)
X(8) = X,(2) + W $X (2) + WX, (2)
Figure E6.28 shows the radix-3 decimation-in-time FFT flow
diagram for N = 9. Here, we have repeated the 3-point cat’s cradle
structure as we had repeated butterflies in the radix-2 case. The input
sequence appears in digit-reversed order.

Develop DIT FFT algorithms for decomposing the

DFT for N = 6 and draw the flow diagrams for (a) N = 2.3 and
(b) N = 3.2. (c) Also, by using the FFT algorithm developed in part (b);
evaluate the DFT values for x(n) = {1, 2, 3, 4, 5, 6}.
Solution
(a) For N = 6 = 2 . 3, where m; = 2 andN,=3 , Eq. 6.40 becomes
2 2
X(k)= F ax(2n) W3" + Y x(2n +) werrd*
n=0 n=0
2 2
= Yan) W3 +W Y x(n +1) W3
n=0 n=0

AlsoX;
,(k + 3) = X; (k)
2
X,(k) = F x(2n) W3"
n=0

= x(0) + x(2)W24+ x(4)W4t

X (0) = x(0) + x(2) + x(4)
X \(1) = x(0) WE + x(2)W2 + x(4)W§ = x(0) + x(2)W 3 + (DWG
X,(2) = x(0) WE + x(2)W4 + x(4)W§ = x(0) + x(2)W4 + x(4)W2
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
358 Digital Signal Processing

X(4) = X (0) + W§X,(0) + W2X5(0)

= 5 + (-0.5 + j0.866) (7) + (-0.5 - j0.866) (9)
=-3 - 71.732
X(5) = X,(1) + WX (1) + WX (D
= —3 + (0.5 + 70.866) (—3) + (-0.5 + j0.866) (-3)
=-3 { 1 + 0.5 + 0.866 —0.5 + 70.866 }
= -3{1+ 1.732}
=-3-75.196
The calculated values of DFT are also shown in Fig. E6.29(b).

x(0) =19-— OSs ps0 X(0)=21

x(3)=4° 2 X(1)=-3+/5.196

x(2)=3°

gs We
x(5)=6° PIY O AN X(5)=-3-/5.196
(5) wg %(1)=-3
We
Fig. E6.29(b) DIT FFT Flow Diagram for N = 6 = 3.2

|Example 6.30 |Develop the DIT FFT algorithm for decomposing the
DFT for N = 12 and draw the flow diagram.
Solution For N = 12 = 3 . 4, where m, = 3 and N, = 4, Eq. 6.40
becomes
. i ;
Xh) = Yx(8n) Wy" + Yx(Gn + DWG”
n=0 n=0
3
+ ¥x(n +2)wiyrt?*
n=0
3 3
= Y a(n) Wit + Wk Y x(8n+)W;*
n= 0 n=0
3
+ Wa J x(n +2)wy*
n=0

=X (k) + W$,
Xa(k) + WÈ
Xa(k)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
362 Digital Signal Processing

X(15) = X,(3) + W 1X3) + W 30X33) + W 1X408)

Figure E6.31 shows the radix-4 decimation-in-time FFT flow diagram
for N = 16.

x (0)=0 Qs aoe x(0)=8

x(4)=0 O w= N X)=-2.6112- 2.6112

x(8)=0 Xi2}=0

x(12)=0 p X(3)=0
x(1)=1 i> Xx(4)=0

zim ; SX RRA J1,0824

KLEF SER, X(8)=~ 1.0-896
>

E A>O x(6)=0
x@)=1 of
; LA2SANII A
LL I
ESS
SP
A NA
4
A SSBKEF Ce.
ASOs
aSLALA
x(13)=1 } SP A0
7
OY» RO VLN
cy ZSA
— PRY
K KY
BROW
OAA See
<>
x(6)=0 Vi SEVA
SA) CO
BE R S
NE x X(9)=2. +j2.6112
'9)=2.. 6112+/2
>
0 >» WX

x(10)=0 ; = KIENT x100

x(14)=0 E KIT SZ > SR i0
Y
x(3)=1 O AA LS <> NY X(12)=0

x(7)=1 = - Lh < SÐ X(13)=1.0848

+j1.0848

x{11)=1 rn ` RSD X(14)20

Fig. E6.31 Radix-4 DIT FFT Flow Diagram for N = 16

To determine the DFT of the given 16 — point sequence

x(n) = {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1}
X (k) = x(0) + (4) W$È + 2(8)WSk + (12)W 12"
=0+0+0+0=0
X a(k) = x(1) + x(5)W48 + x(9)W8% + x(13)W 33"
=141-j) + 1-1) + 1j) = -2j
Xg(k) = x(2) + x(6)W$k + x(10)W8 + x(14)W 124
=0
X,(0) = X,\(1) = X,(2) = X,(3) = 0
X(0)=14+1+1+1=4
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
366 Digital Signal Processing

(b) To develop DIF FFT algorithm for N = 2.3

5 1 3 5
Xik)= Vix(n) Wgt =J an) Wg + Vix Wh + Sanw
n=0 n=0 n=2 n=4

1
= Dlr) + x(n + 2) WH + x(n + 4) WS") wg
n=0

i
X(3k) = Y [x(n) + x(n + 2) + x(n + 4] WS"
n=0

1
X(3k + 1) = Z [zn +x(n+ 2)W2 +x(n+ a)wijwg wgn
n=0

1
X(3k + 2)= > [x(n)+ x(n+2)Wé +x(n+ 4) we ]we” wink
n=0

X(2)

x(5)
we we We
Fig. E6.32 (b) DIF FFT Flow Diagram for Decomposing the DFT for N = 6 = 2.3

fin) = x(n) + x(n + 2) + x(n + 4)

gin) = x(n) + x(n + 2)W2 + x(n + 4)W6

h(n) = x(n) + x(n + 2)W4 4 x(n + 4)W2
F(O) = x(0) + x(2) + x(4), FA) = x(1) + x(3) + x(5)

g(0) =x(0) + 2(2)W24+2(4)Wé, g(1) = x(1) + xt3)W3 + x(5)W4

h(0) = x(0) + x(2)W4 + x(4)W2, AD) = x(1) + x(3)W4 + x(5)W3
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
370 Digital Signal Processing

(d) An L-point inverse FFT is performed on the product sequence

obtained in step (c).
(e) The first (M — 1) IFFT values obtained in step (d) is overlapped
with the last (M — 1) IFFT values for the previous block. Then
addition is done to produce the final convolution output sequence
y(n).
(f) For the next data block, go to step (b).

An FIR digital filter has the unit impulse response

sequence, A(n) = {2, 2, 1}. Determine the output sequence in response
to the input sequence x(n) = {3, 0, -2, 0, 2, 1, 0, -2, -1, 0} using the
overlap-add convolution method.
Solution The impulse response h(n) has the length, M = 3 . The
length of the FFT/IFFT operation is selected as L = 2™ = 2°=8 .
Then,N=L-M+ 1=8-3+1=6, and the segmentation of the input
sequence with the required zero padding is given in Fig. E6.34(a).

1 [ees at a A Oe Freee, Om 10. FTL 126 1S ||

=e |
|

Fig. E6.34(a)
Steps (b), (c) and (d) are described below using the direct
implementation of circular convolution.
Circular convolution of data blocks x,(n) and x(n) with h(n)
padded with (N — 1), i.e. five zeros is given in Fig. E6.34(b) and (c).

CIEE
on Bye oe EE
pat (c)
Fig. E6.34 (b) and (c)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
374 Digital Signal Processing

r= DY x(n+l yin), 120, #1, 2 2,... (6.42)

where index/ is the (time) shift (or lag) parameter and the subscripts xy
on the cross-correlation sequence r,,(/) show the sequences being
correlated. The order of the subscripts, with x preceding y in Eq. 6.41,
indicates that x(n) is kept unshifted and y(n) is shifted by / units in
time, to the right for/ positive and to the left for /negative. Similarly, in
Eq. 6.42, y(n) is kept unshifted and x(n) is shifted by / units in time, to
the left for / positive and to the right for/ negative.
When the roles of x(n) and y(n) are reversed, the cross-correlation
sequence becomes

ryeD= DY yin)xn-1) (6.43)

n=~e

or, equivalently,

r= J, yn4+l) x(n) (6.44)

na-u
Comparing Eq. 6.41 with Eq. 6.44 or Eq. 6.42 with Eq. 6.43, we find
that
ry (l) =r pD) (6.45)
This means that r,,(/) is the folded version ofr ,,(l) , with respect to
l=0 . Therefore, r,,(/) gives exactly the same information as r yl).
Autocorrelation Sequences
When y(n) = x(n), the cross-correlation function becomes the auto-
correlation function. As a result, y(n) is replaced by x(n) in Eq. 6.41
and Eq. 6.42 gives the autocorrelation function, r,,(l), which is defined
as

re(= J$, x(n)x(n-1), 1=0,41, +2, ... (6.46)

or, equivalently,

reM=s YF x(n+l)x(n), 1=0,£1,+2,... (6.47)

nee
From the above equations, it is clear that the maximum
autocorrelation value occurs at/ = 0 because of an in-phase relationship
between the two sequences. As / increases, the autocorrelation value
increases.

Determine the cross-correlation values of the two

sequences x(n) = {1, 0, 0, 1} and h(n) = {4, 3, 2, 1).
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
378 Digital Signal Processing

6.38 Compute the DFTs of the following sequences, where N = 4 using

DIT algorithm.
(a) x(n) = 2", (b) x(n) = 27"

(c) x(n) == sin(

sin| z }
2* (d) x(n) = cos ( P )
na

Ans: (a) X(k) = {15, -3 + j6, -5, -3-j6}

156 3 .35 3,_ .33)
nai e Ie ge Ts
X(k) = (2, 2- -—,-,—+

(c) X(k) = (0, -j2, 0, j2)

(d) X(k) = (1, 1-jV2,1,1+jV2)
6.39 Draw the butterfly line diagram for 8-point FFT calculation
and briefly explain. Use decimation-in-time.
6.40 Find the DFT of the following sequence x(n) using DIT FFT.
x(n) = (1, -1, -1, -1, 1, 1, 1, -1)
Ans: (0, -V2 + j3.4142, 2-j2, J2 - j0.5858, 4, J2 + j0.5858,
24 j2,-J2 -j3.4142)
6.41 Compute the 16-point DFT of the sequence
x(n) = cos(x/2), 0 Sn <15 using DIT algorithm.
6.42 Find DFT (8-point) for a continuous time signal
x(t) = sin(2zft) with f= 100 Hz
6.43 Compute the DFof Tthe sequence x(n) =a", where N = 8 and a = 3.
6.44 Compute the FFT for the sequence x(n) = n? + 1 where N = 8
using DIT algorithm.
Ans: X(k) = 100 (1.48, -0.4686 + j0.7725, -0.24 + j0.32,
-0.2731 + j0.1325, -0.28, -0.2731 - j0.1325,
-0.24 — j0.32, -0.4686 - j0.7725)
6.45 Compute the FFT for the sequence x(n) =n + 1 whereN=8
using DIT algorithm.
Ans: X(k) = (86, -4 + j9.656, -4 + j4, -4 + j 1.6568, -4,
-4 - j1.6568, -4 - j4, -4 - j9.656)
6.46 Compute the DFT coefficients of a finite duration sequence (0, 1,
2, 3, 0, 0, 0, 0).
Ans: X(k) = (6, -V2 - j4.8284, -2 + j2, J2 - j0.8284, -2,
V2 +] 0.8284, -2 - j2, -J2 + j4.8284)
6.47 Draw the flow graph of an 8-point DIF FFT and explain.
6.48 Draw the butterfly line diagram for 8-point FFT calculation
and briefly explain. Use decimation-in-frequency.
6.49 Repeat Q.No.38 using DIF algorithm.
Ans: Same as in Q.No. 38.
6.50 Draw the butterfly diagram for 16-point FFT calculation and
briefly explain. Use decimation in frequency.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
382 Digital Signal Processing

The group delay is defined as the delayed response of the filter as a

function of œ to a signal.
Linear phase filters are those filters in which the phase delay and
group delay are constants, i.e. independent of frequency. Linear phase
filters are also called constant time delay filters. Let us obtain the
conditions FIR filters must satisfy in order to have constant phase and
group delays and hence obtain the conditions for having a linear phase.
For the phase response to be linear

SO or —-NSOS+T
o
Therefore,
O(o)=-
at
where tis a constant phase delay expressed in number of samples. Using
Eq. 7.2,

Dlo)
-1 Im
= tan! ———_-
H(e?”) = - wt
Re H(e’*)
or
M-1
»y h(n)sinan
-1 n=0
ot = tan) o
¥ h(n oson
n=0

or
M-1
2, h(n) sinon

tan t= fy
¥ h(n) cos wn
n=0

Simplifying, we get
M-1
¥ A(n) sin (ot - wn) =0 (7.3)
n=0

and a solution to Eq.7.3 is given by :

& (M -1)
T 3 (7.4)
7.4

and
h(n)
= A(M - 1- n) for
0 < n < M-1 (7.5)
If Eqs 7.4 and 7.5 are satisfied, then the FIR filter will have constant
phase and group delays and thus the phase of the filter will be linear.
The phase and group delays ofthe linear phase FIR filter are equa! and
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
386 Digital Signal Processing

where

HAf)= 5 hin)cos2rnfT (7.11)

and re

Hf) =5 h(n)sin2rnfT (7.12)

fae
From Eq. 7.11 and 7.12 we infer that H, (f) is an even function and
Hí f) is an odd function of frequency. If h(nT) is an even sequence, the
imaginary part of the transfer function, H{ f), will be zero and if h(nT)
is an odd sequence, the real part of the transfer function, H,( f), will be
zero. Thus an even unit impulse response yields a real transfer function
and an odd unit impulse response yields an imaginary transfer function.
A real transfer function has 0 or + x radians phase shift, while an
imaginary transfer function has + r/2 radians phase shift. Therefore, by
making the unit impulse response either even or odd, one can generate
a transfer function that is either real or imaginary.
In the design of digital filters two interesting situations are often
sought after.
(i) For filtering applications, the main interest is in the amplitude
response of the filter, where some portion of the input signal spectrum
is to be attenuated and some portion is to be passed to the output with
no attenuation. This should be accomplished without phase distortion.
Thus the amplitude response is realised by using only a real transfer
function. That is
He!®)=Hif)
and
Af) = 0
(ii) For filtering plus quadrature phase shift. the applications include
integrators, differentiators and Hilbert transform devices. For all these
applications the desired transfer function is imaginary. Thus, the
required amplitude response is realised by using only H;(f ). That is
He!®)=j HAF)
and
H,(f) =0
Design Equations
The term H(e/”) is periodic in the sampling frequency and hence both
Hf) and Hf) are also periodic in the sampling frequency. Both H,(/)
and H,( f ) can be expanded in a Fourier series. Since the real part of the
transfer function, H,(f), is an even function of frequency, its Fourier
series will be of the form

Hf) =ao+ >, a, cos (2nnfT) (7.13)

n=1
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
390 Digital Signal Processing

type-I design and type-II design. In the Type-I design, the set of
frequency samples includes the sample at frequency œ = 0. In some
cases, it may be desirable to omit the sample at œ = 0 and use some
other set of samples. Such a design procedure is referred to as the Type-
II design.
Type-I Design
The samples are taken at the frequency
27k
mAr =0,1,...,M-1
Ok M’ (7.22 )
The samples of the desired frequency response at these frequencies
are given by

A (k)=Ha (t)a REOL, M1

M),
= H;(e?™* k=0,1,... M-1 (7.23)
This set of points can be considered as DFT samples, then the filter
coefficients h(n) can be ae using the IDFT,

h(n) = Š p H(k) e/?*"*/M n=0,1,..,M-1 (7.24)

k=0

If these numbers are all reai, then these can be considered as the
impulse response coefficients of an FIR filter. This can happen when all
the complex terms appear in complex conjugate pairs, and then all the
terms can be matched by comparing the exponentials. The term Hk)
e/?""*/M should be matched with the term that has the exponential
e/?xnk/M as a factor. The matching terms are then H (k) e/?*"*/™ and
H (M — k) e/?*"(M-0M since 2nn (M-k/M = 2nn — (2nnk/M). These
terms are complex conjugates if H (0) is real and
(i) For M odd:
H (M-k) = H“(k), k = 1, 2,...,(M-1)/2 (7.25)
(ii) For M even:
H(M-k) = Ë (k), k= 1, 2,..., M/2-1 (7.26)
H(M/2) =
The desired frequency response H; (e/®) is chosen such that it satisfies
the Eqs 7.25 and 7.26 for M odd or even, respectively. The filter
coefficients can then be written as

A(n) = paoa" A"nof wen], M odd (7.27)

M/2-1
and A(n) - āo 2 Re[A teen Meven (7.28)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Finite Impulse Response (FIR) Filters 397

7.4.3.3 Hanning Window Function

The window function of a causal Hanning window is given by

as 2nn
p OSnsM-1 (7.44)
0, otherwise
The window function of a non-causal Hanning window is expressed by

0.5 + 0.5 cosSE" 0<|nj<4—*

w Hann() = | -1
0, otherwise
The width of the main lobe is approximately 8x/M and the peak of the
first side lobe is at -32dB.
7.4.3.4 Blackman Window Function
The window function of a causal Blackman window is expressed by
2rn 4nn
z 0.42 - 0.5 ——— + 0.08 , OsnsM-1
wm= cosi sH
0, otherwise
The window function of a non-causal Blackman window is given by
2nrn 4nn M-1
0.42 + 0.5 cos + 0.08 cos r for |n|<
waln) = | 0,
M=1 M-i otherwise
2 (1.45)
The width of the main lobe is appreximately 12x/M and the peak of the
first side-lobe is at —58dB.
7.4.3.5 Bartlett Window Function
The window function of a non-causal Bartlett window is expressed by

l-n, l<n<

Table 7.1 gives the important frequency-domain characteristics of some

window functions.

Table 7.1 Frequency-Domain Characteristics of Some Window Functions

Type of Window Approximate Minimum Stopband Peak of first
Transition Width Attenuation Sidelobe
of Main Lobe (dB) (dB)
Rectangular 4n/M -21 -13
Bartlett 8n/ M -25 -27
Hanning 8x/M 44 -32
Hamming 8x/M -53 - 43
Blackman 12n/M -74 -58
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Finite Impulse Response (FIR) Filters 401

Solution The filter coefficients are given by

1 R 7 i 1 3x/4 , :
hg(n) =— Ha (ei?) eo” do-— fee ei” do
2n ia 2T sais
sin 3n (n — 3)/4 3
h a(n) = x(n-3)
=, n +33 and and h,(8)
3)= =4

The filter coefficients are,

ha (0) = 0.0750, hg (1) = — 0.1592, hy (2) = 0.2251, hg (3) = 0.75
h4 (4) = 0.2251, hg (5) = — 0.1592, h 4 (6) = 0.0750
The Hamming window function is,
2rn
bie 0.54 — 0.46 cos 7—1’ O<nsM-1

0, otherwise
Therefore, with M = 7,
w(0) = 0.08, w(1) = 0.31, w(2) = 0.77, w(3) = 1,w(4) = 0.77,
w(5) = 0.31, w(6) = 0.08.
The filter coefficients of the resultant filter are then,
h(n) =hy(n).w(n) n=0, 1, 2,3, 4, 5, 6.
Therefore,
h(0) = 0.006, A(1) = — 0.0494, h(2) = 0.1733, h(3) = 0.75,
h(4) = 0.1733, h(5) =- 0.0494 and h(6) = 0.006.
The frequency response is given by
6
H(e/®) = YAM) ejen
n=0

=e [ h(3) + 2h(0) cos3

œ + 2h (1) cos2 œ + 2h(2) cos œ)
= e/3 [0.75 + 0.3466 cos w — 0.0988 cos2 w + 0.012 cos œ)

TZA Design an FIR digital filter to approximate an ideal

ow-pass filter with passband gain of unity, cut-off frequency of 850
Hz and working at a sampling frequency of f, = 5000 Hz. The length
of the impulse response should be 5. Use a rectangular window.
Solution The desired response of the ideal low-pass filter is given by
1, 0< f <850 Hz
Haee]
0, f > 850 Hz
The above response can be equivalently specified in terms of the
normalised w,. The normalised œ, = 2nf,/ f, = 27 (850)/(5000) = 1.068
rad/sec. Hence, the desired response is
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
406 Digital Signal Processing

High-Pass FIR Filter

-2E ERLE fain ot
hana] SE) Bante (7.65)
1-—< forn =0

where
f,=0.5(f,+f,) and AF=f,-f, (7.66)
Bandpass FIR Filter

= [sin(2nn f2/F)- sin(2rn f /P)} for n>0

han) = 2 (7.67)
pfe = fab forn =0

where
AF AF
fa=fa zy fee =fp2+ KE

AF, =fpr fey AF, =fs2-fp2 (7.68)

AF = min [A F; AF,]
Bandstop FIR Filter

— [sin(2xnf.1/F) -sin(2xnf.9/F)|, forn>0

hyg(n)= (7.69)
Shes - fe2)+1, for n =0

where
AF AF
fer =fp1+ => fee ie

AF, =f, fp1 AF, =fp2-fs2 (7.70)

AF = min [AF, A F,]

Design a low-pass digital FIR filter using Kaiser

window satisfy? 1g the specifications given below.
Passband cut-off frequency, f, = 150 Hz, stopband cut-off
frequency, f, = 250 Hz, passband ripple, A, = 0.1 dB, stopband
attenuation, A, = 40 dB and sampling frequency, F = 1000 Hz.
Solution A computer program can be written for the design of
Kaiser window digital filter using the functions given in Appendix.
The computer output is given below.
From Eq. 7.53, 5 = 0.005756.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
410 Digital Signal Processing

There are four different cases that result in a linear phase FIR filter,
viz., (i) symmetric unit impulse response and the length of the filter, M
odd, (ii) symmetric unit impulse response and M even, (iii) anti-
symmetric unit impulse response and M odd, and (iv) antisymmetric
unit impulse response and M even. The first case is discussed below in
detail and other cases are listed in Table 7.2.
In the symmetric unit impulse response case, h(n) = H(M - 1- n). The
real-valued frequency response characteristics |H(e/)| = |H,(e/®)|,
given in Eq. 7.14, is
(M -3)

|H(es®| = (2 +2 $ hin)c08o(# t-n)

£ = 2 -/
(7.74)
n=0

Let k = (M - 1)/2 —n. Then Eq. 7.74 can be written as

(M -1)
, 2
|He®)| = Y a(k)cos wk (7.75)
k=0

where

a(0)=h (==) (7.76)

alk) = 2h(4—t- a)for 1 <k s W
The magnitude response for the other cases are similarly converted
to a compact form as given in Table 7.2.

Table 7.2 Magnitude Response Functions for Linear Phase FIR Filters

Case (i) - Symmetric and M odd “Faw cos WR

A(ny=h(M-1-n)

Case (ii) -Symmetric and M even “Sow coawk

A(n) = h(M-1-n)

Case (iii) -Antisymmetric and M odd (M -3V2

hin)
=— (M—1-n) Arao
E m
Case (iv) -Antisymmetric and M even ya(Ayank
hin)=-h(M-1-n)

From Table 7.2, it can be seen that the magnitude response function
can be written as given in Eq. 7.77, for the four different cases.
|H(e/®)| = Q(@) Po) (7.77)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
416 Digital Signal Processing

7.28 What is an FIR half-band digital filter ? Explain with a suitable

illustration.
7.29 What is an optimal linear phase FIR filter ? What parameters
are optimised in these filters ?
7.30 State and explain the alternation theorem.
7.31 What are extra ripple filters ?
7.32 What are maximal ripple filters ?
.33. Explain the Remez exchange algorithm used in the design of
optimal filters.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
420 Digital Signal Processing

fa tm(z)

Re(z)

Unit Circle

s-plane
Fig. 8.2 The Mapping of Eq. 8.5 into the z-plane

It can be seen that the mapping of Eq. 8.5 takes the left-half plane
of s-domain into the corresponding points inside the circle of radius 0.5
and centre at z = 0.5, and the right-half of the s-plane is mapped outside
the unit circle. As a result, this mapping results in a stable analog filter
transformed into a stable digital filter; however, as the locations of poles
in the z-domain are confined to smaller frequencies, this design method
can be used only for transforming analog low-pass filters and bandpass
filters having smaller resonant frequencies. Neither a high-pass filter
nor a band reject filter can be realised using this technique.
The forward difference can be substituted for the derivative instead
of the backward difference. This gives,
dy) _ y(nT+T)-y(nT)
dt T
vad va) (8.12)
The transformation formula will be
z-1
s = -—
T (8.13 )

or,
z=1l+sT (8.14)
The mapping of Eq. 8.14 is shown in Fig. 8.3. This results in a worse
situation than the backward difference substitution for the derivative.
When s = j Q, the mapping of these points in the s-domain results in a
straight line in the z-domain with coordinates (Zreai Zimag) = (1,27).
Consequently, stable analog filters do not always map into stable digital
filters.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Infinite Impulse Response (IIR) Filters 425
jQ Im Z

Unit Circle

Fig. 8.4 The Mapping of z = e!

| Example 8.4|Convert the analog filter into a digital filter whose

system function is
$+0.2
H(s)=
(s + 0.2)? +9
Use the impulse invariant technique. Assume T = 1s.
Solution The system response of the analog filter is of the standard
form
sta
H(s)=
(s +a)? +b?
where a = 0.2 and b = 3. The system response of the digital filter can
be obtained using Eq. 8.27.

Hie) = 1-e7*" (cosb T) z~!

Buel
km ania oar 2
1-2e°* (cosbT) 2-1 +e?" z7?
_ 1-e°9?7 (cog3T) 2}
=
1-2¢°°?7
-0.27
(cos3T) 27! + e794? z7?
- -0. -

Taking T = 1s,

H(z)= u z
1- (0.8187) ( 220) z -1 =
1- 2(0.8187) (- 0.99) z~- + 0.6703 z
That is,

H(@)= 1+ (0.8105) z -1
1+ 16210 z~} + 0.6703 z~?
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
430 Digital Signal Processing

[Hw | 8 ô

|H(jQ)|A 4p

Q, a
Fig. 8.5 Relationship Between @ and Q as Given in Eq. 8.40

The sampling period is obtained from the above equation using

= tan % =2 tan £ =0.2768

Q. 2 3 8
Using bilinear transformation,
H(z) = H (s)|, 22-0
(z+1)

2zn +0.1
H(z) = a O
[2 e—»
T (z+)
+oa ] +9
_ __(2/T) (2-1) (z+) +0-1(2+ 1)?
[(2/T) (z - 1) + 0.1(z+ D? +9 (z + 1)?
Substituting T = 0.276 s,

1+ 0.027 z~! - 0.973 z~?

H( )= Pa, a es
8.572 — 1184 27° + 8.177 z

|Example 8.8| Apply bilinear transformation to

a ee
He)= leek
with T =0.1s.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
438 Digital Signal Processing

Step (ii) Determination of the order of the filter.

From Eq. 8.46,

y > 1 8 {10/83 - 1/85 - DI}

2 log (Q3/Q,)
log {24/0.2346}
_ =1 ———_— = 2.62
52 log (2.414) :
Let N = 3.
Step (iii) Determination of — 3 dB cut-off frequency.
From Eq. 8.47,

TEE Q E A2 ETy
[cavaz) - 1°" [v09 - 3]
Step (iv) Determination of H, (s).
From Eq. 8.49,
His)= BQ, (N-1/2 B, Q?

8+09Q., py 8° +b S+ Q?
-( By Q. | B, Q2 )

$+)Q, )\ 8? +b,Q.8+0,Q?
From Eq. 8.50,

b,=2sin = =1, Co=1 cy=1

Bo B, = 1. Therefore By = B; = 1.
Therefore,

H)=( 2.5467 )l 6.4857 )

s + 2.5467 }\ s* + 2.5467s+ 6.4857
Step (v) Determination of H (z).
H(z)= H(s)|, seit 1)
+1)

That is,

2.5467 6.4857
Mers GED (z-1)
RE
zg ar [2auf
- + 2.54675 + 6.4857
Simplifying we get,

H(z) = — 16.5171
a (z + 1)
70.832° + 31.1205z? + 27.2351z + 2.948
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Infinite Impulse Response (IIR) Filters 445

e? Ch (Q,/Q)
<8, 222, (8.63b)
1+e? CÈ (Q,/9)
When Q = Q, Eq. 8.63b becomes

a= —
2 +e?
Rearranging,
öz
£= Gna (8.64)

When Q = Q., Eq. 8.63a becomes,

08 = e? Cz (Q,/Q,)
1+e? CR (Q2/Q,)
0.5 + 0.5 £? C2 (Q,/2,) =e? CZ (Q,/Q,)
or, simplifying

Cy (Q4/2,) = > (8.65)

Using Eq. 8.53,

cosh [N cosh! (2,/2,)] = 3 (8.66)

From Eqs. 8.66 and 8.64 we can get the order of the filter, N.
0.5
cosh”! È - 1|
cosh™?! (1/£) __ 33
N= (8.67)
cosh”! (Q,/Q,) cosh™? (Q4/Q,)
The value of N is chosen to be the nearest integer greater than the
value given by Eq. 8.67.

8.8 ELLIPTIC FILTERS

The elliptic filter is sometimes called the Cauer filter. This filter has
equiripple passband and stopband. Among the filter types discussed so
far, for a given filter order, passband and stopband deviations, elliptic
filters have the minimum transition bandwidth. The magnitude
response of an odd ordered elliptic filter is shown in Fig. 8.9. The
magnitude squared response is given by
1
|H(jQ)|? = 1+e?Uy (Q/2Q,) (8.68)
where Uy(x) is the Jacobian elliptic function of order N and e is a
constant related to the passband ripple.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Infinite Impulse Response (IIR) Filters 451

8.12 An analog filter has the following system function. Convert this
filter into a digital filter using the impulse invariant technique.
1
H(s) = —~~
ý (s+0.1)7 +9
8.13 Convert the analog filter to a digital filter whose system
function is

H(s) = ——_,
D= rap
8.14 Convert the analog filter to a digital filter whose system
function is

H(s) =(s+0.1)?
-—*8 +36
The digital filter should have a resonant frequency of œ, = 0.2 n.
Use impulse invariant mapping.
8.15 What is bilinear transformation ?
8.16 Compare bilinear transformation with other transformations
based on their stability.
8.17 Obtain the transformation formula for the bilinear
transformation.
8.18 An analog filter has the following system function. Convert this
filter into a digital filter using bilinear transformation.
1
Hi= yaa) +6
8.19 Convert the analog filter to a digital filter whose system
function is
r 1
H(s)
~ (s+2)? (s+1)
using bilinear transformation.
8.20 Convert the analog filter to a digital filter whose system
function is

H(s) =
36
(s +0.1)? +36
The digital filter should have a resonant frequency of œ, = 0.27.
Use bilinear transformation.
8.21 What is meant by frequency warping ? What is the cause of this
effect ?
8.22 Describe Butterworth filters ?
8.23 Comment on the passband and stopband characteristics of
Butterworth filters.
8.24 Describe Chebyshev filters ?
8.25 Describe inverse Chebyshev filters ?
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Realisation of Digital Linear Systems 455

For example, the difference equation of the first-order digital system

may be written as
y(n) =a, y(n - 1) + x(n) + b,x (n - 1)
The basic realisation block diagram for this equation and the
corresponding structure of the signal flow graph are shown in Figs 9.2
(a) and (b). Here, it is clear that there is direct correspondence between
branches in the digital realisation structure and branches in the signal
flow graph. But in the signal flow graph, the nodes represent both
branch points and adders in the digital realisation block diagram.
Source Sink
x(n) y(n) moder 1 2 3 node
+ dae o- > —> ——9
°

A7
z
|
i
P
Xz) )
gkbrranchanthy y z z by
noor
Y(z)

ay
a b |
4
(a) (b)
Fig. 9.2 (a) Basic Realisation Block Diagram Representing a First-order
Digital System and
(b) Its Corresponding Signal Flow Graph.

Advantages of representing the digital system in block diagram

form
(i) Just by inspection, the computation algorithm can be easily
written
(ii) The hardware requirements can be easily determined
(iii) A variety of equivalent block diagram representations can be
easily developed from the transfer function
(iv) The relationship between the output and the input can be
determined.
9.2.1 Canonic and Non-Canonic Structures
If the number of delays in the realisation block diagram is equal to the
order of the difference equation or the order of the transfer function of a
digital filter, then the realisation structure is called canonic.
Otherwise, it is a non-canonic structure.

9.3 BASIC STRUCTURES FOR IIR SYSTEMS

Causal IIR systems are characterised by the constant coefficient
difference equation of Eq. 9.1 or equivalently, by the real rational
transfer function of Eq. 9.2. From these equations, it can be seen that
the realisation of infinite duration impulse response (IIR) systems
involves a recursive computational algorithm. In this section, the most
important filter structures namely direct Forms I and II, cascade and
parallel realisations for IIR systems are discussed.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
460 Digital Signal Processing

The direct Form II realisation requires only the larger of M or N

storage elements. When compared to direct Form I realisation, the
direct Form II uses the minimum number of storage elements and hence
said to be a canonic structure. However, when the addition is performed
sequentially, the direct Form II needs two adders instead of one adder
required for the direct form I.
The Direct Form II realisation network structures are shown in
Figs 9.6 and 9.7.
Though the direct Forms I and II are commonly employed, they have
two drawbacks, viz. (i) they lack hardware flexibility and (ii) due to finite
precision arithmetic, as to be discussed in Chapter ten, the sensitivity of
the coefficients to quantisation effects increases with the order of the
filter. This sensitivity may change the coefficient values and hence the
frequency response, thereby causing the filter to become unstable. To
overcome these effects, the cascade and parallel realisations can be
implemented.

|Example 9.2|Determine the direct Forms I and II realisations for a

third-order IIR transfer function.

0.282” + 0.319z+ 0.04

A): — > nn
0.52" +0.32° +0.17z-0.2
Solution Multiplying the transfer function numerator and
denominator by 2z`?, we obtain the standard form of the transfer
function.

0.56 27! + 0.63827? + 0.08273

A(z)=
1+ 0.627! + 0.3427 -0.427%
The direct Forms I and II realisations of the above transfer function
are shown in Figs E 9.2(a) and (b) respectively.

(a)
Fig. E9.2
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
464 Digital Signal Processing
The cascade realisation of the system transfer function is shown
in Fig. E9.4.

9.3.3 Parallel Realisation of IIR Systems

By using the partial fraction expansion, the transfer function of an IIR
system can be realised in a parallel form. A partial fraction expansion of
the transfer function in the form given below will lead to the parallel
form.

Fig. 9.9 Parallel Form Realisation Structure With the Real and Complex
Poles Grouped in Pairs
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
472 Digital Signal Processing

ETEN Obtain the cascade and parallel realisations for the

system function given by

Hia) = ———___4+—___~
(1+4z Tra +r
$ *)( i j 3 a

Solution
Cascade Realisation To obtain the cascade realisation, the
transfer function is broken into a product of two functions as
H(z) = H,(z) H(z)

14427 1
where H,(z)= —— and H,(z)= te
1+=2z7! 1+=2724+=27
2 2 4
The cascade realisation structure for this system function is shown in
Fig. E9.8(a).
x(n)

x(n)

Fig. E9.8(b)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
498 Digital Signal Processing

The largest error occurs when all the discarded bits are one’s. When
the number x is negative, truncation results in reduction of the
magnitude only. However, because of the negative sign, the resulting
number will be greater than the original number. For example, let the
number be x = — 0.375. That is, in sign magnitude form it is represented
asx = 1011 and after truncation of one bit, Q(x) = 1 01. This is equivalent
to — 0.25 in decimal. But — 0.25 is greater than — 0.375. Therefore, the
truncation error is positive and its range is
O<ep (28-274) (10.3)
The overall range of the truncation error for the sign magnitude
representation is
— (2-8 -2°*) Seps (27-2) (10.4)
(ii) Truncation error for two’s complement representation When the
input number is positive, truncation results in a smaller number, as in
the case of sign magnitude numbers. Hence, the truncation error is
negative and its range is same as that given in Eq. 10.2. If the number is
negative, truncation of the number in two’s complement form results in
a smaller number and the error is negative. Thus the complete range of
the truncation error for the two’s complement representation is
-(2%-2+")<e,<0 (10.5)
äi) Round-off error for sign magnitude and two’s complement
representation The rounding of a binary number involves only the
magnitude of the number and is independent of the type of fixed-point
binary representation. The error due to rounding may be either positive

or negative and the peak value

.is —{
(23 -9-4) The round-off error
,is
symmetric about zero and its range is
-B -L -B _o-L
88) e ge a (10.6)
2 2
where epg is the round-off error.
4 Q(x) Q(x)

(a) Rounding (b) Truncation in 2's (c) Truncation in

complement sign magnitude

Fig. 10.1 Quantization Error in Rounding and Truncation

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Effects of Finite Word Length in Digital Filters 503

where E represents the statistical expectation. Using Eq. 10.18 in

Eq. 10.19, we get

koe (Mm) = E 5 h(k) e* (n

— k)- 5 h(k) e(n+m-—k)
k=0 k=0

= J, AP(kh) E [e* (n-k) -e(n + m-b)]

k=0

Yoo eo (M) = X, h? (k) Yee (M) (10.20)

k=0

It has been assumed that the noise resulting from the quantisation
process is a white noise. For this case, we have
Voe (M)=02, and %e(m)=02 (10.21)
where oĉ, is the output noise power (or power of the output error) and
6? is the input noise power. Using Eq. 10.21 in Eq. 10.20 and replacing
the variable & with n,

o2,=02 È h?n) (10.22)

n=0

Using Parseval’s relation (see Example 10.1),

Y= f HOH ez dz
n=0 2n J Cc

in Eq. 10.22 we get,

2
22
Seo = 2 §
Ty c
He Hez dz (10.23)
where the closed contour of integration is around the unit circle |z| = 1.
This integration is evaluated using the method of residues, taking only
the poles that lie inside the unit circle.

| Example 10.1) Prove that

5 x° (n) = -+4 X (z) X (z)271 dz

HHO 2nJ c

Solution The z-transform of x (n) is

X@)= $ xmz” (E1)

n=0
Taking the z-transform of x(n),

Zim = F x(n) x(n)2z" = F enz” (E2)

n=0 n=0
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
508 Digital Signal Processing

10.6 COEFFICIENT QUANTISATION IN DIRECT FORM

REALISATION OF FIR FILTERS
The statistical bounds on the error in the frequency response due to
coefficient quantisation (rounding) is given here. The frequency
response of a linear phase FIR filter is given by
(M-3)/2 |(2> t-n)o}}
Banaran (2 1+2 £ h(n) cos
n=0
=e /% Mo) (10.33)
For a linear phase FIR filter, h (n) =h (M — 1 — n). The term e/™ =
e/0(M~- 12 in the above expression represents the delay and is unaffected
by quantisation. Hence, the quantisation effect is solely on the
pseudomagnitude term M (œ). Let {A,(n)} be the sequence resulting from
rounding [A(n)} to a quantisation step size of 2°. Therefore,
A,(n)= h(n) + e(n) (10.34)
and h,(n) = h,(M -1-n) for 0 <n < (M - 1)/2. Let e(n) be a random
~B a
sequence and uniformly distributed over the range -4 and at Let

H, (z) be the z-transform of (h, (n)} and M, (œ) be the pseudomagnitude

of the quantised linear phase FIR filter. The error function is defined to
be
E (e/®) = M (0) - M (0) (10.35)
(M -3)/2
or B(e*)=e(4=*)+2 2 e(n) cos(= -n)o]
2 nat 2

where E (e/“) is the frequency response of a linear phase FIR filter that
has {e(7)} as the impulse response for the first half and the second half
can be obtained using e(n) = e(M — 1 —- n). Thus, the filter with its
coefficients rounded-off can be considered as a parallel connection of the
ideal filter (infinite precision) with a filter whose frequency response is
E (e/%) e12 ™ -2 Since the error e (n) due to rounding of the filter
-B :

coefficients is always lesser than or equal to 2 a bound on |E (e/®)|

can be obtained as shown below.

|E (e/®)| <
(E +2
E emils (4 -n)a]
J le(n)|
n=0
(10.36)

Letting 7 =<. n = k in the second term of the above expression,

Eag
[E jjj (e7®)|
< sJe 7 (=> +a) |cosk œ|
k=1
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
512 Digital Signal Processing

0.5
ls (0 <bz< 1) (10.48)
1-Jbl

OVERFLOW LIMIT CYCLES

Limit cycles can also occur due to overflow in digital filters implemented
with finite precision arithmetic. The amplitudes of such overflow
oscillations are much more serious in nature than zero input limit cycle
oscillations. Consider a causal, all-pole second-order IIR digital filter
implemented using two’s complement arithmetic with a rounding of the
sum of the products by a single quantiser. The difference equation
describing the system is given by
F (n)=Qr l-a ¥ (n-1)-a, ¥ (n - 2) + x (n))
where Qp [.] represents the rounding operation and F (n) is the actual
output of the filter. The filter coefficients are represented by signed
4-bit fractions. Let a, = 1, 0 0 1 = — 0.8751, and a, = 0, 111 = 0.875, and
the initial conditions be 7 (- 1) = 0 , 1 1 0 = 0.75 ¿and ¥ (-2)=1,010
=- 0.754. For zero input, i.e. x (n) = 0 and for n 2 0, we get the values for
y (n) as shown below.

Table 10.3
Fin-D y (n-2 a, Y (n~ 1)-az 9 n-2)| F (M=Qell| Fn)
in decimal
¥ (-1)20,110 | F (-2)= 1,010 1,010100
¥(O)=1,011 | ¥ (-1)=0,110 10,110011
y (1)=0,110 | ¥ ()=1,011 1,001101
ï (2)=1,010 | ¥ (1)=0,110 10,101100
J (3)=0,110 | F (2)=1,010 14010100

For n = 1, the sum of two products has resulted in a carry bit to the
left of the sign bit that is automatically lost, resulting in a positive
number. The same thing happens forn = 3 and also for other values ofn.
It can be noted from the above table that the output swings between
positive and negative values and the swing of oscillations is also large.
Such limit cycles are referred to as overflow limit cycle oscillations.
The study of limit cycles is important for two reasons. In a
communication environment, when no signal is transmitted, limit cycles
can occur which are extremely undesirable. For example, in a telephone
no one would like to hear unwanted noise when no signal is put in from
the other end. Consequently, when digital filters are used in telephone
exchanges, care must be taken regarding this problem. The: second
reason for studying limit cycles is that this effect can be effectively used
in digital waveform generators. By producing desirable limit cycles in a
reliable manner, these limit cycles can be used as a source in digital
signal processing.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Multirate Digital Signal Processing 525

rs
|
t=nT

x(t) x(n)

x(t)

0 t

s(t)

sell CRA DR G
J [i PA
T O T 2T 3T 4T 5T 6T t

x(nT)

-T 0 T 2T 37 47 ST 6T nT

Fig. 11.2 Periodic Sampling of x(t)

11.3 SAMPLING RATE CONVERSION

Sampling rate conversion is the process of converting the sequence x(n)
which is got from sampling the continuous time signal x(t) with a period
T, to another sequence y(k) obtained from sampling x(t) with a period T’.
The new sequence y(k) can be obtained by first reconstructing the
original signal x(t) from the sequence x(n) and then sampling the
reconstructed signal with a period T”.
Figure 11.3 shows the reconstruction of the original signal with a
D/A converter, low-pass filter and resampler with sampling period T”.

s(t) x(0 yt) á

Ideal D/A Low-pass
converter filter
Fig. 11.3 Conversion of a Sequence x(n) to Another Sequence y(k)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
532 Digital Signal Processing

F F=LF F
x(n) re | x(t) yim)
Sampling rate
expander/
up sampler

ix(e!®)j

2r e!

Iwe”)

0 nil x 2r w
[Y (01)

T
0 niL x 2n a

Fig. 11.7 Interpolation of x(n) by a Factor L

Sadie Pa n = multiples of M
0, otherwise
where M = 2.
x(n)
=0, 1, 2, 3, 4, 5,...
y(n)
= 0, 0, 1, 0, 2, 0, 3, 0, 4, 0, 5, 0,...
In general, to obtain the expanded signal y(n) by a factor M,
(M — 1) zeros are inserted between the samples of the original signal
x(n).
The z-transform of the expanded signal is
Y) =X(z™),
M=2.
The input and output signals are shown in Fig. E 11.2.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
536 Digital Signal Processing

(f) Sampling rate expander

/up sampler

te y(m) = (x(mL),
0 otherwise

(g) Modulator

x(n) y(n) = x(n) s(n)

s(n)
Fig. 11.9 Branch Operations in Signal Flow Graphs

node of the branch. External signals enter the input branches and
signals at the output branches are terminal signals. The sum of the
signals entering the node is equal to the sum of the signals leaving the
node. Based on the signal flow graph of Fig. 11.10, the network
equations can be written as follows.
At the input node,
r(n) = x(n) + a,r(n — 1) + agr(n — 2) (11.21)
At the output node,
y(n) = r(n) + birin — 1) + barin - 2) (11.22)
Combining both the equations,
y(n) = x(n) + byx(n — 1) +b, x(n — 2) + a yin — 1) + aQy(n—-2) (11.23)

r(n)

x(n) y(n)

r(n-2)

Fig. 11.10 Signal Flow Graph for a Second-order System

11.4.1 Manipulation of Signal Flow Graphs

Manipulation of signal flow graphs which is shown in Fig. 11.11,
corresponds to the ways how the set of network equations are
represented. In multirate systems, it is easier to modify the.signal flow
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
540 Digital Signal Processing

With this property, the number of multiplications can be reduced by a

factor of two. If N is even,
Ny
y(n) = Saw fx (n-k) + x (n-(N-1-k))}
k=0
11.5.2 WR Direct Form Structure
Consider an IIR filter with the difference equation represented by,
D N-1
yin)= Ý, agy(n—k)+ J, Oyx(n- hk) (11.25)
k=l k=0
Figure 11.13 shows the signal flow graph for the IIR filter. The
system equation of the IIR filter is given by
N-1

(11.26)

= 4 E >
an-1 by

Fig. 11.13 The Signal Flow Graph for an IIR Filter

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Multirate Digital Signal Processing 545

Solution
H(z) = Ey(z*) +27) Elz”)
where Ey (2°), E (Z 2) are polyphase components.
_ 1-427
H(z)
1+5271
(1-427!) (1-527)
© (14527?) (1-527)
_ 1-927! +202?
1- 2527
1+2027 ., -9
so tz? CO
1- 2527? 1- 2527?
The polyphase components are

E (z) =
1+ 202%
1-2572 and E2) = ia.
1- 2527?
11.6.1 General Polyphase Framework
The z-transform of an anti-aliasing filter shown in Fig. 11.17 (a) with
impulse response A(n) is given by

x(n) ol hin) PE > jm | aa _

Decimation
filter

Fig. 11.17 (a) Decimation by a factor M

H(z) = x h(n) 2”
n=0
= A(0) + A(1) 271+ h(2)277 +...
which can be partitioned into M sub-signals where M represents
decimation factor. Hence,
H(z) = A(O) + A(M) z~™ + hM) 2-2 +...
+z AC) + AM + 1) 27! 4+ (QM +1277 +...)
+27 Dih(M-1)+h(Q2M-1)z-™ +...} (11.32)
Equation 11.32 can be written as
M-1 ~
He)= > F himM + ky zr
k=0 m=0
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
550 Digital Signal Processing

Each multiplier gets 772 sec

to perform one multiplication

x(n) 4
i
i

~ r} ” | m |e á
Fig. 11.18(e)
Polyphase
inplementation
x(n) — > te} ARoz?) |

y(n)

y(n)

Each multiplier gets T sec to

perform one multiplication
Fig. 11.18(f) Polyphase implementation of an interpolator by a Factor of two

IIR Structures for Decimators

The IIR filter is represented by the difference equation,
D N-1
y(n)= Yay(n-k)+ $ bxin-k) (11.34)
k=1 k=0
The system function for the above difference equation is given by,
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Multirate Digital Signal Processing 557

The demerits of the systems are that proper control structure is

required in implementing the system and proper values of J should be
chosen.

Implement a two-stage decimator for the following

specifications.
Sampling rate of the input signal = 20,000 Hz
M = 100
Passband = 0 to 40 Hz
Transition band = 40 to 50 Hz
Passband ripple = 0.01
Stop band ripple = 0.002
Solution
x(n) yim)
h(n) 100
20,000 Hz ' 200 Hz
LPF
IH)

ä ao]
0 40 50

a LPF 1
[ioA ni|
400 Hz LPF 2

lA) |LPF 1 HAI= LPF2

f(Hz) E AE | f(Hz)
w El ais areal and Two-stage Network he Sones

The implementation of the system is shown in Fig. E11.5(a).

F, = 40 Hz
F,= 50 Hz
5, = 0.01
5, = 0.002
Fr = 20 KHz
M = 100
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Multirate Digital Signal Processing 563

eW | T ~ him) || fh=5f
> y(m)

Hie]
|H(e?°) |
A Passband Stopband
x ar, 2 Fs,
> je — m

—- — n and

Fp N N
f 2h,
Fig. 11.28 The Comb Filter for N = 5

The number of multiplications per second is given by

Ru om 12 x 2000 = 12,000
Ry, p = 30x 20% = 60,000
The overall number of multiplications per second for a three-stage
realisation is given by
Ru, c + Ryu,s + Ru, p= 1,25,500
The number of multiplications per second for a three-stage realisation
is more than that of a two-stage realisation. Hence higher than two-
stage realisation may not lead to an efficient realisation.
Comb Filters
The impulse response of a comb filter (FIR filter) is given by,
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Multirate Digital Signal Processing 581

H(z) 14 t4 Gy(z)

c ta G,(z) y(n)

t2 G,(z)

Fig. 11.41 (c) Equivalent Three-channel QMF Filter Bank Realization

On similar lines, we can derive a four-channel QMF filter bank from

a three-channel QMF filter bank. These structures come under the class
of non-uniform QMF filter banks. These types of filters find application
in speech and image coding.

REVIEW QUESTIONS
11.1 What is the need for multirate signal processing ?
11.2 Give some examples of multirate digital systems.
11.3 Explain the interpolation process with an example.
11.4 Explain the decimation process with an example.
11.5 Write the input-output relationship for a decimation processing
a factor of five.
11.6 With an example explain the sampling process.
11.7 What is meant by aliasing ?
11.8 How can aliasing be avoided ?
a", n>0
11.9 The signal x(n) is defined by x(n) = {
0, otherwise
(a) Obtain the decimated signal with a factor of three.
(b) Obtain the interpolated signal with a factor of three.
1.10 Explain polyphase decomposition process.
1.11 How can sampling rate be converted by a rational factor M/L ?
1.12 Draw the block diagram of a multistage decimator and
integrator.
1.13 What are the characteristics of a comb filter ?
1.14 Explain with block diagram the general polyphase framework
for decimators and interpolators.
1.15 What is a signal flow graph ?
(Contd.)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
586 Digital Signal Processing

=f, $ X(f-kf,) (12.7)

k=-%
If aliasing is avoided, i.e. x(t) is band limited to a frequency less than
1/27, then,
X(P=f, XP (12.8)
Let x(n) be the sampled version of x(t) . The Fourier transform of x(n)
is given by

xXp= $, xei (12.9) `

n=-œ

The autocorrelation of the sampled signal x(n) is given by

ra (k= J, x* (n) x(n +k) (12.10)

na-w

The Fourier transform ofr ,,(k) from the Wiener-Khintchine theorem is

Saif = p> Tax (k) e? (12.11)

The other method for computing the energy density spectrum is
obtained from the Fourier transform of x(n),
SaPA = XA P
2
5 x(n) eJ28fn
(12.12)
n=
~æ

Since finite energy signals possess Fourier transform, spectral

analysis is done with the energy spectral density function.

12.3 ESTIMATION OF THE AUTOCORRELATION AND

POWER SPECTRUM OF RANDOM SIGNALS
Consider signals which do not have finite energy. For these signals,
Fourier transform is not possible. But these signals have finite average
power. For these signals, spectral analysis is done with power spectral
density function.
Let x(t) be a stationary random process. The statistical autocorre-
lation function for this signal is,
Yex (1) =E [x + (t) x (t +1) (12.13)
The Fourier transform of the autocorrelation function of a stationary
random process gives the power density spectrum,
Tlf) = F Ye (D)

= | vame Pdr (12.14)

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
600 Digital Signal Processing

Variance

Var a(fii=Tair)=|1(FE2E
L) |
_ 1 (sinnfN .
where =Waar f) = 77
sin nf
When N > œ,
v2
E [P xf Tax(f) | Waan(@)d@
-1/2
= Wart (0) T,, A)
=T,,(f)
var [P,,(f lo Tr f)
This is asymptotically unbiased estimate, but not consistent as
variance does not approach zero when N > œ.
The quality factor is,

-GO
v=
which is constant and independent of N specifies the poor quality.
Bartlett Power Spectrum Estimate
Mean
1/2
EL PR" M]= f TOW pon (f- Odo
-1/2

Variance

ver Pi on] =z eo aR |
x 2

r 2
where Waa (f) = Š (rr)

As N>, M>, k= A (fixed)

1/2

E [PRA] >a f Waar Af

-1/2

=T,, F)Waer (0)

=T,ff)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Spectral Estimation 605

N-1 N-1 N-1 N-1

Z32322 E [x(n )x (na) x(ng)x(n4))
e 77A -n9
“WZ =0 ng =0 ng =0 ng =0

EIP (Fy) Px f)

z1 zi z z -1 [E [x (m) x(nq)] E [x (ng) x(n4)]+

“at Xz: E[x (n1) x(ng)] E [x (n2) x(m4)] +
m=0ng=0n=0n4=0 | E [x (n;) x (n4)] E [x (n2) x (n3))
7 J2 Alay -ng)
e- J2xfa(mg~m4) *
-1 N-1
5 £ ot + pA S ote ~ §2x(f, ~fa) (ar ng)
me A ny =0
ng =O ny =0
ng =0
N? N-1N-1
N +> £ ofe -j2n(fy + fa) (ny ng)

ny =0 ng=0

N-1 Rai
24 x e712" fi- fa) > ef 2h - fans
4
Ss ny=0 ng =0
~ Nè N-i ’ N-1 |.
$ X eizh + f)ni 5 ei 2h + fh) ng
nysO ng =0

ii 1- e712- ha) N ‘ 1- ef? h-iad) N wcities

1- ef 28(h- fa) 1—ef2h-h) N?

1- -j2
e712"
+f) N ‘ 1-e/ j2 nifi + fo) N >i
1-e IA -eht N?
|2e IA- N _ gJ2R(h-h)N E
ot) nA
t 2-e RAe ye *
nae Q—e7J2thith)N _ QJ2elhthyN | 1
g-e Iht) _pi2hth) Ne

=o! 1+ 1 (Peete fr +1 (2222n( 00f +8f)N

"| OUN? C2-2e08 2n(f, - N? \2-2c0s 2n(f, + fy) J!
weil ie sinehifN) (darth GW)
N sin x(f, - fa) N sin n(f, + f2)
(b)
var [Pa (f)) =E [P2 (f)] - {E [P2 (AP
But, E[ P,,(P) = 02
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
614 Digital Signal Processing

where
r,, (k) is the autocorrelation sequence of x(n)
ry (k) is the crosscorrelation sequence of x(n) and d (n)
If d(n) = x(n),
N
Y, ay (k) ralk -D = ry D1 = 1, 2, -n N
k=1

The above set of equations are called normal equations or the Yule-
Walker equation.
Solve the equations recursively. First consider a predictor of order
one, which is given by
Ny D
a, (1) =
Ty (0)
The least squares error becomes

&= Y [ x(n) + a, (1) x(n)]?, since x(n) = dín)

=%

È x(n)+2a, Y xinxin-1) +a? Y x(n-1)

n=- n=-% n=-

£1 = ra (0) + 2a; (1) ry,(1) + a? (1) rys(0)

substituting the value for r,, (1) in terms of a, (1), we get
£1 = ry (0) + 2a; (1) (-a, (1) r,, (0)) + a? (1) r,, (0)
= Py,(0) — 2a? (1) r,, (0) + a? (1) r,,(0)
= ry,(0) [1-a?(1)]
Similarly considering the second-order predictor, we get
ay (1) r,, (0) + ag (2) roy (1) = -ry (1)
ag (1) roy (1) + ag (2) r,,. (0) = -r,, (2)
On solving,
z s- [regret
a Tez (0) — a? (1) 7, (0)
_ Tex (2) + a, (D ra (D)
£1
and a, (1) = @, (1) + a3 (2) a, (1)
Now the second-order predictor coefficients are expressed in terms of
first-order predictor coefficients.
In general, the m™ order predictor coefficients can be expressed in
terms of (M — 1) order predictor coefficients.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Spectral Estimation 623

È o eiki

In matrix form
E"(fi)a=1
where Efi) =11 e771... e/?*Pfy
Minimise the variance o? subjected to the constraint specified, yields
an FIR filter which allows the f, frequency components undistorted.
Other frequency components are attenuated. This yields,
â =TQ EU VE (fi)Ta Ef)
The variance becomes
ite
ee Ss
mn” ETETEA)
The minimum variance power spectrum estimate at frequency f; is
represented in the above equation. By varying the frequency f; from 0 to
0.5, the power spectrum estimate can be obtained. Even if f, changes,
Tz is computed only once. The denominator of o2;,, can be computed
using single DFT. If R, is the estimate of Tx, Rẹ can replace I,.,and the
minimum variance power spectrum estimate of |Capon’s method iis

Pm (f) = ee: VER

E' (f) Rj} E(f)
This estimate results in spectral peaks estimate proportional to the
power at that frequency.
12.6.8 The Pisarenko Harmonic Decomposition Method
This method provides the estimate for signal components which are
sinusoids corrupted by additive white noise.
A real sinusoid signal can be obtained from the difference equation
x(n) = -a x(n — 1) - a; x (n - 2)
where a, =2 cos 2 nf,
a,=1
initial conditions,
x(-1) =-1
x(-2) =
This system has complex-conjugate poles atf =f, and f =—f,, obtaining
the sinusiod x(n) = cos 2nf,n, n 20.
Consider p sinusoid components available in the signal,
2
x(n) =- y a,x(n —m)
m=1

The system function is given by

 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
648 Digital Signal Processing

Ey = $ w"~! jey (l,n) |? (13.37)

1=0

where

ey (I, n) =d (n)-d (l-n)

=d (n)— hy (n) Xy (2)
and w is the weighting factor lying in the range 0 < w < 1.
The weighting factor is used to give more weightage to the most
recent points so that the filter coefficients can be properly adapted to
the time varying statistical characteristics. Otherwise a finite duration
sliding window with uniform weightage can be used.
Minimisation of ©,, with respect to hy (n)

Min {ey} = Min [šw" A me

k=0

= Min » w"! d (1) - At, (n) xuw} (13.38)

l=0

The minimisation of £ọ results in

Ry (n) hy (n) = Dy (n)
where

Ry (n)= > w" `! Xy OXLD

1=0
—estimated signal correlation matrix
n

Dy (n) = }, w`’ Xy O
1=0
—estimated crosscorrelation vector
The solution can be obtained as
hy = Rj} (n) Dy (n) (13.39)
Ry (n) and Dy (n) can be computed recursively by
Ry (n)= w Ry (n — 1) + Xy (n) X(n) (13.40)
This is known as the time update equation for Ry (n)
Dy (n)
= w Dy (n - 1) + d (n)Xy (n) (13.41)
Matrix Inversion Lemma
Let A and B be two positive definite M x M matrices, D is a positive
definite N x N matrix and C is an M x N matrix.
A=B'+¢C-D'cT (13.42)
A`! can be obtained from matrix inversion lemma as
A` =B -BC [D + C” B C)’ C7 B (13.43)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Adaptive Filters 653

Dw Y wo, GD (13.65)

n-1 ¥
Km- = È w” fan- O bm- 0- V+ fp 1) By 1 0-1)
i=l

n-i
x w) i faa ibm- li-l) + fp_-1 (2) by (n1)
i=1
=w Ky —1(n-1)+ fy 1 (2) bm (0-1) (13.66)
Similarly,
EQ _,()=wE® _, fi, 1) vse
and
E®_\(n)=wE®_,(n-1)+62,_,()
(13.68)

Order update recursions The prediction errors are given by

fn (i) = fry 1 D+YL (by, G-D, lsisn
bm (i) = by 1 G- D+ YO fn- @, Lsi<n (13.69)
where
y (n) — forward reflection coefficient of m™ stage
y (n) - backward reflection coefficient of m™ stage
and yf (n) and y® (n) are considered as constants for the
time interval 1 <i <n.
Estimate for forward and backward prediction error
R

ED (n= F wi Pw
isl

‘ ye [fm 1) +P Mba- i- DY

= Su tf? (i) + 2y P (n)

i=l

Eur fa D bn-li- DHP

n

;
F E urio- i=l
2
= EY? Bae iy. | sae |E® (n-1)
m1 BO mn- D Ewin-D
= EP (n) -ram
K?
(13.70)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 cin 14
Applications ofDigital Signal
Processing

14.1 INTRODUCTION
Digital signal processing techniques are used in a variety of areas which
include speech, radar, sonar, image, etc. These techniques are applied
in spectral analysis, channel vocoders, homomorphic processing
systems, speech synthesisers, linear prediction systems, analysing the .
signals in radar tracking, etc.

14.2 VOICE PROCESSING

There are different areas in voice processing like encoding, synthesis
and recognition. In a speech signal some amount of redundancy is
present, which can be removed by encoding. Synthesis is required at the
receiver side because in the transmitter, compression/coding is done.
Recognition involves recognising both the speech and the speaker.
14.2.1 Speech Signal
A speech signal consists of periodic sounds, interspersed with bursts of
wide band noise and sometimes short silences. The vocal organs are in
motion continuously, hence the signal generated is not stationary. But
short segments of 50 ms are treated to be approximately stationary.
These signals are generated by the vibration of the vocal cords. The
muscles in the larynx stretch these cords, which vibrate when air is
forced, thus producing sound in the form of a pulse train. This passes
through the pharynx cavity and tongue and is expelled either at the
mouth or nasal cavity depending on the position of velum. Fig. 14.1 shows
the mechanism of human speech production.
When air passes through narrow constrictions in the vocal tract, a
turbulent flow is produced. Otherwise pressure is built up along the
tract behind a point of total constriction.
The vocal system can be modelled with a periodic signal excitor, a
variable filter representing the vocal tract, switch to pass the signal
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
662 Digital Signal Processing

The short time transform at a finite set of N frequencies, over the

band 0 < wT < 2z. h(nT ) is the impulse response of an FIR filter. It is
non-zero for 0 <n < M — 1 and the centre frequencies are
2nk
=
=F<"*) b=0,1,1,-...,N-
N 1 (14.6)
n

Therefore, X(@,, nT) = 2; x(rT) A(nT —rT) eT

r=n-M+1

[l
= >
n-an
E x MAMT-rT)eF"? — (14.7)
m=0 r=a-~(m-DN+1

where [M/N] + 1 is the greatest integer less than or equal to M/N.

Letl=n-mN-r
XO, 2T) =

yji w-1 P
Y YLxtal-rT-mNT)ACT
-mNT) een?
m=0 l=0

an Ne eae IT- e]
l=0 m=0 Adt+mNT) eFr)

hAT+mNT) F)" (14.8)

_ {2% a Nal 2x

X(w,, nT) =e Aye) 5 gil,n)e lm)" (14.9)

1=0

where

[$]
gil, n)= > x(nT-1IT-mNT)hUT
+ mNT) (14.10)
m=0

This analysis is done usually with a bank of digital filters.

14.2.4 Speech Analysis Synthesis System
The main objective is to measure the outputs of bandpass filter banks
and reconstruct the speech from these signals. Figure 14.5 shows the
analysis-synthesis system. Let x(nT') be the speech input and y (nT)be
the reconstructed synthetic waveform. The impulse response of the
bandpass filter bank is h,(nT), k = 1, 2, ..., M. y(nT) is obtained by
summing the individual bandpass filter outputs
y, (nT), k =1, 2,...,M
h, (nT ) = h (nT ) cos (nT ) (14.11)
where A (nT ) is the impulse response of a lowpass filter.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Applications of Digital Signal Processing 673

i - wavelength
Pulsed Doppler signals can be used to obtain both range and velocity
resolution.
14.3.1 Signal Design
Transmitting narrow pulse provides good range but poor velocity
measurement. A wide pulse of single frequency gives good velocity but
bad range information.
Consider the radar model shown in Fig. 14.16. Let the signal be
generated digitally and transmitted through an analog filter. The
transmitted signal is s(t). The received signal is s(t — t) e?“ -© which
is delayed and frequency shifted. The received signal is passed through

To transmitter From
and antenna receiver
Y

| Analog rlAD Digital

filter matched
filter

s(nt) sih s(t-t)e/?™t—) s(nTs-

1) Prnt)

Fig. 14.16 Block Diagram of a Radar Model

an analog filter, A/D converter and then through a digital matched

filter. The input signal to the matched filter is s(n Ts — 1) e/?*/ Ts -*),
A long duration signal is required for preserving radar power. But for
preserving range resolution, narrow signals are required. This problem
can be resolved by designing long duration signals with short duration
correlation functions. When the received signal is passed through the
appropriate matched filter, a sharp pulse will be available at the filter
output. The digital filter can be matched to the signal return for zero
range and zero Doppler. Hence its impulse response can be s*(—n 7,).

14.4 APPLICATIONS TO IMAGE PROCESSING

2D signal processing is helpful in processing the images. The different
processing techniques are image enhancement, image restoration and
image coding.
Image enhancement focuses mainly on the features of an image. The
various feature enhancements are sharpening the image, edge
enhancement, filtering, contrast enhancement, etc. Linear filtering
emphasises some spectral regions of the signal. Histogram modification
is done on pixel-by-pixel basis and finds its application in contrast
equalisation or enhancement. Figure 14.17 shows some image
enhancement operations.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Applications of Digital Signal Processing 681

in time and centered at the location of the Dirac. For small a’s, the
transform “zooms-in” to the Dirac with good localisation for very small
scales.
Frequency localisation
Consider the sinc wavelet, i.e. a perfect bandpass filter. Its magnitude
spectrum is 1 for || between n and 2r. Consider a complex sinusoid of
unit magnitude and at frequency @). The highest frequency wavelet that
passes the sinusoid having a scale factor of m/w, (gain of Vx/«,) while
the low frequency wavelet that passes the sinusoid having a scale factor
of 27/0 (gain of /2n/a,).
(viii) Reproducing Kernel
The CWT is a very redundant representation since it is a 2-D expansion
of a 1-D function. Consider the space V of a square integrable function
over the plane (a, b) with respect to da db /a?. Only a subspace H of V
corresponds to wavelet transforms of functions from L?(R).
If a function W (a, b) belongs to H, i.e. it is the wavelet transform of
f(t), then W (a, b) satisfies
1 da db
W (ap, bo) = Z SJE oboa, b) W (a, b) a? (14.34)
where Klao, bo, a, b) = < Wa, bẹ Wa, b > is the reproducing kernel.

Discrete Wavelet Transform

The discrete wavelet transform (DWT) corresponding to a CWT
function W (a, b ) can be obtained by sampling the co-ordinates (a, b ) on
a grid as shown in Fig. 14.22. This process is called the dyadic
sampling because the consecutive values of discrete scales as well as
the corresponding sampling intervals differs by a factor of two. Then
the dilation takes the values of the form a = 2" and translation takes
the values of the form b = 2* 1 where k and l are integers. The values
a b —>

al
1

4
Fig. 14.22 Time-Frequency Cells that Correspond to Dyadic Sampling
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
704 Digital Signal Processing

% Program for the design of Butterworth analog Bandpass filter

cle?
close all;clear all;
format long
rp=input (‘enter the passband ripple...’);
rs=input (‘enter the stopband ripple...’);
wp=input (‘enter the passband freq...’);
ws=input (‘enter the stopband freq...');
fs=input (‘enter the sampling freq...’);
wi=2*wp/fs;w2=2*ws/fs;
{n]=buttord(wl,w2,rp,rs,'s’);
wn= [w1 w2];
[b,a]=butter (n,wn, ‘bandpass’,‘s‘);
w=0:.01:pi;
{h,om])=freqs(b,a,w) ;
m=20*1log10(abs(h));
an=angle (h);
subplot
(2 ,1, 1);plot(om/pi,m);
ylabel (‘Gain indB-->’);xlabel('(a) Normalised frequency -->’);
subplot (2,1,2);plot(om/pi,an);
xlabel (‘ (b) Normalised frequency -->’);
ylabel (‘Phase in radians ~-->’);
As an example,
enter the passband ripple... 0.36
enter the stopbandripple... 36
enter the passband freq... 1500
enter the stopband freq... 2000
enter the sampling freq... 6000
The amplitude and phase responses of Butterworth bandpass analog
filter are shown in Fig. 15.9.

1 a 8 T

Gain
dB
in
——>-

0 0.1 02 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(a) Normalised frequency —>
a
Fig. 15.9 (Contd.)
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
726 Digital Signal Processing

\1 $38
dB
in
Gain
——>
$
$

1 N

radians
in
Phase
—>
(4 0.1 0.2 0.3 0.4 05 06 07 0.8 09 1
(b) Normalised frequency —>

Fig. 15.23 Chebyshev Type - | Low-pass Digital Filter

(a) Amplitude Response and (b) Phase Response

(n,wn)=cheblord(wl,w2,rp,rs);
[b,a)=chebyl(n,rp,wn, ‘high’);
w=0:.01/pi:pi;
[h, omJ =freqz(b,a,w);
m=20*10g10
(abs (h) );
an=angle(h);
subplot (2,1, 1);plot(om/pi,m);
ylabel (‘Gain indB-->’);xlabel(‘(a) Normalised frequency -->’) ;
subplot(2 ,1,2);plot(om/pi,an) ;
xlabel (` (b) Normalised frequency -->');
ylabel (‘Phase in radians -->’);

As an example,
enter the passband ripple... 0.3
enter the stopband ripple... 60
enter the passband freq... 1500
enter the stopband freq... 2000
enter the sampling freq... 9000
The amplitude and phase responses of Chebyshev type - 1 high-pass
digital filter are shown in Fig. 15.24.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 MATLAB Programs T31

an=angle(h);
subplot (2,1, 1);plot(om/pi,m);
ylabel (‘Gain indB-->');xlabel(‘(a) Normalised frequency -->’);
subplot (2 ,1,2);plot(om/pi, an);
xlabel(* (b) Normalised frequency -->’);
ylabel (‘Phase in radians -->’);
As an example,
enter the passband ripple... 0.35
enter the stopband ripple... 35
enter the passband freg... 1500
enter the stopband freq... 2000
enter the sampling freq... 8000
The amplitude and phase responses of Chebyshev type - 2 low-pass
digital filter are shown in Fig. 15.27.

dB
in
Gain
———>

-100 ——— j ee | =) a a ee
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Normalised frequency ——>

4 oe =k —ae borar F- a. a ]

o
ee

9
Phase
in
radians
——-> L
a r a S EE S S,N |
orm 0.1 0.2 0.3 0.4 05 06 07 08 09 1
(b) Normalised frequency —->

Fig. 15.27 Chebyshev Type - 2 Low-pass Digital Filter

(a) Amplitude Response and (b) Phase Response
15.13.2 High-pass Filter
Algorithm
1. Get the passband and stopband ripples
2. Get the passband and stopband edge frequencies
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 MATLAB Programs 743

%HIGH-PASS FILTER
b=firl(n-1,wp, ‘high’,y);
{h,o]=freqz(b,1, 256);
m=20*1log10 (abs (h));
subplot (2,2,2) ;plot(o/pi,m) ;ylabel (‘Gain in dB -->’);
xlabel(* (b) Normalised frequency -->‘);
%BAND-PASS FILTER
wn= [wp ws];
b=firl(n-1l,wn,y);
(h,o]=freqz(b,1,256);
m=20*1lo0g10
(abs (h));
subplot (2,2,3);plot(o/pi,m) ;ylabel(‘GainindB-->’);
xlabel (* (c) Normalised frequency -->’);
%BAND-STOP FILTER
b=firl(n-1,wn, ‘stop’,y);
(h,o]=freqz(b,1,256);
m=20*1log10(abs(h));
subplot (2,2,4);plot(o/pi,m) ;ylabel (‘Gain in dB -->’);
xlabel ( * (d) Normalised frequency -->');
As an example,
enter the passband ripple 0.03
enter the stopband ripple 0.02
enter the passband freq 1800
enter the stopband freq 2400
enter the sampling freq 10000
enter the ripple value(in dBs) 40
The gain responses of low-pass, high-pass, bandpass and bandstop
filters using Chebyshev window are shown in Fig. 15.34.
15.14.5. Hamming Window
Algorithm
1. Get the passband and stopband ripples
2 . Get the passband and stopband edge frequencies
3. Get the sampling frequency
4. Calculate the order of the filter
5. Find the window coefficients using Eq. 7.40
Geis Draw the magnitude and phase responses.
%Program for the design of FIR Low pass, High pass, Band pass
and Bandstop filters using Hamming window
cle;clear all;closeall;
rp=input (‘enter the passband ripple’);
rs=input (‘enter the stopband ripple’);
fp=input (‘enter the passband freq’);
fs=input (‘enter the stopband freq’);
f =input (‘enter the sampling freq’);
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
758 Digital Signal Processing

a=input (‘enter the donomiator polynomials=’ ) ;

c=input (‘enter the gain of the filter=');
{n,m]=size(b) ;
aza(:,2:3);b=b(:,2,:3);
u=zeros(n,2);
for i=l :length(x),
for k=1 :n,
unew=x (i)-sum(u(k,:).*a(k,:));
x(i)=-unew+sum(u(k,:).*b(k,:))
u(k,:)=(unew,u(k,1)];
end
y(i)=c*x(i);
end

15.31 DECIMATION BY POLYPHASE DECOMPOSITION

%Program for computing convolution and m-fold decimation by
polyphase decomposition
function y = ppdec(x,h,M);
x=input (‘enter the input sequence=');
h=input (‘enter the FIR filter coeficients="');
M=input (‘enter the decimation factor=");
lh = length(h); 1p = floor ((1h-1)/M) +1;
p = reshape([reshape(h,1,1h),zeros(1,1p*M-1lh)],M,1p);
lx=length(x); ly = floor ((1x + 1h-2) /M)+1;
lu=floor((1x+M-2)/M)+1; length of decimated sequences
u = [zeros(1,M-1),reshape(x,1,1x), zeros (1,M*1lu-1x-M+1) ];
y = zeros (1, 1u+1p-1);
form=1:M, y=y+conv(u(m,:),p(m,:));end
y=y(l:ly);

15.32 MULTIBAND FIR FILTER DESIGN

%Program for the design of multiband FIR filters
functionh = firdes(N,spec,win) ;
N = input (‘enter the length of the filter=’);
spec = input (‘enter the low, high cutoff frequencies and
gain=’);
win = input (‘enter the window length=’);
flag = rem(N, 2);
[K,m] = size(spec) ;
n= (0:N) -N/2;
if (~flag), n(N/2+1) =1;
end, h=zeros (1,N+1);
for k=1:K,
temp = (spec (k, 3) /pi) * (sin(spec (k, 2) *n)-sin(spec(k,1)*n))./n;
if(~flag) , temp(N/2+1) =spec(k,3) * (spec (k,2)-spec(k,1))/pi;
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
764 Digital Signal Processing

rsl=filter(h,1l,y);
efor i=1;length (ns),
% rsl(i)=rs6(i);
send
for i=l:lengthi(ns),
rs(i)=rsl(i);
end
subplot (2,2,2);plot(rs) ;title(‘noise signal’);
xlabel (‘Time ——') ;ylabel (‘Amplitude ——’);
%Far end signal
fs1=(552525252525255555555§822222222222225);
trs=sign(rs2);
%Far end signal is digitally modulated and plotted
zl = dmod(fs1, fc, fd, fs,’psk’);
for i =1 :length (ns),
z(i) =z1(i);
end
subplot (2, 2, 3);plot(z);title (‘far-end signal’);
xlabel (‘Time ——;) ;ylabel (‘Amplitude ——’);
%Echo and the far end modulated signal is added in the hybrid
ql=z1 + rs1;
for i=1:length (ns),
a(i)=ql(i);
end
subplot (2,2,4);plot(q) ;title(‘received signal’);
xlabel (‘Time ——’); ylabel (‘Amplitude ——’);
q2=xcorr(q);
+Auto correlation is taken for the near end signal
ar=xcorr(ns);
%cross correction is taken for the near end and far end signal
erd=xcorr(rs,ns);
ll=length(ar) ;j=1;
for i=round(11/2):11,
arl(j)=ar(i)
j=j+1;
end
$Toeplitz matrix is taken for the auto correlated signal
r=toeplitz (arl);
12=length(cr ;j=
d)1;
for isround (12/2):12,
eral (j)=crd(i);
j=j+1;
end
p=crdl';
%Maximum and minimum eigen values are calculated from the
toeplitz matrix
lam=max(eig(r));la=min(eig(r));l=lam/la;
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
770 Digital Signal Processing

(c) Repeat parts (a) and (b) using the Hamming window
(d) Repeat parts (a) and (b) using the Bartlett window.
15.20 Design an FIR Linear Phase, bandstop filter having the ideal
frequency response
1, for|a|< 2/6
0, for 2/6 <| s 2/3
Hal) = |1, for 213 <|alsa
(a) Determine the coefficient of a 25 tap filter based on the
window method with a rectangular window.
(b) Determine and plot the magnitude and phase response of
the filter.
(c) Repeat parts (a) and (b) using the Hamming window
(d) Repeat parts (a) and (b) using the Bartlett window.
15.21 A digital low-pass filter is required to meeth the following
specfications:
Passband ripple <1 dB
Passband edge 4 KHz
Stopband attenuation 240 dB
Stopband edge 6 KHz
Sample rate 24 KHz
The filter is to be designed by performing a bilinear
transformation on an analog system function. Determine what
order Butterworth, Chebyshev and elliptic analog design must
be used to meet the specifications in the digital
implementation.
15.22 An IIR digital low-pass filter is required to meet the following
specfications
Passband ripple < 0.5 dB
Passband edge 1.2 KHz
Stopband attenuation 240 dB
Stopband edge 2 KHz
Sample rate 8 KHz
Use the design formulas to determine the filter order for
(a) Digital Butterworth filter
(b) Digital Chebyshev filter
(c) Digital elliptic filter
15.23 An analog signal of the form x,t) = a(t) cos(2000 mt) is
bandlimited to the range 900 <F < 1100 Hz. It is used as an
input to the system shown in Fig. Q15.23.
n)

R, =|2500 cos (0.8 xn)

Fig. Q15.23
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Appendix C 787

if (M< =0)
{
M = fabs
(M) ;
}
M1 = fabs (20 * log10(M));
if (( F>=edge[1]) && (F< =edge[2]) && (fx[1] ==1))
{M1 =0;}
else if ( (F >=edge[3]) && (F<=edge[4]) && (fx[2] ==1))
{M1 =0;}
else if ((F>=edge[5]) && (F< =edge[6]) && (fx[3] ==1))
{M1 =0;}
if ( M1 > 85)
{M1 = 85;}
x1 =25 +1000 *F;
Y1 = 25 + M1;
X3 = 25 +1000 * F;
Y3=170-15*N;
line(X1,Y1,X2,Y2);
line (X3,Y3,X4,Y4);
}
getch( );
closegraph( );
if (nfilt !=0)
{ goto agl;}
return ;
}

/* FUNCTION FOR REMEZ EXCHANGE ALGORITHM */

remez (edge, nband)
int nband;
float edge[20];
{
int j, k, 1, nz, nmi, neg, kup, lband;
int luck, itrmax, kkk;
int jm1, jpl, nzz, niter, nu, nut, nutl, jet;
double a[66], p[66], q[66];
double dnum, dden, dtemp;
double d(), gee();
float yl, err, jchnge, devl, delf, fsh, gtemp, cn;
float k1, kn, ynz, knz, klow, comp, aa, bb, ft, xt, xe;
/* The max.no.of iterations is 25*/
itrmax = 25;
devil = -1.0;
nz=nfcns +1;
nzz =nfcns +2;
niter =0;
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 a
You have either reached a page that is unavailable for viewing or reached your viewing limit for this
book.
 Index

A/D converters, 523 AR process and Linear prediction,

Adaptive equalisation, 633 617
Adaptive filters, 631 Area under F(f), 74
w
s.
Adaptive FIR filter, 631 Area under f(t), 74
Adaptive line enhancer, 634 ARMA model, 607
Adaptive noise canceling, 634 Audio compression, 683
Advantages of digital system, 455 Autocorrelation, 374, 586, 694
Advantages of FIR filters, 381 Averaging modified periodogram, 595
Alias free QMF filter bank, 575 Averaging periodogram, 574, 594
Alias free realization, 574
Aliasing, 34, 526, 528 Bandpass filter, 703, 709, 715, 721,
Alternation theorem, 412 727, 732
Amplitude spectrum, 15 Bandstop filter, 710, 716, 722, 729,
Amplitude, 15 734
Analog frequency transformation, Bartlett Method, 594
446, 449 Bartlett power spectrum estimate,
Analog signals 1, 2 600, 602
Analog to Digital converters, 28, 29 Bartlett window, 397, 737
Analysis filter bank, 565, 759 Basic realization, 453
Analysis of speech signal, 660 Basic structures for IIR systems, 455
Antenna beamwidth, 672 Basic wavelet, 678
Anti-aliasing filter, 34 Basics of AR, MA and ARMA models,
Anti-aliasing, 528, 548 607
Anti-causal, 221 Basis function, 607
Anti-imaging filter, 531, 549 BIBO stability, 250
Aperiodic signal, 4, 5, 306: Bilinear transformation, 427
Application to Radar, 671 Binomial series, 776
Applications of Laplace transform, Blackman and Tukey method, 596
127 Blackman window, 397, 739
Applications of Multirate DSP, 523 Blackman-Tukey power spectrum
Applications of wavelet transform, estimate, 602, 603
682 Burg method, 620
Applications to Image processing, Butterworth filters, 432
673
Approximation of derivatives Canonic structures, 455
method, 418 Capon method, 622
AR model, 607 Cascade form, 462
 Index 803

Cascade realization of FIR systems, Decimation-in-time (DIT) algorithm,

484 320
Cascade realization of IIR systems, Decimators, 547
462 Definite integral, 780
Causal and non-causal systems, 20 Definition of inverse z-transform,
Causal system, 20 196
Causality, 243 Denoising, 684
Channel vocoders, 664 Design of IIR filters, 418
Chebyshev filters, 439, 724 Design of optimal filters, 409
Chebyshev window, 741 Design techniques for FIR filters,
Circular convolution, 283, 310 385
Circular correlation of periodic Design using Kaiser window
sequences, 310, 375 function, 404
Classification of signals, 1, 3 Deterministic and Non-deterministic
Coding, 663 signals, 4
Coefficient quantization, 505, 508 DFT in spectral estimation, 591
Comb filter, 563 Differentiation, 207
Comparison of Fourier and Wavelet Digital filter banks, 565
transform, 682 Digital filter design, 551
Complex conjugation, 74 Digital FM stereo, 670
Complex form of Fourier series, 52 Digital frequency transformation,
Composite Radix FFT, 352 450
Compression, 663 Digital representation of speech
Computation of filter coefficients, signals, 638
649 Digital Signal Processing, 2
Computational requirements, 523, Digital signals 1, 2
602 Dirac delta function, 94
Continuous time systems, 3 Direct - Form realization of FIR
Continuous Wavelet Transform systems, 483
(CWT), 677 Direct - Form realization of IIR
Continuous-time periodic signals, 4 systems, 456
Continuous-time signal, 3, 584 Dirichlet's conditions, 41, 56
Continuous-time wavelet, 678 Discrete convolution, 279
Convolution integral, 138, 140 Discrete Fourier Transform(DFT),
Convolution, 66, 144, 208, 279 305
Correlation, 210, 373 Discrete wavelet transform, 677, 681
Cosine function, 130 Discrete-Time Fourier Transform,
Cross-correlation, 373, 693 305
Discrete-time periodic signals, 6
D/A converters, 523 Discrete-time signals, 3, 21, 585
Damped Hyperbolic sine and cosine Doppler effect, 672
functions, 131 Doppler filtering, 672
Damped sine and cosine functions, Down sampler, 523
131 Down sampling, 34
Data Compression, 682 Duality, 65, 74
Data transmission, 636 Dyadic sampling, 681
Dead band, 510 Dynamic systems, 17
Decimation filter, 523, 526
Decimation-in-frequency (DIF) Echo cancellation, 635, 636, 639
algorithm, 334 Elliptic filters, 445
804 Index

Encoder, 29 Fourier transform, 62, 127

Encoding, 36 Fractional sampling rate alternation,
Energy density spectrum, 262, 584 533
Energy signal, 104 Frequency and Time domain
Energy spectrum, 63 characteristics, 552
Equivalent structures, 467 Frequency convolution, 65, 70
Erros in QMF filter bank, 573 Frequency differentiation, 64, 73,
Estimation of autocorrelation, 586 143
Estimation of Power density Frequency domain, 397
spectrum, 587, 589 Frequency integration, 65, 73, 144
Even and odd signals, 7 Frequency response, 260, 261, 384
Even functions, 44 Frequency sampling method, 389
Exponential form of Fourier series, Frequency shifting (modulation), ‘65,
52 712
Exponential function, 129 Frequency transformation, 446, 448
Exponential pulse, 87 Frequency warping, 429
Exponential series, 776
Extra ripple filter, 412 Gate function, 82
Fast convolution, 368 Gaussian pulse, 92
Fast Fourier Transform, 319 General polyphase framework, 545
Geometric series, 777
Filter banks with equal pass bands, Geometrical construction method,
579 269
Filter banks with unequal pass Gibbs phenomenon, 388
bands, 580 Gradient Adaptive Lattice method,
Filter design for FIR decimator and 654
interpolator, 554 Group delay, 380
Filter design for IIR decimator and
interpolator, 555 Half-band filters, 408
Filter structures, 539 Hamming window, 395, 743
Final value theorem, 138, 144, 164, Hanning window, 397, 745
212 High pass filter, 702, 707, 713, 716,
Finger print compression, 684 725, 731
Finite Impulse Response (FIR) Hyperbolic sine and cosine function,
system., 257 130
Finite summation formulae, 775 IIR direct form structure, 540
Finite word length effects, 496 IIR filter bank, 577
Finite-Impulse Response (FIR) IIR structures for decimators, 550
Filters, 380
FIR Decimators and Interpolators, Impulse function (Unit impulse), 94
541 Impulse invariant method, 423
FIR direct form structure, 539 Impulse response, 25, 147, 157, 158,
FIR Half-band digital filters, 408 159, 236
FM stereo transmitter, 670 Infinite Impulse Response (IIR)
Folding, 21 filters, 417
Forward-Backward Lattice method, Infinite Impulse Response (IIR)
650 system, 257, 417
Fourier series method, 385 Infinite summation formulae, 776
Fourier series, 40, 385 Initial value theorem, 137, 144, 164,
Fourier transform pair, 63 211
 Index 805

Interpolation, 523, 530, 547 Magnitude response of digital filters,

Inverse Chebyshev filters, 444 381
Inverse CWT, 679 Magnitude response of Elliptic
Inverse Discrete-Time Fourier filters, 446
Transform, 305 Magnitude response of Inverse
Inverse Fourier transform, 63 Chebyshev filters, 445
Inverse Laplace transform, 128 Magnitude response, 149
Inverse z-transform, 196 Manipulation of signal flow graphs,
536
Kaiser window, 402, 748 Marr wavelet, 684
Kirchoff's voltage law, 162 Matlab Programs, 688
Matrix inversion lemma, 648
Ladder structures, 475 Maximal ripple filter, 412
Lag window, 597 Mean, 588
Laplace transform of periodic Minimization of J(hM), 644, 645, 646
functions, 154 Minimum mean square error
Laplace transform pairs, 128, 142 criterion, 643
Laplace transform, 127, 193 Modeling Voiced and unvoiced speech
L-channel QMF filter bank, 577 sounds, 641
Levinson-Durbin algorithm, 612 Models of vocal organs, 660
Limit cycle oscillations, 510 Modulation, 72
Limitations of non-parametric Mother wavelet, 678
method, 604 Multilevel filter banks, 578
Linear and non-linear systems, 18 Multiplication by tn, 143
Linear convolution verses circular Multiplication of two sequences, 310
convolution, 284 Multiplication theorem, 71
Linear convolution, 284, 291, 691 Multirate Signal Processing
Linear difference equations, 24, 256 advantages, 523
Linear differential equation, 127 Multistage decimators and
Linear filtering, 631 interpolators, 555
Linear phase filters, 384
Linear phase Lth band filters, 571 Network transfer function, 146
Linear prediction, 617 Non parametric methods, 593
Linear systems, 18 Non-canonic structures, 455
Linear time-invariance systems, 236 Non-causal systems, 20
Linear time-invariant system, 147 Non-deterministic, 4
Linearity, 64, 65, 143, 203, 239 Non-linear systems, 18
Logarithmic series, 777 Non-periodic waveforms, 41
Long division method, 213 Nyquist filters, 569
Low pass filter, 700, 706, 712, 718, Nyquist period, 524
724, 729 Nyquist rate, 33, 524
Lth band filters, 569
Odd functions, 44
MA model, 607 Odd signals, 7
MacLaurin series, 776 One-sided Laplace transform, 128
Magnitude and Phase responses, 148 One-sided z-transform, 196, 213
Magnitude and phase spectrum, 262 Optimal linear-phase FIR filter, 409
Magnitude response of Chebyshev Output noise power, 502
filters, 439 Overflow limit cycles, 512
Overlap add method, 369
 806
n
Index
r aaaaaaaaħÃĂă

Overlap save method, 371 Quality of power spectrum estimator,

599
Parallel realization of IIR system, Quanitization effect in A to D
464 conversion, 499
Parametric methods, 606 Quantization errors, 519, 520
Parseval's identity, 58 Quantization, 36
Parseval's theorem, 64, 65, 310 Quantized boxcar signal, 36
Partial fraction expansions, 144, 216
Periodic and aperiodic signals, 4 Radix-2 FFT, 320
Periodic convolution, 283 Radix-3 FFT, 352
Periodic gate function, 99 Radix-4 FFT, 352
Periodic samping, 525 Ramp sequence, 689
Periodic waveforms, 41 Random signals, 4
Periodogram, 589, 599 Range resolution 672
Phase delay, 380 Rayleigh's energy theorem, 64
Phase response, 149, 381 Real time function, 65
Phase spectra, 15 Realization of Linear phase FIR
Pisarenko Harmonic Decomposition systems, 485
method, 623 Reconstruction filter, 31, 36
Poles of a Butterworth filters, 434 Rectangular pulse, 84
Poles of a Chebyshev filters, 441 Rectangular window, 394, 736
Polyphase decomposition, 541, 544 Recursive least square algorithm,
Polyphase FIR structures, 547 647
Polyphase Framework, 545 Region of convergence (ROC), 128,
Polyphase IIR Decimators, 550 196
Polyphase IIR filter structures for Relationship between DFT and other
Decimators, 551 transforms, 306
Polyphase parallel decimators, 550 Remez exchange algorithm, 413, 782
Polyphase Type Decimators, 547 Representation of discrete-time
Polyphase Type Interpolators, 548 signals in terms of impulse, 280
Poly-Wiener criterion, 244 Representation of signals, 13
Power signal, 103 Representation of systems, 24
Power spectral estimation, 591, 610 Residue method, 223
Power spectrum, 59, 586 Response of LTI systems to arbitrary
Product quantization, 513 inputs, 280
Properties of a DSP system, 238 Rounding errors, 496
Properties of convolution, 281 Routh-Hurwitz stability criterion,
Properties of CWT, 679 148
Properties of DFT, 308
Properties of Fourier transform, 64 Sampler, 29
Properties of frequency response, 260 Sampling by impulse function, 32
Properties of Laplace transform, 143 Sampling function, 29
Properties of ROC, 197 Sampling instants, 28
Properties of unit impulse function, Sampling of continuous time signals,
u 29
Properties of z-transform, 203 Sampling rate conversion, 525
Pulse code modulation, 674, 675 Sampling theorem, 34
Pulse train signal, 99 Sampling, 524
Scalar multiplication, 143
Quadrature Mirror filter bank, 572 Scale change, 143
 Index 807

Scaling, 66, 207, 518 Taylor series, 776

Sectioned convolution, 368 Thermal noise, 4
Shift invariance, 244 Thermal noise, 4
Shifting, 21 Time advance, 213
Short time Fourier transform Time convolution, 65, 67
(STFT), 675, 676 Time delay, 143, 149, 213, 263
Short time spectrum analysis, 660 Time differentiation, 64, 73, 143
Signal design, 673 Time integration, 65, 73, 143
Signal flow graphs, 535 Time periodicity, 144
Signal quantisation and encoding, 36 Time reversal, 64, 74, 204
Signal reconstruction, 35 Time scaling, 64,
Signal to Noise Ratio (SNR), 500 Time shifting, 64, 71, 205
Signum function, 97 Time-frequency representation, 675
Sinc function, 130 Time-invariance, 242
Singularity functions, 9 Time-variant and time-invariant
Sinusoidal function, 96 systems, 20
Smoothing the periodogram, 596 tn function, 132
Sound synthesis, 683 Toolboxes, 688
Spectral analysis, 584 Transformation of the independent
Spectral estimation, 585 variable, 21
Spectrum estimation, 604 Transposed forms, 467
Speech analysis synthesis system, Triangular pulse, 90
662 Trigonometric Fourier series, 41
Speech signal, 658 Trigonometric identities, 774
s-plane poles and zeros, 147, 150 Trigonometric series, 777
Stability, 148, 247 Truncation errors, 496
Stable and unstable systems, 20 Two channel quadrature Mirror filter
State equations for Discrete time bank, 572
systems, 28 Two-sided Laplace transform, 128
State matrix, 26 Two-sided z-transform, 196,
State space structures, 479
State space, 26 Unconstrained Least square method,
State vector, 26 621
State-variable technique, 26 Uniform DFT filter bank, 565
Static and dynamic systems, 17 Unilateral Laplace transform, 128
Step and impulse responses, 147, Unit impulse train, 102
157, 158 Unit impulse, 94
Step response of Parallel R-L-C Unit pulse function, 11
circuit, 161 Unit step function, 10, 95, 129
Step response of Series R-L-C circuit, Unit step response, 238
160 Unit-impulse function, 10
Steps in the RLS algorithm, 650 Unit-pulse signal, 688
Structures for FIR decimators and Unit-ramp function, 10
interpolators, 541 Unit-step sequence, 689
Sub band coding, 523, 665 Unstable systems, 20
Sub sampling, 523 Up sampler, 523, 531
Symmetry, 43, 65
Synthesis filter bank, 565 Variance, 588
System modeling, 631 Video compression, 682
 DIGITAL SIGNAL
PROCESSING
This book comprehensively covers the undergraduate course on
Digital Signal Processing. Computer usage is integrated into the
text in the form of problem solving using MATLAB. Solved
examples and critical-thinking exercises and review questions
enhance the reader’s camprehension of the concepts.
i

Salient Features
P> Overview of Signals a Systems concepts.

p> Comprehensive coverage of key topics, including Infinite Impulse

Response Filters (IIR), Finite Impulse Response Filters (FIR),
effects of finite word length in digital filters, and multirate digital
signal processing.

> Review of the essential mathematical concepts such as Fourier

Analysis, Laplace Transforms and Z Transforms.

> A chapter devoted to Applications of Digital Signal Processing in

Speech Processing, Image Processing and RADAR.

> Packed with numerous solved examples (221), 576 review

questions and practice exercises.

p> A chapter on MATLAB featuring numerous examples illustrating

its application to signal processing.

The McGraw-Hill Companies

INA) Tata McGraw-Hill

== Publishing Company Limited _
7 West Patel Nagar, New Delhi 110 008