Principles of Communication (By Prof. v. V Rao, IIT Madras)

Principles of Communication Prof. V.
Venkata Rao

Indian Institute of Technology Madras
Acknowledgements

The author would like to place on record the help received from the
following sources.

Dr. M. S. Ananth, Director, IIT-M, and chairman, PIC
1
, NPTEL
2
and
Dr. K. Mangala Sunder, Professor, department of Chemistry, IIT-M and the
national coordinator, Web based courseware, NPTEL have taken great deal of
interest in this national program and have been steering this program extremely
well. Their infectious enthusiasm in this academic activity made me to give my
best to the assignment taken up by me. Though I retired from the institute
services while the work was in progress, they have done everything possible to
help me complete this work to my satisfaction. A big THANK YOU to them.

Mr. Prabhakar Rao and Mr. J aya Chandran of the Communication
laboratory , EE dept., IIT-M have played a major role in developing the material
used in flash animation and the audio demonstrations. Their theoretical
understanding and experimental skills in communication engineering have been
extremely useful in this context. Also, they have been responsible for generating
a number of experimental waveforms
3
shown in the text. It is a pleasure to
acknowledge their help.

Mr. Vamsi Mohan, M.S. scholar, EE Dept., IIT-M helped in generating the
data used in the demonstration of the Gibbs phenomena, generation and
demodulation of SSB signals and effect of aliasing in speech and music. This is
very much appreciated.

1
Program Implementation Committee
2
National Program for Technology Enhanced Learning
3
All these waveforms have been generated using the analog communication experimenter
system, ACES 301, Shreyaas lab Technologies, Chennai - 17.
Principles of Communication Prof. V. Venkata Rao

The two reviewers for this course, Dr. R. Aravind, associate professor, EE
Dept, IIT-M and Mr. M. S. Narayanan, professor, Hindustan College of
Engineering, Padur, TN - 603103 have gone through the material very carefully.
Their comments and suggestions have been very useful in improving the quality
of the course material.

Dr. Bhaskar Ramamurthy, Dean planning, IIT-M, did find the time (at my
request) to go through the first three chapters (Part I of the course). I very much
appreciate the suggestions and comments made by him.

Dr. S. Srinivasan, Head, EE Dept., IIT-M extended all the facilities of the
department (including the use of my pre-retirement office) there by creating a
very congenial atmosphere to carry out this assignment. This goodwill gesture
certainly deserves a word of appreciation.

Lastly, Ms. Manasa, Project associate, NPTEL program has been
extremely efficient and very co-operative in entering course material (text,
figures, animation, sound tracks, etc.) into the computer, thereby greatly
facilitating my job. She certainly deserves a special mention in these
acknowledgements.

CHAPTER 1
Representation of Signals

1.1 Introduction 1.1
1.2 Periodic Signals and Fourier Series 1.2
1.2.1 Periodic signals 1.2
1.2.2 Fourier series 1.4
1.2.3 Convergence of Fourier series and Gibbs phenomenon 1.19
1.3 Aperiodic Signals and Fourier Transform 1.22
1.3.1 Fourier transform 1.24
1.3.2 Dirichlet conditions 1.29
1.4 Properties of the Fourier Transform 1.31
1.5 Unified Approach to Fourier Transform 1.49
1.5.1 Unit impulse (Dirac delta function) 1.49
1.5.2 Impulse response and convolution 1.60
1.5.3 Signum function and unit step function 1.69
1.6 Correlation Functions 1.78
1.6.1 Cross-correlation functions 1.78
1.6.2 Autocorrelation function 1.85
1.7 Hilbert Transform 1.95
1.7.1 Properties of the Hilbert transform 1.100
1.8 Bandpass Signals 1.104
1.8.1 Pre-envelope 1.107
1.8.2 Complex envelope 1.111
1.9 Bandpass (BP) Systems 1.119
Appendix A1.1: Tabulation of ( ) sinc 1.122
Appendix A1.2: Fourier transform of ( )
p p
x y
R 1.124
Appendix A1.3: Complex envelope of the output of a BP System 1.127
References 1.130
CHAPTER 2
PROBABILITY AND RANDOM VARIABLES

2.2 Basics of Probability 2.3
2.2.1 Terminology 2.3
2.2.2 Probability of an event 2.5
2.3 Random Variables 2.13
2.3.1 Distribution function 2.13
2.3.2 Probability density function 2.17
2.3.3 J oint distribution and density function 2.19
2.3.4 Conditional density 2.22
2.3.5 Statistical independence 2.23
2.4 Transformation of Variables 2.27
2.4.1 Functions of one random variable 2.28
2.4.2 Functions of two random variables 2.37
2.5 Statistical Averages 2.48
2.5.1 Variance 2.50
2.5.2 Covariance 2.53
2.6 Some Useful Probability Models 2.59
2.6.1 Discrete random variables 2.59
2.6.2 Continuous random variables 2.63
Appendix A2.1: Proof of Eq. 2.36 2.75
Appendix A2.2: ( ) Q Function Table 2.78
Appendix A2.3: Proof that
( )
2
,
X X
N m is a valid PDF 2.80
References 2.81
CHAPTER 3
Random Signals and Noise

3.2 Definition of a Random Process 3.3
3.3 Stationarity 3.8
3.4 Ensemble Averages 3.13
3.4.1 Properties of ACF 3.16
3.4.2 Cross-correlation 3.26
3.5 Systems with Random Signal Excitation 3.28
3.6 Power Spectral Density 3.30
3.6.1 Properties of Power Spectral Density 3.31
3.7 Cross Spectral Density 3.37
3.8 Gaussian Process 3.41
3.9 Electrical Noise 3.46
3.9.1 White noise 3.50
3.10 Narrowband Noise 3.61
3.10.1 Representation of narrowband noise 3.64
3.10.2 Properties of narrowband noise 3.66
3.10.3 PDF of the envelope of narrowband noise 3.72
3.10.4 Sine wave plus narrowband noise 3.74
Appendix A3.1 Properties of Narrowband Noise: Some proofs 3.78
References 3.82
CHAPTER 4
LINEAR MODULATION

4.2 DSB-SC Modulation 4.5
4.2.1 Modulation 4.5
4.2.2 Coherent Demodulation 4.14
4.2.3 Carrier recovery for coherent demodulation 4.25
4.3 DSB-LC modulation (or AM) 4.29
4.3.1 Tone modulation 4.32
4.3.2 Spectra of AM Signals 4.36
4.4 Generation of AM and DSB-SC signals 4.41
4.4.1 Generation of AM 4.42
4.4.2 Generation of DSB-SC 4.47
4.5 Envelope Detector 4.52
4.6 Theory of Single Sideband 4.57
4.7 Generation of SSB Signals 4.64
4.7.1 Frequency discrimination method 4.64
4.7.2 Phase discrimination method 4.68
4.8 Demodulation of SSB 4.71
4.9 Vestigial SideBand (VSB) Modulation 4.80
4.9.1 Frequency domain description of VSB 4.80
4.9.2 Time domain description of VSB 4.86
4.10 Envelope Detection of VSB+C 4.92
4.11 Superheterodyne Receiver 4.97
Appendix A4.1: Analysis of ED with Tone Modulation 4.106
References 4.111
CHAPTER 5
Angle Modulation

5.2 Bandwidth of FM 5.13
5.2.1 NarrowBand FM (NBFM) 5.14
5.2.2 WideBand FM (WBFM) 5.15
5.3 Tone Modulation 5.19
5.3.1 NBFM 5.19
5.3.2 WBFM 5.21
5.4 Phase Modulation 5.30
5.5 Generation of FM 5.36
5.5.1 Narrowband FM 5.37
5.5.2 WBFM: Indirect and direct methods 5.37
5.6 Demodulation of FM 5.49
5.6.1 FM-to-AM conversion 5.49
5.6.2 Phase shift discriminator 5.57
5.6.3 Zero-crossing detection 5.61
5.6.4 FM demodulation using PLL 5.63
5.7 Bandpass Limiter (BPL) 5.68
5.8 Broadcast FM 5.71
5.8.1 Monophonic FM reception 5.71
5.8.2 Two-channel (stereo) FM 5.73
Appendix A5.1: Table of Bessel Functions 5.76
Appendix A5.2: Phase Shift Discriminator 5.77
A5.2.1: Foster-Seely discriminator 5.77
A5.2.2: The ratio detector 5.82
Appendix A5.3: Multitone FM 5.85
Appendix A5.4: RMS Bandwidth of WBFM 5.86
Appendix A5.5: Modulation techniques in TV 5.91
References 5.97
CHAPTER 6

DIGITAL TRANSMISSION OF ANALOG SIGNALS:
PCM, DPCM AND DM

6.2 The PCM system 6.2
6.3 Sampling 6.4
6.3.1 Ideal impulse sampling 6.4
6.3.2 Sampling with a rectangular pulse train 6.9
6.3.3 Flat topped sampling 6.11
6.3.4 Undersampling and the problem of aliasing 6.13
6.4 Quantization 6.26
6.4.1 Uniform quantization 6.26
6.4.2 Quantization noise 6.31
6.4.3 Non-uniform quantization and companding 6.41
6.5 Encoding 6.56
6.6 Electrical waveform representation of binary sequences 6.59
6.7 Bandwidth requirements of PCM 6.61
6.7.1 Unipolar format 6.62
6.7.2 Polar format 6.63
6.7.3 Bipolar format 6.64
6.7.4 Manchester format 6.65
6.8 Differential Pulse Code Modulation (DPCM) 6.70
6.9 Delta Modulation 6.80
6.9.1 Linear Delta Modulation (LDM) 6.81
6.9.2 Adaptive delta modulation 6.92
Appendix A6.1: PSD of a waveform process with discrete amplitudes 6.97
References 6.99
CHAPTER 7

Noise Performance of Various Modulation Schemes

7.2 Receiver Model and Figure of Merit: Linear Modulation 7.2
7.2.1 Receiver model 7.2
7.2.2 Figure-of-merit 7.3
7.3 Coherent Demodulation 7.8
7.3.1 DSB-SC 7.8
7.3.2 SSB 7.11
7.4 Envelope Detection 7.12
7.4.1 Large predetection SNR 7.13
7.4.2 Weak predetection SNR 7.14
7.5 Receiver Model: Angle Modulation 7.21
7.6 Calculation of FOM 7.22
7.6.1 Large predetection SNR 7.22
7.6.2 Weak predetection SNR 7.33
7.7 Pre-Emphasis and de-Emphasis in FM 7.40
7.8 Noise Performance of a PCM System 7.49
Appendix A7.1: PSD of noise for angle modulated signals 7.59
References 7.63
1
ACF Auto Correlation Function
ADM Adaptive Delta Modulation
AM Amplitude Modulation
BLWN Band Limited White Noise
BM Balanced Modulator
BP Band Pass
BPF Band Pass Filter
BPL Band Pass Limiter
CCF Cross Correlation Function
CDF Cumulative Distribution Function (Same as DF)
DC Direct Current
DE De-Emphasis
DF Distribution Function
DM Delta Modulation
DPCM Differential Pulse Code Modulation
DSB Double Side Band
DSB-LC Double Side Band-Large Carrier
DSB-SC Double Side Band-Suppressed Carrier
ED Envelope Detector
FM Frequency Modulation
FOM` Figure of Merit
FT Fourier Transform
HPF High Pass Filter
2
HT Hilbert Transform/ Hilbert Transformer
IFT Inverse Fourier Transform
LDM Linear Delta Modulation
LO Local Oscillator
LPF Low Pass Filter
LSB Lower Side Band
LTI Linear Time Invariant
NBBP Narrow Band Band Pass
NBFM Narrow Band FM
NBN Narrow Band Noise
PB Pass Band
PCM Pulse Code Modulation
PDF Probability Density Function
PE Pre-Emphasis
PLL Phase Lock Loop
PM Phase Modulation
PSD Power Spectral Density
QAM Quadrature Amplitude Modulation
QCM Quadrature Carrier Multiplexing
QD Quadrature Detector
RC-LPF RC-Low Pass Filter
RV Random Variable
SB Stop Band
3
SBF Side Band Filter
SNR Signal-to-Noise Ratio
SSB Single Side Band
SSB+C Single Side Band plus Carrier
TB Transition Band
TV Tele Vision
USB Upper Side Band
VCO Voltage Controlled Oscillator
VSB Vestigial Side Band
VSB+C Vestigial Side Band plus Carrier
WBFM Wide Band FM
ZOH Zero Order Hold


1.1
1 CHAPTER 1

Representation of Signals

1.1 Introduction
The process of (electronic) communication involves the generation,
transmission and reception of various types of signals. The communication
process becomes fairly difficult, because:
a) the transmitted signals may have to travel long distances (there by
undergoing severe attenuation) before they can reach the destination i.e.,
the receiver.
b) of imperfections of the channel over which the signals have to travel
c) of interference due to other signals sharing the same channel and
d) of noise at the receiver input
1
.

In quite a few situations, the desired signal strength at the receiver input
may not be significantly stronger than the disturbance component present at that
point in the communication chain. (But for the above causes, the process of
communication would have been quite easy, if not trivial). In order to come up
with appropriate signal processing techniques, which enable us to extract the
desired signal from a distorted and noisy version of the transmitted signal, we
must clearly understand the nature and properties of the desired and undesired
signals present at various stages of a communication system. In this lesson, we
begin our study of this aspect of communication theory.

1
Complete statistical characterization of the noise will be given in chapter 3, namely, Random
Signals and Noise.

1.2
Signals physically exist in the time domain and are usually expressed as a
function of the time parameter
1
. Because of this feature, it is not too difficult, at
least in the majority of the situations of interest to us, to visualize the signal
behavior in the Time Domain. In fact, it may even be possible to view the signals
on an oscilloscope. But equally important is the characterization of the signals in
the Frequency Domain or Spectral Domain. That is, we characterize the signal
in terms of its various frequency components (or its spectrum). Fourier analysis
(Fourier Series and Fourier Transform) helps us in arriving at the spectral
description of the pertinent signals.

1.2 Periodic Signals and Fourier Series
Signals can be classified in various ways such as:
a) Power or Energy
b) Deterministic or Random
c) Real or Complex
d) Periodic or Aperiodic etc.

Our immediate concern is with periodic signals. In this section we shall
develop the spectral description of these signals.

1.2.1 Periodic signals
Def. 1.1: A signal ( )
p
x t is said to be periodic if
( ) ( )
p p
x t x t + T = , (1.1)
for all t and some T .

( denotes the end of definition, example, etc.)

1
We will not discuss the multi-dimensional signals such as picture signals, video signals, etc.

1.3
Let T
0
be the smallest value of T for which this is possible. We call T
0
as
the period of ( )
p
x t .

Fig. 1.1 shows a few examples of periodic signals.

Fig. 1.1: Some examples of periodic signals

1.4
The basic building block of Fourier analysis is the complex exponential,
namely,
( )
j f t
A e
2 +
or ( ) A j f t + exp 2

, where
A : Amplitude (in Volts or Amperes)
f : Cyclical frequency (in Hz)
: Phase angle at t 0 = (either in radians or degrees)

Both A and f are real and non-negative. As the radian frequency, (in
units of radians/ sec), is equal to 2 f , the complex exponential can also written
as
( )
j t
A e
+
. We use subscripts on A, f (or ) and to denote the specific
values of these parameters.

Fourier analysis uses t cos or sin cos
2
t t

=

in the represen-
tation of real signals. From Eulers relation, we have, cos sin
j t
e t j t
= + .

As t cos is the Re
j t
e

, where
[ ] x Re denotes the real part of x , we
have
( )
cos
2
j t j t
e e
t

+
= (denotes the complex conjugate)

2
j t j t
e e

+
=
The term
j t
e

or
2 j f t
e

is referred to as the complex exponential at
the negative frequency (or f ).

1.2.2 Fourier series

1.5
Let ( )
p
x t be a periodic signal with period T
0
. Then
0
0
1
f
T
= is called the
fundamental frequency and
0
nf is called the
th
n harmonic, where n is an
integer (for 0 n = , we have the DC component and for the DC singal,
T
0

is not defined;

1 n = results in the fundamental). Fourier series
decomposes ( )
p
x t in to DC, fundamental and its various higher harmonics,
namely,
( )
0
j 2 nf t
p n
n
x t x e

=
= (1.2a)
The coefficients { }
n
x constitute the Fourier series and are related to
( )
p
x t as
( )
0
0
0
j 2 nf t
n p
T
1
x x t e dt
T

=

(1.2b)
where
0
T
denotes the integral over any one period of ( )

p
x t . Most often, we
use the interval
0 0
,
2 2
T T

or
( )
0
0 , T . Eq. 1.2(a) is referred to as the
Exponential form of the Fourier series.

The coefficients { }
n
x are in general complex; hence
n
j
n n
x x e

= (1.3)
where
n
x denotes the magnitude of the complex number and
n
, the argument
(or the angle). Using Eq. 1.3 in Eq. 1.2(a), we have,
( )
( ) n
j nf t
p n
n
x t x e
0
2
+
=
=

Eq. 1.2(a) states that ( )
p
x t , in general, is composed of the frequency
components at DC, fundamental and its higher harmonics.
n
x is the magnitude

1.6
of the component in ( )
p
x t at frequency
0
nf and
n
, its phase. The plot of
n
x
vs. n (or
0
nf ) is called the magnitude spectrum, and
n

vs. n (or
0
nf ) is called
the phase spectrum. It is important to note that the spectrum of a periodic signal
exists only at discrete frequencies, namely, at
0
nf
,
n 0, 1, 2, = , etc.
Let ( )
p
x t be real; then
( )
j nf t
n p
T
x x t e d t
T
0
0
2
0
1
=

n
x
=
That is, for a real periodic signal, we have the two symmetry properties, namely,
n n
x x
= (1.4a)
- n n
= (1.4b)

Properties of Eq. 1.4 are part of an if and only if (iff) relationship. That is, if
( )
p
x t is real, then Eq. 1.4 holds and if Eq. 1.4 holds, then ( )
p
x t has to be real.
This is because the complex exponentials at
( )
0
nf and
( )
0
nf can be combined
into a cosine term. As an example, let the only nonzero coefficients of a periodic
signal be x , x , x
2 1 0
.
*
X
0
= x
0
implies, x
0
is real and let
2
2
j
4
2
x e x
= =
and
j
3
x e x
1 1
3

= = and
0
1 x = . Then,
( )
0 0 0 0
4 2 2 4
3 3 4 4
3 1 3 2
j j j j
j f t j f t j f t j f t
p
x t 2e e e e e e e e

+ + + + =

Combining the appropriate terms results in,
( )
p
x t f t f t
0 0
4 cos 4 6cos 2 1
4 3

= + +

which is a real signal. The above form of representing ( )
p
x t , in terms of cosines
is called the Trigonometric form of the Fourier series.

1.7

We shall illustrate the calculation of the Fourier coefficients using the
periodic rectangular pulse train (This example is to be found in almost all the
textbooks on communication theory).

Example 1.1
For the unit amplitude rectangular pulse train shown in Fig. 1.2, let us
compute the Fourier series coefficients.

Fig. 1.2: Periodic rectangular pulse train

( )
p
x t has a period
0
4 T = milliseconds and is ON for half the period and
OFF during the remaining half. The fundamental frequency
0
f =250 Hz.
From Eq. 1.2(b), we have
T
j n f t
n
T
x e dt
T
0
0
0
2
2
0
2
1

=

0
0
0
4
2
0
4
1
T
j nf t
T
e dt
T

=

0 0
2
sin
1
n
T n f

=


1.8

sin
2
n
n

=

As can be seen from the equation for
n
x , all the Fourier coefficients are
real but could be bipolar (+ve or ve). Hence
n
is either zero or for all n .
Fig. 1.3 shows the plots of magnitude and phase spectrum.

Fig. 1.3: Magnitude and phase spectra for the
p
x (t) of example 1.1

From Fig. 1.3, we observe:

1.9
i)
0
x , the average or the DC value of the pulse train is
1
2
. For any periodic
signal, the average value is ( )
0
0
2
0
2
1
T
p
T
x t dt
T
.
ii) spectrum exists only at discrete frequencies, namely,
0
f nf = , with
0
250 f Hz = . Such a spectrum is called the discrete spectrum (or line
spectrum).
iii) the curve drawn with broken line in Fig. 1.3(a) is the envelope of the
magnitude spectrum. The envelope consists of several lobes and the
maximum value of each lobe keeps decreasing with increase in frequency.

iv) the plot of
n
x vs. frequency is symmetric and the plot of
n
vs. frequency is
anti-symmetric. This is because ( )
p
x t is real.

v)
n
at n , 2 4 = etc. is undefined as 0
n
x = for these n . This is indicated
with a cross on the phase spectrum plot.

One of the functions that is useful in the study of Fourier analysis is the
( ) sinc function defined by
( )
( ) sin
sin sin c c

= =
(1.5)

A plot of the sinc vs. along with a table of values are given in
appendix A1.1, at the end of the chapter.


1.10
In terms of ( ) sinc , the Fourier coefficient of example 1.1 can be written
as,
1
sin
2 2
n
n
x c

=

.

Spectrum analyzer is an important laboratory instrument, which can be
used to obtain the magnitude spectrum of periodic signals (frequency resolution,
frequency range over which the spectrum can be measured etc. depend on that
particular instrument). We have given below a set of four waveforms (output of a
function generator) and their line spectra, as indicated by a spectrum analyzer.

The spectral plots 1 to 4 give the values of
10
1
20log
n
x
x

for the
waveforms 1 to 4 respectively. The units for the above quantities are in decibels
(dB).

Exercise 1.1
For the ( )
p
x t of Fig .1.4, show that
( )
n
x c nf
T
0
0
sin

=

Fig. 1.4: ( )
p
x t of Exercise 1.1

1.11

1.

Waveform 1

Spectral Plot 1

1.12

Waveform 1: A cosine signal (frequency 10 kHz).
Comments on spectral plot 1: Waveform 1 has only two Fourier coefficients,
namely,
1
x
and
1
x . Also, we have
1 1
x x
= . Hence the spectral plot has only

two lines, namely at 10 kHz, and their values are
x
x
1
10
1
20log 0 = dB.

2.

Waveform 2


1.13

Waveform 2: Periodic square wave with
T
0
1
5
= ; T
0
0.1 = msec.
Comments on Spectral Plot 2: Values of various spectral components are:
i) fundamental: 0 dB
ii) second harmonic:
( )
( )
10
sin 0.4
20log
sin 0.2
c
c

10
20 log 0.809 0.924dB = =
iii) third harmonic:
( )
( )
10
sin 0.6
20log
sin 0.2
c
c

10
0.504
20 log
0.935

=

10
20log (0.539) dB =
5.362 =
iv) fourth Harmonic:
( )
( )
10
sin 0.8
20log
sin 0.2
c
c

12.04dB =
v) fifth harmonic:
( )
( )
10
sin 1
20log
sin 0.2
c
c

( )
10
20log 0 = =

because of the limitations of the instrument, we see a small spike at 60 dB.
Similarly, the values of other components can be calculated.

1.14

3.

Waveform 3

Values of Spectral Components: Exercise

1.15

4.

Waveform 4

Values of Spectral Components: Exercise

1.16

The Example 1.1 and the periodic waveforms 1 to 4 all have fundamental
as part of their spectra. Based on this, we should not surmise that every periodic
signal must necessarily have a nonzero value for its fundamental. As a counter to
the conjuncture, let ( ) ( ) ( )
p
x t t t cos 20 cos 2000 = .

This is periodic with period 100 msec. However, the only spectral
components that have nonzero magnitudes are at 990 Hz and 1010 Hz. That is,
the first 99 spectral components (inclusive of DC) are zero!

Let ( )
p
x t be a periodic voltage waveform across a 1 resistor or a
current waveform flowing in a 1 resistor. We now define its (normalized)
average power, denoted by
p
x
P , as
( )
p
x p
T
P x t d t
T
0
2
0
1
=

Parsevals (Power) Theorem relates
p
x
P to
n
x as follows:
p
x n
n
P x
2
=
=

(The proof of this relation is left as an exercise.)

If ( )
p
x t consists of a single complex exponential, ie,

1.17
( )
( ) j nf t
p n
x t x e
0
2 +
=
then,
p
x n
P x
2
=
In other words, Parsevals power theorem implies that the total average power in
( )
p
x t is superposition of the average powers of the complex exponentials
present in it.

When the periodic signal exhibits certain symmetries, Fourier coefficients
take special forms. Let us first define some of these symmetries (We assume
( )
p
x t to be real).

Def. 1.2(a): A periodic signal ( )
p
x t is even, if ( ) ( ) =
p p
x t x t (1.6a)
Def. 1.2(b): A periodic signal ( )
p
x t is odd, if ( ) ( ) =
p p
x t x t (1.6b)
Def. 1.2(c): A periodic signal ( )
p
x t has half-wave symmetry, if
( )

=

p p
T
x t x t
0
2

(1.6c)

With respect to the symmetries defined by Eq. 1.6, we have the following
special forms for the coefficients
n
x :
( )
p
x t even:
n
x s are purely real and even with respect to n
( )
p
x t odd:
n
x s are purely imaginary and odd with respect to n
( )
p
x t half-wave symmetric:
n
x s are zero for n even.

A proof of these properties is as follows:
i) ( )
p
x t even:

1.18
( )
T
j nf t
n p
T
x x t e d t
T
0
0
0
2
2
2
0
1

=

( )
T
j nf t j nf t
p p
T
x t e d t x (t)e d t
T
0
0 0
0
2
2
0
2 2
0
0
1

= +

Changing t to - t in the first integral, and noting ( ) ( )
p p
x t x t = ,
( ) ( )
T T
j nf t j nf t
p p
x t e d t x t e d t
T
0 0
0 0
2 2
2 2
0
0 0
1

= +

( ) ( )
T
p
x t nf t d t
T
0
/2
0
0
0
2
cos2

=

The above integral is real and as ( ) n f t nf t
0 0
cos 2 cos(2 ) =

,
n n
x x
= .

ii) ( )
p
x t odd:
( ) ( )
T
j nf t j nf t
n p p
T
x x t e d t x t e d t
T
0
0 0
0
2
2
0
2 2
0
0
1

= +

Changing t to - t , in the first integral and noting that ( ) ( )
p p
x t x t = ,
we have
( ) ( )
T T
j nf t j nf t
p p
x t e d t x t e d t
T
0 0
0 0
2 2
2 2
0
0 0
1

= +

( )
T
j nf t j nf t
p
x t e e d t
T
0
0 0
2
2 2
0
0
1

=

( ) ( )
T
p
j
x t nf t d t
T
0
2
0
0
0
2
sin 2 =

Hence the result.

iii) ( )
p
x t has half-wave symmetry:

1.19
( ) ( )
T
j nf t j nf t
n p p
T
x x t e d t x t e d t
T
0
0 0
0
0
0
/2
2 2
0
/2
1

= +

In the first integral, replace t by
0
t + T /2 .
( )
( )
0
T T
j n t n j n t
n p p 0
x x (t + T /2) e d t x t e d t
T
0
0 0
0
/2 /2
0
0
1 +

= +

The result follows from the relation
1,
1,
j n
n odd
e
n even

=

1.2.3 Convergence of Fourier series and Gibbs phenomenon
As seen from Eq. 1.2, the representation of a periodic function in terms of
Fourier series involves, in general, an infinite summation. As such, the issue of
convergence of the series is to be given some consideration.

There is a set of conditions, known as Dirichlet conditions that guarantee
convergence. These are stated below.

i) ( )
p
T
x t
0
<

That is, the function is absolutely integrable over any period. It is easy to
verify that the above condition results in
n
x < for any n .
ii) ( )
p
x t has only a finite number of maxima and minima over any period
0
T .
iii) There are only finite number of finite discontinuities over any period
1
.

Let ( )
M
j nf t
M n
n M
x t x e
0
2
=
=

(1.7)

1
For examples of periodic signals that do not satisfy one or more of the conditions i) to iii), the
reader is referred to [1, 2] listed at the end of this chapter.

1.20
and ( ) ( ) ( )
M p M
e t x t x t =
then ( )
M
M
x t lim

converges uniformly to ( )
p
x t wherever ( )
p
x t is continuous;
that is ( )
M
M
e t lim 0

= for all t.
Dirichlet conditions are sufficient but not necessary. Later on, we shall
have examples of Fourier series for functions that voilate some of the Dirichet
conditions.

If ( )
p
x t is not absolutely integrable but square integrable, that is,
( )
p
T
x t dt
0
2
<
, then the series converges in the mean. That is,

( )
M
M
T
e t d t
0
2
lim 0

=
(1.8)

Note that Eq. 1.8 does not imply that ( )
M
M
e t lim
is zero. There could be

nonzero values in ( )
M
M
e t lim
; but they occur at isolated points, resulting in the

integral of Eq. 1.8, being equal to zero.

The limiting behavior of ( )
M
x t at the points of discontinuity in ( )
p
x t is
somewhat interesting, regardless of ( )
p
x t being absolutely integrable or square
integrable. This is illustrated in Fig. 1.5(a). From the figure, we see that


1.21

Fig. 1.5(a): Convergence behavior of ( )
M
x t at a discontinuity in ( )
p
x t

( )
M
x t passes through the mid-point of the discontinuity and has a peak
overshoot (as well as undershoot) of amplitude 0.09A (We assume M to be
sufficiently large). The period of oscillations (whose amplitudes keep decreasing
with increasing t , 0 t > ) is
0
2
T
M
. These oscillations (with the peak overshoot as
well as the undershoot of amplitude 0. 09 A) persist even as M . In the
limiting case, all the oscillations converge in location to the point t t
1
= (the point
of discontinuity) resulting in what is called as Gibbs ears as shown in Fig. 1.5(b).

Fig. 1.5(b): Gibbs ears at t =t
1

1.22
(In 1898, Albert Michelson, a well-known name in the field of optics,
developed an instrument called Harmonic Analyzer (HA), which was capable of
computing the first 80 coefficients of the Fourier series. HA could also be used a
signal synthesizer. In other words, it has the ability to self-check its calculations
by synthesizing the signal using the computed coefficients. When Michelson tried
this instrument on signals with discontinuities (with continuous signals, close
agreement was found between the original signal and the synthesized signal), he
observed a strange behavior: synthesized signal, based on the 80 coefficients,
exhibited ringing with an overshoot of about 9% of the discontinuity, in the vicinity
of the discontinuity. This behavior persisted even after increasing the number of
terms beyond 80. J . W. Gibbs, professor at Yale, investigated and clarified the
above behavior by taking the saw-tooth wave as an example; hence the name
Gibbs Phenomenon.)

The convergence of the Fourier series and the corresponding
Gibbs oscillations can be seen from the animation that follows. You have been
provided with three options with respect to the number of harmonics M to be
summed. These are: M 10, 25 and 75 = .

1.3 Aperiodic Signals and Fourier Transform
Aperiodic (also called nonperiodic) signals can be of finite or infinite
duration. A few of the aperiodic signals occur quote often in theoretical studies.
Hence, it behooves us to introduce some notation to describe their behavior.
i) Rectangular pulse,
t
ga
T

1,
2
0,
2
T
t
t
ga
T T
t
<

=

>
(1.9a)

1.23
[Rectangular pulse is sometimes referred to as a gate pulse; hence the
symbol ( ) ga ]. In view of the above
t
A ga
T

refers to a rectangular pulse
of amplitudeA and duration T , centered at 0 t = .
ii) Triangular pulse,
t
tri
T

t
t t T
tri
T
T
outside
1 ,
0 ,
(1.9b)
iii) One-sided (decaying) exponential pulse, 1
t
ex
T

:

t
T
e t
t
ex t
T
t
, 0
1
1 , 0
2
0 , 0
>

= =

<
(1.9c)
iv) Two-sided (symmetrical) exponential pulse, 2
t
ex
T

:
t
T
t
T
e t
t
ex t
T
e t
, 0
2 1 , 0
, 0
>

= =
<
(1.9d)
Fig. 1.6 illustrates specific examples of these pulses.


1.24

Fig. 1.6: Examples of pulses defined by Eq. 1.9

Let ( ) x t be any aperiodic signal. We define its normalized energy
x
E , as
( )
x
E x t dt
2
=

(1.10)
An aperiodic signal with
x
E 0 < < is said to be an energy signal.
(When no specific signal is being referred to, we use the symbol E without any
subscript to denote the energy quantity.)

1.3.1 Fourier transform
Like periodic signals, aperiodic signals also can be represented in the
frequency domain. However, unlike the discrete spectrum of the periodic case,
we have a continuous spectrum for the aperiodic case; that is, the frequency
components constituting a given signal ( ) x t lie in a continuous range (or
ranges), and quite often this range could be ( ) , . Eq. 1.2(a) expresses ( )
p
x t
as a sum over a discrete set of frequencies. Its counterpart for the aperiodic case
is an integral relationship given by

1.25
( ) ( )
j f t
x t X f e d f
2
=

(1.11a)
where ( ) X f is the Fourier transform of ( ) x t .

Eq. 1.11(a) is given the following interpretation. Let the integral be treated
as a sum over incremental frequency ranges of width f . Let ( )
i
X f f be the
incremental complex amplitude of
2
i
j f t
e

at the frequency
i
f f = . If we sum a
large number of such complex exponentials, the resulting signal should be a very
good approximation to ( ) x t . This argument, carried to its natural conclusion,
leads to signal representation with a sum of complex exponentials replaced by an
integral, where a continuous range of frequencies, with the appropriate complex
amplitude distribution will synthesize the given signal ( ) x t .

Eq. 1.11(a) is called the synthesis relation or Inverse Fourier Transform
(IFT) relation. Quite often, we know ( ) x t and would want ( ) X f . The companion
relation to Eq. 1.11(a) is
( ) ( )
2 j f t
X f x t e dt

=

(1.11b)
Eq. 1.11(b) is referred to as the Fourier Transform (FT) relation or, the
analysis equation, or forward transform relation. We use the notation
( ) ( ) X f F x t =

(1.12a)
( ) ( )
1
x t F X f
=

(1.12b)

Eq. 1.12(a) and Eq. 1.12(b) are combined into the abbreviated notation, namely,
( ) ( ) x t X f . (1.12c)
( ) X f is, in general, a complex quantity. That is,
( ) ( ) ( )
R I
X f X f j X f = +

1.26
( )
( ) j f
X f e

=
( ) ( )
R
X f Real part of X f = ,
( ) ( )
I
X f Imaginary part of X f =
( ) ( ) X f magnitude of X f =
( ) ( )
2 2
R I
X f X f = +
( ) ( )
( )
( )
arg tan
I
R
X f
f X f arc
X f

= =

Information in ( ) X f is usually displayed by means of two plots: (a)
( ) . X f vs f , known as magnitude spectrum and (b) ( ) . f vs f , known as the
phase spectrum.

Example 1.2
Let ( )
t
x t A ga
T

=

. Let us compute and sketch ( ) X f .

( )
2
2
2
sin ( )
T
j ft
T
X f A e dt AT c f T

= =
,
where
( )
( )
c
sin
sin

=

Appendix A1.1 contains the tabulated values of ( ) c sin . Its behavior is shown in
Fig. A1.1. Note that ( ) c
1, 0
sin
0, 1, 2 etc.
=
=

=

Fig. 1.7(a) sketches the magnitude spectrum of the rectangular pulse for 1 AT = .


1.27

Fig. 1.7: Spectrum of the rectangular pulse

Regarding the phase plot, ( ) sinc f T is real. However it could be bipolar.
During the interval,
1 m m
f
T T
+
< < , with m odd, ( ) sinc f T is negative. As the
magnitude spectrum is always positive, negative values of ( ) sinc f T are taken
care of by making ( )
0
180 f = for the appropriate ranges, as shown in Fig.
1.7(b).
Remarkable balancing act: A serious look at the magnitude and phase plots
reveals a very charming result. From the magnitude spectrum, we find that a
rectangular pulse is composed of frequency components in the range
f < < , each with its own amplitude and phase. Each of these complex
exponentials exist for all t . But when we synthesize a signal using the complex
exponentials with the magnitudes and phases as given in Fig. 1.7, they add up to

1.28
a constant for
2
T
t < and then go to zero forever. A very fascinating result
indeed!

Fig. 1.7(a) illustrates another interesting result. From the figure, we see
that most of the energy, (that is, the range of strong spectral components) of the
signal lies in the interval
1
f
T
< , where T is the duration of the rectangular pulse.
Hence, if T is reduced, then its spectral width increases and vice versa (As we
shall see later, this is true of other pulse types, other than the rectangular). That
is, more compact is the signal in the time-domain, the more wide-spread it would
be in the frequency domain and vice versa. This is called the phenomenon of
reciprocal spreading.

Example 1.3
Let ( ) ( ) 1 x t ex t = . Let us find ( ) X f and sketch it.

( )
2
0
1
1 2
t j f t
X f e e dt
j f

= =
+

Hence, ( )
( )
2
1
1 2
X f
f
=
+

( ) ( ) tan 2 f arc f =
A plot of the magnitude and the phase spectrum are given in Fig. 1.8.


1.29

Fig. 1.8: (a) Magnitude spectrum of the decaying exponential
(b) Phase spectrum

1.3.2 Dirichlet conditions
Given ( ) X f , Eq. 1.11(a) enables us to synthesize the signal ( ) x t . Now
the question is: Is the synthesized signal, say
( )
( )
s
x t , identical to ( ) x t ? This
leads to the topic of convergence of the Fourier Integral. Analogous to the
Dirichlet conditions for the Fourier series, we have a set of sufficient conditions,
(also called Dirichlet conditions) for the existence of Fourier transform, which are
stated below:

1.30
(i) ( ) x t be absolutely integrable; that is,
( ) x t dt

<

This ensures that ( ) X f is finite for all f , because
( ) ( )
2 j f t
X f x t e dt

=

( ) ( )
2 j f t
X f x t e dt

=

( ) ( )
2 j f t
x t e dt x t dt

= <

(ii) ( ) x t is single valued and has only finite number of maxima and minima
with in any finite interval.
(iii) ( ) x t has a finite number of finite discontinuities with in any finite interval.

If
( )
( ) ( )
2
lim
W
s j f t
W
W
x t X f e df
=

, then
( )
( )
s
x t converges to ( ) x t
uniformly wherever ( ) x t is continuous.

If ( ) x t is not absolutely integrable but square integrable, that is,
( )
2
x t dt

<
(Finite energy signal), then we have the convergence in the

mean, namely
( )
( )
( )
2
0
s
x t x t dt


1.31
Regardless of whether ( ) x t is absolutely integrable or square integrable,
( )
( )
s
x t exhibits Gibbs phenomenon at the points of discontinuity in ( ) x t , always
passing through the midpoints of the discontinuities.

1.4 Properties of the Fourier Transform
Fourier Transform has a large number of properties, which are developed
in the sequel. A thorough understanding of these properties, and the ability to
make use of them appropriately, helps a great deal in the analysis of various
signals and systems.

P1) Linearity
Let ( ) ( )
1 1
x t X f and ( ) ( )
2 2
x t X f .
Then, for all constants
1
a and
2
a , we have
( ) ( ) ( ) ( )
1 1 2 2 1 1 2 2
a x t a x t a X f a X f + +
It is very easy to see the validity of the above transform relationship. This
property will be used quite often in the development of this course material.

P2a) Area under ( ) x t
If ( ) ( ) x t X f , then
( ) ( ) 0 x t dt X

The above property follows quite simply by setting 0 f = in Eq. 1.11(b). As an
example of this property, we have the transform pair
( ) sin
t
ga T c f T
T

By inspection, area of the time function is T , which is equal to ( )
0
sin |
f
T c f T
=
.
P2b) Area under ( ) X f

1.32
If ( ) ( ) x t X f , then ( ) ( ) x X f df 0

=

The proof follows by making 0 t = in Eq. 1.11(a).
As an illustration of this property, we have
( ) ( ) ( )
1
1
1 2
x t ex t X f
j f
= =
+

Hence ( )
( )
2
1 1 2
1 2
1 2
j f
X f df df df
j f
f

= =
+
+

Noting that
( )
2
2
0
1 2
f
df
f
=
+
, we have
( )
( )
2
1 1
2
1 2
X f df df
f

= =
+

, which is the value of ( )
0
1 |
t
ex t
=

P3) Time Scaling
If ( ) ( ) x t X f , then ( )
1 f
x t X

, where is a real
constant.
Proof: Exercise

The value of decides the behavior of ( ) x t . If 1 = , ( ) x t is a time
reversed version of ( ) x t . If 1 > , ( ) x t is a time compressed version of ( ) x t ,
where as if 0 1 < < , we have a time expanded version of ( ) x t . Let ( ) x t be as
shown in Fig. 1.9(a). Then ( ) 2 x t would be as shown in Fig. 1.9(b).


1.33

Fig. 1.9: A triangular pulse and its time compressed and reversed version

For the special case 1 = , we have the transform pair
( ) ( ) x t X f . That is, both the time function and its Fourier transform
undergo reversal. As an example, we know that
( )
1
1
1 2
ex t
j f
+

Hence ( )
1
1
1 2
ex t
j f

( ) ( ) ( ) 1 1 2 ex t ex t ex t + =

Using the linearity property of the Fourier transform, we obtain the
transform pair
( ) ( )
1 1
2 exp
1 2 1 2
ex t t
j f j f
= +
+

( )
2
2
1 2 f
=
+

Consider ( ) x t with 2 = . Then,
( )
1
2
2 2
f
x t X


1.34
Let us compare ( ) X f with
1
2 2
f
X

by taking ( ) ( ) 1 x t ex t = , and
( ) ( ) 1 2 y t ex t = . That is,
( )
2
, 0
1
, 0
2
0 , 0
t
e t
y t t
t
>
= =
<

( )
( )
1 1
2 1 2 2
Y f
j f

=

+

1
2 2 j f
=
+

Let ( ) ( )
( )
x
f
X f X f e
= and
( ) ( )
( )
y
f
Y f Y f e
= , where
( )
2 2
1
2 1
Y f
f
=
+

( ) ( ) tan
y
f arc f =

Fig. 1.10 gives the plots of ( ) X f and ( ) Y f . In Fig. 1.10(a), we have the
plots ( ) . X f vs f and ( ) . Y f vs f . In Fig. 1.10(b), we have the plot of ( ) Y f
normalized to have the maximum value of unity. This plot is denoted by ( )
N
Y f .
Fig. 1.10(c) gives the plots of ( )
x
f and ( )
y
f . From Fig. 1.10(b), we see that
i) ( ) ( ) 2 y t x t = has a much wider spectral width as compared to the spectrum
of the original signal. (In fact, if ( ) X f is band limited to W Hz, then
2
f
X

will be band limited to W 2 Hz.)


1.35

Fig 1.10: Spectral plots of ( ) 1 ex t and ( ) 1 2 ex t

1.36
(ii) Let ( ) ( ) ( ) E f X f Y f = . Then the value of ( ) E f is dependent on f ; that
is, the original spectral magnitudes are modified by different amounts at
different frequencies (Note that ( ) ( ) Y f k X f where k is a constant).
In other words, ( ) Y f is a distorted version of ( ) X f .
(iii) From Fig. 1.10(c), we observe that ( ) ( )
y x
f f and their difference is a
function of frequency; that is ( )
y
f is a distorted version of ( )
x
f .

In summary, time compression would result either in the introduction of
new, higher frequency components (if the original signal is band limited) or
making the latter part of the original spectrum much more significant; the
remaining spectral components are distorted (both in amplitude and phase). On
the other hand, time expansion would result either in the loss or attenuation of
higher frequency components, and distortion of the remaining spectrum.

Let ( ) x t represent an audio signal band limited to 10 kHz. Then ( ) 2 x t will
have a spectral components upto 20 kHz. These higher frequency components
will impart shrillness to the audio, besides distorting the original signal. Similarly,
if the audio is compressed, we have loss of sharpness in the resulting signal
and severe distortion. This property of the FT will now be demonstrated with the
help of a recorded audio signal.

P4a) Time shift
If ( ) ( ) x t X f then,
( ) ( )
0
2
0
j f t
x t t e X f

If
0
t is positive, then ( )
0
x t t is a delayed version of ( ) x t and if
0
t is
negative, the ( )
0
x t t is an advanced version of ( ) x t . In any case, time shifting

1.37
will simply result in the multiplication of ( ) X f by a linear phase factor. This
implies that ( ) x t and ( )
0
x t t have the same magnitude spectrum.
Proof: Let ( )
0
t t = . Then,
( ) ( )
( )
0
2
0
j f t
F x t t x e d
+

=

( )
0
2 2 j f t j f
e x e d

( )
0
2 j f t
e X f

=

P4b) Frequency Shift
( ) ( )
2
c
j f t
c
e x t X f f

where
c
f is a real constant. (This property is also known as modulation theorem).
Proof: Exercise
As an application of the above result, let us consider the spectrum of
( ) ( ) ( ) 2cos 2
c
y t f t x t = ; that is, we want the Fourier transform of
( )
2 2
c c
j f t j f t
e e x t

+

. From the frequency shift theorem, we have
( ) ( ) ( ) ( ) ( ) ( ) 2cos 2
c c c
y t f t x t Y f X f f X f f = = + + . If ( ) X f is as shown
in Fig. 1.11(a), then ( ) Y f will be as shown in Fig. 1.11(b) for
c
f W = .

Fig.1.11: Illustration of modulation theorem

1.38
P5) Duality
If we look fairly closely at the two equations constituting the Fourier
transform pair, we find that there is a great deal of similarity between them. In
Eq.1.11a, ' ' f is the variable of integration where as in Eq. 1.11b, it is the variable
' ' t . The sign of the exponent is positive in Eq. 1.11a where as it is negative in
Eq. 1.11b. Both t and f are variables of the continuous type. This results in an
interesting property, namely, the duality property, which is stated below.

( ) ( ) X t x f and ( ) ( ) X t x f
Note: This is one instance, where the variable t is associated with a function
denoted using the upper case letter and the variable f is associated with a
function denoted using a lower case letter.
Proof: ( ) ( )
2 j f t
x t X f e df

=

( ) ( )
2 j f t
x t X f e df

=

The result follows by interchanging the variables t and f . The proof of the
second part of the property is similar.

Duality theorem helps us in creating additional transform pairs, from the
given set. We shall illustrate the duality property with the help of a few examples.

Example 1.4
If ( ) sin 2 z t A c W t = , let us use duality to find ( ) Z f .


1.39
We look for a transform pair, ( ) ( ) x t X f where in ( ) X f , if we
replace f by t , we have ( ) sin 2 z t A c W t = .
We know that if ( )
2 2
A t
x t ga
W W

=

, then,
( ) ( ) sin 2 X f A c Wf = and
( ) ( ) ( ) sin 2 X t A c Wt z t = =
( ) ( )
2 2
A f
Z f x f ga
W W

= =

As ( ) ( ) ga f ga f = , we have
( )
2 2
A f
Z f ga
W W

=

.

Example 1.5
Find the Fourier transform of ( )
2
2
1
z t
t
=
+
.

We know that if ( )
t
y t e
= , then ( )
( )
2
2
1 2
Y f
f
=
+

Let ( ) ( ) x t y t = , with 2 = .
Then ( )
1
2 2
f
X f Y

=

2
1 2
2 1 f
=
+

or ( )
2
2
2
1
X f
f
=
+

As ( )
2
2
1
z t
f
=
+
with f being replaced by t , we have
( ) ( ) 2 Z f x f =

2
2
f
e

=

1.40
Hence
2
2
2
2
1
f
e
t

+

P6) Conjugate functions
If ( ) ( ) x t X f , then ( ) ( ) x t X f

Proof: ( ) ( )
2 j f t
X f x t e dt

=

( ) ( )
2 j f t
X f x t e dt

=

( ) ( )
2 j f t
X f x t e dt

=

Hence the result.

From the time reversal property, we get the additional relation, namely
( ) ( ) x t X f

Def. 1.3(a): A signal ( ) x t is called conjugate symmetric, if ( ) ( ) x t x t
= .
If ( ) x t is real, then ( ) x t is even if ( ) ( ) x t x t = .
Exercise 1.2: Let ( ) ( ) ( )
R I
x t x t j x t = +
where ( )
R
x t is the real part and ( )
I
x t is the imaginary part of ( ) x t . Show
that

( ) ( ) ( )
1
2
R
x t X f X f

+

( ) ( ) ( )
1
2
I
x t X f X f
j


1.41
Def. 1.3(b): A signal ( ) x t is said to be conjugate anti-symmetric if
( ) ( ) x t x t
= .
If ( ) x t is real, then ( ) x t is odd if ( ) ( ) x t x t = .

If ( ) x t is real, then ( ) ( ) x t x t
= .
As a result, ( ) ( ) X f X f
= or ( ) ( ) X f X f
= .
Hence, the spectrum for the negative frequencies is the complex conjugate of the
positive part of the spectrum. This implies, that for real signals,
( ) ( ) X f X f =
( ) ( ) f f =

Going one step ahead, we can show that if ( ) x t is real and even, then
( ) X f is also real and even. (Example:
( )
2
2
1 2
t
e
f
+
). Similarly, if ( ) x t
is real and odd, its transform is purely imaginary and odd (See Example 1.7).

P7a) Multiplication in the time domain
If ( ) ( )
1 1
x t X f
( ) ( )
2 2
x t X f
then, ( ) ( ) ( ) ( ) ( ) ( )
1 2 1 2 2 1
x t x t X X f d X X f d

=

The integrals on the R.H.S represent the convolution of ( )
1
X f and ( )
2
X f . We
denote the convolution of ( )
1
X f and ( )
2
X f by ( ) ( )
1 2
X f X f .
(Note that in between two functions represents the convolution of the two
quantities where as a superscript, it denotes the complex conjugate)

1.42
Proof: Exercise

P7b) Multiplication of Fourier transforms
Let ( ) ( )
1 1
x t X f
( ) ( )
2 2
x t X f
then, ( ) ( ) ( ) ( )
1
1 2 1 2
F X f X f x x t d

=

( ) ( )
2 1
x x t d

As any one of the above integrals represent the convolution of ( )
1
x t and ( )
2
x t ,
we have
( ) ( ) ( ) ( )
1 2 1 2
x t x t X f X f
Proof: Let ( ) ( ) ( )
3 1 2
x t x t x t =
That is, ( ) ( ) ( )
3 1 2
x t x x t d

( ) ( ) ( ) ( ) ( )
j f t j f t
X f F x t x t e dt x x t d e dt
2 2
3 3 3 1 2

= = =

Rearranging the integrals,
( ) ( ) ( )
j f t
X f x x t e dt d
2
3 1 2

=

But the bracketed quantity is the Fourier transform of ( )
2
x t . From the
property P4(a), we have
( ) ( )
j f
x t e X f
2
2 2

Hence,

1.43
( ) ( ) ( )
2
3 1 2
j f
X f x X f e d

( ) ( )
2
2 1
j f
X f x e d

( ) ( )
1 2
X f X f =

Property P7(b), known as the Convolution theorem, is one of the very useful
properties of the Fourier transform.

The concept of convolution is very basic in the theory of signals and
systems. As will be shown later, the input and output of a linear, time- invariant
system are related by the convolution integral. For a fairly detailed treatment of
the properties of systems, convolution integral etc. the student is advised to refer
to [1 - 3].

Example 1.6
In this example, we will find the Fourier transform of
t
T tri
T

.

t
T tri
T

can be obtained as the convolution of
t
ga
T

with itself. That is,
t t t
ga ga T tri
T T T

=

As ( ) sin
t
ga T c f T
T

, we have
( )
2
sin
t
T tri T c f T
T


1.44
P8) Differentiation
P8a) Differentiation in the time domain
Let ( ) ( ) x t X f ,
then, ( ) ( ) 2
d
x t j f X f
dt

Generalizing,
( )
( ) ( ) 2
n
n
n
d x t
j f X f
dt

Proof: We shall prove the first part; generalization follows as a consequence
this. We have,
( ) ( )
2 j f t
x t X f e df

=

( ) ( )
2 j f t
d d
x t X f e df
dt dt

Interchanging the order of differentiation and integration on the RHS,
( ) ( )
2 j f t
d d
x t X f e df
dt dt

( )
2
2
j f t
j f X f e df

=

From the above, we see that ( )
1
2 F j f X f

is ( )
d
x t
dt
. Hence the property.

Example 1.7
Let us find the FT of the doublet pulse ( ) x t shown in Fig. 1.12 below.


1.45

Fig 1.12: ( ) x t of Example 1.7

( )
d t
x t T tri
dt T

=

. Hence,
( ) 2
t
X f j f F T tri
T

=

( ) ( )
2 2
2 sin j f T c f T = , (using the result of example 1.6)
( )
( )
( )( )
2
2
sin
2
f T
j f T
f T f T
=

( )
( )
( )
sin
2 sin
f T
j T f T
f T

( ) ( ) 2 sin sin j T c f T f T =

As a consequence of property P8(a), we have the following interesting result.

Let ( ) ( ) ( )
3 1 2
x t x t x t =
Then ( ) ( ) ( )
3 1 2
X f X f X f =
( ) ( ) ( ) ( ) ( )
3 1 2
2 2 j f X f j f X f X f =

( ) ( )
1 2
2 X f j f X f =

That is,

1.46
( ) ( ) ( ) ( )
1 2 1 2
d d
x t x t x t x t
dt dt
=

( ) ( )
1 2
d
x t x t
dt
=

P8b) Differentiation in the frequency domain
Let ( ) ( ) x t X f .
Then, ( ) ( )
( )
2
d X f
j t x t
df

Generalizing, ( ) ( )
( )
2
n
n
n
d X f
j t x t
df

Proof: Exercise
The generalized property can also be written as
( )
( )
2
n
n
n
n
d X f
j
t x t
df

Example 1.8
Find the Fourier transform of ( ) 1
t
x t t ex
T

=

.

We have, ( )
1
1
1 2
ex t
j f
+

Hence,
1
1
1 2
t
ex T
T j f T

+

1
2 1 2
t j d T
t ex
T df j f T

+

( )
( )
2
2
2
1 2
j T
j
T
j f T

+


1.47

( )
2
2
1 2
T
j f T

+

P9) Integration in time domain
This property will be developed subsequently.

P10) Rayleighs energy theorem
This theorem states that,
x
E , energy of the signal ( ) x t , is
( )
x
E X f df
2

=

This result follows from the more general relationship, namely,
( ) ( ) ( ) ( )
1 2 1 2
x t x t dt X f X f df

=

Proof: We have
( ) ( ) ( ) ( )
2
1 2 1 2
j f t
x t x t dt x t X f e df dt

=

( ) ( )
2
2 1
j f t
X f x t e dt df

=

( ) ( )
2 1
X f X f df

=

If ( ) ( ) ( )
1 2
x t x t x t = = , then
( ) ( )
x
x t dt E X f df
2 2

= =

Note: If ( )
1
x t and ( )
2
x t are real, then,

1.48
( ) ( ) ( ) ( ) ( ) ( )
1 2 1 2 2 1
x t x t d t X f X f d f X f X f d f

= =

Property P10 enables us to compute the energy of a signal from its
magnitude spectrum. In a few situations, this may be easier than computing the
energy in the time domain. (Some authors refer to this result as Parsevals
theorem)

Example 1.9
Let us find the energy of the signal ( ) ( ) 2 sin 2 x t AW c W t = .

( )
x
E AW c W t d t
2
2 sin 2

=

In this case, it would be easier to compute
x
E based on ( ) X f . From Example
1.4, ( )
2
f
X f A ga
W

=

. Hence,

W
x
W
E A df W A
2 2
2
= =

More important than the calculation of the energy of the signal, Rayleighs
energy theorem enables to treat ( )
2
X f as the energy spectral density of ( ) x t .
That is, ( )
2
1
X f df is the energy in the incremental frequency interval d f ,
centered at
1
f f = . Let ( )
W
x
W
X f df E
2
0.9
. Then, 90 percent of the energy of

signal is confined to the interval f W . Consider the rectangular pulse
t
ga
T

.
The first nulls of the magnitude spectrum occur at
1
f
T
= . The evaluation of

1.49
( ) ( )
1 1
2
2
1 1
sin
T T
T T
X f df T c f T df

=

will yield the value 0.92T , which is 92
percent of the total energy of
t
ga
T

. Hence, the frequency range
1 1
,
T T

can
be taken to be the spectral width of the rectangular pulse. [The interval
2 2
,
T T

may result in about 95 percent of the total energy].

1. 5 Unified Approach to Fourier Transform
So far, we have represented the periodic functions by Fourier series and
the aperiodic functions by Fourier transform. The question arises: is it possible to
unify these two approaches and talk only in terms of say, Fourier transform? The
answer is yes provided we are willing to introduce Impulse Functions both in
time and frequency domains. This would also enable us to have Fourier
transforms for signals that do not satisfy one or more of the Dirichlets conditions
(for the existence of the Fourier transform).

1.5.1 Unit impulse (Dirac delta function)
Impulse function is not a function in its strict sense [Note that a function
( ) f , takes a number y and a produces another number, ( ) f y ]. It is a
distribution or generalized function. A distribution is defined in terms of its effect
on another function. The symbol ( ) t is fairly common in the technical literature
to denote the unit impulse. We define the unit impulse as any (generalized)
function that satisfies the following conditions:
(i) ( ) 0, 0 t t = (1.13a)
(ii) ( ) t t , 0 = = (1.13b)
(iii) Let ( ) p t be any ordinary function, then

1.50
( ) ( ) ( ) ( ) ( ) 0 , 0 p t t dt p t t dt p

= = >

(1.13c)
( could be infinitesimally small)

If ( ) p t t 1, for = , then we have
( ) ( ) t d t t d t 1

= =

(1.13d)

From Eq. 1.13(c), we see that ( ) t operates on a function such as ( ) p t
and produces the number, namely, ( ) 0 p . As such ( ) t falls between a function
and a transform (A transform operates on a function and produces a function).

A number of conventional functions have a limiting behavior that
approaches ( ) t . We cite a few such functions below:
Let
(a) ( )
1
1 t
p t ga

=

(b) ( )
2
1 t
p t tri

=

(c) ( )
3
1
sin
t
p t c

=

Then, ( ) ( )
0
lim
i
p t t
= , 1, 2, 3 i = . ( )
3
p t is shown below in Fig. 1.13.


1.51

Fig. 1.13: ( ) sinc with the limiting behavior of ( ) t

( )
t
c ga f
1
Note that sin . Hence the area under the time function 1.

=

From the above examples, we see that the shape of the function is not
very critical; its area should remain at 1 in order to approach ( ) t in the limit.

By delaying ( ) t by
0
t and scaling it by A, we have ( )
0
A t t . This is
normally shown as a spear (Fig. 1.14) with the weight or area of the impulse
shown in parentheses very close to it.


1.52

Fig. 1.14: Symbol for ( )
0
A t t

Some properties of unit impulse
P1) Sampling (or sifting) property
Let ( ) p t be any ordinary function. Then for
0
a t b < < ,
( ) ( ) ( )
0 0
b
a
p t t t dt p t =

(This is generalization of condition (iii)). Proof follows from making the
change of variable
0
t t = and noting ( ) is zero for 0 . Note that
for the sampling property, the values of ( ) p t ,
0
t t are of no
consequence.
P2) Replication property
Let ( ) p t be any ordinary function. Then,
( ) ( ) ( )
0 0
p t t t p t t =

The proof of this property follows from the fact, that in the process of
convolution, every value of ( ) p t will be sampled and shifted by
0
t
resulting in ( )
0
p t t .
(Note: Some authors use this property as the operational definition of
impulse function.)


1.53
P3) Scaling Property
( ) ( )
1
, 0 t t =

Proof: (i) Let , 0 t = >
( ) ( )
1 1
t dt d

= =

( )
1
t dt

( )
1
t dt

(ii) Let 0 < ; that is = , and let t = .
( ) ( )
1 1
t dt d

= =

( )
1
t dt

It is easy to show that ( ) ( )
0 0
1
t t t t =

.

Special Case: If 1 = , we have the result ( ) ( ) t t = .
The above result is not surprising, especially if we look at the examples
( )
1
p t to ( )
3
p t , which are all even functions of t . Hence some authors call this
as the even sided delta function. It is also possible to come up with delta
functions as a limiting case of functions that are not even; that is, as a limiting
case of one-sided functions. In such a situation we have a left-sided delta
function or right-sided delta function etc. Left-sided delta function will prove to be
useful in the context of probability density functions of certain random variables,
subject matter of chapter 2.


1.54
Example 1.10
Find the value of
(a) ( )
4
3
4
5 t t dt

(b) ( )
5.1
3
4.9
5 t t dt

(a) ( ) 5 t is nonzero only at 5 t = . The range of integration does not include
the impulse. Hence the integral is zero.
(b) As the range of integration includes the impulse, we have a nonzero value for
the product ( )
3
5 t t . As ( ) 5 t occurs at 5 t = , we can write
( ) ( )
3 3
5 5 5 t t t = .
Hence,
( ) ( )
5.1 5.1
3 3
4.9 4.9
5 5 5 125 t t dt t dt = =

Example 1.11
Let ( )
4
t
p t tri

=

. Find ( ) ( ) 2 1 p t t

.

( )
1 1 1
2 1 2
2 2 2
t t t

= =

( )
1 1 1 1 1 12
2 2 2 2 2 4
t
p t t p t tri

= =

Let us now compute the Fourier transform of ( ) t . From Eq. 1.11(b), we
have,
( ) ( )
2
1
j f t
F t t e dt
= =

(1.14a)

1.55
How do we interpret this result? The spectrum of the unit impulse consists
of frequency components in the range ( ) , , all with unity magnitude and
zero phase shift, a fascinating result indeed! Hence exciting any electric network
or system with a unit impulse is equivalent to exciting the network simultaneously
with complex exponentials of all possible frequencies, all with the same
magnitude (unity in this case) and zero phase shift. That is, the unit impulse
response of a linear network is the synthesis of responses to the individual
complex exponentials and we intuitively feel that the impulse response of a
network should be able to characterize the system in the time domain. (We shall
see a little later that if the network is linear and time invariant, a simple relation
exists between the input to the network, its impulse response and the output).

The dual of the Fourier transform pair of Eq. 1.14(a) gives us
( ) ( ) 1 f f = (1.14b)

Based on Eq. 1.14(a) and Eq. 1.14(b), we make the following observation:
a constant in one domain will transform into an impulse in the other domain.

Eq. 1.14(b) is intuitively satisfying; a constant signal has no time variations
and hence its spectral content ought to be confined to 0 f = ; ( ) f is the proper
quantity for the transform because it is zero for 0 f and its inverse transform
yields the required constant in time (note that only an impulse can yield a
nonzero value when integrated over zero width).
Because of the transform pair,
( ) 1 f ,
we obtain another transform pair (from modulation theorem)
( )
0
2
0
j f t
e f f
(1.15a)
( )
0
2
0
j f t
e f f

+ (1.15b)

1.56
As
0 0
2 2
0
cos
2
j f t j f t
e e
t

+
= , we have,
( ) ( )
0 0 0
1
cos
2
t f f f f + +

(1.16)

Similarly,
( ) ( )
0 0 0
1
sin
2
t f f f f
j
+

(1.17)

0
cos F t

and
[ ]
0
sin F t are shown in Fig. 1.15.

Fig. 1.15: Fourier transforms of (a)
0
cos t and (b)
0
sin t

Note that the impulses in Fig. 1.15(b) have weights that are complex. It is
fairly conventional to show the spectrum of
0
sin t as depicted in Fig. 1.15(b); or
else we can make two separate plots, one for magnitude and the other for phase,
where the magnitude plot is identical to that shown in Fig. 1.15(a) and the phase
plot has values of
2
at
0
f f = and
2
+ at
0
f f = .

In summary, we have found the Fourier transform of ( ) t (a time function
with a discontinuity that is not finite), and using impulses in the frequency
domain, we have developed the Fourier transforms of the periodic signals such

1.57
as
0
2 j f t
e

,
0
cos t and
0
sin t , which are neither absolutely integrable nor
square integrable.

We are now in a position to present both Fourier series and Fourier
transform in a unified framework and talk only of Fourier transform whether the
signal is aperiodic or not. This is because, for a periodic signal ( )
p
x t , we have
the Fourier series relation,
( )
0
2 j nf t
p n
n
x t x e
=
=

Taking the Fourier transform on both the sides,
( ) ( )
0
2 j nf t
p p n
n
F x t X f F x e
=

= =

0
2 j nf t
n
n
x F e
=

=

( )
0 n
n
x f nf
=
=
(1.18)

FT of ( )
p
x t is a function of the continuous variable f , whereas, in the FS
representation of ( )
p
x t ,
n
x is a function of the discrete variable n . However, as
( )
p
X f is purely impulsive, spectral components exist only at
0
f nf = , with
complex weights
n
x . As inversion of ( )
p
X f requires integration, we require
impulses in the spectrum. As such, the differences between the line spectrum of
sec. 1.1 and spectral representation given by Eq. 1.18 are only minor in nature.
They both provide the same information, differing essentially only in notation.

There is an interesting relation between
n
x and the Fourier transform of
one period of a periodic signal. Let,

1.58
( )
( )
0 0
,
2 2
0 ,
p
T T
x t t
x t
outside
< <

( )
0
0
0
2
2
0
2
1
T
j nf t
n p
T
x x t e dt
T

=

( )
0
0
0
2
2
0
2
1
T
j nf t
T
x t e dt
T

( ) ( )
0 0
2
2
0 0
1 1
n
j t
T j nf t
x t e dt x t e dt
T T

= =

The bracketed quantity is ( )
0
n
f
T
X f
=

Hence,
0 0
1
n
n
x X
T T

=

(1.19)

Example 1.12
Find the Fourier transform of the uniform impulse train
( ) ( )
0 p
n
x t t nT
=
=
shown in Fig 1.16 below.

Fig. 1.16: Uniform impulse train


1.59
Let ( ) x t be one period of ( )
p
x t in the interval,
0 0
2 2
T T
t < < . Then,
( ) ( ) x t t = for this example. But as ( ) 1 t , from Eq. (1.19) we have,
0
1
n
x
T
= for all n . Hence,
( ) ( )
0
0
1
p
n
X f f nf
T
=
=
(1.20)
From Eq. 1.20, we have another interesting result:
A uniform periodic impulse train in either domain will transform into
another uniform impulse train in the other domain.

From the transform pair, ( ) 1 t , we have
[ ] ( )
1 2
1
j f t
F e df t

= =

As ( ) ( ) t t = , we have, ( )
2 j f t
e df t

That is, ( )
2 j f t
e df t
(1.21)
Using Eq. 1.21, we show that ( ) x t and ( ) X f constitute a transform pair.
Let ( ) ( )
2
j f t
x t X f e df

=

( )
2 2 j f j f t
x e d e df

=

( )
( ) 2 j f t
x e df d

=

( )
( ) 2 j f t
x e df d

=


1.60
( ) ( ) ( ) x t d x t

= =

As ( ) ( )
x t x t = , we have ( ) X f uniquely representing ( ) x t .

1.5.2 Impulse response and convolution
Let ( ) x t be the input to a Linear, Time-Invariant (LTI) system resulting in
the output, ( ) y t . We shall now establish a relation between ( ) x t and ( ) y t .

From the replication property of the impulse, we have,
( ) ( ) ( ) ( ) ( ) x t x t t x t d

= =

Let ( ) t

R denote the output (response) of the LTI system, when the
input is exited by the unit impulse ( ) t . This is generally denoted by the symbol
( ) h t and is called the impulse response of the system. That is, when ( ) t is
input to an LTI system, its output
( ) ( ) ( ) y t t h t = =

R . As the system is time
invariant,
( ) ( ) t h t =

R .

As the system is linear,
( ) ( ) ( ) ( ) x t x h t =

R
and ( ) ( ) ( ) ( ) ( ) ( ) x t y t x t d x h t d

= =

= R R
That is, ( ) ( ) ( ) y t x t h t = (1.22)

The following properties of convolution can be established:
Convolution operation
P1) is commutative
P2) is associative

1.61
P3) distributes over addition

P1) implies that ( ) ( ) ( ) ( )
1 2 2 1
x t x t x t x t =
That is, ( ) ( ) ( ) ( )
1 2 2 1
x x t d x x t d

=

.
P2) implies that ( ) ( ) ( ) ( ) ( ) ( )
1 2 3 1 2 3
x t x t x t x t x t x t =

, where the
bracketed convolution is performed first. Of course, we assume that every
convolution pair gives rise to bounded output.
P3) implies that
( ) ( ) ( ) ( ) ( ) ( ) ( )
1 2 3 1 2 1 3
x t x t x t x t x t x t x t + = +

.

Note that the properties P1) to P3) are valid even if the independent variable is
other than t .

Taking the Fourier transform on both sides of Eq. 1.22, we have,
( ) ( ) ( ) Y f X f H f = (1.23)
where ( ) ( ) h t H f . The quantity ( ) H f is referred to (quite obviously) as
the frequency response of the system and describes the frequency domain
behavior of the system. (As
( )
( )
( )
Y f
H f
X f
= , it is also referred to as the transfer
function of the LTI system). As ( ) H f is, in general, complex, it is normally shown
as two different plots, namely, the magnitude response: ( ) . H f vs f and the
phase response: ( ) arg . H f vs f

.

If ( ) ( )
1 2
H f H f , we then have ( ) ( )
1 1
1 2
F H f F H f

. That is,
( ) ( )
1 2
h t h t . In other words, the impulse response of any LTI system can be
used to uniquely characterize the system in the time domain.


1.62
Example 1.13
RC-lowpass filter (RC-LPF) is one among the quite often used LTI
systems in the study of communication theory. This network is shown in Fig.
1.17. Let us find its frequency response as well as the impulse response.

Fig. 1.17: The RC-lowpass filter

One of the important properties of any LTI system is: if the input
( )
0
2 j f t
x t e

= , then the output ( ) y t is also a complex exponential given by
( ) ( )

=
j f t
y t H f e
0
2
0
.

But, ( )
0
2
0
0
1
2
1
2
j f t
j f C
y t e
R
j f C
=
+

or ( )
0
2
0
1
1 2
j f t
y t e
j f RC

=
+
, when ( )
0
2 j f t
x t e

=
Generalizing this result, we obtain,
( )
1
1 2
H f
j f RC
=
+
(1.24)
That is,
( )
( )
2
1
1 2
H f
f RC
=
+
(1.25a)
( ) ( ) ( ) arg tan 2 f H f arc f RC = =

(1.25b)

1.63
Let
1
2
F
RC
=
. Then,
( )
2
1
1
H f
f
F
=

+

(1.26)

A plot of ( ) . H f vs f and ( ) arg . H f vs f

is shown in Fig. 1.18.

Fig. 1.18 RC-LPF: (a) Magnitude response
(b) Phase response


1.64
Let us now compute ( ) ( )
1
h t F H f
=

. We have the transform pair
( ) ex t
j f
1
1
1 2
+

Hence,
t
ex
RC RC j f RC
1 1
1
1 2

+

That is,
( )
RC LPF
t
h t ex
RC RC
1
1

=

(1.27)
This is shown in Fig. 1.19.

Fig. 1.19: Impulse response of an RC-LPF

Example 1.14
The input ( ) x t and the impulse response ( ) h t of an LTI system are as
shown in Fig. 1.20. Let us find the output ( ) y t of the system.


1.65

Fig. 1.20: The input ( ) x t and the impulse response ( ) h t of an LTI system

Let ( ) ( ) ( ) y t h x t d

To compute ( ) h t we have to perform the following three steps:
i) Obtain ( ) h and
( ) x t for a given
1
t t = .
ii) Take the product of the quantities in (i).
iii) Integrate the result of (ii) to obtain ( )
1
y t .
( ) h is the same as ( ) h t with the change of variable from t to . Note
that is the variable of integration. ( ) x t

is actually ( ) x t

; that is,
we first reverse ( ) x to get ( ) x and then shift by t , the time instant for which
( ) y t is desired. This completes the operations in step (i). The operations
involved in steps (ii) and (iii) are quite easy to understand.

In quite a few situations, where convolution is to be performed, it would be
of great help to have the plots of the quantities in step (i). These have been

1.66
shown in Fig. 1.21 for three different values of t , namely 0 t = , 1 t = and
2 t = .

From Fig.1.21(c), we see that if 1 t < , then ( ) h and ( ) x t do not
overlap; that ( ) 0 y t = , for 1 t < . For 1 0 t < , overlap of ( ) h and
( ) x t increases as t increases and the integral of the product (which is
positive) increases linearly reaching a value of 1 for 0 t = . For 0 2 t < , net
area of the product ( ) ( ) h x t , keeps decreasing and at 2 t = , we have,
( ) ( ) 0 h x t d
.


1.67

Fig. 1.21: A few plots to implement step (i) of convolution:
(a): ( ) h
(b), (c), (d): ( ) x t for 0 t = , 1 t = and 2 t = respectively.


1.68
Using the similar arguments, ( ) y t can be computed for 2 t > . The result
of the convolution is indicated in Fig.1.22.

Fig. 1.22: Complete output ( ) y t of Example 1.14

(Sometimes, computing ( ) ( ) ( ) y t x t h t = , could be very tricky and might
even be sticky
1
.)

1
Some people claim that convolution has driven many electrical engineering students to
contemplate theology either for salvation or as an alternative career (IEEE Spectrum, March
1991, page 60). For an interesting cartoon expressing the student reaction to the convolution
operation, see [3].
Exercise 1.3
Let ( )
, 0
0 ,
t
e t
x t
otherwise

( )
, 0
0 ,
t
e t
h t
otherwise

where , 0 > . Find ( ) ( ) ( ) y t x t h t = for
(i) = and (ii)

1.69

1.5.3 Signum function and unit step function
Def. 1.4: Signum Function
We denote the signum function by ( ) sgn t and define it as,
( )
1 , 0
sgn 0, 0
1, 0
t
t t
t
>
= =
<
(1.28)

Def. 1.5: Unit Step Function
We denote the unit step function by ( ) u t , and define it as,
Exercise 1.4
Find ( ) ( ) ( ) y t x t h t = where
( )
2 , 2
0,
t
x t
outside
<
=

( )
2 , 0
0 ,
t
e t
h t
outside

Exercise 1.5
Let ( )
1
, 950 1050
5
1
, 1050 950
10
0 ,
f Hz
X f f Hz
elsewhere
< <
= < <

(a) Compute and sketch ( ) ( ) ( ) Y f X f X f =
(b) Let
1
50 f Hz = ,
2
2050 f Hz = and
3
20 f Hz =
Verify that ( ) ( ) ( )
1 2 3
2 and 1 Y f Y f Y f = = = .

1.70
( )
1 , 0
1
, 0
2
0 , 0
t
u t t
t
>
= =
<
(1.29)
We shall now develop the Fourier transforms of ( ) sgn t and ( ) u t .

( )

sgn F t :
Let ( ) ( ) ( )
t t
x t e u t e u t

= (1.30)
where is a positive constant. Then
( ) ( )
0
sgn lim t x t

= , as can be seen from
Fig. 1.23.

Fig. 1.23: ( ) sgn t as a limiting case of ( ) x t of Eq. 1.30

( )
( )
2
2
1 1 4
2 2
2
j f
X f
j f j f
f

= =
+
+

( ) ( )
0
1
sgn lim F t X f
j f

= =

(1.31)

( )

F u t
As
( ) ( )
1
1 sgn
2
u t t = +

, we have

1.71
( ) ( )
1 1
2 2
U f f
j f
= +
(1.32)

We shall now state and prove the FT of the integral of a function.

Let ( )
t
p t ga
1
1
=

. Then, we know that
( ) ( )
1
0
lim p t t

= .
But ( ) ( )
0 0
1 1
0
2
1
lim
2
p t dt p t dt

= =

Properties of FT continued...
P9) Integration in the time domain
Let ( ) ( ) x t X f
Then, ( ) ( )
( )
( )
0
1
2 2
t
X
x d X f f
j f

+
(1.33)
Proof:
Consider ( ) ( ) ( ) ( ) x t u t x u t d
. As ( ) 0 u t = for t > ,
( ) ( ) ( ) ( ) ( ) ( ) ( ) . But
t
x t u t x d F x t u t X f U f

= =

.
Hence, ( ) ( )
( ) 1
2 2
t
f
x d X f
j f

+

Here there are two possibilities:
(i) ( ) 0 0 X = ; then ( )
( )
2
t
X f
x d
j f

(ii) ( ) 0 0 X ; then ( )
( ) ( ) ( ) 0
2 2
t
X f X f
x d
j f


1.72
and ( ) ( )
2
1 1
0
2
lim 1 p t dt p t dt
= =

.
That is,
( ) ( )
t
d u t

=
(1.34a)
or
( )
( )
d u t
t
d t
= (1.34b)

We shall now give an alternative proof for the FT relation,
( ) ( ) u t f
j f
1 1
2 2
+
.
As
( )
( )
d u t
t
d t
= ,
( ) j f U f 2 1 =
or ( ) U f
j f
1
2
=

But this is valid only for f 0 because of the following argument.
( ) ( ) u t u t 1 + = . Therefore,
( ) ( ) ( ) U f U f f + =
As ( ) f is nonzero only for f 0 = , we have
( ) ( ) ( ) ( ) U U U f 0 0 2 0 + = = or
( ) ( ) U f
1
0
2
= . Hence,
( )
( ) f f
U f
f
j f
1
, 0
2
1
, 0
2


1.73
Example 1.15
(a) For the scheme shown in Fig. 1.24, find the impulse response (This system is
referred to as Zero-Order-Hold, ZOH).
(b) If two such systems are cascaded, what is the overall impulse response
(cascade of two ZOHs is called a First-Order-Hold, FOH).

Fig. 1.24: Block schematic of a ZOH

a) ( ) h t of ZOH:
When ( ) ( ) x t t = , we have
( ) ( ) ( ) v t t t T =
Hence ( ) ( )
ZOH
y t h t =

, the impulse response of the ZOH, is
( ) ( ) ( ) ( )
2
t t
T
t
d T d u t u t T ga
T

= =

b) Impulse response of two LTI systems in cascade is the convolution of the
impulse responses of the constituents. Hence,
( )
2 2
FOH
T T
t t
h t ga ga
T T

=

t T
T tri
T

=


1.74
Eq. 1.34(a) can also be established by working in the frequency domain.
From Eq. 1.33, with ( ) ( ) x t t = and ( ) 1 X f = ,
( ) ( ) ( )
1 1
. That is,
2 2
t
d f U f
j f

+ =

( ) ( )
t
d u t

=

Eq. 1.34(b) helps in finding the derivatives of signals with discontinuities.
Consider the pulse ( ) p t shown in Fig. 1.25(a).

Fig. 1.25: (a) A signal with discontinuities
(b) Derivative of the signal at (a)

( ) p t can be written as
( ) ( ) ( ) ( ) 2 2 1 3 p t u t u t u t = + + +


1.75
Hence,
( )
( ) ( ) ( ) 2 2 1 3
d p t
t t t
d t
= + + +
which is shown in Fig. 1.25(b). From this result, we note that if there is a step
discontinuity of size A at
1
t t = in the signal, its derivative will have an impulse
of weight A at
1
t t = .

Example 1.16
Let ( ) x t be the doublet pulse of Example 1.7 (Fig.1.12). We shall find
( ) X f
from
( ) d x t
dt
.

( )
( ) ( ) ( ) 2
d x t
t T t t T
dt
= + +
Taking Fourier transform on both the sides,
( )
2 2
2 2
j f T j f T
j f X f e e

= +

( )
2
j f T j f T
e e

=
( )
( ) ( )
2
2 2
j f T j f T j f T j f T
e e e e
X f j T
j f T j

=

( ) ( ) 2 sin sin j T c f T f T =

Example 1.17
Let ( ) x t , ( ) h t and ( ) y t denote the input, impulse response and the
output respectively of an LTI system. It is given that,
( ) ( )
2t
x t t e u t
= and ( ) ( )
4t
h t e u t
= .
Find a) ( ) Y f

1.76
b) ( ) ( )
1
y t F Y f
=

c) ( ) ( ) ( ) y t x t h t =

a) From Eq. 1.22, we obtain
( ) ( ) ( ) Y f X f H f =
If ( ) ( )
2t
z t e u t
= , then ( )
1
2 2
Z f
j f
=
+
.
As ( ) ( ) x t t z t = , we have
( )
( )
( )
2
1
2
2 2
d Z f
j
X f
df
j f

= =

+

( )
1
4 2
H f
j f
=
+

Hence, ( )
( )
2
1 1
4 2
2 2
Y f
j f
j f
=
+
+

(b) Using partial fraction expansion,
( )
( )
( )
2
1 1 1
4 2 4
2 2 4 2
2 2
Y f
j f j f
j f
= + +
+ +
+

Hence, ( ) ( ) ( ) ( )
2 2 4
1 1 1
4 2 4
t t t
y t e u t t e u t e u t

= + +

(c) ( ) ( ) ( ) y t x h t d

( )
( )
( )
4 2 t
e u e u t d

( ) 0 y t = for 0 t < because for t negative, ( ) u t is 1 only for negative; but
then ( ) 0 u = . As ( ) 0 u t = for t > , we have
( )
( ) 4 2
0
t
t
y t e e d

=


1.77

4 2
0
t
t
e e d

=

2 2
4
0
2 2
t
t
e e
e d

2 2
4
0
2 4
t
t
t
e e
e t

=

2 2
4
1
, 0
2 4
t t
t
e e
t e t
=

0 = for 0 t <

Exercise 1.6
Let ( )
( )
0
0
0
1 cos 2 ,
2
0 ,
2
T
f t t
x t
T
t
+ <
>

where
0
T is the period of the cosine signal and
0
0
1
f
T
= .
(a) Show that
( )
( )
( )
3
2
0 0
0
3
2
2 2
d x t d x t T T
f t t
dt d t

= +

(b) Taking the FT of the equation in (a) above, show that
( ) ( )
0
0
2 2
0
sin
f
X f c f T
f f
=


1.78
1.6 Correlation Functions
Two basic operations that arise in the study of communication theory are:
(i) convolution and (ii) correlation. As we have some feel for the convolution
operation by now, let us develop the required familiarity with the correlation
operation.

When we say that there is some correlation between two objects, we imply
that there is some similarity between them. We would like to quantify this intuitive
notion and come up with a formal definition for correlation so that we have a
mathematically consistent and physically meaningful measure for the correlation
of the objects of interest to us.

Our interest is in electrical signals. We may like to quantify, say, the
similarity between a transmitted signal and the received signal or between two
different transmitted or received signals etc. We shall first introduce the cross
correlation functions; this will be followed by the special case, namely, auto-
correlation function.

In the context of correlation functions, we have to distinguish between the
energy signals and power signals. Accordingly, we make the following definitions.

1.6.1 Cross-correlation functions (CCF)
Let ( ) x t and ( ) y t be signals of the energy type. We now define their
cross-correlation functions, ( ) ( ) and
x y y x
R R .
Def. 1.6(a): The cross-correlation function ( )
x y
R is given by
( ) ( ) ( )
x y
R x t y t d t
(1.35a)
Def. 1.6(b): The cross-correlation function ( )
y x
R is given by

1.79
( ) ( ) ( )
y x
R y t x t d t
(1.35b)
In Eq. 1.35(a), ( ) y t
is a conjugated and shifted version of ( ) y t ,

accounting for the shift in ( ) y t
. Note that the variable of integration in Eq. 1.35

is t ; hence ( )
x y
R as well as ( )
y x
R is a function of , the shift parameter ( is
also called the scanning parameter or the search parameter).

Let ( ) x t and ( ) y t be the signals of the power type.
Def. 1.7(a): The cross-correlation function, ( )
x y
R is given by
( ) ( ) ( )
2
2
1
lim
T
x y
T
T
R x t y t dt
T

(1.36a)
Def. 1.7(b): The cross-correlation function, ( )
y x
R is given by
( ) ( ) ( )
2
2
1
lim
T
y x
T
T
R y t x t dt
T

(1.36b)

The power signals that we have to deal with most often are of the periodic
variety. For periodic signals, we have the following definitions:
Def. 1.8(a): ( ) ( ) ( )
0
0
2
0
2
1
p p
T
x y p p
T
R x t y t dt
T

(1.37a)
Def. 1.8(b): ( ) ( ) ( )
0
0
2
0
2
1
p p
T
y x p p
T
R y t x t dt
T

(1.37b)
Let t = in Eq. 1.35(a). Then, t = + and dt d = . Hence,
( ) ( ) ( )
x y
R x y d

= +

( )
y x
R
= (1.38)

1.80
As ( ) ( )
x y y x
R R , cross-correlation, unlike convolution is not in
general, commutative. To understand the significance of the parameter ,
consider the situation shown in Fig. 1.26.

Fig. 1.26: Waveforms used to compute ( )
y x
R and ( )
z x
R

If we compute ( ) ( ) x t y t dt
, we find it to be zero as ( ) x t and ( ) y t do not

overlap. However, if we delay ( ) x t by half a unit of time, we find that
1
2
x t

and ( ) y t start overlapping and for
1
2
> , we have nonzero value for the
integral. For the value of 2.75 , we will have a positive value for

1.81
( ) ( ) y t x t dt
, which would be about the maximum of ( )

y x
R for any .
For values of 2.75 > , ( )
y x
R keeps decreasing, becoming zero for 5 > .
Similarly, if we compute ( )
z x
R , we find that ( )
max
z x
R would be much smaller
than ( )
max
y x
R (Note that for 2.75 , the product quantity, ( ) ( ) y t x t , is
essentially positive for all t ). If ( ) y t is the received signal of a communication
system, then we are willing to accept ( ) x t as the likely transmitted signal (we can
treat the received signal as a delayed and distorted version of the transmitted
signal) whereas if ( ) z t is received, it would be difficult for us to accept that ( ) x t
could have been the transmitted signal. Thus the parameter helps us to find
time-shifted similarities present between the two signals.

From Eq. 1.35(a), we see that computing ( )
x y
R for a given , involves
the following steps:
(i) Shift ( ) y t
by
(ii) Take the product of ( ) x t and ( ) y t

(iii) Integrate the product with respect to t .
The above steps closely resemble the operations involved in convolution. It is not
difficult to see that
( ) ( ) ( )
x y
R x y
= , because
( ) ( ) ( ) ( ) x y x t y t dt

=

( ) ( ) x t y t dt
( )
x y
R = (1.39a)
Let ( ) ( )
x y x y
E f F R =

.
Then, ( ) ( ) ( )
x y
E f F x y

=

( ) ( ) X f Y f
= (1.39b)

1.82
Example 1.18
Let ( ) ( ) exp1 x t t = and ( )
2
t
y t ga

=

.
Let us find (a) ( )
x y
R and (b) ( )
y x
R .
From the results of (a) and (b) above, let us verify Eq. 1.38.
(a) ( )
x y
R :
( ) x t and ( ) y t are sketched below (Fig. 1.27).

Fig. 1.27: Waveforms of Example 1.18

(i) 1 < :
( ) x t and ( ) y t do not overlap and the product is zero. That is,
( ) 0
x y
R = for 1 < .
ii) 1 1 < :
( )
( )
1
1
0
1
t
x y
R e dt e
+
+
= =


1.83

(iii) 1 :
( )
( ) ( )
1
1 1
1
t
x y
R e dt e e
+
+

= =

1
e e
e

=

(b) ( )
y x
R :
(i) For 1 > , ( ) y t and ( ) x t do not overlap. Hence, ( ) 0
y x
R = for 1 > .
(ii) For 1 1 < ,
( )
( )
1
t
y x
R e dt

=

1
e e e

=

( ) 1
1 e

=
(iii) For 1 ,
( )
( )
1
1
t
y x
R e dt

=

1
e e
e

=

( )
x y
R and ( )
y x
R are plotted in Fig. 1.28.


1.84

Fig. 1.28: Cross correlation functions of Example 1.18

From the plots of ( )
x y
R and ( )
y x
R , it is easy to see that
( ) ( )
y x x y
R R =

Def. 1.9: Two signals ( ) x t and ( ) y t are said to be orthogonal if
( ) ( ) 0 x t y t dt
(1.40)
Eq. 1.41 implies that for orthogonal signals, say ( ) x t and ( ) y t
( )
0
0
x y
R
=
= (1.41a)
We have the companion relation to Eq. 1.41(a), namely
( ) 0 0
y x
R = , if ( ) x t and ( ) y t are orthogonal. (1.41b)

Let ( )
p
x t and ( )
p
y t be periodic with period
0
T . Then, from Eq. 1.37(a),
for any integer n ,

1.85
( ) ( )
0
p p p p
x y x y
R nT R + =
That is, ( )
p p
x y
R is also periodic with the same period as ( )
p
x t and ( )
p
y t .
Similarly, we find ( )
p p
y x
R . Derivation of the FT of ( )
p p
x y
R is given in appendix
A1.2.

1.6.2 Autocorrelation function (ACF)
ACF can be treated as a special case of CCF. In Eq. 1.35(a), let
( ) ( ) x t y t = . Then, we have
( ) ( ) ( )
x x
R x t x t dt
(1.42a)
Instead of ( )
x x
R , we use somewhat simplified notation, namely, ( )
x
R which is
called the auto correlation function of ( ) x t .

ACF compares ( ) x t with a shifted and conjugated version of itself. If
( ) x t and ( ) x t
are quite similar, we can expect large value for ( )

x
R ,
whereas as a value of ( )
x
R close to zero implies the orthogonality of the two
signals. Hence ( )
x
R can provide some information about the time variations of
the signal.

Let ( ) t = in Eq. 1.42(a). We then have,
( ) ( ) ( )
x
R x x d

= +

( ) ( ) x t x t dt

= +
(1.42b)
Eq. 1.42(b) gives another relation for ( )
x
R . This is quite meaningful
because, assuming positive, ( ) x t
is a right shifted version of ( ) x t
. In Eq.

1.86
1.42(a), we keep ( ) x t fixed and move ( ) x t
to the right by and take the

product; in Eq. 1.42(b), we keep ( ) x t
fixed and move ( ) x t to the left by . This

does not change the integral of Eq. 1.42(a), because if we let
1
t t = and
1
= ,
the product of Eq. 1.42(a) is ( ) ( )
1 1 1
x t x t
. This product is obtained from Eq.

1.42(b) for
1 1
t t = . For a given
1
= , as t is varied, all the product
quantities are obtained and hence the integral for a given
1
remains the same.
(Note that shifting a function does not change its area.)

If ( ) x t is a power signal, then the ACF is special case of Eq. 1.36(a). That
is, for the power signals, we have
( ) ( ) ( )
2
2
lim
T
x
T
T
R x t x t dt
(1.43a)
It can easily be shown that,
( ) ( ) ( )
2
2
lim
T
x
T
T
R x t x t dt
= +
(1.43b)

For power signals that are periodic, we have
( ) ( ) ( )
0
0
2
0
2
1
T
x p p
T
R x t x t dt
T

(1.44a)
( ) ( )
0
0
2
0
2
1
T
p p
T
x t x t dt
T

= +
(1.44b)

Properties of ACF (energy signals)
P1) ACF exhibits conjugate symmetry. That is,
( ) ( )
x x
R R
= (1.45a)

1.87
Proof: Exercise

Eq. 1.45(a) implies that the real part of ( )
x
R is an even function of
where as the imaginary part is an odd function of .
P2) ( ) ( )
2
0
x x
R x t d t E

= =
(1.45b)
where
x
E is the energy of the signal ( ) x t (Eq. 1.10).
P3) Maximum value of ( )
x
R occurs at the origin. That is, ( ) ( ) 0
x x
R R .
Proof: The proof of the above property follows from Schwarzs inequality;
which is stated below.
Let ( )
1
g t and ( )
2
g t be two energy signals.
Then, ( ) ( ) ( ) ( )
2
2 2
1 2 1 2
g t g t dt g t dt g t dt

.
Let ( ) ( )
1
g t x t = and ( ) ( )
2
g t x t
= .
From the Schwarzs Inequality,
( ) ( ) ( ) ( )
2
2
2
x t x t dt x t dt x t dt

( ) ( )
2 2
0
x x
R R

or
( ) ( ) 0
x x
R R
(1.45c)
P4) Let ( )
x
E f denote the Energy Spectral Density (ESD) of the signal ( ) x t .
That is, ( )
x x
E f df E

Then, ( ) ( )
x x
R E f
Proof
From Eq. 1.39(b),
( ) ( ) ( )
x y
E f X f Y f
=

1.88
Let ( ) ( ) x t y t = ; then ( ) ( )
2
x x
E f X f = . That is,
( ) ( )
2
x
R X f
But ( )
2
X f is the ESD of ( ) x t . That is,
( ) ( ) ( )
2
x x x
E f E f X f = =
Hence,
( ) ( )
x x
R E f
(1.45d)

It is to be noted that ( )
x
E f depends only on the magnitude spectrum,
( ) X f . Let ( ) x t and ( ) y t be two signals such that ( ) ( ) X f Y f = . Then,
( ) ( )
x y
R R = . Note that if ( ) ( )
x y
f f , ( ) x t may not have any
resemblance to ( ) y t ; but their ACFs will be the same. In other words, ACF does
not provide a unique description of the signal. Given ( ) x t , its ACF is unique;
but given some ACF, we can find many signals that have the given ACF.
P5) Let ( ) ( ) ( ) x t y t v t = + . Then,
( ) ( ) ( ) ( ) ( )
x y v y v v y
R R R R R = + + + (1.45e)
Proof: Exercise
If ( ) ( ) 0
y v v y
R R = (that is, ( ) y t and ( ) v t
are orthogonal for all

), then, ( ) ( ) ( )
x y v
R R R = + (1.45f)
In such a situation, ( )
x
R is the superposition of the ACFs of the components of
( ) x t . This also leads to the superposition of the ESDs; that is
( ) ( ) ( )
x y v
E f E f E f = + (1.45g)

Properties of ACF (periodic signals)
We list below the properties of the ACF of periodic signals. Proofs of these
properties are left as an exercise.

1.89
P1) ACF exhibits conjugate symmetry
( ) ( )
p p
x x
R R
= (1.46a)
P2) ( ) ( )
0
0
2
2
0
0
2
1
p p
T
x p x
T
R x t d t P
T =
= =
(1.46b)
where
p
x
P denotes the average power of ( )
p
x t (Sec. 1.2.2, Pg. 1.16).
P3) ( ) ( ) 0
p
x x
R R (1.46c)
That is, the maximum value of ( )
p
x
R occurs at the origin.
P4) ( ) ( )
0
, 1, 2, 3, . . .
p p
x x
R nT R n = = (1.46d)
where
0
T is the period of ( )
p
x t . That is, the ACF of a periodic signal is also
periodic with the same period as that of the signal.
P5) Let ( )
p
x
P f denote the Power Spectral Density (PSD) of ( )
p
x t . That is,
( )
p p
x x
P f df P
. Then,
( ) ( )
p p
x x
R P f (1.46e)
As ( )
p
x
R is periodic, we expect the PSD to be purely impulsive.

Example 1.19
a) Let ( ) x t be the signal shown in Fig. 1.29. Compute and sketch ( )
x
R .
Exercise 1.7
Let ( )
( )
0 0
,
2 2
0 ,
p
T T
x t t
x t
outside
< <
and ( ) ( ) x t X f .
Show that ( ) ( )
2
2
0 0 0
1
p p
x x
n
n n
F R P f X f
T T T
=

= =


1.90
b) ( ) x t of part (a) is given as the input to an LTI system with the impulse
response ( ) h t . If the output ( ) y t of the system is ( ) 3
x
R t , find ( ) h t .

Fig. 1.29: ( ) x t of Example 1.19

a) Computation of ( )
x
R :
We know that the maximum value of the ACF occurs at the origin; that is, at
0 = and ( ) 0
x x
R E = .
( )
1
0 1. 1 . 2 1.5
4
x
R = + =
Consider the product ( ) ( ) x t x t for 0 1 < . As increases in this range,
the overlap between the positive parts of the pulses ( ) x t and ( ) x t (and also
between the negative parts and these pulses) decreases, which implies a
decrease in the positive value for the integral of the product. In addition, a part of
( ) x t that is positive overlaps with the negative part of ( ) x t , there by further
reducing the positive value of the ( ) ( ) x t x t
. (The student is advised to

make a sketch of ( ) x t and ( ) x t .) This decrease is linear (with a constant
slope) until 1 = . The value of ( ) 1
x
R is
1
4

. For 1 2 < < , the positive part
of ( ) x t fully overlaps with the negative part of ( ) x t ; this makes ( )
x
R further
negative and the ACF reaches its minimum value at 2 = . As can easily be

1.91
checked, ( )
1
2
2
x
R

=

. As increases beyond 2, the ( )
x
R becomes less
and less negative and becomes zero at 3 = . As ( ) ( )
x x
R R = , we have all
the information necessary to sketch ( )
x
R , which is shown in Fig. 1.30.

Fig. 1.30: ACF of the signal of Example 1.19

b) Calculating ( ) h t :
We have ( ) ( ) ( ) y t x t h t =
( ) 3
x
R t =
( ) ( ) 3
x
R t t =
But from Eq. 1.39(a), ( ) ( ) ( )
x
R t x t x t = .

Hence, ( ) ( ) ( ) ( ) 3 y t x t x t t =

( ) ( ) 3 x t x t =

That is, ( ) ( ) 3 h t x t =

which is sketched in Fig. 1.31.

1.92

Fig. 1.31 Impulse response of the LTI system of Example 1.19

Example 1.20
Let ( ) ( )
0
cos x t A t = + . We will find ( )
x
R .

Method 1:
( ) ( ) ( )
0
0
2
2
0 0
0
2
1
cos cos
T
x
T
R A t t dt
T
= + +

( ) { }
0
0
2 2
0 0 0
0
2
1
cos 2 2 cos
2
T
T
A
t dt
T
= + +

2
0
cos
2
A
=
We find that:
(i) ( )
x
R is periodic with the same period as ( ) x t .
(ii) Its maximum value occurs at 0 =
(iii) The maximum value is
2
2
A
which is the average power of the signal.
(iv) ( )
x
R is independent of .


1.93
Method 2:
( )
( ) ( )
0 0
0
cos
2 2
j t j t
A A
A t e e
+ +
+ = +

( ) ( ) v t w t = +
As ( ) ( ) ( ) x t v t w t = + , we have
( ) ( ) ( ) ( ) ( )
x v w v w wv
R R R R R = + + + .
It is not difficult to see that ( ) ( ) 0
v w wv
R R = = for all . Hence,
( ) ( ) ( )
x v w
R R R = +
But ( ) ( )
w v
R R
= . Hence,
( ) ( ) 2Re
x v
R R =

( )
( ) ( )
0 0
0 0 0
0 0
2 2 2 2
0 0
2 2
1
4 4
T T
j t j t j
v
T T
A A
R e e dt e dt
T T
+ +

= =

0
2
4
j
A
e

=
Hence ( ) ( )
2
0
cos
2
x
A
R =

Example 1.21
Let ( ) ( ) sin 2 , 0 2 x t t t = . Let us find its ACF and sketch it.

In Fig. 1.32, we show ( ) x t and ( ) x t for 0 2 < < . Note that if 2 > ,
( ) ( ) 0 x t x t = which implies ( ) 0
x
R = for 2 > .


1.94

Fig. 1.32: (a) Sinusoidal pulse of Example 1.21 and (b) its shifted version

( ) ( ) ( )
2
sin 2 sin 2
x
R t t dt
=

( ) ( )
2
cos 2 cos 4 2
2
t
dt

=

( )
2
2
sin 4 2
cos 2
2 4
t
t

=

( )
( ) ( ) ( )
=
cos 2 sin 8 2 sin 2

cos 2
2 4

( )
cos 2 sin 2
cos 2 , 0 2
2 2

= +

As ( ) ( )
x x
R R = , we have
( )
( )
( ) ( )
,
sin 2 cos 2
cos 2 , 2
2 2
0
x
R
otherwise

+
=

( )
x
R is plotted in Fig. 1.33.


1.95

Fig. 1.33: ( )
x
R of Example 1.21

1.7 Hilbert Transform
Let ( ) x t be the input of an LTI system with the impulse response ( ) h t .
Then, the output ( ) y t is
( ) ( ) ( ) y t x t h t =
or ( ) ( ) ( ) Y f X f H f =
That is, ( ) ( ) ( ) Y f X f H f = (1.47a)
( ) ( ) ( )
y x h
f f f = + (1.47b)

From Eq. 1.47 we see that an LTI system alters, in general, both the magnitude
spectrum and the phase spectrum of the input signal. However, there are certain
networks, called all pass networks, which would alter only the (input) phase
spectrum. That is, if ( ) H f is the frequency response of an all pass network, then

1.96
( ) ( ) Y f X f =
( ) ( ) ( )
y x h
f f f = +
An interesting case of all-pass network is the ideal delay, with the impulse
response ( ) ( )
d
h t t t = . Though ( ) ( )
y x
f f , phase shift imparted to each
input spectral component is proportional to the frequency, the proportionality
constant being 2
d
t . Another interesting network is the Hilbert transformer. Its
output is characterized by:
(i) ( ) ( ) Y f X f = , 0 f and
(ii) ( )
( )
( )
, 0
2
, 0
2
x
y
x
f f
f
f f

+ >
+ <

That is, a Hilbert transform is essentially a
2

phase shifter.

Hence, we define the Hilbert transformer in the frequency domain, with the
frequency response function
( ) ( ) sgn H f j f = (1.48a)
where ( )
1 , 0
sgn 0 , 0
1 , 0
f
f f
f
>
= =
<

As ( )
1
sgn j f
t

,
( )
1
h t
t
=
(1.48b)
When ( ) x t is the input to a Hilbert transformer, we denote its output as ( )
x t
where
( ) ( )
1
x t x t
t
=


1.97

( )
( )
1
x
d
t
1
(1.49a)
and ( ) ( ) ( )
sgn X f j f X f = (1.49b)
( )
x t is called the Hilbert transform of ( ) x t .
Note: Unlike other transforms, both ( ) x t and ( )
x t are functions of the same
variable (t in our case).

Hilbert Transform (HT) will prove quite useful later on in the study of
bandpass signals and single sideband signals. For the present, let us look at
some examples of HT.

Example 1.22
Hilbert transform ( ) t .
Let ( ) ( ) x t t = . Let us find ( )
x t .

As ( ) 1 t , we have
( ) ( )
sgn F t j f

=

( )
1
t
t
=

This also establishes the relation, ( )
1 1
!! t
t t

=

Example 1.23
HT of a cosine signal.
Let ( ) ( )
0
cos 2 x t f t = . Let us find ( )
x t .

1
This integral is actually Cauchys principal value.

1.98
( ) ( ) ( )
0 0
1
2
X f f f f f = + +

( ) ( ) ( )
sgn X f j f X f = . That is,

( ) ( ) ( )
0 0
1
2
X f f f f f
j
= +

That is, ( ) ( )
0
sin 2 x t f t = .
Alternatively,
if ( )
0
2
1
j f t
x t e

= , then ( )
( ) 0
2
2
1
j f t
x t e

=
and if ( )
0
2
2
j f t
x t e

= , then ( )
( ) 0
2
2
2
j f t
x t e

=
Hence,
( ) ( ) ( ) ( )
= = +

x t f t x t x t
0 1 2
1
cos 2
2
has ( )
0
cos
2
x t t

=

( )
0
sin t =
Similarly, we can show that if ( ) ( )
0
sin 2 x t f t = , then ( )
0
cos x t t = .

Example 1.24
Let ( )
2
1
1
x t
t
=
+
. Let us find ( )
x t .

( ) ( )
1
x t x t
t
=

( )
( )
2
1 1
1
d
t

( )
2 2
1 1
1 1
t d
d
t t

+
= +
+ +

As
2
1
0
1
d d
t

= =
+

, we have

1.99
( )
2 2
1 1
1 1
t
x t d
t

=
+ +

As the bracketed integral is equal to , we have
( )
2
1
t
x t
t
=
+

Example 1.25
Let ( ) ( ) cos 2
c
x t m t f t = where ( ) m t is a lowpass signal with
( ) 0 M f = for f W > and
c
f W > . We will show that
( ) ( ) ( ) ( )

( )
sin 2 cos 2
c c
x t m t f t m t f t = = .

( ) ( ) ( )
1
2
c c
X f M f f M f f = + +

( ) ( ) ( ) ( )
1
sgn
2
c c
X f M f f M f f j f = + +

But ( )
c
M f f

is nonzero only for 0 f >
and ( )
c
M f f +

is nonzero only for 0 f < .
Hence,
( )
( )
( )
2
2
1
, 0
2
1
, 0
2
j
c
j
c
M f f e f
X f
M f f e f
>
+ <

Exercise 1.8
(a) Let ( )
1
sin t
x t
t
= . Show that ( )
1
1 cos
t
x t
t
=
(b) Let ( ) ( )
2
x t ga t = . Show that ( )
2
1
1
2
ln
1
2
t
x t
t
+


1.100
Consider ( ) ( ) ( ) ( ) ( )
1
sin 2
2
c c c
m t f t M f f f f f
j
+

( ) ( ) ( )
2 2
1
2
j j
j
c c
M f f f e e e f f

+ +

( ) ( ) ( )
2 2
1
2
j j
c c
M f f f e e f f

+ +

That is,
( ) ( )
( )
( )
2
2
1
, 0
2
sin 2
1
, 0
2
j
c
c
j
c
M f f e f
F m t f t
M f f e f
>
+ <

As ( ) ( )
sin 2
c
F m t f t X f =

, we have
( ) ( ) ( ) ( )

( )
sin 2 cos 2
c c
x t m t f t m t f t = =

Note: It is possible to establish even a stronger result, which is stated below.

Let ( ) ( ) ( ) x t m t v t = where ( ) m t is a lowpass signal with ( ) 0 M f = for
f W > and ( ) v t is a high-pass signal with ( ) 0 V f = for f W < . Then
( ) ( ) ( )
x t m t v t = . [We assume that there are no impulses in either ( ) M f or
( ) V f .

1.7.1 Properties of Hilbert transform
Our area of application of HT is real signals. Hence, we develop the
properties of HT as applied to real signals. We assume that the signals under
consideration have no impulses in their spectra at 0 f = .
P1) A signal ( ) x t and its HT, ( )
x t , have the same energy.
Proof: ( )
2
x
E X f d f

=


1.101
As ( ) ( ) ( )
sgn X f j f X f =
( ) ( )
X f X f = . Hence =
x x
E E

Note: Though ( )
0
f
X f
=
is zero, it will not change the value of the integral; and
hence the energy.

P2) ( )
{ }
( )
HT x t x t =

That is, applying the HT twice on a given signal ( ) x t changes the sign of the
signal. Intuitively, this is satisfying. Each time we perform HT, we change the
phase of a spectral component in ( ) X f by
0
90 . Hence, performing the
transformation twice results in a phase shift of
0
180 .
Proof: ( ) ( ) ( )
sgn x t j f X f
( ) ( ) ( ) ( )
sgn sgn HT x t j f j f X f

( ) ( ) ( ) ( )
2
sgn j f X f X f = =

Example 1.26
Let ( )
1
x t
t
= . We shall find ( )
x t .

From Example 1.22, we have ( )
1
t
t
=
.
Hence,
( ) ( )
1
HT t t HT
t

= =

. That is,
( )
1
HT t
t

=

P3) A signal ( ) x t and its HT, ( )
x t , are orthogonal.

1.102
Proof: We know that
( ) ( ) ( ) ( )
x t x t d t X f X f d f

=

(See the note, property P10, Sec. 1.4). Hence,
( ) ( ) ( ) ( ) ( )
sgn x t x t d t X f j f X f d f

=

As ( ) x t is real, ( ) ( ) X f X f
= . Therefore,
( ) ( ) ( ) ( )
2
sgn x t x t d t j f X f d f

=

0 = (Note that the integrand on the RHS is odd)

P4) ( ) HT x t

, where is a nonzero constant is ( ) ( )
sgn x t

.
Proof: We will first establish that if ( ) ( ) ( ) y t x t h t = and
( ) ( ) ( ) z t x t h t = , then ( ) ( )
1
z t y t =
.
If ( ) ( ) ( ) z t x t h t = , then
( )
2
1 f f
Z f X H
a a

=

Also, ( )
1 f
y t Y

, where

f f f
Y X H

=

( )
2
Z f =
Hence, ( ) ( ) ( )
2 1
y t Z f Z f =

That is, ( )
( ) y t
z t

=


1.103
( ) ( )
1
HT x t x t
t
=

( )
1
x t
t

=

( )
1
x t
t

=

As ( ) ( )
1
x t x t
t

=

, we have ( )
( )
1
x t
x t
t

=

.
Hence
( ) ( ) ( ) ( )
sgn HT x t x t x t
= =

.

As a simple illustration of the property, let ( )
2
1
1
x t
t
=
+
and 2 = . Then
( )
2
1
1 4
x t
t
=
+
. Let us obtain
2
1
1 4
HT
t

+

using P4.
As ( )
2
1
t
x t
t
=
+
, ( )
( )
2 2 2
1 2 2
sgn 2
1 4 1 4
1 2
t t
HT
t t
t

= =

+ + +

If 2 = , ( )
2
1
1 4
x t
t
=
+
which is the same as with 2 = .
With 2 = , ( ) ( )
2
2
sgn
1 4
t
x t
t

=

+

2
2
1 4
t
t
=
+
, as required

Exercise 1.9
Using the result
sin 1 cos t t
HT
t t

=

,
find the ( ) sin HT c t

.

1.104
P5) The cross correlation function of ( ) x t and ( )
x t , ( )
x x
R is the negative of
the HT of ( )
x
R . That is,
( ) ( )
x x x
R R =
Proof: ( ) ( ) ( )
x x
F R F x x
=

( ) ( ) ( ) sgn X f j f X f =

( ) ( ) ( ) ( )
2 2
sgn sgn X f j f X f j f = =

That is,
( ) ( )
x x x
R R =

1.8 Bandpass Signals
Consider a communication system that transmits the signal
( ) ( ) cos 2
c
s t m t f t = , where ( ) m t is a (lowpass) message signal and
( ) cos 2
c
f t is the (high-frequency) carrier term. Then the spectrum ( ) S f of the
transmitted signal is ( ) ( ) ( )
1
2
c c
S f M f f M f f = + +

. If ( ) M f is as shown in
Fig 1.34(a), then for a carrier frequency 100
c
f k Hz = , ( ) S f will be as shown in
Fig 1.34(b).

Exercise 1.10
Show that ( ) ( )
x x x
R R = .

1.105

Fig. 1.34: A typical narrowband, bandpass spectrum

We see that the spectrum of ( ) s t is confined to the frequency interval
95 105 f k Hz . Where as ( ) m t is a lowpass signal, ( ) s t is a bandpass
signal. Moreover ( ) s t is a narrowband, bandpass signal because the spectral
width of ( ) S f , ( ) 105 95 10 k Hz = , is quite small in comparison with the carrier
frequency
c
f of 100 k Hz . Hence, we call ( ) s t as a Narrowband, Bandpass
(NBBP) signal. Fig 1.35 shows some more spectra that represent NBBP signals.


1.106

Fig. 1.35: A few more examples of NBBP spectra

NBBP signals play an important role in the communication process. Let us
assume that the rest of the communication system (channel, a part of the
receiver etc.) is also of the bandpass variety. The study of such transmission-
reception schemes becomes a little complicated because of the presence of the
carrier term in some form or the other (The term cos
c
t in ( ) cos
c
m t t is
meant only to carry the information ( ) m t and is not part of the information) If
we develop tools to study bandpass signals and bandpass systems, independent
of the carrier, the analysis of the communications schemes would become
somewhat simplified. (That is, bandpass signals and bandpass systems are
studied in terms of their lowpass equivalents.) The mathematical concepts of pre-
envelope and complex envelope have been developed for this purpose. We shall
make use of these concepts in our studies on linear modulation and angle
modulation.


1.107
1.8.1 Pre-envelope
Def. 1.10: Let ( ) x t be any real signal with the FT, ( ) X f . We define its pre-
envelope as,
( ) ( ) ( )
pe
x t x t j x t = +
(1.50a)
Taking the Fourier transform of Eq. 1.54(a), we have
( ) ( ) ( ) ( ) sgn
pe
X f X f j j f X f = +

( ) ( ) ( ) sgn X f f X f = +

That is,
( )
( )
( )
2 , 0
0 , 0
0 , 0
pe
X f f
X f X f
f
>
= =
<
(1.50b)

(We assume that ( ) X f has no impulse at 0 f = ). That is, ( )
pe
x t has spectrum
only for 0 f , even though ( ) X f is two sided (As ( )
pe
x t has spectral
components only for 0 f , some authors use the symbol ( ) x t
+
to denote the
pre-envelope of ( ) x t ). Of course, ( )
pe
x t
will have spectrum only for 0 f ).

Consider the signals ( )
1
x t and ( )
2
x t whose spectra are shown in Fig. 1.36.


1.108

Fig. 1.36: Typical two-sided spectra

The corresponding ( )
1, pe
X f and ( )
2, pe
X f are as shown in Fig. 1.37(a) and (b)
respectively.

Fig. 1.37: Fourier transform of (a) ( )
1, pe
x t and (b) ( )
2, pe
x t of the signals in


1.109
Because of Eq. 1.50(b), we can write
( ) ( )
2
0
2
j f t
pe
x t X f e d f
=

(1.51)

Example 1.27
Let ( )
2
1
1
x t
t
=
+
. Let us find ( )
pe
X f and ( )
pe
x t .

First, let us compute ( ) X f . From Example 1.5, we know that
2
2
1
1
f
e
t

+
.
Hence ( ) ( )
2
2
f
pe
X f e u f

= , where ( )
1 , 0
1
, 0
2
0 , 0
f
u f f
f
>
= =
<
.
We require ( )
1
pe
F X f

.
As ( ) ex t
j f
1
1
1 2
+

( ) ex t
j f
1 1
1 2
2 1

+
.
From duality, ( ) ( )
f
ex f e u f
j t
2
1
2 1 2 2
1

=
.
That is, ( )
2
1 1
1 1
pe
j t
x t
j t t
+
= =
+

2 2
1
1 1
t
j
t t
= +
+ +
.
Then, ( )
2
1
t
x t
t
=
+
, which is a known result.


1.110
As can be seen from the above discussion, the spectrum of ( )
pe
x t is still
bandpass (though one-sided) if ( ) x t is a bandpass signal. Let ( ) x t be a
bandpass signal with ( )
pe
X f centered with respect to 102
c
f k Hz = as shown
in Fig. 1.38(a).

Fig 1.38: A typical ( )
pe
X f and shifted version

(We can treat
c
f to be the center frequency in Fig. 1.38(a), by taking the
bandpass spectrum from 94 to 110 k Hz , though the spectrum is zero for the
frequency range 94 to100 k Hz .) From Fig 1.38(b), we see that ( )
pe c
X f f + is a
lowpass spectrum, nonzero in the frequency range ( ) 2 to 8 k Hz .


1.111
1.8.2 Complex envelope
Def. 1.11: We now define the complex envelope of ( ) x t , denoted ( )
c e
x t as
( ) ( )
2
c
j f t
c e pe
x t x t e

=
(1.52a)
Eq. 1.52(a) implies
( ) ( )
c e pe c
X f X f f = + (1.52b)
(We assume that we know the center frequency
c
f and it is such that
( )
pe c
X f f + is lowpass in nature).
Equation 1.52(a) also implies
( ) ( )
2
c
j f t
pe c e
x t x t e

= (1.53)
( )
c e
x t is also referred to as the equivalent lowpass signal of the bandpass signal
( ) x t . In general ( )
c e
x t is complex. Let ( )
c
x t be the real part and ( )
s
x t its
imaginary part. Then,
( ) ( ) ( )
c e c s
x t x t j x t = + (1.54)
We will show a little later that, both ( )
c
x t and ( )
s
x t are lowpass in nature. Using
Eq. 1.54 in Eq. 1.53, we obtain,
( ) ( ) ( )
c
j f t
pe c s
x t x t j x t e
2
= +

As the real bandpass signal ( ) x t is the real part of ( )
pe
x t , we have
( ) ( ) ( ) ( ) ( )
c c s c
x t x t f t x t f t cos 2 sin 2 = (1.55)
Eq. 1.55 is referred to as the canonical representation of the bandpass signal.
( )
c
x t , which is the coefficient of the cosine term, is usually referred to as the in-
phase component and ( )
s
x t , the coefficient of the sine term, as the quadrature
component. Note that ( ) sin 2
c
f t is in phase quadrature to ( ) cos 2
c
f t . (It is
also common in the literature to use the symbol ( )
I
x t for the in-phase
component and ( )
Q
x t for the quadrature component.)


1.112
To express Eq. 1.54 in polar form, let
( ) ( ) ( )
c s
A t x t x t
2 2
= + (1.56a)
( )
( )
( )
s
c
x t
t
x t
1
tan

=

(1.56b)
Then,
( ) ( )
( ) j t
ce
x t A t e

= (1.56c)
( ) ( ) ( )
2
Re Re
c
j f t
pe c e
x t x t x t e

= =

(1.57)
( )
( ) 2
Re
c
j t j f t
A t e e

=

(1.58)
Eq. 1.58 resembles phasor representation of a sinusoid. We know that,
( )
2
cos 2 Re
c
j f t j
c
A f t A e e

+ =

(1.59)
j
A e

is generally referred to the as the phasor associated with the sinusoidal
signal ( ) cos 2
c
A f t + . (The phasor is a complex number providing information
about the amplitude and phase (at 0 t = ) of the sinusoid.) The quantity
2
c
j f t j
A e e

can be treated as a rotating vector as shown in Fig. 1.39(a).
Comparing Eq. 1.58 with Eq. 1.59, we find that they have a close resemblance.
Phasor of the monochromatic (single frequency) signal has constant amplitude
A and a fixed phase . In the case of the complex envelope (of a narrowband
signal) both ( ) A t and ( ) t are, (slowly) time-varying. This is shown in Fig
1.39(b). (Note that a single frequency sinusoid is the extreme case of a
narrowband signal with zero spectral width!)


1.113

Fig. 1.39: (a) Phasor representation of
( ) 2
c
j f t
A e
+

(b) Complex envelope as a (slowly) varying amplitude and phase

In other words, complex envelope can be treated as a generalization of
the phasor representation used for single frequency sinusoids; the generalization
permits the amplitude and phase to change as a function of time. Note, however,
that for a given time
1
t t = , ( )
( )
1
1
j t
A t e

is a complex number, having the
necessary information about the narrowband signal. As we shall see later,
different modulation schemes are basically different methods of controlling either
( ) A t or ( ) t (or both) as a function of the message signal ( ) m t .

If ( ) x t is a NBBP signal with spectrum confined to the frequency range
c
f f W , then ( )
c
x t and ( )
s
x t are lowpass signals with spectrum confined

1.114
to f W . This is because ( ) ( )
c e pe c
X f X f f = + is nonzero only for f W .
As ( )
c
x t is the real part of ( )
c e
x t , we have
( )
( ) ( )
c e c e
c
x t x t
x t
2

+

=
or ( ) ( )
( ) ( )
c e c e
c c
X f X f
F x t X f
2
+
= =

As ( )
c e
X f and ( )
c e
X f
are zero for f W > , ( )

c
X f is also zero for f W > .
Similarly, ( )
s
x t is also a lowpass signal with ( )
s
X f 0 = for f W > .

The scheme shown below (Fig 1.40) enables us to obtain ( )
c
x t and ( )
s
x t
from ( ) x t .

Fig. 1.40: Scheme for the recovery of ( )
c
x t and ( )
s
x t from ( ) x t

In Fig 1.40, ( ) x t a NBBP signal, with the spectrum confined to the interval
c
f f W , where
c
W f << .
( ) ( )
1
2 cos
c
v t x t t =
( ) ( ) ( )
c c s c c
x t t x t t t 2 cos sin cos =
( ) ( )
c c s c
x t t x t t
2
2 cos sin 2 =

1.115
( ) ( )
c
c s c
t
x t x t t
1 cos 2
2 sin 2
2
+
=

( ) ( ) ( ) ( )
c c c s c
x t x t t x t t cos 2 sin 2 = + (1.60)

In Fig. 1.40, ( )
l p
H f is an Ideal Lowpass Filter (ILPF) with the frequency
response.
( )
1 ,
0,
l p
f W
H f
outside

( )
c
x t and ( )
s
x t are lowpass signals, band limited to W Hz. ( ) ( )
c c
x t t cos 2
and ( ) ( )
s c
x t t sin 2 have bandpass spectra centered at
c
f . These quantities
will be filtered out by the ILPF and at the output of the top channel, we obtain
( )
c
x t . Similar analysis will show that the output of the bottom channel is ( )
s
x t .

From Eq. 1.58, we have,
( ) ( ) ( ) cos
c
x t A t t t = +

(1.61)
Eq. 1.61 is referred to as the envelope and phase representation of ( ) x t . ( ) A t
is called the natural envelope (or simply the envelope) of ( ) x t and ( ) t , its
phase. As we assume that
c
f is known, the information about ( ) x t is contained
in either of the quantities ( ) ( ) ( )
c s
x t x t , or ( ) ( ) , A t t

and these are lowpass in
nature. Note that
( ) ( ) ( )
ce pe
A t x t x t = = (1.62)
and is always non-negative.

We shall now illustrate the concepts of ( )
pe
x t , ( )
ce
x t , ( ) A t and ( ) t with
the help of a few examples.


1.116
Example 1.28
Let ( ) [ ]
cos 2
c
x t f t = . Let us find ( )
pe
x t , ( )
ce
x t , ( ) A t and ( ) t .

( ) ( ) ( )
1
2
c c
X f f f f f = + +

Hence ( ) ( ) ( )
2
c
j f t
pe c pe
X f f f x t e

= =
As ( ) ( )
ce
X f f = , we obtain ( ) 1
ce
x t =
As ( )
ce
x t is real and positive, ( ) 0 t = and ( ) ( ) 1
ce
A t x t = = .

Example 1.29
Let ( ) cos
c
t
x t ga t
T

=

. Assume that 1
c
f T >> so that ( ) x t can be
taken as a NBBP signal. We shall find ( )
pe
x t , ( )
ce
x t and ( ) A t .

Method 1 (Frequency domain):
Because of the assumption 1
c
f T >> , we can take ( ) X f approximately as
( )
( )
( )
sin , 0
2
sin , 0
2
c
c
T
c f f T f
X f
T
c f f T f
>

+ <

From Eq. 1.50(b),
( )
( ) sin , 0
0 ,
c
pe
T c f f T f
X f
otherwise
>

Hence,
( )
2
c
j f t
pe
t
x t ga e
T

=


1.117
( ) ( )
2
c
j f t
ce pe
x t x t e

=

t
ga
T

=

As ( )
ce
x t is real, we have ( ) ( ) = =
s
x t t 0 0.
( ) ( )
ce
t
A t x t g a
T

= =

Note that for Examples 1.28 and 1.29, the complex envelope is real valued and is
equal to the envelope, ( ) A t .

Method 2 (Time domain):
Comparing the given ( ) x t with Eq. 1.55, we find that it is already in the
canonic form with ( )
c
t
x t ga
T

=

, ( )
s
x t 0 =
Hence, ( )
ce
t
x t ga
T

=

And ( ) ( )
2
c
j f t
pe ce
x t x t e

=

2
c
j f t
t
ga e
T

=

( )
t
A t ga
T

=

and ( ) 0 t =

Example 1.30
( ) x t is a NBBP signal with ( ) X f as shown in Fig 1.41. Let us find ( )
c
x t
and ( )
s
x t .


1.118

Fig. 1.41: Spectrum of the NBBP signal of Example 1.30

From the given spectrum, we can obtain ( )
ce
X f , which is shown below in
Fig. 1.42.

Fig. 1.42: Spectrum of the complex envelope of the signal of Example 1.30

As shown in the Fig 1.42, we can take ( )
ce
X f as the sum of A and B.
Inverse Fourier transform of A : ( )
100
200 sin 100
j t
c t e

Inverse Fourier transform of B : ( )
50
50 sin 50
j t
c t e

Hence ( ) ( ) ( )
100 50
200 sin 100 50 sin 50
j t j t
ce
x t c t e c t e

= +
( ) ( ) ( ) ( ) ( )
c
x t c t t c t t 200 sin 100 cos 100 50 sin 50 cos 50 = +
( ) ( ) ( ) ( ) ( ) =
s
x t c t t c t t 50 sin 50 sin 50 200 sin 100 sin 100


1.119

1.9 Bandpass (BP) Systems
Let a signal ( ) x t be input to an LTI system with impulse response ( ) h t . If
( ) y t is the output, then ( ) ( ) ( ) Y f X f H f = and
( ) ( ) ( )
2 2
Y f X f H f
2
=
If ( )
x
E f is the energy spectral density of ( ) x t , then
( ) ( ) ( )
2
y x
E f E f H f = (1.63a)
Let ( ) ( )
2
h
H f R be the Fourier transform pair where ( )
h
R is the ACF
of the impulse response ( ) h t . Then,
( ) ( ) ( )
y x h
R R R =
( ) ( ) ( )
x
R h h

=

(1.63b)
Exercise 1.11
Find the pre-envelope of ( ) ( ) sin x t c t = .
Hint: use the result of Exercise 1.9.

Exercise 1.12
Let ( ) ( ) ( )
c
x t ex t t 1 sin = +

where
c
>> .
Find ( )
ce
x t .

Exercise 1.13
Find the expression for the envelope of
( ) ( ) ( ) 1 cos cos
m c
x t k t t = +

where k is a real constant and
c m
f f >> . Sketch it for 0.5 and 1.5 k = .

1.120
Let ( ) x t be a bandpass signal with ( ) X f zero for
c
f f W > . A
bandpass signal is usually processed a bandpass system; that is, a system with
passband in the interval
c
f f B < where B W . We would like to study the
effect of a BP system on a BP input. This study is again greatly facilitated if we
were to use complex envelopes of the signals involved.

In appendix A1.3 (Eq A1.3.8), it is shown that
( ) ( ) ( )
1
2
ce ce ce
y t x t h t =

(1.64)
We shall now give an example to illustrate the use of Eq. 1.64.

Example 1.31
Let ( ) x t be a sinusoidal pulse given by
( )
( )
6
2 cos 2 10 , 0 1 sec
0 ,
t t m
x t
outside

( ) x t is the input to an LTI system with impulse response ( ) ( ) h t x T t = ,
where 1 sec T m = . Find ( )
ce
y t and ( ) y t .

( ) x t can be taken as a NBBP signal. Its complex envelope,
( ) 2, 0 1 sec
ce
x t t m =
( )
( )
( )
6
2 cos 2 10 h t T t

=

( ) ( ) ( ) ( ) { }
6 6 6 6
2 cos 2 10 cos 2 10 sin 2 10 sin 2 10 T t T t = +

( ) ( ) ( ) { }
3 6 3 6
2 cos 2 10 cos 2 10 sin 2 10 sin 2 10 t t

= +

( )
6
2 cos 2 10 , 0 1 sec t t m

=

Again, we can treat ( ) h t as a NBBP signal, with

1.121
( )
2, 0 1 sec
0,
ce
t m
h t
outside

=

That is, ( ) ( )
2
2
ce ce
T
t
x t h t ga
T

= =

From Eq.1.64,
( ) ( ) ( )
( )
3
1
2 10
2
ce ce ce
t T
y t x t h t tri
T

= =

( ) ( ) ( ) cos 2
ce c
y t y t f t =
Obtaining ( ) y t directly as ( ) ( ) x t h t

would be quite cumbersome.

Exercise 1.14
Let ( ) ( )
( )
6
sin 200 cos 2 10 x t c t t

=

be the input to an NBBP
system with ( )
( ) ( )
6 6
1 1
1
10 10
2
H f H f H f
j

= +

(1.65)
where ( )
1
H f is as shown in Fig. 1.43.

Fig. 1.43: ( )
1
H f of Eq. 1.65

Find ( )
ce
y t , the complex envelope of the output.

1.122
Appendix A1.1
Tabulation of
( )
sinc
( ) sin c ( ) sin c ( ) sin c
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.45
0.50
0.55
0.60
0.65
0.70
0.75
0.80
0.85
0.90
0.95
1.00
1.05
1.10
1.15
1.20
1.25
1.30
1.35
1.40
1.45
1.50
1.55
1.60
1.65
1.000000
0.995893
0.983631
0.963397
0.935489
0.900316
0.858393
0.810331
0.756826
0.698645
0.636619
0.571619
0.504550
0.436331
0.367882
0.300104
0.233871
0.170010
0.109291
0.052414
-0.000001
-0.047424
-0.089422
-0.125661
-0.155915
-0.180064
-0.198091
-0.210086
-0.216236
-0.216821
-0.212203
-0.202833
-0.189207
-0.171888
1.70
1.75
1.80
1.85
1.90
1.95
2.00
2.05
2.10
2.15
2.20
2.25
2.30
2.35
2.40
2.45
2.50
2.55
2.60
2.65
2.70
2.75
2.80
2.85
2.90
2.95
3.00
3.05
3.10
3.15
3.20
3.25
3.30
3.35
-0.151481
-0.128616
-0.103943
-0.078113
-0.051770
-0.025536
0.000000
0.024290
0.046840
0.067214
0.085045
0.100035
0.111964
0.120688
0.126138
0.128323
0.127324
0.123291
0.116435
0.107025
0.095377
0.081847
0.066821
0.050705
0.033919
0.016880
0.000000
-0.016326
-0.031730
-0.045876
-0.058468
-0.069255
-0.078036
-0.084661
3.40
3.45
3.50
3.55
3.60
3.65
3.70
3.75
3.80
3.85
3.90
3.95
4.00
4.05
4.10
4.15
4.20
4.25
4.30
4.35
4.40
4.45
4.50
4.55
4.60
4.65
4.70
4.75
4.80
4.85
4.90
4.95
5.00
-0.089038
-0.091128
-0.090946
-0.088561
-0.084092
-0.077703
-0.069600
-0.060021
-0.049237
-0.037535
-0.025222
-0.012607
-0.000000
0.012295
0.023991
0.034821
0.044547
0.052960
0.059888
0.065199
0.068802
0.070650
0.070736
0.069097
0.065811
0.060993
0.054791
0.047385
0.038979
0.029796
0.020074
0.010059
0.000000


1.123

Fig. A1.1: Plot of ( ) sinc for 5


1.124
Appendix A1.2
Fourier transform of ( )
p p
x y
R
As ( )
p p
x y
R is periodic, we expect the spectrum to be purely impulsive.
We have only to decide the weights of these impulses.

( ) ( ) ( )
0
0
2
0
2
1
p p
T
x y p p
T
R x t y t dt
T

Let ( )
( )
0 0
,
2 2
0 ,
p
T T
x t t
x t
outside
< <

then, ( ) ( )
0 p
n
x t x t nT
=
=

Similarly, ( ) ( )
0 p
n
y t y t nT

=
=

where ( )
( )
0 0
,
2 2
0 ,
p
T T
y t t
y t
outside

< <

( ) ( ) ( )
0
0
2
*
0
0
2
1
p p
T
x y
n T
R x t y t nT dt
T

( ) ( )
0
0
/ 2
*
0
0
/ 2
1
T
n
T
x t y t nT dt
T
=

=

As ( ) 0 x t = for
0
2
T
t > ,

1.125
( ) ( ) ( )
0
0
1
p p
x y
n
R x t y t nT dt
T
=

=

( )
0
0
1
x y
n
R nT
T
=
= +

Taking the Fourier transform on both sides,
( ) ( )
0
0
1
F
p p
x y x y
n
R F R nT
T
=

= +

As ( ) ( ) ( )
x y
R X f Y f
,
( ) ( ) ( )
0
2
0
1
F
p p
j nf T
x y
n
R X f Y f e
T

=

=

( ) ( )
0
2
0
1
j nf T
n
X f Y f e
T

=

=

(A1.2.1)

But from Example 1.12, we have
( ) ( )
1
0 0
0
1
m m
t mT F f nf
T

= =

=

(A1.2.2)

0
2
0
1
j mf t
m
e
T
=

=

Replacing t by f ,
0
T by
0
f , we get the dual relation
( )
0
2
0
0
1
j mf T
m m
f mf e
f

= =

=

As
0
0
1
f
T
= , we have
0
2
0 0
1
j mf T
m m
m
f e
T T

= =

=

(A1.2.3)

1.126
Using Eq. A1.2.3 in Eq. A1.2.1, we obtain,
( )
2
0 0 0
0
1
F
p p
x y
m
m m m
R X Y f
T T T
T
=

=

(A1.2.4)
where ( )
0
0
m
f
T
m
X X f
T
=

=

and ( )
0
0
m
f
T
m
Y Y f
T

=

=

.


1.127
Appendix A1.3
Complex envelope of the output of a BP system
Let ( ) x t , a BP signal, be applied as input to a BP system with impulse response
( ) h t . Let the resulting output be denoted ( ) y t , which is also a BP signal. We
shall derive a relation for ( )
c e
y t in terms of ( )
c e
x t and ( )
c e
h t .

We know that,
( ) ( ) ( )
c c s c
x t x t t x t t cos sin = ,
( ) ( ) ( )
ce c s
x t x t j x t = + ,
( ) ( )
2
Re
c
j f t
ce
x t x t e

=

,
( ) ( ) ( )
{ }
1
2
ce c ce c
X f X f f X f f
= + +

. (A1.3.1)
Similarly, let
( ) ( ) ( )
ce c s
h t h t j h t = + , (A1.3.2)
( ) ( ) ( )
c c s c
h t h t t h t t cos sin = , (A1.3.3)
( ) ( )
2
Re
c
j f t
ce
h t h t e

=

,
( )
2 2
2
c c
j f t j f t
ce ce
h t h e h e

= + . (A1.3.4)
Taking the FT of Eq. A1.3.4, we have
( ) ( ) ( ) 2
ce c ce c
H f H f f H f f
= + +

(A1.3.5)
But ( ) ( )
2
Re
c
j f t
ce
y t y t e

=

.
Therefore,
( ) ( ) ( )
( ) ( )
2
ce c ce c
Y f f Y f f
Y f X f H f

+ +

= = (A1.3.6)
Because of Eq. A1.3.1 and A1.3.5, Eq. A1.3.6 becomes
( ) ( ) ( ) ( ) ( ) ( )
{ }
1
4
ce c ce c ce c ce c
X f H f H f f H f f X f f X f f

= + + + +

Consider the product term

1.128
( ) ( )
ce c ce c
H f f X f f
+

( )
ce c
H f f has spectrum confined to the range ( ) ,
c c
f B f B + . ( )
ce c
X f f
+

has nonzero spectral components only in the range ( ) ( ) { }
,
c c
f W f W + .
That is, the spectra ( )
ce c
H f f and ( )
ce c
X f f
+

do not overlap; hence the
product is zero. Similarly, ( ) ( ) 0
ce c ce c
H f f X f f
+ =

. Hence,

( ) ( )
( ) ( )
( ) ( )
1
2 4
1
4
ce c ce c
ce c ce c
ce c ce c
Y f f Y f f
H f f X f f
H f f X f f

+ +

=
+ + +

( )
ce c
Y f f has nonzero spectral components only in the range ( ) ,
c c
f B f B + .
That is,
( ) ( ) ( )
1 1
2 4
ce c ce c ce c
Y f f H f f X f f =

,
and ( ) ( ) ( ) ( ) ( )
1 1
2 4
ce c ce c ce c
Y f f H f f X f f

+ = + +

.
In other words,
( ) ( ) ( )
1
2
ce ce ce
Y f X f H f = (A1.3.7)
Therefore,
( ) ( ) ( )
1
2
ce ce ce
y t x t h t =

(A1.3.8)
From Eq. A1.3.8, we obtain the equations for ( )
c
y t and ( )
s
y t .
( ) ( ) ( ) ( ) ( )
{ }
ce c s c s
y t x t j x t h t j h t
1
2
= + +

Therefore,
( ) ( ) ( ) ( ) ( ) { }
c c c s s
y t x t h t x t h t
1
2
= (A1.3.9)
( ) ( ) ( ) ( ) ( ) { }
s c s s c
y t x t h t x t h t
1
2
= + (A1.3.10)
and, ( ) ( ) ( )
ce c s
y t y t j y t = + .

1.129

Exercise A1.3.1
Given the pairs ( ) ( ) ( )
c s
x t x t , and ( ) ( )
c s
h t h t ,

suggest a scheme to
recover ( )
c
y t and ( )
s
y t .

1.130
References
1. Oppenheim, A. V, Willisky, A. S, with Hamid Nawab, S., Signals and Systems
(2
nd
Edition), PHI, 1997
2. Ashok Ambardar, Analog and Digital Signal Processing (2
nd
Edition),
Brooks/Cole Publishing Company, Thomson Asia Pvt. Ltd., Singapore, 1999
3. Lathi, B. P., Signal Processing and Linear systems, Berkeley-Cambridge
Press, 1998

Note: The above three books have a large collection of problems. The student is
advised to try to solve them.

Other Suggested Books
1. Couch II, L. W., Digital and Analog Communication Systems (6
th
Edition)
Pearson Asia, 2001
2. Carlson, A. B., Communication Systems (4
th
Edition), Mc Graw-Hill, 2003
3. Lathi, B. P., Modern Digital and Analog Communication Systems (3
rd
Edition),
Oxford University Press, 1998

2.1
2 UU CHAPTER 2

Probability and Random Variables

2.1 Introduction
At the start of Sec. 1.1.2, we had indicated that one of the possible ways
of classifying the signals is: deterministic or random. By random we mean
unpredictable; that is, in the case of a random signal, we cannot with certainty
predict its future value, even if the entire past history of the signal is known. If the
signal is of the deterministic type, no such uncertainty exists.

Consider the signal
( ) ( )
1
cos 2 x t A f t = + . If A, and
1
f are known,
then (we are assuming them to be constants) we know the value of
( ) x t for all t .
( A, and
1
f can be calculated by observing the signal over a short period of
time).

Now, assume that
( ) x t is the output of an oscillator with very poor
frequency stability and calibration. Though, it was set to produce a sinusoid of
frequency
1
f f = , frequency actually put out maybe f
1
'
where ( ) f f f
1 1 1
'
.
Even this value may not remain constant and could vary with time. Then,
observing the output of such a source over a long period of time would not be of
much use in predicting the future values. We say that the source output varies in
a random manner.

Another example of a random signal is the voltage at the terminals of a
receiving antenna of a radio communication scheme. Even if the transmitted

2.2
(radio) signal is from a highly stable source, the voltage at the terminals of a
receiving antenna varies in an unpredictable fashion. This is because the
conditions of propagation of the radio waves are not under our control.

But randomness is the essence of communication. Communication
theory involves the assumption that the transmitter is connected to a source,
whose output, the receiver is not able to predict with certainty. If the students
know ahead of time what is the teacher (source +transmitter) is going to say
(and what jokes he is going to crack), then there is no need for the students (the
receivers) to attend the class!

Although less obvious, it is also true that there is no communication
problem unless the transmitted signal is disturbed during propagation or
reception by unwanted (random) signals, usually termed as noise and
interference. (We shall take up the statistical characterization of noise in
Chapter 3.)

However, quite a few random signals, though their exact behavior is
unpredictable, do exhibit statistical regularity. Consider again the reception of
radio signals propagating through the atmosphere. Though it would be difficult to
know the exact value of the voltage at the terminals of the receiving antenna at
any given instant, we do find that the average values of the antenna output over
two successive one minute intervals do not differ significantly. If the conditions of
propagation do not change very much, it would be true of any two averages (over
one minute) even if they are well spaced out in time. Consider even a simpler
experiment, namely, that of tossing an unbiased coin (by a person without any
magical powers). It is true that we do not know in advance whether the outcome
on a particular toss would be a head or tail (otherwise, we stop tossing the coin
at the start of a cricket match!). But, we know for sure that in a long sequence of
tosses, about half of the outcomes would be heads (If this does not happen, we
suspect either the coin or tosser (or both!)).

2.3
Statistical regularity of averages is an experimentally verifiable
phenomenon in many cases involving random quantities. Hence, we are tempted
to develop mathematical tools for the analysis and quantitative characterization
of random signals. To be able to analyze random signals, we need to understand
random variables. The resulting mathematical topics are: probability theory,
random variables and random (stochastic) processes. In this chapter, we shall
develop the probabilistic characterization of random variables. In chapter 3, we
shall extend these concepts to the characterization of random processes.

2.2 Basics of Probability
We shall introduce some of the basic concepts of probability theory by
defining some terminology relating to random experiments (i.e., experiments
whose outcomes are not predictable).

2.2.1. Terminology
Def. 2.1: Outcome
The end result of an experiment. For example, if the experiment consists
of throwing a die, the outcome would be anyone of the six faces,
1 6
,........, F F

Def. 2.2: Random experiment
An experiment whose outcomes are not known in advance. (e.g. tossing a
coin, throwing a die, measuring the noise voltage at the terminals of a resistor
etc.)

Def. 2.3: Random event
A random event is an outcome or set of outcomes of a random experiment
that share a common attribute. For example, considering the experiment of
throwing a die, an event could be the 'face
1
F ' or 'even indexed faces'
(
2 4 6
, , F F F ). We denote the events by upper case letters such as A, B or
A A
1 2
,


2.4
Def. 2.4: Sample space
The sample space of a random experiment is a mathematical abstraction
used to represent all possible outcomes of the experiment. We denote the
sample space by S .

Each outcome of the experiment is represented by a point in S and is
called a sample point. We use s (with or without a subscript), to denote a sample
point. An event on the sample space is represented by an appropriate collection
of sample point(s).
Def. 2.5: Mutually exclusive (disjoint) events
Two events A and B are said to be mutually exclusive if they have no
common elements (or outcomes).Hence if A and B are mutually exclusive, they
cannot occur together.

Def. 2.6: Union of events
The union of two events A and B, denoted A B , {also written as
( ) A B + or ( A or B)}is the set of all outcomes which belong to A or B or both.
This concept can be generalized to the union of more than two events.

Def. 2.7: Intersection of events
The intersection of two events, A and B, is the set of all outcomes which
belong to A as well as B. The intersection of A and B is denoted by
( ) A B
or simply
( ) AB . The intersection of A and B is also referred to as a joint event
A and B. This concept can be generalized to the case of intersection of three or
more events.

Def. 2.8: Occurrence of an event
An event A of a random experiment is said to have occurred if the
experiment terminates in an outcome that belongs to A.


2.5
Def. 2.9: Complement of an event
The complement of an event A, denoted by A is the event containing all
points in S but not in A.

Def. 2.10: Null event
The null event, denoted , is an event with no sample points. Thus = S
(note that if A and B are disjoint events, then AB = and vice versa).

2.2.2 Probability of an Event
The probability of an event has been defined in several ways. Two of the
most popular definitions are: i) the relative frequency definition, and ii) the
classical definition.

Def. 2.11: The relative frequency definition:
Suppose that a random experiment is repeated n times. If the event A
occurs
A
n times, then the probability of A, denoted by
( ) P A , is defined as
( ) lim
A
n
n
P A
n

=

(2.1)
A
n
n

represents the fraction of occurrence of A in n trials.

For small values of n , it is likely that
A
n
n

will fluctuate quite badly. But
as n becomes larger and larger, we expect,
A
n
n

to tend to a definite limiting
value. For example, let the experiment be that of tossing a coin and A the event
'outcome of a toss is Head'. If n is the order of 100,
A
n
n

may not deviate from

2.6
1
2
by more than, say ten percent and as n becomes larger and larger, we
expect
A
n
n

to converge to
1
2
.
Def. 2.12: The classical definition:
The relative frequency definition given above has empirical flavor. In the
classical approach, the probability of the event A is found without
experimentation. This is done by counting the total number N of the possible
outcomes of the experiment. If
A
N of those outcomes are favorable to the
occurrence of the event A, then
( )
A
N
P A
N
= (2.2)
where it is assumed that all outcomes are equally likely!

Whatever may the definition of probability, we require the probability
measure (to the various events on the sample space) to obey the following
postulates or axioms:
P1)
( ) 0 P A (2.3a)
P2)
( ) 1 P = S (2.3b)
P3) ( ) AB = , then
( ) ( ) ( ) P A B P A P B + = + (2.3c)

(Note that in Eq. 2.3(c), the symbol + is used to mean two different things;
namely, to denote the union of A and B and to denote the addition of two real
numbers). Using Eq. 2.3, it is possible for us to derive some additional
relationships:
i) If AB , then
( ) ( ) ( ) ( ) P A B P A P B P AB + = + (2.4)

ii) Let
1 2
, ,......,
n
A A A be random events such that:
a)
i j
A A = , for i j and (2.5a)

2.7
b)
1 2
......
n
A A A + + + = S . (2.5b)
Then,
( ) ( ) ( ) ( )
1 2
......
n
P A P AA P AA P AA = + + + (2.6)
where A is any event on the sample space.
Note:
n
A A A
1 2
, , , are said to be mutually exclusive (Eq. 2.5a) and exhaustive
(Eq. 2.5b).

iii)
( ) ( ) 1 P A P A = (2.7)
The derivation of Eq. 2.4, 2.6 and 2.7 is left as an exercise.

A very useful concept in probability theory is that of conditional
probability, denoted
( ) | P B A ; it represents the probability of B occurring, given
that A has occurred. In a real world random experiment, it is quite likely that the
occurrence of the event B is very much influenced by the occurrence of the
event A. To give a simple example, let a bowl contain 3 resistors and 1
capacitor. The occurrence of the event 'the capacitor on the second draw' is very
much dependent on what has been drawn at the first instant. Such dependencies
between the events is brought out using the notion of conditional probability .

The conditional probability
( ) | P B A can be written in terms of the joint
probability
( ) P AB and the probability of the event
( ) P A . This relation can be
arrived at by using either the relative frequency definition of probability or the
classical definition. Using the former, we have
( ) lim
AB
n
n
P AB
n

=

( ) lim
A
n
n
P A
n

=


2.8
where
AB
n is the number of times AB occurs in n repetitions of the experiment.
As
( ) | P B A refers to the probability of B occurring, given that A has occurred,
we have

Def 2.13: Conditional Probability
( ) | lim
AB
n
A
n
P B A
n

=

( )
( )
( ) lim , 0
AB
n
A
n
P AB
n
P A
n
P A
n

= =

(2.8a)
or
( ) ( ) ( ) | P AB P B A P A =
Interchanging the role of A and B, we have
( )
( )
( )
( ) | , 0
P AB
P A B P B
P B
= (2.8b)
Eq. 2.8(a) and 2.8(b) can be written as

( ) ( ) ( ) ( ) ( ) | | P AB P B A P A P B P A B = = (2.9)
In view of Eq. 2.9, we can also write Eq. 2.8(a) as
( )
( ) ( )
( )
|
|
P B P A B
P B A
P A
= , ( ) P A 0 (2.10a)
Similarly
( )
( ) ( )
( )
P A P B A
P A B
P B
|
| = , ( ) P B 0 (2.10b)
Eq. 2.10(a) or 2.10(b) is one form of Bayes rule or Bayes theorem.

Eq. 2.9 expresses the probability of joint event AB in terms of conditional
probability, say
( ) | P B A and the (unconditional) probability
( ) P A . Similar
relation can be derived for the joint probability of a joint event involving the
intersection of three or more events. For example
( ) P ABC can be written as

2.9
( ) ( ) ( ) | P ABC P AB P C AB =

( ) ( ) ( ) | | P A P B A P C AB = (2.11)

Another useful probabilistic concept is that of statistical independence.
Suppose the events A and B are such that
( ) ( ) | P B A P B = (2.13)
That is, knowledge of occurrence of A tells no more about the probability of
occurrence B than we knew without that knowledge. Then, the events A and B
are said to be statistically independent. Alternatively, if A and B satisfy the
Eq. 2.13, then

( ) ( ) ( ) P AB P A P B = (2.14)

Either Eq. 2.13 or 2.14 can be used to define the statistical independence
of two events. Note that if A and B are independent, then
( ) ( ) ( ) P AB P A P B = , whereas if they are disjoint, then
( ) 0 P AB = . The notion
of statistical independence can be generalized to the case of more than two
events. A set of k events
1 2
, ,......,
k
A A A are said to be statistically independent
if and only if (iff) the probability of every intersection of k or fewer events equal
the product of the probabilities of its constituents. Thus three events , , A B C are
independent when
Exercise 2.1
Let
1 2
, ,......,
n
A A A be n mutually exclusive and exhaustive events
and B is another event defined on the same space. Show that

( )
( ) ( )
( ) ( )
1
|
|
|
j j
j
n
j j
i
P B A P A
P A B
P B A P A
=
=
(2.12)
Eq. 2.12 represents another form of Bayes theorem.

2.10

( ) ( ) ( ) P AB P A P B =

( ) ( ) ( ) P AC P A P C =

( ) ( ) ( ) P BC P B P C =
and
( ) ( ) ( ) ( ) P ABC P A P B P C =

We shall illustrate some of the concepts introduced above with the help of
two examples.

Example 2.1
Priya (P1) and Prasanna (P2), after seeing each other for some time (and
after a few tiffs) decide to get married, much against the wishes of the parents on
both the sides. They agree to meet at the office of registrar of marriages at 11:30
a.m. on the ensuing Friday (looks like they are not aware of Rahu Kalam or they
dont care about it).

However, both are somewhat lacking in punctuality and their arrival times
are equally likely to be anywhere in the interval 11 to 12 hrs on that day. Also
arrival of one person is independent of the other. Unfortunately, both are also
very short tempered and will wait only 10 min. before leaving in a huff never to
meet again.

a) Picture the sample space
b) Let the event A stand for P1 and P2 meet. Mark this event on the sample
space.
c) Find the probability that the lovebirds will get married and (hopefully) will
live happily ever after.

a) The sample space is the rectangle, shown in Fig. 2.1(a).

2.11

Fig. 2.1(a): S of Example 2.1

b) The diagonal OP represents the simultaneous arrival of Priya and
Prasanna. Assuming that P1 arrives at 11: x , meeting between P1 and P2
would take place if P2 arrives within the interval a to b, as shown in the
figure. The event A, indicating the possibility of P1 and P2 meeting, is
shown in Fig. 2.1(b).

Fig. 2.1(b): The event A of Example 2.1

2.12
c)
Shaded area 11
Probability of marriage
Total area 36
= =

Example 2.2:
Let two honest coins, marked 1 and 2, be tossed together. The four
possible outcomes are
1 2
T T ,
1 2
T H ,
1 2
H T ,
1 2
H H . (
1
T indicates toss of coin 1
resulting in tails; similarly
2
T etc.) We shall treat that all these outcomes are
equally likely; that is the probability of occurrence of any of these four outcomes
is
1
4
. (Treating each of these outcomes as an event, we find that these events
are mutually exclusive and exhaustive). Let the event A be 'not
1 2
H H ' and B be
the event 'match'. (Match comprises the two outcomes
1 2
T T ,
1 2
H H ). Find
( ) | P B A . Are A and B independent?

We know that ( )
( )
( )
|
P AB
P B A
P A
= .
AB is the event 'not
1 2
H H ' and 'match'; i.e., it represents the outcome
1 2
T T .
Hence ( )
1
4
P AB = . The event A comprises of the outcomes
1 2
T T ,
1 2
T H and
1 2
H T ; therefore,
( )
3
4
P A =
( )
1
1
4
|
3
3
4
P B A = =
Intuitively, the result ( )
1
|
3
P B A = is satisfying because, given 'not
1 2
H H the
toss would have resulted in anyone of the three other outcomes which can be

2.13
treated to be equally likely, namely
1
3
. This implies that the outcome
1 2
T T given
'not
1 2
H H ', has a probability of
1
3
.
As ( )
1
2
P B = and ( )
1
|
3
P B A = , A and B are dependent events.

2.3 Random Variables
Let us introduce a transformation or function, say X , whose domain is the
sample space (of a random experiment) and whose range is in the real line; that
is, to each
i
s S , X assigns a real number,
( )
i
X s , as shown in Fig.2.2.

Fig. 2.2: A mapping
( ) X from S to the real line.

The figure shows the transformation of a few individual sample points as
well as the transformation of the event A, which falls on the real line segment
[ ]
1 2
, a a .

2.3.1 Distribution function:
Taking a specific case, let the random experiment be that of throwing a
die. The six faces of the die can be treated as the six sample points in S ; that is,

2.14
, 1, 2, ...... , 6
i i
F s i = = . Let ( )
i
X s i = . Once the transformation is induced,
then the events on the sample space will become transformed into appropriate
segments of the real line. Then we can enquire into the probabilities such as
( ) { }
P s X s a
1
:

<

( ) { }
P s b X s b
1 2
:

<

or
( ) { }
: P s X s c

=

These and other probabilities can be arrived at, provided we know the
Distribution Function of X, denoted by
( )
X
F which is given by
( ) ( ) { }
X
F x P s X s x :

=

(2.15)
That is,
( )
X
F x is the probability of the event, comprising all those sample points
which are transformed by X into real numbers less than or equal to x . (Note
that, for convenience, we use x as the argument of
( )
X
F . But, it could be any
other symbol and we may use
( )
X
F ,
( )
1 X
F a etc.) Evidently,
( )
X
F is a function
whose domain is the real line and whose range is the interval
[ ]
0, 1 ).

As an example of a distribution function (also called Cumulative
Distribution Function CDF), let S consist of four sample points,
1
s to
4
s , with
each with sample point representing an event with the probabilities ( )
1
1
4
P s = ,
( )
2
1
8
P s = , ( )
3
1
8
P s = and ( )
4
1
2
P s = . If
( ) 1.5, 1, 2, 3, 4
i
X s i i = = , then
the distribution function
( )
X
F x , will be as shown in Fig. 2.3.


2.15

Fig. 2.3: An example of a distribution function

( )
X
F satisfies the following properties:
i)
( ) 0,
X
F x x < <
ii)
( ) 0
X
F =
iii)
( ) 1
X
F =
iv) If a b > , then
( ) ( ) ( ) { }
:
X X
F a F b P s b X s a
= <

v) If a b > , then
( ) ( )
X X
F a F b

The first three properties follow from the fact that
( )
X
F represents the
probability and
( ) 1 P = S . Properties iv) and v) follow from the fact
( ) { } ( ) { } ( ) { }
: : : s X s b s b X s a s X s a < =

Referring to the Fig. 2.3, note that
( ) 0
X
F x = for 0.5 x < whereas
( )
X
F
1
0.5
4
= . In other words, there is a discontinuity of
1
4
at the point
x 0.5 = . In general, there is a discontinuity in
X
F of magnitude
a
P at a point
x a = , if and only if
( ) { }
:
a
P s X s a P

= =

(2.16)

2.16
The properties of the distribution function are summarized by saying that
( )
x
F is monotonically non-decreasing, is continuous to the right at each point
x TP
1
PT, and has a step of size
a
P at point a if and if Eq. 2.16 is satisfied.

Functions such as
( ) X for which distribution functions exist are called
Random Variables (RV). In other words, for any real x ,
( ) { }
: s X s x should
be an event in the sample space with some assigned probability. (The term
random variable is somewhat misleading because an RV is a well defined
function from S into the real line.) However, every transformation from S into
the real line need not be a random variable. For example, let S consist of six
sample points,
1
s to
6
s . The only events that have been identified on the sample
space are:
{ } { } { }
1 2 3 4 5 6
, , , , and A s s B s s s C s = = = and their probabilities
are ( ) ( ) ( ) = = = P A P B P C
2 1 1
, and
6 2 6
. We see that the probabilities for the
various unions and intersections of A, B and C can be obtained.

Let the transformation X be
( )
i
X s i = . Then the distribution function
fails to exist because
[ ] ( )
4
: 3.5 4.5 P s x P s < = is not known as
4
s is not an event on the sample
space.

TP
1
PT Let x a = . Consider, with 0 > ,
( ) ( ) ( ) lim lim
0 0
X X
P a X s a F a F a

< + = +

We intuitively feel that as 0 , the limit of the set ( ) { } : s a X s a < + is the null set and
can be proved to be so. Hence,
( ) ( ) 0
X X
a F a F
+
= , where ( )
0
lim a a
+

= +
That is, ( )
X
F x is continuous to the right.

2.17

2.3.2 Probability density function
Though the CDF is a very fundamental concept, in practice we find it more
convenient to deal with Probability Density Function (PDF). The PDF,
( )
X
f x is
defined as the derivative of the CDF; that is
( )
( )
X
X
d F x
f x
d x
= (2.17a)
or
( ) ( )
x
X X
F x f d

=
(2.17b)

The distribution function may fail to have a continuous derivative at a point
x a = for one of the two reasons:
i) the slope of the
( )
x
F x is discontinuous at x a =
ii)
( )
x
F x has a step discontinuity at x a =
The situation is illustrated in Fig. 2.4.

Exercise 2.2
Let S be a sample space with six sample points,
1
s to
6
s . The events
identified on S are the same as above, namely, { } ,
1 2
A s s = ,
{ }
3 4 5
, , B s s s = and { }
6
C s = with ( ) ( ) ( )
1 1 1
, and
3 2 6
P A P B P C = = = .
Let ( ) Y be the transformation,
( )
1, 1, 2
2, 3, 4, 5
3, 6
i
i
Y s i
i
=
= =

Show that ( ) Y is a random variable by finding ( )
Y
F y . Sketch ( )
Y
F y .

2.18

Fig. 2.4: A CDF without a continuous derivative

As can be seen from the figure,
( )
X
F x has a discontinuous slope at
1 x = and a step discontinuity at 2 x = . In the first case, we resolve the
ambiguity by taking
X
f to be a derivative on the right. (Note that
( )
X
F x is
continuous to the right.) The second case is taken care of by introducing the
impulse in the probability domain. That is, if there is a discontinuity in
X
F at
x a = of magnitude
a
P , we include an impulse
( )
a
P x a in the PDF. For
example, for the CDF shown in Fig. 2.3, the PDF will be,

( )
1 1 1 1 1 3 1 5
4 2 8 2 8 2 2 2
X
f x x x x x

= + + + +

(2.18)

In Eq. 2.18,
( )
X
f x has an impulse of weight
1
8
at
1
2
x = as
1 1
2 8
P X

= =

. This impulse function cannot be taken as the limiting case of
an even function (such as
1 x
ga

) because,
( )
1 1
2 2
0 0
1 1
2 2
1 1 1
lim lim
8 2 16
X
f x dx x dx

=


2.19
However, ( )

=
1
2
0
1
2
1
lim
8
X
f x dx . This ensures,
( )

<
<
2 1 1
,
8 2 2
3 1 3
,
8 2 2
X
x
F x
x

Such an impulse is referred to as the left-sided delta function.
As
X
F is non-decreasing and
( ) 1
X
F = , we have
i)
( ) 0
X
f x (2.19a)
ii) ( ) 1
X
f x dx
(2.19b)

Based on the behavior of CDF, a random variable can be classified as:
i) continuous (ii) discrete and (iii) mixed. If the CDF,
( )
X
F x , is a continuous
function of x for all x , then X TP
1
PT is a continuous random variable. If
( )
X
F x is a
staircase, then X corresponds to a discrete variable. We say that X is a mixed
random variable if
( )
X
F x is discontinuous but not a staircase. Later on in this
lesson, we shall discuss some examples of these three types of variables.

We can induce more than one transformation on a given sample space. If
we induce k such transformations, then we have a set of k co-existing random
variables.

2.3.3 Joint distribution and density functions
Consider the case of two random variables, say X and Y . They can be
characterized by the (two-dimensional) joint distribution function, given by
( ) ( ) ( ) { }
,
, : ,
X Y
F x y P s X s x Y s y

=

(2.20)

TP
1
PT As the domain of the random variable ( ) X is known, it is convenient to denote the variable
simply by X .

2.20
That is,
( )
,
,
X Y
F x y is the probability associated with the set of all those
sample points such that under X , their transformed values will be less than or
equal to x and at the same time, under Y , the transformed values will be less
than or equal to y . In other words,
( )
, 1 1
,
X Y
F x y is the probability associated with
the set of all sample points whose transformation does not fall outside the
shaded region in the two dimensional (Euclidean) space shown in Fig. 2.5.

Fig. 2.5: Space of
( ) ( ) { }
, X s Y s corresponding to
( )
, 1 1
,
X Y
F x y

Looking at the sample space S , let A be the set of all those sample
points s S such that
( )
1
X s x . Similarly, if B is comprised of all those
sample points s S such that
( )
1
Y s y ; then
( )
1 1
, F x y is the probability
associated with the event AB .

Properties of the two dimensional distribution function are:
i)
( )
,
, 0, ,
X Y
F x y x y < < < <
ii)
( ) ( )
, ,
, , 0
X Y X Y
F y F x = =
iii)
( )
,
, 1
X Y
F =
iv)
( ) ( )
,
,
X Y Y
F y F y =
v)
( ) ( )
,
,
X Y X
F x F x =

2.21
vi) If
2 1
x x > and
2 1
y y > , then
( ) ( ) ( )
, 2 2 , 2 1 , 1 1
, , ,
X Y X Y X Y
F x y F x y F x y

We define the two dimensional joint PDF as
( ) ( )
=

X Y X Y
f x y F x y
x y
2
, ,
, , (2.21a)
or ( ) ( )
, ,
, ,
y x
X Y X Y
F x y f d d

=

(2.21b)
The notion of joint CDF and joint PDF can be extended to the case of k random
variables, where 3 k .

Given the joint PDF of random variables X and Y , it is possible to obtain
the one dimensional PDFs,
( )
X
f x and
( )
Y
f y . We know that,
( ) ( )
, 1 1
,
X Y X
F x F x = .
That is, ( ) ( )
1
1 ,
,
x
X X Y
F x f d d

=

( ) ( )
1
1 ,
1
,
x
X X Y
d
f x f d d
d x

=

(2.22)
Eq. 2.22 involves the derivative of an integral. Hence,
( )
( )
( )
1
1 , 1
,
X
X X Y
x x
d F x
f x f x d
d x

=
= =

or ( ) ( )
,
,
X X Y
f x f x y dy

=

(2.23a)
Similarly, ( ) ( )
,
,
Y X Y
f y f x y dx

=

(2.23b)
(In the study of several random variables, the statistics of any individual variable
is called the marginal. Hence it is common to refer
( )
X
F x as the marginal

2.22
distribution function of X and
( )
X
f x as the marginal density function.
( )
Y
F y and
( )
Y
f y are similarly referred to.)

2.3.4 Conditional density
Given
( )
,
,
X Y
f x y , we know how to obtain
( )
X
f x or
( )
Y
f y . We might also
be interested in the PDF of X given a specific value of
1
y y = . This is called the
conditional PDF of X , given
1
y y = , denoted ( )
X Y
f x y
| 1
| and defined as
( )
( )
( )
X Y
X Y
Y
f x y
f x y
f y
, 1
| 1
1
,
| = (2.24)
where it is assumed that
( )
1
0
Y
f y . Once we understand the meaning of
conditional PDF, we might as well drop the subscript on y and denote it by
( )
|
|
X Y
f x y . An analogous relation can be defined for
( )
|
|
Y X
f y x . That is, we have
the pair of equations,
( )
( )
( )
,
|
,
|
X Y
X Y
Y
f x y
f x y
f y
= (2.25a)
and ( )
( )
( )
,
|
,
|
X Y
Y X
X
f x y
f y x
f x
= (2.25b)
or
( ) ( ) ( )
, |
, |
X Y X Y Y
f x y f x y f y = (2.25c)

( ) ( )
|
|
Y X X
f y x f x = (2.25d)
The function
( )
|
|
X Y
f x y may be thought of as a function of the variable x with
variable y arbitrary, but fixed. Accordingly, it satisfies all the requirements of an
ordinary PDF; that is,

( )
|
| 0
X Y
f x y (2.26a)
and ( )
|
| 1
X Y
f x y dx
(2.26b)


2.23
2.3.5 Statistical independence
In the case of random variables, the definition of statistical independence
is somewhat simpler than in the case of events. We call k random variables
1 2
, , ......... ,
k
X X X statistically independent iff, the k -dimensional joint PDF
factors into the product
( )
1
i
k
X i
i
f x
=
(2.27)
Hence, two random variables X and Y are statistically independent, iff,
( ) ( ) ( )
,
,
X Y X Y
f x y f x f y = (2.28)
and three random variables X , Y , Z are independent, iff
( ) ( ) ( ) ( )
, ,
, ,
X Y Z X Y z
f x y z f x f y f z = (2.29)
Statistical independence can also be expressed in terms of conditional
PDF. Let X and Y be independent. Then,

( ) ( ) ( )
,
,
X Y X Y
f x y f x f y =
Making use of Eq. 2.25(c), we have

( ) ( ) ( ) ( )
|
|
X Y Y X Y
f x y f y f x f y =
or
( ) ( )
|
|
X Y X
f x y f x = (2.30a)
Similarly,
( ) ( )
|
|
Y X Y
f y x f y = (2.30b)
Eq. 2.30(a) and 2.30(b) are alternate expressions for the statistical independence
between X and Y . We shall now give a few examples to illustrate the concepts
discussed in sec. 2.3.

Example 2.3
A random variable X has
( )
2
0 , 0
, 0 10
100 , 10
X
x
F x K x x
K x
<
>

i) Find the constant K
ii) Evaluate
( ) 5 P X and
( ) 5 7 P X <

2.24
iii) What is
( )
X
f x ?

i) ( )
X
F K K
1
100 1
100
= = = .
ii)
( ) ( )
1
5 5 25 0.25
100
X
P x F

= = =

( ) ( ) ( ) 5 7 7 5 0.24
X X
P X F F < = =
( )
( )
0 , 0
0.02 , 0 10
0 , 10
X
X
x
d F x
f x x x
d x
x
<
= =
>

Note: Specification of
( )
X
f x or
( )
X
F x is complete only when the algebraic
expression as well as the range of X is given.

Example 2.4
Consider the random variable X defined by the PDF
( ) ,
b x
X
f x ae x
= < < where a and b are positive constants.

i) Determine the relation between a and b so that
( )
X
f x is a PDF.
ii) Determine the corresponding
( )
X
F x
iii) Find
[ ]
1 2 P X < .

i) As can be seen
( ) 0
X
f x for x < < . In order for
( )
X
f x to
represent a legitimate PDF, we require
0
2 1
b x b x
ae dx ae dx

= =

.
That is,
0
1
2
b x
ae dx
; hence 2 b a = .
ii) the given PDF can be written as
( )
, 0
, 0.
b x
X
b x
a e x
f x
a e x
<


2.25
For 0 x < , we have
( )
1
2 2
x
b b x
X
b
F x e d e

= =
.
Consider 0 x > . Take a specific value of 2 x =
( ) ( )
2
2
X X
F f x d x

=

( )
2
1
X
f x d x
=

But for the problem on hand, ( ) ( )
X X
f x d x f x d x
2
2

=

Therefore, ( )
1
1 , 0
2
b x
X
F x e x
= >
We can now write the complete expression for the CDF as
( )
1
, 0
2
1
1 , 0
2
b x
X
b x
e x
F x
e x
<

iii) ( ) ( ) ( )
2
1
2 1
2
b b
X X
F F e e

=

Example 2.5
Let ( )
,
1
, 0 , 0 2
,
2
0 ,
X Y
x y y
f x y
otherwise

Find (a) i)
( )
|
|1
Y X
f y and ii)
( )
|
|1.5
Y X
f y
(b) Are X and Y independent?

(a) ( )
( )
( )
,
|
,
|
X Y
Y X
X
f x y
f y x
f x
=

2.26
( )
X
f x can be obtained from
, X Y
f by integrating out the unwanted variable y
over the appropriate range. The maximum value taken by y is 2; in
addition, for any given , x y x . Hence,
( )
2
1
1
2 2
X
x
x
f x d y = =
, 0 2 x
Hence,
(i) ( )
|
1
1 , 1 2
2
|1
1
0,
2
Y X
y
f y
otherwise

= =

(ii) ( )
|
1
2, 1.5 2
2
|1.5
1
0,
4
Y X
y
f y
otherwise

= =

b) the dependence between the random variables X and Y is evident from
the statement of the problem because given a value of
1
X x = , Y should
be greater than or equal to
1
x for the joint PDF to be non zero. Also we see
that
( )
|
|
Y X
f y x depends on x whereas if X and Y were to be independent,
then
( ) ( )
|
|
Y X Y
f y x f y = .


2.27
Exercise 2.3
For the two random variables X and Y , the following density functions
have been given. (Fig. 2.6)

Fig. 2.6: PDFs for the exercise 2.3
Find
a) ( )
,
,
X Y
f x y
b) Show that
( )
, 0 10
100
1
, 10 20
5 100
Y
y
y
f y
y
y
<

2.4 Transformation of Variables
The process of communication involves various operations such as
modulation, detection, filtering etc. In performing these operations, we typically
generate new random variables from the given ones. We will now develop the
necessary theory for the statistical characterization of the output random
variables, given the transformation and the input random variables.


2.28
2.4.1 Functions of one random variable
Assume that X is the given random variable of the continuous type with
the PDF,
( )
X
f x . Let Y be the new random variable, obtained from X by the
transformation
( ) Y g X = . By this we mean the number
( )
1
Y s associated with
any sample point
1
s is
( ) ( ) ( )
1 1
Y s g X s =
Our interest is to obtain
( )
Y
f y . This can be obtained with the help of the following
theorem (Thm. 2.1).
Theorem 2.1
Let
( ) Y g X = . To find
( )
Y
f y , for the specific y , solve the equation
( ) y g x = . Denoting its real roots by
n
x ,
( ) ( )
1
,......, , .......
n
y g x g x = = = = ,
we will show that
( )
( )
( )
( )
( )
1
1
........... .........
' '
X X n
Y
n
f x f x
f y
g x g x
= + + + (2.31)
where
( ) ' g x is the derivative of
( ) g x .

Proof: Consider the transformation shown in Fig. 2.7. We see that the equation
( ) y g x
1
= has three roots namely,
1 2 3
, and x x x .


2.29

Fig. 2.7: X Y transformation used in the proof of theorem 2.1

We know that
( ) [ ]
Y
f y d y P y Y y d y = < + . Therefore, for any given
1
y , we
need to find the set of values x such that
( )
1 1
y g x y d y < + and the
probability that X is in this set. As we see from the figure, this set consists of the
following intervals:

1 1 1 2 2 2 3 3 3
, , x x x d x x d x x x x x x d x < + + < < +
where
1 3 2
0, 0, but 0 d x d x d x > > < .
From the above, it follows that
[ ] [ ] [ ]
[ ]
< + = < + + + <
+ < +
P y Y y d y P x X x d x P x dx X x
P x X x d x
1 1 1 1 1 2 2 2
3 3 3
This probability is the shaded area in Fig. 2.7.

[ ] ( )
( )
1 1 1 1 1 1
1
,
'
X
d y
P x X x d x f x d x d x
g x
< + = =

[ ] ( )
( )
2 2 2 2 2 2
2
,
'
X
d y
P x d x x x f x d x d x
g x
+ < = =

[ ] ( )
( )
3 3 3 3 3 3
3
,
'
X
d y
P x X x d x f x d x d x
g x
< + = =


2.30
We conclude that
( )
( )
( )
( )
( )
( )
( )
= + +
X X X
Y
f x f x f x
f y d y d y d y d y
g x g x g x
1 2 3
1 3 2
' ' '
(2.32)
and Eq. 2.31 follows, by canceling d y from the Eq. 2.32.

Note that if
( )
1
g x y = = constant for every x in the interval
( )
0 1
, x x , then
we have
[ ] ( ) ( ) ( )
1 0 1 1 0 X X
P Y y P x X x F x F x = = < = ; that is
( )
Y
F y is
discontinuous at
1
y y = . Hence
( )
Y
f y contains an impulse,
( )
1
y y of area
( ) ( )
1 0 X X
F x F x .

We shall take up a few examples.

Example 2.6
( ) Y g X X a = = + , where a is a constant. Let us find ( )
Y
f y .

We have
( ) ' 1 g x = and x y a = . For a given y , there is a unique x
satisfying the above transformation. Hence,
( )
( )
( ) '
X
Y
f x
f y d y d y
g x
= and as
( ) ' 1 g x = , we have
( ) ( )
Y X
f y f y a =
Let ( )
1 , 1
0 ,
X
x x
f x
elsewhere

and 1 a =
Then,
( ) 1 1
Y
f y y = +
As 1 y x = , and x ranges from the interval
( ) 1, 1 , we have the range of y
as
( ) 2, 0 .

2.31
Hence ( )
1 1, 2 0
0 ,
Y
y y
f y
elsewhere
+

( )
X
F x and
( )
Y
F y are shown in Fig. 2.8.

Fig. 2.8:
( )
X
F x and
( )
Y
F y of example 2.6

As can be seen, the transformation of adding a constant to the given variable
simply results in the translation of its PDF.

Example 2.7
Let Y b X = , where b is a constant. Let us find ( )
Y
f y .

Solving for X , we have
1
X Y
b
= . Again for a given y , there is a unique
x . As
( ) ' g x b = , we have ( )
1
Y X
y
f y f
b b

=

.
Let ( )
X
x
x
f x
otherwise
1 , 0 2
2
0 ,

and 2 b = , then

2.32
( )
1
1 , 4 0
2 4
0 ,
Y
y
y
f y
otherwise

+

( )
X
f x and
( )
Y
f y are sketched in Fig. 2.9.

Fig. 2.9:
( )
X
f x and
( )
Y
f y of example 2.7

Example 2.8

2
, 0 Y a X a = > . Let us find ( )
Y
f y .

Exercise 2.4
Let Y a X b = + , where a and b are constants. Show that
( )
1
Y X
y b
f y f
a a

=

.
If ( )
X
f x is as shown in Fig.2.8, compute and sketch ( )
Y
f y for 2 a = , and
1 b = .

2.33

( ) ' 2 g x a x = . If 0 y < , then the equation
2
y a x = has no real solution.
Hence
( ) 0
Y
f y = for 0 y < . If 0 y , then it has two solutions,
1
y
x
a
= and
2
y
x
a
= , and Eq. 2.31 yields
( )
1
, 0
2
0 ,
X X
Y
y y
f f y
a a
y
f y
a
a
otherwise

+

Let 1 a = , and ( )
2
1
exp ,
2
2
X
x
f x x

= < <

(Note that
( ) exp is the same as e
)
Then ( )
1
exp exp
2 2
2 2
Y
y y
f y
y

= +

y
y
y
otherwise
1
exp , 0
2 2
0 ,

=

Sketching of
( )
X
f x and
( )
Y
f y is left as an exercise.

Example 2.9
Consider the half wave rectifier transformation given by
0, 0
, 0
X
Y
X X

=

>

a) Let us find the general expression for
( )
Y
f y
b) Let ( )
X
x
f x
otherwise
1 1 3
,
2 2 2
0,
< <

We shall compute and sketch ( )
Y
f y .


2.34
a) Note that
( ) g X is a constant (equal to zero) for X in the range of
( ) , 0 .
Hence, there is an impulse in
( )
Y
f y at 0 y = whose area is equal to
( ) 0
X
F . As Y is nonnegative,
( ) 0
Y
f y = for 0 y < . As Y X = for 0 x > ,
we have
( ) ( )
Y X
f y f y = for 0 y > . Hence
( )
( ) ( ) ( ) ( ) 0
0,
X X
Y
f y w y F y
f y
otherwise
+

where ( )
1 , 0
0,
y
w y
otherwise

=

b) Specifically, let ( )
1 1 3
,
2 2 2
0 ,
X
x
f x
elsewhere
< <

Then,
( )
( )
1
, 0
4
1 3
, 0
2 2
0 ,
Y
y y
f y y
otherwise
= <

( )
X
f x and
( )
Y
f y are sketched in Fig. 2.10.

Fig. 2.10:
( )
X
f x and
( )
Y
f y for the example 2.9


2.35
Note that X , a continuous RV is transformed into a mixed RV, Y .

Example 2.10
Let
X
Y
X
1, 0
1, 0
<
=

+

a) Let us find the general expression for
( )
Y
f y .
b) We shall compute and sketch ( )
Y
f x assuming that
( )
X
f x is the same as
that of Example 2.9.

a) In this case, Y assumes only two values, namely 1 . Hence the PDF of Y
has only two impulses. Let us write
( )
Y
f y as
( ) ( ) ( )
1 1
1 1
Y
f y P y P y
= + + where
[ ] [ ]
1 1
0 0 P P X and P X
= <
b) Taking
( )
X
f x of example 2.9, we have
1
3
4
P = and
1
1
4
P
= . Fig. 2.11
has the sketches
( )
X
f x and
( )
Y
f y .

Fig. 2.11: ( )
X
f x and ( )
Y
f y for the example 2.10

Note that this transformation has converted a continuous random variable X into
a discrete random variable Y .


2.36

Exercise 2.5
Let a random variable X with the PDF shown in Fig. 2.12(a) be the
input to a device with the input-output characteristic shown in Fig. 2.12(b).
Compute and sketch ( )
Y
f y .

Fig. 2.12: (a) Input PDF for the transformation of exercise 2.5
(b) Input-output transformation

Exercise 2.6
The random variable X of exercise 2.5 is applied as input to the
X Y transformation shown in Fig. 2.13. Compute and sketch ( )
Y
f y .

2.37
We now assume that the random variable X is of discrete type taking on
the value
k
x with probability
k
P . In this case, the RV,
( ) Y g X = is also
discrete, assuming the value
( )
k k
Y g x = with probability
k
P .

If
( )
k
y g x = for only one
k
x x = , then
[ ] [ ]
k k k
P Y y P X x P = = = = . If
however,
( )
k
y g x = for
k
x x = and
m
x x = , then
[ ]
k k m
P Y y P P = = + .

Example 2.11
Let
2
Y X = .
a) If ( ) ( )
X
i
f x x i
6
1
1
6
=
=
, find ( )
Y
f y .
b) If ( ) ( )
X
i
f x x i
3
2
1
6
=
=
, find ( )
Y
f y .
a) If X takes the values
( ) 1, 2, ......., 6 with probability of
1
6
, then Y takes the
values
2 2 2
1 , 2 , ......., 6 with probability
1
6
. That is, ( ) ( )
Y
i
f y x i
6
2
1
1
6
=
=
.
b) If, however, X takes the values 2, 1, 0, 1, 2, 3 with probability
1
6
, then
Y takes the values 0, 1, 4, 9 with probabilities
1 1 1 1
, , ,
6 3 3 6
respectively.
That is,
( ) ( ) ( ) ( ) ( )
1 1
9 1 4
6 3
Y
f y y y y y = + + +

2.4.2 Functions of two random variables
Given two random variables X and Y (assumed to be continuous type),
two new random variables, Z and W are defined using the transformation

( ) , Z g X Y = and
( ) , W h X Y =

2.38
Given the above transformation and
( )
,
,
X Y
f x y , we would like to obtain
( )
,
,
Z W
f z w . For this, we require the J acobian of the transformation, denoted
,
,
z w
J
x y

where
,
,
z z
x y
z w
J
x y w w
x y

=

z w z w
x y y x

=

That is, the J acobian is the determinant of the appropriate partial derivatives. We
shall now state the theorem which relates
( )
,
,
Z W
f z w and
( )
,
,
X Y
f x y .

We shall assume that the transformation is one-to-one. That is, given

( )
1
, g x y z = , (2.33a)

( )
1
, h x y w = , (2.33b)
then there is a unique set of numbers,
( )
1 1
, x y satisfying Eq. 2.33.

Theorem 2.2: To obtain
( )
,
,
Z W
f z w , solve the system of equations

( )
1
, g x y z = ,
( ) h x y w
1
, = ,
for x and y . Let
( )
1 1
, x y be the result of the solution. Then,
( )
( )
, 1 1
,
1 1
,
,
,
,
X Y
Z W
f x y
f z w
z w
J
x y
=

(2.34)

Proof of this theorem is given in appendix A2.1. For a more general version of
this theorem, refer [1].

2.39
Example 2.12
X and Y are two independent RVs with the PDFs ,
( )
1
, 1
2
X
f x x =
( )
1
, 1
2
Y
f y y =
If Z X Y = + and W X Y = , let us find (a)
( )
,
,
Z W
f z w and (b)
( )
Z
f z .

a) From the given transformations, we obtain ( )
1
2
x z w = + and
( )
1
2
y z w = . We see that the mapping is one-to-one. Fig. 2.14(a)
depicts the (product) space A on which
( )
,
,
X Y
f x y is non-zero.

Fig. 2.14: (a) The space where
, X Y
f is non-zero
(b) The space where
, Z W
f is non-zero


2.40
We can obtain the space B (on which
( )
,
,
Z W
f z w is non-zero) as follows:
space A space B
The line 1 x =
( )
1
1 2
2
z w w z + = = +
The line 1 x =
( ) ( )
1
1 2
2
z w w z + = = +
The line 1 y =
( ) = = z w w z
1
1 2
2

The line 1 y =
( )
1
1 2
2
z w w z = = +

The space B is shown in Fig. 2.14(b). The J acobian of the transformation
is

1 1
,
2
, 1 1
z w
J
x y

= =

and
( ) 2 J = .
Hence ( )
,
1
1
, ,
4
,
8
2
0 ,
Z W
z w
f z w
otherwise
= =

B

b) ( ) ( )
,
,
Z Z W
f z f z w dw

=

From Fig. 2.14(b), we can see that, for a given z ( 0 z ), w can take
values only in the to z z . Hence
( )
1 1
, 0 2
8 4
z
Z
z
f z dw z z
+
= =

For z negative, we have
( )
1 1
, 2 0
8 4
z
Z
z
f z dw z z
= =


2.41
Hence ( )
1
, 2
4
Z
f z z z =

Example 2.13
Let the random variables R and be given by,
2 2
R X Y = + and
tan
Y
arc
X

=

where we assume 0 R and < < . It is given that
( )
2 2
,
1
, exp , ,
2 2
X Y
x y
f x y x y
+
= < <

.
Let us find
( )
,
,
R
f r
.

As the given transformation is from cartesian-to-polar coordinates, we can
write cos x r = and sin y r = , and the transformation is one -to -one;
cos sin
1
sin cos
r r
x y
J
r
r r
x y

= = =

Hence, ( )
R
r r
r
f r
otherwise
2
,
exp , 0 ,
, 2 2
0 ,

< <

=

It is left as an exercise to find
( )
R
f r ,
( ) f
and to show that R and are

independent variables.

Theorem 2.2 can also be used when only one function
( ) , Z g X Y = is
specified and what is required is
( )
Z
f z . To apply the theorem, a conveniently
chosen auxiliary or dummy variable W is introduced. Typically W X = or
W Y = ; using the theorem
( )
,
,
Z W
f z w is found from which
( )
Z
f z can be
obtained.


2.42
Let Z X Y = + and we require
( )
Z
f z . Let us introduce a dummy variable
W Y = . Then, X Z W = , and Y W = .
As J 1 = ,

( ) ( )
, ,
, ,
Z W X Y
f z w f z w w =
and ( ) ( )
,
,
Z X Y
f z f z w w dw
(2.35)
If X and Y are independent, then Eq. 2.35 becomes
( ) ( ) ( )
Z X Y
f z f z w f w dw
(2.36a)
That is,
Z X Y
f f f = (2.36b)

Example 2.14
Let X and Y be two independent random variables, with
( )
1
, 1 1
2
0 ,
X
x
f x
otherwise

( )
1
, 2 1
3
0 ,
Y
y
f y
otherwise

If Z X Y = + , let us find
[ ]
2 P Z .

From Eq. 2.36(b),
( )
Z
f z is the convolution of
( )
X
f z and
( )
Y
f z . Carrying
out the convolution, we obtain
( )
Z
f z as shown in Fig. 2.15.


2.43

Fig. 2.15: PDF of Z X Y = + of example 2.14

[ ]
2 P Z is the shaded area
1
12
= .

Example 2.15
Let
X
Z
Y
= ; let us find an expression for
( )
Z
f z .

Introducing the dummy variable W Y = , we have
X Z W =
Y W =
As
1
J
w
= , we have
( ) ( )
,
,
Z X Y
f z w f zw w dw

=

Let ( )
,
1
, 1, 1
,
4
0 ,
X Y
x y
x y
f x y
elsewhere
+

Then ( )
2
1
4
Z
zw
f z w dw

+
=


2.44

( )
,
,
X Y
f x y is non-zero if
( ) , x y A . where A is the product space
1 x and 1 y (Fig.2.16a). Let
( )
,
,
Z W
f z w be non-zero if
( ) , z w B . Under
the given transformation, B will be as shown in Fig. 2.16(b).

Fig. 2.16: (a) The space where
, X Y
f is non-zero
(b) The space where
, Z W
f is non-zero

To obtain
( )
Z
f z from
( )
,
,
Z W
f z w , we have to integrate out w over the
appropriate ranges.
i) Let 1 z < ; then
( )
1 1 2 2
1 0
1 1
2
4 4
Z
zw zw
f z w dw w dw
+ +
= =

1
1
4 2
z
= +

ii) For 1 z > , we have
( )
1
2
2 3
0
1 1 1 1
2
4 4 2
z
Z
zw
f z w dw
z z
+
= = +

iii) For 1 z < , we have
( )
1
2
2 3
0
1 1 1 1
2
4 4 2
z
Z
zw
f z w dw
z z
+
= = +


2.45
Hence ( )
2 3
1
1 , 1
4 2
1 1 1
, 1
4 2
Z
z
z
f z
z
z z

+

+ >

In transformations involving two random variables, we may encounter a
situation where one of the variables is continuous and the other is discrete; such
cases are handled better, by making use of the distribution function approach,
as illustrated below.

Example 2.16
The input to a noisy channel is a binary random variable with
[ ] [ ]
1
0 1
2
P X P X = = = = . The output of the channel is given by Z X Y = +
Exercise 2.7
Let Z X Y = and W Y = .
a) Show that ( )
,
1
,
Z X Y
z
f z f w dw
w w

b) X and Y be independent with
( )
2
1
, 1
1
X
f x x
x
=
+

and ( )
2
2
, 0
0 ,
y
Y
y e y
f y
otherwise

Show that ( )
2
2
1
,
2
z
Z
f z e z
= < <
.

2.46
where Y is the channel noise with ( )
2
2
1
,
2
y
Y
f y e y
= < <
. Find
( )
Z
f z .

Let us first compute the distribution function of Z from which the density
function can be derived.

( ) [ ] [ ] [ ] [ ]
| 0 0 | 1 1 P Z z P Z z X P X P Z z X P X = = = + = =
As Z X Y = + , we have
[ ] ( ) | 0
Y
P Z z X F z = =
Similarly
[ ] ( ) ( ) | 1 1 1
Y
P Z z X P Y z F z = = =

Hence ( ) ( ) ( )
1 1
1
2 2
Z Y Y
F z F z F z = + . As ( ) ( )
Z Z
d
f z F z
d z
= , we have
( )
( )
2
2
1
1 1 1
exp exp
2 2 2
2 2
Z
z
z
f z

= +

The distribution function method, as illustrated by means of example 2.16, is a
very basic method and can be used in all situations. (In this method, if
( ) Y g X = , we compute
( )
Y
F y and then obtain ( )
( )
Y
Y
d F y
f y
d y
= . Similarly, for
transformations involving more than one random variable. Of course computing
the CDF may prove to be quite difficult in certain situations. The method of
obtaining PDFs based on theorem 2.1 or 2.2 is called change of variable
method.) We shall illustrate the distribution function method with another
example.

Example 2.17
Let Z X Y = + . Obtain
( )
Z
f z , given
( )
,
,
X Y
f x y .


2.47

[ ] [ ]
P Z z P X Y z = +

[ ]
P Y z x =
This probability is the probability of
( ) , X Y lying in the shaded area shown in Fig.
2.17.

Fig. 2.17: Shaded area is the
[ ]
P Z z

That is,
( ) ( )
,
,
z x
Z X Y
F z d x f x y d y

=

( ) ( )
,
,
z x
Z X Y
f z d x f x y d y
z

( )
,
,
z x
X Y
d x f x y d y
z

( )
,
,
X Y
f x z x d x
(2.37a)
It is not too difficult to see the alternative form for
( )
Z
f z , namely,

2.48
( ) ( )
,
,
Z X Y
f z f z y y d y
(2.37b)
If X and Y are independent, we have
( ) ( ) ( )
Z X Y
f z f z f z = . We note that Eq.
2.37(b) is the same as Eq. 2.35.

So far we have considered the transformations involving one or two
variables. This can be generalized to the case of functions of n variables. Details
can be found in [1, 2].

2.5 Statistical Averages
The PDF of a random variable provides a complete statistical
characterization of the variable. However, we might be in a situation where the
PDF is not available but are able to estimate (with reasonable accuracy) certain
(statistical) averages of the random variable. Some of these averages do provide
a simple and fairly adequate (though incomplete) description of the random
variable. We now define a few of these averages and explore their significance.

The mean value (also called the expected value, mathematical
expectation or simply expectation) of random variable X is defined as

[ ] ( )
X X
m E X X x f x d x

= = =

(2.38)
where E denotes the expectation operator. Note that
X
m is a constant. Similarly,
the expected value of a function of X ,
( ) g X , is defined by
( ) ( ) ( ) ( )
X
E g X g X g x f x d x

= =

(2.39)
Remarks: The terminology expected value or expectation has its origin in games
of chance. This can be illustrated as follows: Three small similar discs, numbered
1,2 and 2 respectively are placed in bowl and are mixed. A player is to be

2.49
blindfolded and is to draw a disc from the bowl. If he draws the disc numbered 1,
he will receive nine dollars; if he draws either disc numbered 2, he will receive 3
dollars. It seems reasonable to assume that the player has a '1/3 claim' on the 9
dollars and '2/3 claim' on three dollars. His total claim is 9(1/3) +3(2/3), or five
dollars. If we take X to be (discrete) random variable with the PDF
( ) ( ) ( )
1 2
1 2
3 3
X
f x x x = + and
( ) 15 6 g X X = , then
( ) ( ) ( ) 15 6 5
X
E g X x f x d x

= =

That is, the mathematical expectation of
( ) g X is precisely the player's claim or
expectation [3]. Note that
( ) g x is such that
( ) 1 9 g = and
( ) 2 3 g = .
For the special case of
( )
n
g X X = , we obtain the n-th moment of the
probability distribution of the RV, X ; that is,
( )
n n
X
E X x f x d x

=

(2.40)
The most widely used moments are the first moment ( 1 n = , which results in the
mean value of Eq. 2.38) and the second moment ( 2 n = , resulting in the mean
square value of X ).
( )
2 2 2
X
E X X x f x d x

= =

(2.41)
If
( ) ( )
n
X
g X X m = , then ( ) E g X

gives n-th central moment; that is,
( ) ( ) ( )
n n
X X X
E X m x m f x d x

=

(2.42)

We can extend the definition of expectation to the case of functions of
( ) 2 k k random variables. Consider a function of two random variables,
( ) , g X Y . Then,

2.50
( ) ( ) ( )
,
, , ,
X Y
E g X Y g x y f x y d x d y

=

(2.43)
An important property of the expectation operator is linearity; that is, if
( ) , Z g X Y X Y = = + where and are constants, then Z X Y = + .
This result can be established as follows. From Eq. 2.43, we have
[ ] ( ) ( )
,
,
X Y
E Z x y f x y d x dy

= +

( ) ( ) ( ) ( )
, ,
, ,
X Y X Y
x f x y d x dy y f x y d x dy

= +

Integrating out the variable y in the first term and the variable x in the second
term, we have

[ ] ( ) ( )
X Y
E Z x f x d x y f y d y

= +

X Y = +

2.5.1 Variance
Coming back to the central moments, we have the first central moment
being always zero because,
( ) ( ) ( )
X X X
E X m x m f x d x

=

0
X X
m m = =
Consider the second central moment

( )
2
2 2
2
X X X
E X m E X m X m

= +

From the linearity property of expectation,

[ ]
2 2 2 2
2 2
X X X X
E X m X m E X m E X m + = +

2 2 2
2
X X
E X m m = +

2
2 2 2
X
X m X X

= =


2.51
The second central moment of a random variable is called the variance and its
(positive) square root is called the standard deviation. The symbol
2
is
generally used to denote the variance. (If necessary, we use a subscript on
2
)

The variance provides a measure of the variable's spread or randomness.
Specifying the variance essentially constrains the effective width of the density
function. This can be made more precise with the help of the Chebyshev
Inequality which follows as a special case of the following theorem.

Theorem 2.3: Let
( ) g X be a non-negative function of the random variable X . If
( ) E g X

exists, then for every positive constant c ,
( )
( ) E g X
P g X c
c

(2.44)

Proof: Let
( ) { }
: A x g x c = and B denote the complement of A.
( ) ( ) ( )
X
E g X g x f x d x

=

( ) ( ) ( ) ( )
X X
A B
g x f x d x g x f x d x = +

Since each integral on the RHS above is non-negative, we can write
( ) ( ) ( )
X
A
E g X g x f x d x

If x A , then ( ) g x c , hence
( ) ( )
X
A
E g X c f x d x

But ( ) [ ] ( )
X
A
f x d x P x A P g X c = =

That is, ( ) ( ) E g X c P g X c

, which is the desired result.

Note: The kind of manipulations used in proving the theorem 2.3 is useful in
establishing similar inequalities involving random variables.

2.52
To see how innocuous (or weak, perhaps) the inequality 2.44 is, let
( ) g X
represent the height of a randomly chosen human being with
( ) 1.6 E g X m =

.
Then Eq. 2.44 states that the probability of choosing a person over 16 m tall is at
most
1
!
10
(In a population of 1 billion, at most 100 million would be as tall as a
full grown Palmyra tree!)

Chebyshev inequality can be stated in two equivalent forms:
i)
2
1
, 0
X X
P X m k k
k
>

(2.45a)
ii) < >

X X
P X m k
k
2
1
1 (2.45b)
where
X
is the standard deviation of X .
To establish 2.45(a), let ( ) ( )
2
X
g X X m = and
2 2
X
c k = in theorem
2.3. We then have,
( )
2
2 2
2
1
X X
P X m k
k

In other words,
2
1
X X
P X m k
k

which is the desired result. Naturally, we would take the positive number k to be
greater than one to have a meaningful result. Chebyshev inequality can be
interpreted as: the probability of observing any RV outside k standard
deviations off its mean value is no larger than
2
1
k
. With 2 k = for example, the
probability of 2
X X
X m does not exceed 1/4 or 25%. By the same token,
we expect X to occur within the range
( ) 2
X X
m for more than 75% of the
observations. That is, smaller the standard deviation, smaller is the width of the
interval around
X
m , where the required probability is concentrated. Chebyshev

2.53
inequality thus enables us to give a quantitative interpretation to the statement
'variance is indicative of the spread of the random variable'.

Note that it is not necessary that variance exists for every PDF. For example, if
( )
2 2
, and 0
X
f x x
x
= < < >

+
, then 0 X = but
2
X is not finite.
(This is called Cauchys PDF)

2.5.2 Covariance
An important joint expectation is the quantity called covariance which is
obtained by letting
( ) ( ) ( ) ,
X Y
g X Y X m Y m = in Eq. 2.43. We use the
symbol to denote the covariance. That is,
[ ] ( ) ( )
XY X Y
X Y E X m Y m cov , = =

(2.46a)
Using the linearity property of expectation, we have

[ ]
XY X Y
E XY m m = (2.46b)
The
[ ]
X Y cov , , normalized with respect to the product
X Y
is termed as the
correlation coefficient and we denote it by . That is,

[ ]
X Y
XY
X Y
E XY m m
=

(2.47)
The correlation coefficient is a measure of dependency between the variables.
Suppose X and Y are independent. Then,

[ ] ( )
,
,
X Y
E XY x y f x y d x d y

=

( ) ( )
X Y
x y f x f y d x d y

=

( ) ( )
X Y X Y
x f x d x y f y d y m m

= =


2.54
Thus, we have
XY
(and
XY
) being zero. Intuitively, this result is appealing.
Assume X and Y to be independent. When the joint experiment is performed
many times, and given
1
X x = , then Y would occur sometimes positive with
respect to
Y
m , and sometimes negative with respect to
Y
m . In the course of
many trials of the experiments and with the outcome
1
X x = , the sum of the
numbers
( )
1 Y
x y m would be very small and the quantity,
sum
number of trials
,
tends to zero as the number of trials keep increasing.

On the other hand, let X and Y be dependent. Suppose for example, the
outcome y is conditioned on the outcome x in such a manner that there is a
greater possibility of
( )
Y
y m being of the same sign as
( )
X
x m . Then we
expect
XY
to be positive. Similarly if the probability of
( )
X
x m and
( )
Y
y m
being of the opposite sign is quite large, then we expect
XY
to be negative.
Taking the extreme case of this dependency, let X and Y be so conditioned
that, given
1
X x = , then
1
Y x = , being constant. Then 1
XY
= . That
is, for X and Y be independent, we have 0
XY
= and for the totally dependent
( ) y x = case, 1
XY
= . If the variables are neither independent nor totally
dependent, then will have a magnitude between 0 and 1.

Two random variables X and Y are said to be orthogonal if
[ ]
E XY 0 = .
If
( )
or 0
XY XY
= , the random variables X and Y are said to be
uncorrelated. That is, if X and Y are uncorrelated, then XY X Y = . When the
random variables are independent, they are uncorrelated. However, the fact that
they are uncorrelated does not ensure that they are independent. As an example,
let

2.55
( )
1
,
2 2
0 ,
X
x
f x
otherwise

< <
,
and
2
Y X = . Then,

2
3 3 3
2
1 1
0 XY X x d x x d x

= = = =

As 0, 0 X X Y = = which means 0
XY
XY XY = = . But X and Y are
not independent because, if
1
X x = , then
2
1
! Y x =

Let Y be the linear combination of the two random variables
1
X and
2
X .
That is
1 1 2 2
Y k X k X = + where
1
k and
2
k are constants. Let
[ ]
i i
E X m = ,
2 2
i
X i
= , 1, 2 i = . Let
12
be the correlation coefficient between
1
X and
2
X .
We will now relate
2
Y
to the known quantities.
[ ]
2
2 2
Y
E Y E Y =

[ ]
1 1 2 2
E Y k m k m = +
2 2 2
1 1 2 2 1 2 1 2
2 E Y E k X k X k k X X = + +

With a little manipulation, we can show that

2 2 2
1 1 2 2 12 1 2 1 2
2
Y
k k k k = + + (2.48a)
If
1
X and
2
X are uncorrelated, then

2 2 2
1 1 2 2 Y
k k = + (2.48b)
Note: Let
1
X and
2
X be uncorrelated, and let
1 2
Z X X = + and
1 2
W X X = .
Then
2 2 2 2
1 2 Z W
= = + . That is, the sum as well as the difference random
variables have the same variance which is larger than
2
1
or
2
2
!

The above result can be generalized to the case of linear combination of
n variables. That is, if

2.56

1
n
i i
i
Y k X
=
=

, then

2 2 2
1
2
n
Y i i i j i j i j
i i j
i j
k k k
=
<
= +

(2.49)
where the meaning of various symbols on the RHS of Eq. 2.49 is quite obvious.

We shall now give a few examples based on the theory covered so far in
this section.

Example 2.18
Let a random variable X have the CDF
( )
2
0 , 0
, 0 2
8
, 2 4
16
1 , 4
X
x
x
x
F x
x
x
x
<

We shall find a) X and b)
2
X

a) The given CDF implies the PDF

( )
0 , 0
1
, 0 2
8
1
, 2 4
8
0 , 4
X
x
x
f x
x x
x
<

Therefore,
[ ]
2 4
2
0 2
1 1
8 8
E X x d x x d x = +


2.57

1 7 31
4 3 12
= + =
b)
[ ]
2
2 2
X
E X E X =

2 4
2 2 3
0 2
1 1
8 8
E X x d x x d x = +

1 15 47
3 2 6
= + =
2
2
47 31 167
6 12 144
X

= =

Example 2.19
Let cos Y X = , where
( )
X
x
f x
otherwise
1 1
1,
2 2
0,
< <

Let us find
[ ]
E Y and
2
Y
.

From Eq. 2.38 we have,
[ ] ( )
1
2
1
2
2
cos 0.0636 E Y x d x
= = =

( )
1
2
2 2
1
2
1
cos 0.5
2
E Y x d x
= = =

Hence
2
2
1 4
0.96
2
Y
= =

Example 2.20
Let X and Y have the joint PDF
( )
,
, 0 1, 0 1
,
0 ,
X Y
x y x y
f x y
elsewhere
+ < < < <
=


2.58
Let us find a)
2
E XY

and b)
XY

a) From Eq. 2.43, we have
( )
2 2
,
,
X Y
E XY x y f x y dx dy

=

( )
1 1
2
0 0
17
72
x y x y dx dy = + =

b) To find
XY
, we require
[ ] [ ] [ ]
, , E XY E X E Y ,
X
and
Y
. We can easily
show that
[ ] [ ]
2 2
7 11
,
12 144
X Y
E X E Y = = = = and
[ ]
48
144
E XY =
Hence
1
11
XY
=

Another statistical average that will be found useful in the study of
communication theory is the conditional expectation. The quantity,
( ) ( ) ( )
|
| |
X Y
E g X Y y g x f x y dx

= =

(2.50)
is called the conditional expectation of
( ) g X , given Y y = . If
( ) g X X = , then
we have the conditional mean, namely,
[ ]
| E X Y .

[ ] [ ] ( )
|
| | |
X Y
E X Y y E X Y x f x y dx

= = =

(2.51)
Similarly, we can define the conditional variance etc. We shall illustrate the
calculation of conditional mean with the help of an example.

Example 2.21
Let the joint PDF of the random variables X and Y be

2.59
( )
,
1
, 0 1, 0
,
0 ,
X Y
x y x
f x y
x
outside
< < < <
.
Let us compute
[ ]
| E X Y .

To find
[ ]
| E X Y , we require the conditional PDF, ( )
( )
( )
,
|
,
|
X Y
X Y
Y
f x y
f x y
f y
=
( )
1
1
ln , 0 1
Y
y
f y d x y y
x
= = < <

( )
|
1
1
| , 1
ln ln
X Y
x
f x y y x
y x y
= = < <

Hence ( )
1
1
|
ln
y
E X Y x d x
x y

=

1
ln
y
y

Note that
[ ]
| E X Y y = is a function of y .

2.6 Some Useful Probability Models
In the concluding section of this chapter, we shall discuss certain
probability distributions which are encountered quite often in the study of
communication theory. We will begin our discussion with discrete random
variables.

2.6.1. Discrete random variables
i) Binomial:
Consider a random experiment in which our interest is in the occurrence or non-
occurrence of an event A. That is, we are interested in the event A or its
complement, say, B. Let the experiment be repeated n times and let p be the

2.60
probability of occurrence of A on each trial and the trials are independent. Let X
denote random variable, number of occurrences of A in n trials'. X can be
equal to 0, 1, 2, ........., n . If we can compute
[ ]
, 0, 1, ........., P X k k n = = , then
we can write
( )
X
f x .
Taking a special case, let 5 n = and 3 k = . The sample space
(representing the outcomes of these five repetitions) has 32 sample points, say,
1 32
, ..........., s s . The sample point
1
s could represent the sequence ABBBB. The
sample points such as ABAAB, AAABB etc. will map into real number 3 as
shown in the Fig. 2.18. (Each sample point is actually an element of the five
dimensional Cartesian product space).

Fig. 2.18: Binomial RV for the special case 5 n =

( ) ( ) ( ) ( ) ( ) ( ) P AB AAB P A P B P A P A P B = , as trails are independent.

( ) ( ) ( )
2
2 3
1 1 1 p p p p p p = =
There are ( )
5
10 sample points for which 3
3
X s

= =

.
In other words, for 5 n = , 3 k = ,
[ ] ( )
2
3
5
3 1
3
P X p p

= =

Generalizing this to arbitrary n and k , we have the binomial density, given by

2.61
( ) ( )
0
n
X i
i
f x P x i
=
=
(2.52)
where ( ) 1
n i
i
i
n
P p p
i

=

As can be seen,
( ) 0
X
f x and
( ) ( )
0 0
1
n n
n i
i
X i
i i
n
f x d x P p p
i
= =

= =

( ) 1 1
n
p p = + =

It is left as an exercise to show that
[ ]
E X n p = and
( )
2
1
X
n p p = . (Though
the formulae for the mean and the variance of a binomial PDF are simple, the
algebra to derive them is laborious).
We write X is
( ) , b n p to indicate X has a binomial PDF with parameters
n and p defined above.

The following example illustrates the use of binomial PDF in a
communication problem.

Example 2.22
A digital communication system transmits binary digits over a noisy
channel in blocks of 16 digits. Assume that the probability of a binary digit being
in error is 0.01 and that errors in various digit positions within a block are
statistically independent.
i) Find the expected number of errors per block
ii) Find the probability that the number of errors per block is greater than or
equal to 3.

Let X be the random variable representing the number of errors per block.
Then X is
( ) 16, 0.01 b .
i)
[ ]
16 0.01 0.16; E X n p = = =

2.62
ii)
( ) [ ]
3 1 2 P X P X =
( ) ( )
2
16
0
16
1 0.1 1
i i
i
p
i

=

=

0.002 =

ii) Poisson:
A random variable X which takes on only integer values is Poisson
distributed, if
( ) ( )
0
!
m
X
m
e
f x x m
m

=
(2.53)
where is a positive constant.
Evidently
( ) 0
X
f x and ( ) 1
X
f x d x
because
0
!
m
m
e
m
.
We will now show that

[ ]
2
X
E X = =
Since,

0
!
m
m
e
m
=

, we have

( )
1
0 1
1
! !
m m
m m
d e
m
e m
d m m
= =

= = =

[ ]
1
!
m
m
m e
E X e e
m

=
= = =

Differentiating the series again, we obtain,
Exercise 2.8
Show that for the example 2.22, Chebyshev inequality results in
3 0.0196 0.02 P X

. Note that Chebyshev inequality is not very
tight.

2.63

2 2
E X = +

. Hence
2
X
= .

2.6.2 Continuous random variables
i) Uniform:
A random variable X is said to be uniformly distributed in the interval
a x b if,
( )
1
,
0 ,
X
a x b
b a f x
elsewhere
(2.54)
A plot of
( )
X
f x is shown in Fig.2.19.

Fig.2.19: Uniform PDF

It is easy show that

[ ]
2
a b
E X
+
= and
( )
2
2
12
X
b a
=
Note that the variance of the uniform PDF depends only on the width of the
interval
( ) b a . Therefore, whether X is uniform in
( ) 1, 1 or
( ) 2, 4 , it has the
same variance, namely
1
3
.

ii) Rayleigh:
An RV X is said to be Rayleigh distributed if,

2.64
( )
2
exp , 0
2
0 ,
X
x x
x
f x b b
elsewhere

=

(2.55)
where b is a positive constant,

A typical sketch of the Rayleigh PDF is given in Fig.2.20. (
( )
R
f r of
example 2.12 is Rayleigh PDF.)

Fig.2.20: Rayleigh PDF

Rayleigh PDF frequently arises in radar and communication problems. We will
encounter it later in the study of narrow-band noise processes.


2.65

iii) Gaussian
By far the most widely used PDF, in the context of communication theory
is the Gaussian (also called normal) density, specified by
( )
( )
2
2
1
exp ,
2
2
X
X
X
X
x m
f x x

= < <

(2.56)
where
X
m is the mean value and
2
X
the variance. That is, the Gaussian PDF is
completely specified by the two parameters,
X
m and
2
X
. We use the symbol
( )
2
,
X X
N m to denote the Gaussian density
1
PT. In appendix A2.3, we show that
( )
X
f x as given by Eq. 2.56 is a valid PDF.

As can be seen from the Fig. 2.21, The Gaussian PDF is symmetrical with
respect to
X
m .

TP
1
PT In this notation, ( ) 0, 1 N denotes the Gaussian PDF with zero mean and unit variance. Note that
if X is
( )
2
,
X X
N m , then
X
X
X m
Y

=

is ( ) 0, 1 N .
Exercise 2.9
a) Let ( )
X
f x be as given in Eq. 2.55. Show that
( )
0
1
X
f x d x
. Hint: Make
the change of variable
2
x z = . Then,
2
d z
x d x = .
b) Show that if X is Rayleigh distributed, then
[ ]
2
b
E X

= and
2
2 E X b =


2.66

Fig. 2.21: Gaussian PDF

Hence ( ) ( ) 0.5
X
m
X X X
F m f x d x

= =

Consider
[ ]
P X a . We have,
[ ]
( )
2
2
1
exp
2
2
X
X a X
x m
P X a d x

This integral cannot be evaluated in closed form. By making a change of variable
X
X
x m
z

=

, we have

[ ]
2
2
1
2
X
X
z
a m
P X a e d z

X
X
a m
Q

=

where ( )
2
1
exp
2
2
y
x
Q y d x
(2.57)
Note that the integrand on the RHS of Eq. 2.57 is
( ) 0, 1 N .

( ) Q function table is available in most of the text books on
communication theory as well as in standard mathematical tables. A small list is
given in appendix A2.2 at the end of the chapter.

2.67
The importance of Gaussian density in communication theory is due to a
theorem called central limit theorem which essentially states that:

If the RV X is the weighted sum of N independent random components,
where each component makes only a small contribution to the sum, then
( )
X
F x
approaches Gaussian as N becomes large, regardless of the distribution of the
individual components.

For a more precise statement and a thorough discussion of this theorem,
you may refer [1-3]. The electrical noise in a communication system is often due
to the cumulative effects of a large number of randomly moving charged
particles, each particle making an independent contribution of the same amount,
to the total. Hence the instantaneous value of the noise can be fairly adequately
modeled as a Gaussian variable. We shall develop Gaussian random
processes in detail in Chapter 3 and, in Chapter 7, we shall make use of this
theory in our studies on the noise performance of various modulation schemes.

Example 2.23
A random variable Y is said to have a log-normal PDF if ln X Y = has a
Gaussian (normal) PDF. Let Y have the PDF,
( )
Y
f y given by,
( )
( )
2
2
ln
1
exp , 0
2
2
0 ,
Y
y
y
f y
y
otherwise

where and are given constants.
a) Show that Y is log-normal
b) Find
( ) E Y
c) If m is such that
( ) 0.5
Y
F m = , find m.

a) Let ln X Y = or ln x y = (Note that the transformation is one-to-one)

2.68
1 1 d x
J
d y y y
= =
Also as 0, y x and as , y x
Hence ( )
( )
2
2
1
exp ,
2
2
X
x
f x x

= < <

Note that X is
( )
2
N ,

b)
( )
2
2
2
1
2
x
X x
Y E e e e d x

= =

( )
2
2
2
2
2
2
1
2
x
e e d x

+

+

=

As the bracketed quantity being the integral of a Gaussian PDF
between the limits
( ) , is 1, we have

2
2
Y e

+
=
c)
[ ] [ ]
ln P Y m P X m =
Hence if
[ ]
0.5 P Y m = , then
[ ]
ln 0.5 P X m =
That is, ln m = or m e

iv) Bivariate Gaussian
As an example of a two dimensional density, we will consider the bivariate
Gaussian PDF,
( )
,
,
X Y
f x y , , x y < < given by,
( )
( ) ( ) ( ) ( )
2 2
, 2 2
1 2
1 1
, exp 2
X Y X Y
X Y
X Y X Y
x m y m x m y m
f x y
k k

= +

(2.58)
where,

2
1
2 1
X Y
k =

2.69

( )
2
2
2 1 k =
Correlation coefficient between and X Y =
The following properties of the bivariate Gaussian density can be verified:
P1) If and X Y are jointly Gaussian, then the marginal density of or X Y is
Gaussian; that is, X is
( )
2
,
X X
N m and Y is
( )
2
,
Y Y
N m TP
1
PT
P2)
( ) ( ) ( ) , iff 0
X Y X Y
f x y f x f y = =
That is, if the Gaussian variables are uncorrelated, then they are
independent. That is not true, in general, with respect to non-Gaussian
variables (we have already seen an example of this in Sec. 2.5.2).
P3) If Z X Y = + where and are constants and and X Y are jointly
Gaussian, then Z is Gaussian. Therefore
( )
Z
f z can be written after
computing
Z
m and
2
Z
with the help of the formulae given in section 2.5.
Figure 2.22 gives the plot of a bivariate Gaussian PDF for the case of
0 = and
X Y
= .

TP
1
PT Note that the converse is not necessarily true. Let
X
f and
Y
f be obtained from
, X Y
f and let
X
f
and
Y
f be Gaussian. This does not imply
, X Y
f is jointly Gaussian, unless and X Y are
independent. We can construct examples of a joint PDF
, X Y
f , which is not Gaussian but results in
X
f and
Y
f that are Gaussian.

2.70

Fig.2.22: Bivariate Gaussian PDF (
X Y
= and 0 = )

For 0 = and
X Y
= ,
, X Y
f resembles a (temple) bell, with, of course,
the striker missing! For 0 , we have two cases (i) , positive and (ii)
, negative. If 0 > , imagine the bell being compressed along the
X Y = axis so that it elongates along the X Y = axis. Similarly for
0 < .

Example 2.24
Let X and Y be jointly Gaussian with
2 2
1, 1
X Y
X Y = = = = and
1
2
XY
= . Let us find the probability of
( ) , X Y lying in the shaded region D
shown in Fig. 2.23.


2.71

Fig. 2.23: The region D of example 2.24

Let A be the shaded region shown in Fig. 2.24(a) and B be the shaded
region in Fig. 2.24(b).

Fig. 2.24: (a) Region A and (b) Region B used to obtain region D

The required probability =
( ) ( ) , , P x y P x y

A B
For the region A , we have
1
1
2
y x + and for the region B , we have
1
2
2
y x + . Hence the required probability is,

2.72

X X
P Y P Y 1 2
2 2

+ +

Let
X
Z Y
2
= +
Then Z is Gaussian with the parameters,

1 1
2 2
Z Y X = + =

2 2 2
1 1
2.
4 2
Z X Y XY
= + +

1 1 1 3
1 2. .
4 2 2 4
= + =
That is, Z is
1 3
,
2 4
N

. Then
1
2
3
4
Z
W
+
= is
( ) 0, 1 N

[ ]
1 3 P Z P W

=

[ ]
5
2
3
P Z P W

=

Hence the required probability
( )
( )
5
3 0.04 0.001
3
Q Q

=

0.039 =


2.73

Exercise 2.10
X and Y are independent, identically distributed (iid) random variables,
each being ( ) 0, 1 N . Find the probability of , X Y lying in the region A shown in
Fig. 2.25.

Fig. 2.25: Region A of Exercise 2.10

Note: It would be easier to calculate this kind of probability, if the space is a
product space. From example 2.12, we feel that if we transform ( ) , X Y into
( ) , Z W such that Z X Y = + , W X Y = , then the transformed space B
would be square. Find ( )
,
,
Z W
f z w and compute the probability ( ) , Z W B .

2.74

Exercise 2.11
Two random variables X and Y are obtained by means of the
transformation given below.
( ) ( )
1
2
1 2
2log cos 2
e
X U U = (2.59a)
( ) ( )
1
2
1 2
2log sin 2
e
Y U U = (2.59b)
1
U and
2
U are independent random variables, uniformly distributed in the
range
1 2
0 , 1 u u < < . Show that X and Y are independent and each is
( ) 0, 1 N .
Hint: Let
1 1
2log
e
X U = and
1 1
Y X =
Show that
1
Y is Rayleigh. Find ( )
,
,
X Y
f x y using
1
cos X Y = and
1
sin Y Y = , where
2
2 U = .
Note: The transformation given by Eq. 2.59 is called the Box-Muller
transformation and can be used to generate two Gaussian random number
sequences from two independent uniformly distributed (in the range 0 to 1)
sequences.

2.75
Appendix A2.1
Proof of Eq. 2.34
The proof of Eq. 2.34 depends on establishing a relationship between the
differential area d z dw in the z w plane and the differential area d x d y in
the x y plane. We know that

( ) [ ]
,
, ,
Z W
f z w d z d w P z Z z d z w W w d w = < + < +
If we can find d x d y such that

( ) ( )
, ,
, ,
Z W X Y
f z w d z d w f x y d x d y = , then
, Z W
f can be found. (Note that
the variables x and y can be replaced by their inverse transformation
quantities, namely,
( )
1
, x g z w
= and
( )
1
, y h z w
= )
Let the transformation be one-to-one. (This can be generalized to the case of
one-to-many.) Consider the mapping shown in Fig. A2.1.

Fig. A2.1: A typical transformation between the ( ) x y plane and ( ) z w plane

Infinitesimal rectangle ABCD in the z w plane is mapped into the
parallelogram in the x y plane. (We may assume that the vertex A transforms
to A
'
, B to B
'
etc.) We shall now find the relation between the differential area of
the rectangle and the differential area of the parallelogram.


2.76
Consider the parallelogram shown in Fig. A2.2, with vertices P P P
1 2 3
, ,
and P
4
.

Fig. A2.2: Typical parallelogram

Let ( ) x y , be the co-ordinates of P
1
. Then the P
2
and P
3
are given by

g h
P x d z y d z
z z
1 1
2
,

= + +

x y
x d z y d z
z z
,

= + +

= + +

x y
P x dw y dw
w w
3
,

Consider the vectors
1
V and
2
V shown in the Fig. A2.2 where
( ) P P
1 2 1
= V and
( ) = P P
2 3 1
V .
That is,

x y
d z d z
z z
1

= +

V i j

x y
dw dw
w w
2

= +

V i j
where i and j are the unit vectors in the appropriate directions. Then, the area
A of the parallelogram is,
A
1 2
= V V

2.77
As 0 = i i , 0 = j j , and
( ) = = j i i j k where k is the unit vector
perpendicular to both i and j , we have

x y y x
d z dw
z w z w
1 2

=

V V

x y
A J d z dw
z w
,
,

=

That is,
( ) ( )
Z W X Y
x y
f z w f x y J
z w
, ,
,
, ,
,

=

( )
X Y
f x y
z w
J
x y
,
,
,
,
=

.


2.78
Appendix A2.2
( )
Q Function Table
( )
2
2
1
2
x
Q e d x

It is sufficient if we know
( ) Q for 0 , because
( ) ( ) 1 Q Q = . Note
that
( ) 0 0.5 Q = .
y
( ) Q y y
( ) Q y y
( ) Q y
0.05
0.10
0.15
0.20
0.25
0.4801
0.4602
0.4405
0.4207
0.4013
1.05
1.10
1.15
1.20
1.25
0.1469
0.1357
0.1251
0.1151
0.0156
2.10
2.20
2.30
2.40
2.50
0.0179
0.0139
0.0107
0.0082
0.0062
0.30
0.35
0.40
0.45
0.50
0.3821
0.3632
0.3446
0.3264
0.3085
1.30
1.35
1.40
1.45
1.50
0.0968
0.0885
0.0808
0.0735
0.0668
2.60
2.70
2.80
2.90
3.00
0.0047
0.0035
0.0026
0.0019
0.0013
0.55
0.60
0.65
0.70
0.75
0.2912
0.2743
0.2578
0.2420
0.2266
1.55
1.60
1.65
1.70
1.75
0.0606
0.0548
0.0495
0.0446
0.0401
3.10
3.20
3.30
3.40
3.50
0.0010
0.00069
0.00048
0.00034
0.00023
0.80
0.85
0.90
0.95
1.00
0.2119
0.1977
0.1841
0.1711
0.1587
1.80
1.85
1.90
1.95
2.00
0.0359
0.0322
0.0287
0.0256
0.0228
3.60
3.70
3.80
3.90
4.00
0.00016
0.00010
0.00007
0.00005
0.00003
y
( ) Q y
3
10
3.10
3
10
2
3.28
4
10
3.70
4
10
2
3.90
5
10
4.27
6
10
4.78

2.79
Note that some authors use
( ) erfc , the complementary error function which is
given by
( ) ( )
2 2
1 erfc erf e d
= =

and the error function, ( )
2
0
2
erf e d

Hence ( )
1
2
2
Q erfc

=

.


2.80
Appendix A2.3
Proof that
( )
2
,
X X
N m is a valid PDF
We will show that
( )
X
f x as given by Eq. 2.56, is a valid PDF by
establishing ( ) 1
X
f x d x
. (Note that
( ) 0
X
f x for x < < ).
Let
2
2
2 2
y
v
I e dv e d y

= =

.
Then,
2
2
2
2 2
y
v
I e dv e d y

=

2 2
2
v y
e dv d y
+

=

Let cos v r = and sin y r = . Then,
2 2
r v y = + and
1
tan
y
v

=

,
and d x d y r d r d = . (Cartesian to Polar coordinate transformation).

2
2
2
2
0 0
r
I e r d r d

=

2 = or 2 I =
That is,
2
2
1
1
2
v
e dv

(A2.3.1)
Let
X
x
x m
v

=
(A2.3.2)
Then,
x
d x
d v =
(A2.3.3)
Using Eq. A2.3.2 and Eq. A2.3.3 in Eq. A2.3.1, we have the required result.


2.81
References
1) Papoulis, A., Probability, Random Variables and Stochastic Processes,
McGraw Hill (3P
rd
P edition), 1991.
2) Wozencraft, J . M. and J acobs, I. J ., Principles of Communication
Engineering, J ohn Wiley, 1965.
3) Hogg, R. V., and Craig, A. T., Introduction to Mathematical Statistics, The
Macmillan Co., 2P
nd
P edition, 1965.

3.1
CHAPTER 3

Random Signals and Noise

3.1 Introduction
The concept of 'random variable' is adequate to deal with unpredictable
voltages; that is, it enables us to come up with the probabilistic description of the
numerical value of a random quantity, which we treat for convenience to be a
voltage quantity. In the real world, the voltages vary not only in amplitude but
also exhibit variations with respect to the time parameter. In other words, we
have to develop mathematical tools for the probabilistic characterization of
random signals. The resulting theory, which extends the mathematical model of
probability so as to incorporate the time parameter, is generally called the theory
of Random or Stochastic Processes.

Before we get started with the mathematical development of a random
process, let us consider a few practical examples of random processes and try to
justify the assertion that the concept of random process is an extension of the
concept of a random variable.

Let the variable X denote the temperature of a certain city, say, at 9 A.M.
In general, the value of X would be different on different days. In fact, the
temperature readings at 9 A.M. on two different days could be significantly
different, depending on the geographical location of the city and the time
separation between the days of observation (In a place like Delhi, on a cold
winter morning, temperature could be as low as 40 F
whereas at the height of

the summer, it could have crossed 100
F even by 9 A.M.!).


3.2
To get the complete statistics of X , we need to record values of X over
many days, which might enable us to estimate
( )
X
f x .

But the temperature is also a function of time. At 12 P.M., for example, the
temperature may have an entirely different distribution. Thus the random variable
X is a function of time and it would be more appropriate to denote it by
( ) X t . At
least, theoretically, the PDF of
( )
1
X t could be very much different from that of
( )
2
X t for
1 2
t t , though in practice, they may be very much similar if
1
t and
2
t
are fairly closely spaced.

As a second example, think of a situation where we have a very large
number of speakers, each one of them uttering the same text into their individual
microphones of identical construction. The waveforms recorded from different
microphones would be different and the output of any given microphone would
vary with time. Here again, the random variables obtained from sampling this
collection of waveforms would depend on the sampling instants.

As a third example, imagine a large collection of resistors, each having the
same value of resistance and of identical composition and construction. Assume
that all these resistors are at room temperature. It is well known that thermal
voltage (usually referred to as thermal noise) develops across the terminals of
such a resistor. If we make a simultaneous display of these noise voltages on a
set of oscilloscopes, we find amplitude as well as time variations in these signals.
In the communications context, these thermal voltages are a source of
interference. Precisely how they limit our capacity to enjoy, say, listening to
music in an AM receiver, is our concern. The theory of random processes
enables one to come up with a quantitative answer to this kind of problem.


3.3
3.2 Definition of a Random Process
Consider a sample space S (pertaining to some random experiment) with
sample points
1 2
, , ..........., , ........
n
s s s . To every
j
s S , let us assign a real
valued function of time,
( )
,
j
x s t which we denote by
( )
j
x t . This situation is
illustrated in Fig. 3.1, which shows a sample space S with four points and four
waveforms, labeled
( ), 1, 2, 3, 4
j
x t j = .

Now, let us think of observing this set of waveforms at some time instant
1
t t = as shown in the figure.

Since each point
j
s of S has associated with it, a number
( )
1 j
x t and a
probability
j
P , the collection of numbers, ( ) { }
1
, 1, 2, 3, 4
j
x t j = forms a random
variable. Observing the waveforms at a second time instant, say
2
t , yields a
different collection of numbers, and hence a different random variable. Indeed
this set of four waveforms defines a random variable for each choice of the
observation instant. The above situation can easily be extended to the case
where there are infinite numbers of sample points and hence, the number of
waveforms associated with them are correspondingly rich.


3.4

Fig 3.1: A simple random process

The probability system composed of a sample space, an ensemble
(collection) of time functions and a probability measure is called a Random
Process (RP) and is denoted
( ) X t . Strictly speaking, a random process is a
function of two variables, namely s S and
( ) , t . As such, a better
notation would be
( ) , X s t . For convenience, we use the simplified notation
( ) X t
to denote a random process. The individual waveforms of
( ) X t are called
sample functions and the probability measure is such that it assigns a probability
to any meaningful event associated with these sample functions.

Given a random process
( ) X t , we can identify the following quantities:
( ) X t : The random process
( )
j
x t : The sample function associated with the sample point
j
s
( )
i
X t : The random variable obtained by observing the process at
i
t t =
( )
j i
x t : A real number, giving the value of
( )
j
x t at
i
t t = .


3.5

We shall now present a few examples of random processes.

Example 3.1
Consider the experiment of tossing a fair coin. The random process
( ) X t
is defined as follows:
( ) ( ) sin X t t = , if head shows up and
( ) 2 X t t = , if the
toss results in a tail. Sketch the sample functions. We wish to find the expression
for the PDF of the random variables obtained from sampling the process at (a)
0 t = and (b) 1 t = .

There are only two sample functions for the process. Let us denote them
by
( )
1
x t and
( )
2
x t where
( ) ( )
1
sin x t t = and
( )
2
2 x t t = which are shown in
Fig. 3.2.

Fig. 3.2: The ensemble for the coin tossing experiment

As heads and tails are equally likely, we have
( ) ( )
1 2
1
2
P x t P x t = =

.
Let
0
X denote the random variable
( )
0
|
t
X t
=
and
1
X correspond to
( )
1
|
t
X t
=
.
Then, we have ( ) ( )
0
X
f x x = and
( ) ( ) ( )
1
1
2
2
X
f x x x = +

.


3.6
Note that this is one among the simplest examples of RPs that can be used to
construe the concept.
Example 3.2
Consider the experiment of throwing a fair die. The sample space consists
of six sample points,
1 6
, ........, s s corresponding to the six faces of the die. Let
the sample functions be given by
( ) ( )
1
1
2
i
x t t i = + for , 1, ......., 6
i
s s i = = .
Let us find the mean value of the random variable
( )
1
|
t
X X t
=
= .

A few of the sample functions of this random process are shown below
(Fig 3.3).

Fig. 3.3: A few sample functions of the RP of Example 3.2

The PDF of X is
( ) ( )
6
1
1 1
1
6 2
X
i
f x x i
=

= +

[ ] ( )
1
1 3 5 7 9 11 3.0
12
E X = + + + + + =

The examples cited above have two features in common, namely (i) the
number of sample functions are finite (in fact, we could even say, quite small)

3.7
and (ii) the sample functions could be mathematically described. In quite a few
situations involving a random process, the above features may not be present.

Consider the situation where we have at our disposal N identical
resistors, N being a very large number (of the order of a million!).

Let the experiment be 'picking a resistor at random' and the sample
functions be the thermal noise voltage waveforms across these resistors. Then,
typical sample functions might look like the ones shown in Fig. 3.4.

Assuming that the probability of any resistor being picked up is
1
N
, we find
that this probability becomes smaller and smaller as N becomes larger and
larger. Also, it would be an extremely difficult task to write a mathematical
expression to describe the time variation of any given voltage waveform.
However, as we shall see later on in this chapter, statistical characterization of
such noise processes is still possible which is adequate for our purposes.

Fig. 3.4: The ensemble for the experiment 'picking a resistor at random'

One fine point deserves a special mention; that is, the waveforms (sample
functions) in the ensemble are not random. They are deterministic. Randomness

3.8
in this situation is associated not with the waveforms but with the uncertainty as
to which waveform will occur on a given trial.

3.3 Stationarity
By definition, a random process
( ) X t implies the existence of an infinite
number of random variables, one for every time instant in the range,
t < < . Let the process
( ) X t be observed at n time instants,
1 2
, , ...........,
n
t t t . We then have the corresponding random variables
( ) ( ) ( )
1 2
, , ...........,
n
X t X t X t . We define their Joint distribution function by
( ) ( )
( ) ( ) ( ) { }
1
1 2 1 1 , ...........,
, , ..........., , ........
n
n n n X t X t
F x x x P X t x X t x = . Using a
vector notation is quite often convenient and we denote the joint distribution by
( )
( ) F x
X t
where the n-component random vector ( ) ( ) ( ) ( )
1
, ...........,
n
X t X t = X t
and the dummy vector
( )
1 2
, , ...........,
n
x x x = x . The joint PDF of
( ) X t ,
( )
( ) f
X t
x ,
is given by
( )
( )
( )
( )
1 2
........
n
n
f F
x x x
=

X t X t
x x
We say a random process
( ) X t is specified if and only if a rule is given or
implied for determining
( )
( )
X t
F x or
( )
( )
X t
f x for any finite set of observation
instants
( )
1 2
, , ...........,
n
t t t .

In application, we encounter three methods of specification. The first (and
simplest) is to state the rule directly. For this to be possible, the joint density
function must depend in a known way on the time instants. For the second
method, a time function involving one or more parameters is given. For example,
( ) ( ) cos
c
X t A t = + where A and
c
are constants and is a random
variable with a known PDF. The third method of specifying a random process is

3.9
to generate its ensemble by applying a stated operation to the sample functions
of a known process. For example, a random process
( ) Y t may be the result of
linear filtering on some known process
( ) X t . We shall see later on in this lesson,
examples of all these methods of specification.

Once
( )
( )
X t
f x is known, it is possible for us to compute the probability of
various events. For example, we might be interested in the probability of the
random process
( ) X t passing through a set of windows as shown in Fig. 3.5.

Let A be the event:
( ) ( ) ( ) { }
1 1 1 2 2 2 3 3 3
: , , A s a X t b a X t b a X t b = < < <

That is, the event A consists of all those sample points
{ }
j
s such that the
corresponding sample functions ( ) { }
j
x t satisfy the requirement,
( ) , 1, 2, 3
i j i i
a x t b i = . Then the required quantity is
( ) P A . A typical
sample function which would contribute to
( ) P A is shown in the same figure.
( ) P A can be calculated as
( )
( )
( )
1 2 3
1 2 3
b b b
a a a
P A f =
X t
x d x
where ( ) ( ) ( ) ( ) ( )
1 2 3
, , X t X t X t = x t and
1 2 3
d x d x d x = d x

Fig. 3.5: Set of windows and a waveform that passes through them

3.10

The above step can easily be generalized to the case of a random vector
with n-components.

Stationary random processes constitute an important subset of the
general class of random processes. We shall define stationarity here. Let the
process
( ) X t be observed at time instants
1 2
, , ..........,
n
t t t and
( ) X t be the
corresponding random vector.

Def. 3.1: A random process
( ) X t is said to be strictly stationary or Stationary in
a Strict Sense (SSS) if the joint PDF
( )
( ) f
X t
x is invariant to a translation of the
time origin; that is,
( ) X t is SSS, only if

( )
( )
( )
( )
X t T X t
f x f x
+
= (3.1)
where
( ) ( )
1 2
, , ..........,
n
t T t T t T t T + = + + + .

For
( ) X t to be SSS, Eq. 3.1 should be valid for every finite set of time
instants
{ }
, 1, 2, ..........,
j
t j n = , and for every time shift T and dummy vector
x . If
( ) X t is not stationary, then it is called a nonstationary process.

One implication of stationarity is that the probability of the set of sample
functions of this process which passes through the windows of Fig. 3.6(a) is
equal to the probability of the set of sample functions which passes through the
corresponding time shifted windows of Fig. 3.6(b). Note, however, that it is not
necessary that these two sets consist of the same sample functions.


3.11

Fig. 3.6: (a) Original set of Windows (b) The translated set

Example 3.3
We cite below an example of a nonstationary process (It is easier to do
this than give a nontrivial example of a stationary process). Let
( ) ( ) sin 2 X t F t =
where F is a random variable with the PDF
( )
1
, 100 200
100
0 ,
F
f Hz
f f
otherwise

(Note that this specification of
( ) X t corresponds to the second method
mentioned earlier on in this section). We now show that
( ) X t is nonstationary.

( ) X t consists of an infinite number of sample functions. Each sample
function is a sine wave of unit amplitude and a particular frequency f . Over the
ensemble, the random variable F takes all possible values in the range
( ) 100, 200 Hz . Three members of this ensemble, (with 100, 150 f = and 200
Hz) are plotted in Fig. 3.7.

3.12

Fig. 3.7: A few sample functions of the process of example 3.3

To show that
( ) X t is nonstationary, we need only observe that every
waveform in the ensemble is,
zero at 0 t = ,
positive for 0 2.5 sec t m < <
negative for
( ) 2.5 sec 0 m t < < .

3.13
Thus the density function of the random variable
( )
1
X t for
1
1 sec t m = is
identically zero for negative arguments whereas the density function of the RV
( )
2
X t for
2
1 sec t m = is non-zero only for negative arguments (Of course, the
PDF of
( ) 0 X is an impulse). For a process that is SSS, the one-dimensional
PDF is independent of the observation instant, which is evidently not the case for
this example. Hence
( ) X t is nonstationary.

3. 4 Ensemble Averages
We had mentioned earlier that a random process is completely specified,
if
( )
( ) f
X t
x is known. Seldom is it possible to have this information and we may
have to be content with a partial description of the process based on certain
averages. When these averages are derived from the ensemble, they are called
ensemble averages. Usually, the mean function and the autocorrelation function,
(or the auto-covariance function) of the process provide a useful description of
the processes. At times, we require the cross-correlation between two different
processes. (This situation is analogous to the random variable case, wherein we
had mentioned that even if the PDF of the variable is not available, certain
averages such as mean value, variance etc., do provide adequate or useful
information).
Def. 3.2: The Mean Function
The mean function of a process
( ) X t is
( ) ( ) ( )
( )
( )
X x t
m t E X t X t x f x d x

= = =

(3.2)
For example, if
i
X and
j
X are the random variables obtained by sampling
the process at
i
t t = and
j
t t = respectively, then
( ) ( )
i
i x X i
X x f x d x m t

= =


3.14
( ) ( )
j
j x X j
X x f x d x m t

= =

In general,
( )
X
m t is a function of time.
Def. 3.3: The Autocorrelation Function
The autocorrelation function of a random process
( ) X t is a function of
two variables
k
t and
i
t , and is given by
( ) ( ) ( ) ,
X k i k i
R t t E X t X t =

(3.3)
Denoting the joint PDF of the random variables
( )
k
X t and
( )
i
X t by ( )
,
,
k i
X X
f x y
we may rewrite Eq. 3.3 as
( ) ( )
,
, ,
k i
X k i X X
R t t x y f x y d x d y

=

We also use ACF to denote the Auto Correlation Function.
Def. 3.4: The Auto-covariance Function
Let
( ) ,
X k i
C t t denote the auto covariance function of
( ) X t . It is given by
( ) ( ) ( ) ( ) ( ) ( ) ( )
,
X k i k X k i X i
C t t E X t m t X t m t

=

(3.4a)
It is not too difficult to show that
( ) ( ) ( ) ( ) , ,
X k i X k i X k X i
C t t R t t m t m t =
(3.4b)
In general, the autocorrelation and the auto-covariance would be a function of
both the arguments
k
t and
i
t . If the process has a zero mean value (that is,
( ) 0
X
m t = for all t ), then
( ) ( ) , ,
X k i X k i
C t t R t t = .

For a stationary process, the ensemble averages defined above take a
simpler form. In particular, we find that mean function of the process is a
constant. That is,
( )
X X
m t m = (3.5a)
X
m being a constant. In such a case, we can simply mention the mean value of
the process. Also, for a stationary process, we find that the autocorrelation and

3.15
auto-covariance functions depend only on the time difference
( )
k i
t t , rather
than on the actual values of
k
t and
i
t . This can be shown as follows.
( ) ( ) ( ) ,
X k i k i
R t t E X t X t =

( ) ( )
k i
E X t T X t T = + +

, as
( ) X t is stationary.
In particular if
i
T t = , then
( ) ( ) ( ) , 0
X k i k i
R t t E X t t X =

(3.5b)
( ) ( ) , , 0
X k i X k i
R t t R t t =
In order to simplify the notation, it is conventional to drop the second argument
on the RHS of Eq. 3.5(b) and write as
( ) ( ) ,
X k i X k i
R t t R t t =
In view of Eq. 3.4(b), it is not difficult to see that for a stationary process

( ) ( ) ,
X k i X k i
C t t C t t =

It is important to realize that for a process that is SSS, Eq. 3.5(a) and Eq.
3.5(b) hold. However, we should not infer that any process for which Eq. 3.5 is
valid, is a stationary process. In any case, the processes satisfying Eq. 3.5 are
sufficiently useful and are termed Wide Sense Stationary (WSS) processes.

Def. 3.5: Wide Sense Stationarity
A process
( ) X t is WSS or stationary in a wide sense
1
, provided
( )
X X
m t m = (3.6a)
and
( ) ( ) ,
X k i X k i
R t t R t t =
(3.6b)
Wide sense stationarity represents a weak kind of stationarity in that all
processes that are SSS are also WSS; but the converse is not necessarily true.
When we simply use the word stationary, we imply stationarity in the strict sense.

1
For definitions of other forms of stationarity (such as N
th
order stationary) see [1, P302]

3.16
3.4.1 Properties of ACF
The autocorrelation function of a WSS process satisfies certain properties,
which we shall presently establish. ACF is a very useful tool and a thorough
understanding of its behavior is quite essential for a further study of the random
processes of interest to us. For convenience of notation, we define the ACF of a
wide-sense stationary process
( ) X t as
( ) ( ) ( )
X
R E X t X t = +

(3.7)
Note that is the difference between the two time instants at which the process
is being sampled.

P1) The mean-square value of the process is ( )
0
X
R
=
.
This follows from Eq. 3.7 because,
( ) ( )
2
0
X
R E X t =

As
( ) 0
X
R is a constant, we infer that for a WSS process, mean and mean-
square values are independent of time.
P2) The ACF is an even function of ; that is,

( ) ( )
X X
R R =
This is because ( ) ( ) ( ) ( ) E X t X t E X t X t + = +

. That is,
( ) ( )
X X
R t t R t t + = +

which is the desired result.
P3) ACF is maximum at the origin.
Consider the quantity, ( ) ( )
{ }
2
E X t X t +

Being the expectation of a squared quantity, it is nonnegative. Expanding
the squared quantity and making use of the linearity property of the
expectation, we have
( ) ( ) 0 0
X X
R R
which implies ( ) ( ) 0
X X
R R .

3.17
P4) If the sample functions of the process
( ) X t are periodic with period
0
T , then
the ACF is also periodic with the same period.
This property can be established as follows.

Consider ( ) ( ) E X t X t +

for
0
T . As each sample function
repeats with period
0
T , the product repeats and so does the expectation of
this product.

The physical significance of
( )
X
R is that it provides a means of
describing the inter-dependence of two random variables obtained by
observing the process
( ) X t at two time instants seconds apart. In fact if
( ) X t is a zero mean process, then for any
1
= ,
( )
( )
1
0
X
X
R
R

is the
correlation coefficient of the two random variables spaced
1
seconds apart.
It is therefore apparent that the more rapidly
( ) X t changes with time, the
more rapidly
( )
X
R decreases from its maximum value
( ) 0
X
R as
increases (Fig. 3.8). This decrease may be characterized by a de-
correlation time
0
, such that for ( ) ,
X
R
0
remains below some
prescribed value, say
( ) 0
100
X
R
. We shall now take up a few examples to
compute some of the ensemble averages of interest to us.

Fig. 3.8: ACF of a slowly and rapidly fluctuating random process

3.18
Example 3.4
For the random process of Example 3.2, let us compute
( ) 2, 4
X
R .

( ) ( ) ( ) 2, 4 2 4
X
R E X X =

( ) ( )
6
1
2 4
j j j
j
P x x x
=
=

A few of the sample functions of the process are shown below.

Fig. 3.9: Sampling the process of example 3.2 at 2 and 4 t =

As
1
6
j
P x =

for all j , we have
( ) ( ) ( )
6
1
1
2, 4 2 4
6
X j j
j
R x x
=
=

for 1 j = , the two samples that contribute to the ACF have the values 1
(indicated by the on
( )
1
x t ) and 2 (indicated by on
( )
1
x t ). The other sample
pairs are
( ) ( ) ( ) ( ) ( ) 2, 3 , 3, 4 , 4, 5 , 5, 6 , 6, 7 .


3.19
Hence
( ) [ ]
1
2, 4 2 6 12 20 30 42
6
X
R = + + + + +

112
18.66
6
= =

Exercise 3.1
Let a random process ( ) X t consist of 6 equally likely sample
functions, given by ( ) , 1, 2, ......, 6
i
x t i t i = = . Let X and Y be the
random variables obtained by sampling process at 1 t = and 2 t =
respectively.
Find
a)
[ ]
E X and
[ ]
E Y
b) ( )
,
,
X Y
f x y
c) ( ) 1, 2
X
R


3.20

Example 3.5
Let
( ) ( ) cos
c
X t A t = + ,
where A and
c
are constants and is a random variable with the PDF,
( )
1
, 0 2
2
0 ,
f
otherwise
<

Let us compute a)
( )
X
m t and b)
( )
1 2
,
X
R t t .

Exercise 3.2
A random process ( ) X t consists of 5 sample functions, each
occurring with probability
1
5
. Four of these sample functions are given
below.
( ) ( ) ( )
1
cos 2 sin 2 x t t t =
( ) ( ) ( )
2
sin 2 cos 2 x t t t =
( )
3
2 cos x t t =
( )
4
2 sin x t t =
a) Find the fifth sample function ( )
5
x t of the process ( ) X t such that the
process ( ) X t is
i) zero mean
ii) ( ) ( )
1 2 1 2
,
X X
R t t R t t =
b) Let V be the random variable ( )
0 t
X t
=
and W be the random
variable ( )
4 t
X t
=
. Show that, though the process is WSS,
( ) ( )
V W
f v f v .

3.21
(Note that the random processes is specified in terms of a random parameter,
namely , another example of the second method of specification)

The ensemble of
( ) X t is composed of sinusoids of amplitude A and
frequency
c
f but with a random initial phase. Of course, a given sample function
has a fixed value for
1
= , where
1
0 2 < .

Three sample functions of the process are shown in Fig. 3.10 for
6
10
c
f Hz = and 1 A = .


3.22

Fig 3.10: Three sample functions of the process of example 3.5

3.23
a) ( ) ( ) cos
X c
m t E A t = +

. As is the only random variable, we have,

( ) ( )
2
0
cos
2
X c
A
m t t d
= +

0 =
Note that is uniform in the range 0 to 2 .

This is a reasonable answer because, for any given
1
= , the
ensemble has both the waveforms, namely,
( )
1
cos
c
A t + and
( ) ( )
1 1
cos cos
c c
A t A t + + = +

. Both the waveforms are
equally likely and sum to zero for all values of t .

b) ( ) ( ) ( )
1 2 1 2
, cos cos
X c c
R t t E A t A t = + +

( ) ( )
{ }
2
1 2 1 2
cos 2 cos
2
c c
A
t t t t = + + +

As the first term on the RHS evaluates to zero, we have
( ) ( )
2
1 2 1 2
, cos
2
X c
A
R t t t t =

As the ACF is only a function of the time difference, we can write
( ) ( ) ( ) ( )
2
1 2 1 2
, cos
2
X X X c
A
R t t R t t R = = =
Note that
( ) X t is composed of sample functions that are periodic with
period
1
c
f
. In accordance with property P4, we find that the ACF is also
periodic with period
1
c
f
. Of course, it is also an even function of (property
P3).


3.24

Example 3.6
In this example, we shall find the ACF of the random impulse train process
specified by
( ) ( )
n d
n
X t A t nT T =

where the amplitude
n
A is a random variable with
( ) ( )
1
1 1
2
n n
P A P A = = = = .
Successive amplitudes are statistically independent. The time interval
d
t of the
first impulse from the origin has uniform PDF in the range
( ) 0, T . That is,
( )
1
, 0
D
T d d
f t t T
T
= and zero elsewhere. Impulses are spaced T seconds
Exercise 3.3
For the process ( ) X t of example 3.5, let 2
c
= . let Y be the
random variable obtained from sampling ( ) X t and
1
4
t = . Find ( )
Y
f y .

Exercise 3.4
Let ( ) ( ) = + cos
c
X t A t where A and
c
are constants, and is
a random variable, uniformly distributed in the range 0 . Show the
process is not WSS.

Exercise 3.5
Let ( ) ( ) ( ) cos sin
c c
Z t X t Y t = + where X and Y are independent
Gaussian variables, each with zero mean and unit variance. Show that ( ) Z t is
WSS and ( ) ( ) =
Z c
R cos . Let ( )
1
Z t V = . Show that V is ( ) 0, 1 N .


3.25
apart and
n
A is independent of
d
T . (The symbol
n
indicates summation with

respect to n where n , an integer, ranges from
( ) , .)

A typical sample function of the process
( ) X t is shown in Fig. 3.11.

Fig. 3.11: Sample function of the random impulse train process

( ) ( ) ( ) ,
X m d n d
m n
R t t E A t mT T A t nT T

+ = +

( ) ( )
m n d d
m n
E A A t mT T t nT T

= +

As
n
A is independent of
d
T , we can write (after interchanging the order of
expectation and summation),
( ) ( ) ( ) ,
X m n d d
m n
R t t A A t nT T t nT T + = +

But,
1,
0,
m n
m n
A A
otherwise
=
=

This is because when m n = ,
( ) ( )
2 2
2
1 1
1 1 1
2 2
m n m
A A A = = + = . If m n ,
then
m n m n
A A A A = , as successive amplitudes are statistically independent. But
( ) ( )
1 1
1 1 0
2 2
m n
A A = = + = .

Using this result, we have

3.26
( ) ( ) ( ) ,
X d d
n
R t t t nT T t nT T + = +

( ) ( )
0
1
T
d d d
n
t nT t t nT t d t
T
= +

(Note that
d
T is uniformly distributed in the range 0 toT )
Let
d
t nT t x = . Then,
( ) ( ) ( )
( ) 1
1
,
t nT
X
n
t n T
R t t x x d x
T
+
+ = +

( ) ( )
+
= +
x x d x
T
1

Letting y x = , we have
( ) ( ) ( ) ( ) ( )
1 1
,
X
R t t y y d y y y d y
T T

+ = =

( ) ( ) ( )
1 1
T T
= =

That is, the ACF is a function of alone and it is an impulse!

It is to be pointed out that in the case of a random process, we can also
define time averages such as time-averaged mean value or time-averaged ACF
etc., whose calculation is based on the individual sample functions. There are
certain processes, called ergodic processes where it is possible to interchange
the corresponding ensemble and time averages. More details on ergodic
processes can be found in [2].

3.4.2 Cross-correlation
Consider two random processes
( ) X t and
( ) Y t . We define the two
cross-correlation functions of
( ) X t and
( ) Y t as follows:
( ) ( ) ( )
, 1 2 1 2
,
X Y
R t t E X t Y t =


3.27
( ) ( ) ( )
, 1 2 1 2
,
Y X
R t t E Y t X t =

where
1
t and
2
t are the two time instants at which the processes are observed.
Def. 3.6:
Two processes are said to be jointly wide-sense stationary if,
i)
( ) X t and
( ) Y t are WSS and
ii)
( ) ( ) ( )
, 1 2 , 1 2
,
X Y X Y XY
R t t R t t R = =

Cross-correlation is not generally an even function of as is the case with
ACF, nor does it have a maximum at the origin. However, it does obey a
symmetrical relationship as follows:

( ) ( )
XY Y X
R R = (3.8)
Def. 3.7:
Two random process
( ) X t and
( ) Y t are called (mutually) orthogonal if

( )
1 2
, 0
XY
R t t = for every
1
t and
2
t .

Def. 3.8:
Two random process
( ) X t and
( ) Y t are uncorrelated if

( ) ( ) ( ) ( )
1 2 1 2 1 2
, , 0
XY XY X Y
C t t R t t m t m t = = for every
1
t and
2
t .

Exercise 3.6
Show that the cross-correlation function satisfies the following
inequalities.
a) ( ) ( ) ( ) 0 0
XY X Y
R R R
b) ( ) ( ) ( )
+

1
0 0
2
XY X Y
R R R

3.28
3.5 Systems with Random Signal Excitation
In Chapter 1, we discussed the transmission of deterministic signals
through linear systems. We had developed the relations for the input-output
spectral densities. We shall now develop the analogous relationships for the case
when a linear time-invariant system is excited by random signals.
Consider the scheme shown in Fig. 3.12.
( ) h t represents the (known)
impulse response of a linear time-invariant system that is excited by a random
process
( ) X t , resulting in the output process
( ) Y t .

Fig 3.12: Transmission of a random process through a linear filter

We shall now try to characterize the output process
( ) Y t in terms of the
input process
( ) X t and the impulse response
( ) h t [third method of
specification]. Specifically, we would like to develop the relations for
( )
Y
m t and
( )
1 2
,
Y
R t t when
( ) X t is WSS.

Let
( )
j
x t be a sample function of
( ) X t which is applied as input to the
linear time-invariant system. Let
( )
j
y t be the corresponding output where
( )
j
y t
belongs to
( ) Y t . Then,
( ) ( ) ( )
j j
y t h x t d

As the above relation is true for every sample function of
( ) X t , we can write
( ) ( ) ( ) Y t h X t d

Consider first the mean of the output process

3.29
( ) ( ) ( ) ( )
Y
m t E Y t E h X t d

= =

(3.9a)
Provided that ( ) E X t

is finite for all t , and the system is stable, we may
interchange the order of the expectation and integration with respect to in Eq.
3.9(a) and write
( ) ( ) ( )
Y
m t h E X t d

=

(3.9b)
where we have used the fact that
( ) h is deterministic and can be brought
outside the expectation. If
( ) X t is WSS, then ( ) E X t

is a constant
X
m , so that
Eq. 3.9(b) can be simplified as
( ) ( )
Y X
m t m h d

( ) 0
X
m H = (3.10)
where ( ) ( )
0
0
f
H H f
=
= and
( ) H f is the transfer function of the given system.
We note that
( )
Y
m t is a constant.

Let us compute
( ) ,
Y
R t u , where t and u denote the time instants at
which the output process is observed. We have,
( ) ( ) ( ) ( ) ( ) ( ) ( )
1 1 1 2 2 2
,
Y
R t u E Y t Y u E h X t d h X u d

= =

Again, interchanging the order of integration and expectation, we obtain
( ) ( ) ( ) ( ) ( )
1 1 2 2 1 2
,
Y
R t u d h d h E X t X u

=

( ) ( ) ( )
1 1 2 2 1 2
,
X
d h d h R t u

=

If
( ) X t is WSS, then

3.30
( ) ( ) ( ) ( )
1 1 2 2 1 2
,
Y X
R t u d h d h R

= +

(3.11)
where t u = . Eq. 3.11 implies that
( ) ,
Y
R t u is only a function of t u .
Hence, the LHS of Eq. 3.11 can be written as
( )
Y
R . Eq. 3.10 and 3.11 together
imply that if
( ) X t is WSS, then so is
( ) Y t .

3.6 Power Spectral Density
The notion of Power Spectral Density (PSD) is an important and useful
one. It provides the frequency domain description of a stationary (at least WSS)
random process. From Eq. 3.11, we have
( ) ( ) ( ) ( ) ( )
2
1 2 2 1 1 2
0
Y X
E Y t R h h R d d

= =

But, ( ) ( )
1
2
1
j f
h H f e d f

=

Hence, ( ) ( ) ( ) ( )
1
2 2
2 2 1 1 2
j f
X
E Y t H f e d f h R d d

=

( ) ( ) ( )
1
2
2 2 2 1 1
j f
X
H f d f h d R e d

=

Let
2 1
= ; that is,
1 2
= .
( ) ( ) ( ) ( )
2
2 2 2
2 2
j f j f
X
E Y t H f d f h e d R e d

=

The second integral above is
( ) H f
, the complex conjugate of

( ) H f . The third
integral will be a function f , which we shall denote by
( )
X
S f . Then
( ) ( ) ( )
2
2
X
E Y t S f H f d f

=

(3.12)

3.31
where ( ) ( )
2 j f
X X
S f R e d
.

We will now justify that
( )
X
S f can be interpreted as the power spectral
density of the WSS process
( ) X t . Suppose that the process
( ) X t is passed
through an ideal narrowband, band-pass filter with the passband centered at
c
f
and having the amplitude response
( )
1
1,
2
1
0,
2
c
c
f f f
H f
f f f
<
>

Then, from Eq. 3.12, we find that if the filter band-width is sufficiently small
and
( )
X
S f is a continuous function, then the mean square value of the filter
output is approximately,
( ) ( ) ( )
2
2
X c
E Y t f S f

The filter however passes only those frequency components of the input random
process
( ) X t that lie inside the narrow frequency band of width f centered
about
c
f . Thus,
( )
X c
S f represents the frequency density of the average power
in the process
( ) X t , evaluated at
c
f f = . The dimensions of
( )
X
S f are
watts/Hz.

3.6.1 Properties of power spectral density
The PSD
( )
X
S f and the ACF
( )
X
R of a WSS process
( ) X t form a
Fourier transform pair and are given by
( ) ( )
2 j f
X X
S f R e d
(3.13)

3.32
( ) ( )
2 j f
X X
R S f e d f

=

(3.14)
Eq. 3.13 and 3.14 are popularly known as Weiner-Khinchine Relations. Using
this pair of equations, we shall derive some general properties of PSD of a WSS
random process.
P1) The zero frequency value of the PSD of a WSS process equals the total
area under the graph of ACF; that is
( ) ( ) 0
X X
S R d

This property follows directly from Eq. 3.13 by putting 0 f = .
P2) The mean square value of a WSS process equals the total area under the
graph of the PSD; that is,
( ) ( )
2
X
E X t S f d f

=

This property follows from Eq. 3.14 by putting 0 = and noting
( ) ( )
2
0
X
R E X t =

.
P3) The PSD is real and is an even function of frequency; that is,
( )
X
S f is real
and
( ) ( )
X X
S f S f = .
This result is due to the property of the ACF, namely,
( )
X
R is real and
even.
P4) The PSD of a WSS process is always non-negative; that is,
( ) 0
X
S f , for all f .
To establish this, assume that
( )
X
S f is negative, for a certain frequency
interval, say
( )
1 1
, f f f + . Let
( ) X t be the input to a narrowband filter with
the transfer function characteristic,
( )
1 1
1,
0,
f f f f
H f
otherwise
+


3.33
Then from Eq. 3.12, we have ( ) ( ) ( )
2
2
X
E Y t S f H f d f

=

, as a
negative quantity which is a contradiction.

We shall now derive an expression for the power spectral density,
( )
Y
S f ,
of the output of Fig. 3.12.

Using the relation ( ) ( )
1
X X
R F S f
=

, Eq. 3.11 can be written as
( ) ( ) ( ) ( )
( )
1 2
2
1 2 1 2
j f
Y X
R h h S f e d f d d

+

=

( ) ( ) ( )
1 2
2 2 2
1 1 2 2
j f j f j f
X
h e d h e d S f e d f

=

( ) ( ) ( )
2 j f
X
H f H f S f e d f

( ) ( )
2
2 j f
X
H f S f e d f
(3.15)
As
( ) ( )
Y Y
R S f , Eq. 3.15 implies
( ) ( ) ( )
2
Y X
S f S f H f = (3.16)
Note that when the input to an LTI system is deterministic, we have the input-
output FT relationship,
( ) ( ) ( ) Y f X f H f = . The corresponding time domain
relationship is
( ) ( ) ( ) y t x t h t = . Let
( )
h
R denote the ACF of the
( ) h t . Then
( ) ( )
2
h
R H f (see P4, sec 1.6.2). Hence

( ) ( ) ( )
Y x h
R R R =

( ) ( ) ( )
x
R h h
= (3.17a)
If the impulse response is real, then

( ) ( ) ( ) ( )
Y x
R R h h = (3.17b)

3.34
We shall now take up a few examples involving the computation of PSD.

Example 3.7
For the random process
( ) X t of example 3.5, let us find the PSD.

Since ( ) ( )
2
cos
2
X c
A
R = , we have
( ) ( ) ( )
2
4
X c c
A
S f f f f f = + +

Example 3.8: (Modulated Random Process)
Let
( ) ( ) ( ) cos
c
Y t X t t = + where
( ) X t is a WSS process with known
( )
X
R and
( )
X
S f . is a uniformly distributed random variable in the range
( ) 0 2 .
( ) X t and are independent. Let us find the ACF and PSD of
( ) Y t .

( ) ( ) ( )
Y
R E Y t Y t = +

( ) ( ) ( ) ( )
{ }
cos cos
c c
E X t t X t t = + + + +

As
( ) X t and are independent, we have
( ) ( ) ( ) ( ) ( ) ( ) cos cos
Y c c
R E X t X t E t t
= + + + +

( ) ( )
1
cos
2
X c
R =

( ) ( ) ( ) ( )
1
4
Y Y X c X c
S f F R S f f S f f = = + +

Example 3.9: (Random Binary Wave)
Fig. 3.13 shows the sample function
( )
j
x t of a process
( ) X t consisting of
a random sequence of binary symbols 1 and 0. It is assumed that:

3.35
1. The symbols 1 and 0 are represented by pulses of amplitude A + and A
volts, respectively and duration T seconds.
2. The pulses are not synchronized, so that the starting time of the first pulse,
d
t is equally likely to lie anywhere between zero and T seconds. That is,
d
t
is the sample value of a uniformly distributed random variable
d
T with its
probability density function defined by
( )
1
, 0
0 ,
d
d
T d
t T
f t
T
elsewhere

Fig. 3.13: Random binary wave

3. During any time interval
( ) 1
d
n T t t nT , where n is an integer, the
presence of a 1 or 0 is determined by tossing a fair coin; specifically, if the
outcome is heads, we have a 1 and if the outcome is 'tails', we have a 0.
These two symbols are thus equally likely, and the presence of a 1 or 0 in
anyone interval is independent of all other intervals. We shall compute
( )
X
S f and
( )
X
R .

The above process can be generated by passing the random impulse train
(Example 3.6) through a filter with the impulse response,
( )
, 0
0,
A t T
h t
otherwise

=


3.36
( ) ( ) sin H f AT c f T =
Therefore,
( ) ( )
2 2 2
1
sin
X
S f A T c f T
T
=

( )
2 2
sin A T c f T =
Taking the inverse Fourier transform, we have
( )
2
1 , 0
0 ,
X
A T
R T
otherwise

=

The ACF of the random binary wave process can also be computed by direct
time domain arguments. The interested reader is referred to [3].

We note that the energy spectral density of a rectangular pulse
( ) x t of
amplitude A and duration T , is
( ) ( )
2 2 2
sin
x
E f A T c f T =
Hence,
( ) ( )
x x
S f E f T = .

Exercise 3.7
In the random binary wave process of example 3.9, let 1 be
represented by a pulse of amplitude A and duration T sec. The binary zero is
indicated by the absence of any pulse. The rest of the description of the
process is the same as in the example 3.9. Show that
( ) ( )
( )
2
2
2 2
sin
4
X
f T
A
S f f
f T

= +

Exercise 3.8
The input voltage to an RLC series circuit is a WSS process ( ) X t with
( ) 2 X t = and ( )
2
4
X
R e

= + . Let ( ) Y t be the voltage across the
capacitor. Find
a) ( ) Y t
b) ( )
Y
S f

3.37
3.7 Cross-Spectral Density
J ust as the PSD of a process provides a measure of the frequency
distribution of the given process, the cross spectral density provides a measure
of frequency interrelationship between two processes. (This will become clear
from the examples that follow). Let
( ) X t and
( ) Y t be two jointly WSS random
processes with their cross correlation functions
( )
XY
R and
( )
Y X
R . Then, we
define,
( )
XY
S f and
( )
Y X
S f as follows:
( ) ( )
2 j f
XY XY
S f R e d
(3.18)
( ) ( )
2 j f
Y X Y X
S f R e d
(3.19)
That is,
( ) ( )
XY XY
R S f
( ) ( )
Y X Y X
R S f
The cross-spectral density is in general complex. Even if it is real, it need
not be positive. However, as
( ) ( )
XY Y X
R R = , we find
( ) ( ) ( )
XY Y X Y X
S f S f S f
= =
We shall now give a few examples that involve the cross-spectrum.

Example 3.10
Let
( ) ( ) ( ) Z t X t Y t = +
where the random processes
( ) X t and
( ) Y t are jointly WSS and
( ) ( ) 0 X t Y t = = . We will show that to find
( )
Z
S f , we require the cross-
spectrum.

( ) ( ) ( ) ,
Z
R t u E Z t Z u =


3.38
( ) ( ) ( ) ( ) ( ) ( )
E X t Y t X u Y u

= + +

( ) ( ) ( ) ( ) , , , ,
X XY Y X Y
R t u R t u R t u R t u = + + +
Letting t u = , we have

( ) ( ) ( ) ( ) ( )
Z X XY Y X Y
R R R R R = + + + (3.20a)
Taking the Fourier transform, we have
( ) ( ) ( ) ( ) ( )
Z X XY Y X Y
S f S f S f S f S f = + + +
(3.20b)
If
( ) X t and
( ) Y t are uncorrelated, that is,
( ) ( ) ( ) 0
XY
R X t Y t = + =
then,

( ) ( ) ( )
Z X Y
R R R = + and

( ) ( ) ( )
Z X Y
S f S f S f = +
Hence, we have the superposition of the autocorrelation functions as well as the
superposition of power spectral densities.

Example 3.11
Let
( ) X t be the input to an LTI system with the impulse response
( ) h t . If
the resulting output is
( ) Y t , let us find an expression for
( )
Y X
S f .

( ) ( ) ( ) ,
Y X
R t u E Y t X u =

( ) ( ) ( )
1 1 1
E h X t d X u

Interchanging the order of expectation and integration, we have
( ) ( ) ( ) ( )
1 1 1
,
Y X
R t u h E X t X u d

=

If
( ) X t is WSS, then

3.39
( ) ( ) ( )
1 1 1
,
Y X X
R t u h R t u d

Letting t u = gives the result
( ) ( ) ( )
1 1 1
,
Y X X
R t u h R d

( ) ( )
X
h R = (3.21a)
That is, the cross-correlation between the output and input is the convolution of
the input ACF with the filter impulse response. Taking the Fourier transform,

( ) ( ) ( )
Y X X
S f S f H f =
(3.21b)
Eq. 3.21(b) tells us that
( ) X t and
( ) Y t will have strong cross-correlation in
those frequency components where
( ) ( )
X
S f H f is large.

Example 3.12
In the scheme shown (Fig. 3.14),
( ) X t and
( ) Y t are jointly WSS. Let us
compute
( )
V Z
S f .

Fig. 3.14: Figure for the example 3.12

( ) ( ) ( ) ,
V Z
R t u E V t Z u =

( ) ( ) ( ) ( )
1 1 1 1 2 2 2 2
E h X t d h Y u d

=

Interchanging the order of expectation and integration,

3.40
( ) ( ) ( ) ( ) ( )
1 1 2 2 1 2 1 2
,
V Z
R t u h h E X t Y u d d

=

As
( ) X t and
( ) Y t are jointly WSS, we have
( ) ( ) ( ) ( )
1 1 2 2 1 2 1 2
,
V Z XY
R t u h h R d d

= +

where t u = . That is,
( ) ( ) ( ) ,
V Z V Z V Z
R t u R t u R = =
It not too difficult to show that

( ) ( ) ( ) ( )
1 2 V Z XY
S f H f H f S f
(3.22)

We shall now consider some special cases of Example 3.12.
i) Let
( )
1
H f and
( )
2
H f of Fig. 3.14 have non-overlapping passbands. Then
( ) ( )
1 2
0 H f H f
= for all f . That is,

( )
V Z
S f 0 ; this implies
( )
VZ
R 0
and we have
( ) V t and
( ) Z t being orthogonal. In addition, either ( ) V t or
( ) Z t (or both) will be zero (note that ( ) ( )
1
0
X
V t H m = and
( ) ( )
2
0
Y
Z t H m = ), because atleast one of the quantities,
( )
1
0 H or
( )
2
0 H
has to be zero. That is, the two random variables,
( )
i
V t and
( )
j
Z t obtained
from sampling the processes
( ) V t at
i
t , and ( ) Z t at
j
t respectively, are
uncorrelated. That is, the processes
( ) V t and ( ) Z t are uncorrelated.
ii) Let
( ) ( ) X t Y t = and
( ) X t is WSS. Then we have the situation shown in
Fig. 3.15.


3.41

Fig. 3.15: Two LTI systems with a common input

Then from Eq. 3.22, we obtain,

( ) ( ) ( ) ( )
1 2 V Z X
S f H f H f S f
= (3.23)
iii) In the scheme of Fig. 3.15, if
( ) ( ) ( )
1 2
h t h t h t = = , then we have the
familiar result
( ) ( ) ( )
2
V X
S f H f S f =

3.8 Gaussian Process
Gaussian processes are of great practical and mathematical significance
in the theory of communication. They are of great practical significance because
a large number of noise processes encountered in the real world (and that
interfere with the communication process) can be characterized as Gaussian
processes. They are of great mathematical significance because Gaussian
processes possess a neat mathematical structure which makes their analytical
treatment quite feasible.

Let a random process
( ) X t be sampled at time instants
1 2
, , ........,
n
t t t .
Let
( ) ( ) ( )
1 1 2 2
, , ............,
n n
X t X X t X X t X = = = and X denote the row vector
( )
1 2
, , ...........,
n
X X X . The process
( ) X t is called Gaussian, if
( ) f
X
x is an n-
dimensional joint Gaussian density for every 1 n and
( ) ( )
1 2
, , ..........., ,
n
t t t . The n-dimensional Gaussian PDF is given by

3.42
( )
( )
( ) ( )
1
1
2
2
1 1
exp
2
2
T
x X
n
X
f C
C

=

X X
x x m x m (3.24)
where
x
C is the covariance matrix, denotes its determinant and
1
X
C
is its
inverse.
X
m is the mean vector, namely,
( )
1 2
, , ...........,
n
X X X and the
superscript T denotes the matrix transpose. The covariance matrix
x
C is given
by

11 12 1
21 22 2
1 2
n
n
x
nn n n
C

=

where cov ,
i j i j
X X =

( ) ( )
i i j j
E X X X X

=

Similarly, ( )
( )
1 1 2 2
, , ........,
X n n
x X x X x X = x m
Specification of a Gaussian process as above corresponds to the first
method mentioned earlier. Note that an n-dimensional Gaussian PDF depends
only on the means and the covariance quantities of the random variables under
consideration. If the mean value of the process is constant and
( ) ( )
cov ,
i j i j
X t X t

=

depends only on
i j
t t , then the joint PDF is
independent of the time origin. In other words, a WSS Gaussian process is also
stationary in a strict sense. To illustrate the use of the matrix notation, let us take
an example of a joint Gaussian PDF with 2 n = .

Example 3.13
1
X and
2
X are two jointly Gaussian variables with
1 2
0 X X = = and
1 2
= = . The correlation coefficient between
1
X and
2
X is . Write the
joint PDF of
1
X and
2
X in i) the matrix form and ii) the expanded form.


3.43
i) As
2
11 22
= = and
2
12 21
= = , we have
( )
2 2
4 2
2 2
, 1
x X
C C

= =

( )
2 2
1
2 2 4 2
1
1
X
C

=

( )
2 2
1
1
1
1

=

Therefore,
( )
( )
( )
1 2
1
, 1 2 1 2
2 2
2 2
2
1
1 1
, exp
1 2 1
2 1
X X
x
f x x x x
x

=

ii) Taking the matrix products in the exponent above, we have the expanded
result, namely,
( )
( )
1 2
2 2
1 1 2 2
, 1 2 1 2
2 2
2 2
2 1
, exp , ,
2 1
2 1
X X
x x x x
f x x x x

+
= < <

This is the same as the bivariate Gaussian PDF of Chapter 2, section 6 with
1
x x = and
2
x y = , and 0
X Y
m m = = , and
X Y
= = .

Example 3.14
( ) X t is a Gaussian process with
( ) 3
X
m t = and
( )
1 2
0.2 0.2
1 2
, 4 4
t t
X
C t t e e

= = . Let us find, in terms of
( ) Q ,
a) ( ) 5 2 P X

b) ( ) ( ) 8 5 1 P X X


3.44
a) Let Y be the random variable obtained by sampling
( ) X t at 5 t = . Y is
Gaussian with the mean value 3, and the variance
( ) 0 4
X
C = = . That is,
Y is
( ) 3, 4 N .
[ ] ( )
3 1
2 0.5
2 2
Y
P Y P Q

= =

b) Let
( ) ( ) 8 5 Z X X = and
( ) 8 Y X = , and
( ) 5 W X = . Note that
Z Y W = is Gaussian. We have 0 Z = and
2 2 2
,
2
Z Y W Y W Y W
= +

[ ]
2 2
2 cov ,
Y W
Y W = +

0.2 3
4 4 2 4 e

= +

( )
0.6
8 1 3.608 e
= =
[ ] [ ]
1
1 2 0 1 2 1
2
P Z P Z P Z

= = >

( )
1
1 2 1 2 0.52
3.6 3.6
Z
P Q

= > =


3.45

We shall now state (without proof) some of the important properties of n
jointly Gaussian random variables,
( )
1 2
, , ........,
n
X X X = X . (These properties
are generalizations of the properties mentioned for the bivariate case in
Chapter 2).

P1) If
( )
1 2
, , ........,
n
X X X = X are jointly Gaussian, then any subset of them
are jointly Gaussian.
P2) The
i
X s are statistically independent if their covariance matrix is diagonal;
that is,
2
i j i i j
= where
Exercise 3.9
Let X be a zero-mean Gaussian vector with four components,
1 2
, , X X
3 4
and X X . We can show that

1 2 3 4 1 2 3 4 1 3 2 4 1 4 2 3
X X X X X X X X X X X X X X X X = + +
The above formula can also be used to compute the moments such as
4
1
X ,
2 2
1 2
X X , etc.
4
1
X can be computed as
= = X X X X X
4 4
1 1 1 1 1 1
3
Similarly,
( )
2
2 2 2 2
1 2 1 1 2 2 1 2 1 2
2 X X X X X X X X = = +

A zero mean stationary Gaussian process is sampled at
1
t t = and
2
t .
Let
1
X and
2
X denote the corresponding random variables. The covariance
matrix of
1
X and
2
X is

2 1
1 2
X
C

=

Show that
2 2
1 2
6 E X X =


3.46
1,
0,
i j
i j
i j
=
=

P3) Let
( )
1 2
, , ..........,
n
Y Y Y = Y be a set of vectors obtained from
( )
1 2
, , ..........,
n
X X X = X by means of the linear transformation
T T T
A = + Y X a
where A is any n n matrix and
( )
1 2
, , ..........,
n
a a a = a is a vector of
constants
{ }
i
a . If X is jointly Gaussian, then so is Y .

As a consequence of P3) above, we find that if a Gaussian process
( ) X t is input to a linear system, then the output
( ) Y t is also Gaussian
process. We shall make use of this result in developing the properties of
narrowband noise.

3.9 Electrical Noise
Electronic communication systems are made up of circuit elements such
as R , L and C , and devices like diodes, transistors, etc. All these components
give rise to what is known as internal circuit noise. It is this noise that sets the
fundamental limits on communication of acceptable quality.

Two of the important internal circuit noise varieties are: i) Thermal noise
ii) Shot noise

Exercise 3.10
Let
T T T
A = + Y X a
where the notation is from P3) above. Show that

T
Y X
C AC A =

3.47
Historically, J ohnson and Nyquist first studied thermal noise in metallic
resistors; hence it is also known as J ohnson noise or resistance noise. The
random motion of the free electrons in a conductor caused by thermal agitation
gives rise to a voltage
( ) V t at the open ends of the conductor and is present
whether or not an electrical field is applied. Consistent with the central limit
theorem,
( ) V t is a Gaussian process with zero mean and the variance is a
function of R and T , where R is the value of the resistance and T is the
temperature of R . It has also been found that the spectral density of
( ) V t
( )
( )
2
in Volts Hz , denoted
( )
V
S f is essentially constant for
12
10 f Hz, if T is
290 K
or 63 F
(290 K
is taken as the standard room temperature).

12
10 Hz is
already in the infrared region of EM spectrum. This constant value of
( )
V
S f is
given by

( )
2
2
V
S f RkT V Hz = (3.25)
where T : the temperature of R , in degrees Kelvin
( )
K C 273 = +

k : Boltzmans constant
23
1.37 10 J oules K
=

.
It is to be remembered that Eq. 3.25 is valid only upto a certain frequency limit.
However this limit is much, much higher than the frequency range of interest to
us.

If this open circuit voltage is measured with the help of a true RMS
voltmeter of bandwidth B (frequency range: to B B ), then the reading on the
instrument would be 2 2 4 RkT B RkT B V = .
Thermal noise sources are also characterized in terms of available noise
PSD.
Def. 3.9:
Available noise PSD is the maximum PSD that can be delivered by a
source.


3.48
Let us treat the resistor to be noise free. But it is in series with a noise
source with
( ) 2
V
S f RkT = . In other words, we have a noise source with the
source resistance R (Fig. 3.16(a)).

Fig. 3.16: (a) Resistor as noise source (b) The noise source driving a load

We know that the maximum power is transferred to the load, when the load is
matched to the source impedance. That is, the required load resistance is R
(Fig. 3.16(b)). The transfer function of this voltage divider network is
1
2

i.e.
R
R R

+

. Hence,
available
( )
( )
2
V
S f
PSD H f
R
=

2
Watts Hz
4 2
RkT kT
R
= = (3.26)
It is to be noted that available PSD does not depend on R , though the
noise voltage is produced by R .
Because of Eq. 3.26, the available power in a bandwidth of B Hz is,
2
2
a
kT
P B kT B = = (3.27)

In many solid state devices (and of course, in vacuum tubes!) there exists
a noise current mechanism called shot noise. In 1918, Schottky carried out the
first theoretical study of the fluctuations in the anode current of a temperature

3.49
limited vacuum tube diode (This has been termed as shot noise). He showed that
at frequencies well below the reciprocal of the electron transit time (which
extends upto a few Giga Hz), the spectral density of the mean square noise
current due to the randomness of emission of electrons from the cathode is given
by,
( ) ( )
2
2 Amp Hz
I
S f qI =
where q is the electronic charge and I is the average anode current. (Note that
units of spectral density could be any one of the three, namely, (a) Watts Hz,
(b) ( )
2
Volts Hz and (c) ( )
2
Amp Hz. The circuit of Fig. 3.16(a) could be drawn
with
( )
I
S f for the source quantity. Then, we will have the Norton equivalent
circuit with
( )
2
2 2
I
RkT kT
S f
R R
= = in parallel with the resistance R .)

Semiconductor diodes, BJ Ts and J FETs have sources of shot noise in
them. Shot noise (which is non-thermal in nature), occurs whenever charged
particles cross a potential barrier. Here, we have an applied potential and there is
an average flow in some direction. However, there are going to be fluctuations
about this average flow and it is these fluctuations that contribute a noise with a
very wide spectrum.
In p-n junction devices, fluctuations of current occurs because of (i)
randomness of the transit time across space charge region separating p and n
regions, (ii) randomness in the generation and recombination of electron-hole
pairs, and (iii) randomness in the number of charged particles that diffuse etc.

Schottkys result also holds for the semiconductor diode. The current
spectral density of shot noise in a p-n junction diode is given by
( ) ( ) 2 2
I S
S f q I I = + (3.28)
where I is the net DC current and
S
I is the reverse saturation current.


3.50
A BJ T has two semiconductor junctions and hence two sources of shot
noise. In addition, it contributes to thermal noise because of internal ohmic
resistance (such as base resistance).

A J FET has a reverse biased junction between the gate terminal and the
semiconductor channel from the source to the drain. Hence, we have the gate
shot noise and the channel thermal noise in a J FET. (Note that gate shot noise
could get amplified when the device is used in a circuit.)

3.9.1 White noise
Eq. 3.25 and Eq. 3.28 indicate that we have noise sources with a flat
spectral density with frequencies extending upto the infrared region of the EM
spectrum.

The concept of white noise is an idealization of the above. Any noise
quantity (thermal or non-thermal) which has a fiat power spectrum (that is, it
contains all frequency components in equal proportion) for f < < is
called white noise, in analogy with white light. We denote the PSD of a white
noise process
( ) W t as
( )
0
watts Hz
2
W
N
S f = (3.29)
where the factor
1
2
has been included to indicate that half the power is
associated with the positive frequencies and half with negative frequencies. In
addition to a fiat spectrum, if the process happens to be Gaussian, we describe it
as white Gaussian noise. Note that white and Gaussian are two different
attributes. White noise need not be Gaussian noise, nor Gaussian noise need be
white. Only when "whiteness" together with "Gaussianity" simultaneously exists,
the process is qualified as White Gaussian Noise (WGN) process.


3.51
White noise, whether Gaussian or not, must be fictitious (as is the case
with everything that is "ideal") because its total mean power is infinity. The utility
of the concept of white noise stems from the fact that such a noise, when passed
through a linear filter for which
( )
2
H f

<
(3.30)
the filter output is a stationary zero mean noise process
( ) N t that is meaningful
(Note that white noise process, by definition, is a zero mean process).

The condition implied by Eq.3.30 is not too stringent as we invariably deal
with systems which are essentially band limited and have finite value for ( ) H f
within this band. In so far as the power spectrum at the output is concerned, it
makes little difference how the input power spectrum behaves outside of the
pass band. Hence, if the input noise spectrum is flat within the pass band of the
system, we might as well treat it to be white as this does not affect the calculation
of output noise power. However, assuming the input as white will simplify the
calculations.

As
( )
0
2
W
N
S f = , we have, for the ACF of white noise
( ) ( )
0
2
W
N
R = , (3.31)
as shown in Fig. 3.17. This is again a nonphysical but useful result.


3.52

Fig.3.17: Characterization of white noise: a) Power spectral density
b) Auto correlation function

Eq. 3.31 implies that any two samples of white noise, no matter how closely
together in time they are taken, are uncorrelated. (Note that
( ) 0
W
R = for
0 ). In addition, if, the noise process is Gaussian, we find that any two
samples of WGN are statistically independent. In a sense, WGN represents the
ultimate in randomness!

Imagine that white noise is being displayed on an oscilloscope. Though
the waveforms on successive sweeps are different, the display on the
oscilloscope always appears to be the same, no matter what sweep speed is
used. In the case of display of a deterministic waveform (such as a sinusoid)
changing the time-base makes the waveform to expand or contract. In the case
of white noise, however, the waveform changes randomly from one instant to the
next, no matter what time scale is used and as such the display on the scope
appears to be the same for all time instants. If white noise drives a speaker, it

3.53
should sound monotonous because the waveform driving the speaker appears to
be the same for all time instants.

As mentioned earlier, white noise is fictitious and cannot be generated in
the laboratory. However, it is possible to generate what is known as Band Limited
White Noise (BLWN) which has a flat power spectral density for f W , where
W is finite. (BLWN sources are commercially available. The cost of the
instrument depends on the required bandwidth, power level, etc.) We can give
partial demonstration (audio-visual) of the properties of white noise with the help
of these sources.

We will begin with the audio. By clicking on the speaker symbol, you will
listen to the speaker output when it is driven by a BLWN source with the
frequency range upto 110 kHz. (As the speaker response is limited only upto 15
kHz, the input to the speaker, for all practical purposes, is white.)

We now show some flash animation pictures of the spectral density and
time waveforms of BLWN.

1. Picture 1 is the display put out by a spectrum analyzer when it is fed with a
BLWN signal, band limited to 6 MHz. As can be seen from the display, the
spectrum is essentially constant upto 6 MHz and falls by 40 to 45 dB at
about 7MHz and beyond.

2. Picture 2 is time domain waveform (as seen on an oscilloscope) when the 6
MHz, BLWN signal is the input to the oscilloscope. Four sweep speeds
have been provided for you to observe the waveform. These speeds are:
100 sec/div, 200 sec/div, 500 sec/div and 50 sec/div. As you switch
from 100 sec/div to 200 sec/div to 500 sec/div, you will find that the

3.54
display is just about the same, whereas when you switch from 500 sec/div
to 50 sec/div, there is a change, especially in the display level.

Consider a 500 kHz sine wave being displayed on an oscilloscope
whose sweep is set to 50 sec/div. This will result in 25 cycles per division
which implies every cycle will be seen like a vertical line of appropriate
height. As the frequency increases, the display essentially becomes a band
of fluorescence enclosed between two lines. (The spikes in the display are
due to large amplitudes in noise voltage which occur with some non-zero
probability.)

Now consider a 20 kHz sine wave being displayed with the same
sweep speed, namely, 50 sec/div. This will result in 1 cycle/div which
implies we can see the fine structure of the waveform. If the sweep speed
were to be 500 sec/div, then a 20 kHz time will result in 10 cycles/div,
which implies that fine structure will not be seen clearly. Hence when we
observe BLWN, band limited to 6 MHz, on an oscilloscope with a sweep
speed of 50 sec/div, fine structure of the low frequency components could
interfere with the constant envelope display of the higher frequency
components, thereby causing some reduction in the envelope level.

3. Picture 3 is the display from the spectrum analyzer when it is fed with a
BLWN signal, band limited to 50 MHz. As can be seen from the display,
PSD is essentially constant upto 50 MHz and then starts falling, going 40
dB down by about 150 MHz.

4. Picture 4 is the time domain signal, as seen on an oscilloscope, when the
input is the 50 MHz wide, BLWN signal.

You have again the choice of the same four sweep rates. This time
you will observe that when you switch from 500 sec/div to 50 sec/div, the

3.55
change is much less. This is because, this signal has much wider spectrum
and the power in the frequency range of 100 f kHz is 0.002P where P
is the total power of the BLWN signal.

From the above, it is clear that as the BLWN tends towards white noise,
variations in the time waveform keep reducing, resulting in a steady picture on
the oscilloscope no matter what sweep speed is used for the display.

Example 3.15: (White noise through an ILPF)
White Gaussian noise with the PSD of
0
2
N
is applied as input to an ideal
LPF of bandwidth B . Find the ACF of the output process. Let Y be the random
variable obtained from sampling the output process at 1 t = sec. Let us find
( )
Y
f y .

Let
( ) N t denote the output noise process when the WGN process
( ) W t
is the input. Then,
( )
0
,
2
0 ,
N
N
B f B
S f
otherwise

Taking the inverse Fourier transform of
( )
N
S f , we have

( ) ( )
0
sin 2
N
R N B c B =
A plot of
( )
N
R is given in Fig. 3.18.


3.56

Fig 3.18:
( )
N
R of example 3.15

The filter output
( ) N t is a stationary Gaussian process. Hence the PDF is
independent of the sampling time. As the variable Y is Gaussian, what we need
is the mean value and the variance of Y . Evidently
[ ]
0 E Y = and
( )
2 2
0
0
Y Y
Y R N B = = = . Hence Y is
( )
0
0, N N B .

Note that ACF passes through zeros at
2
n
B
= where 1, 2, ..... n = .
Hence any two random variables obtained by sampling the output process at
times
1
t and
2
t such that
1 2
t t is multiple of
1
2B
, are going to be statistically
independent.

Example 3.16: (White Noise through an RC-LPF)
Let
( ) W t be input to an RC-LPF. Let us find the ACF of the output
( ) N t .
If X and Y are two random variables obtained from sampling
( ) N t with a
separation of 0.1 sec, let us find
XY
.

3.57
The transfer function of the RC LPF is given by ( )
1
1 2
H f
j f RC
=
+
.
Therefore
( )
( )
0
2
2
1 2
N
N
S f
f RC
=
+
, and
( )
0
exp
4
N
N
R
RC RC

=

[ ]
2
XY
XY
X Y
E XY
E X
= =

(Note that ( ) 0 N t = and
X Y
= ).
[ ] ( )
0.1sec
N
E XY R
=
=
( )
2 0
0
4
N
N
E X R
RC
=
= =

Hence,
( ) 0.1RC
XY
e

Exercise 3.11
( ) X t is a zero mean Gaussian process with ( )
2
1
1
X
R =
+
. Let
( )

( ) Y t X t = where
( ) X t is the Hilbert transform of ( ) X t . The process ( ) X t

and ( ) Y t are sampled at 1 t = and 2 t = sec. Let the corresponding random
variables be denoted by ( )
1 2
, X X and ( )
1 2
, Y Y respectively.
a) Write the covariance matrix of the four variables
1 2
, X X
1 2
and Y Y .
b) Find the joint PDF of
2
, X
1 2
and Y Y .

3.58

Let
0
denote the de-correlation time where
0
is defined such that if
0
> , then ( ) ( ) 0.01 0
N N
R R . Then for the RC-LPF,
0
4.61RC = . That
is, if the output process ( ) N t is sampled at
1
t and
2
t such that
1 2
4.61 t t RC , then the random variables ( )
1
N t and ( )
2
N t will be
Exercise 3.12
The impulse response of a filter (LTI system) is given by
( ) ( ) ( )
t
h t t e u t

=
where is a positive constant. If the input to the filter is white Gaussian noise
with the PSD of
0
2
N
watts/Hz, find
a) the output PSD
b) Show that the ACF of the output is
( )
0
2 2
N
e

Exercise 3.13
White Gaussian noise process is applied as input to a zero-order-hold
circuit with a delay of T sec. The output of the ZOH circuit is sampled at
t T = . Let Y be the corresponding random variable. Find ( )
Y
f y .

Exercise 3.14
Noise from a 10 k resistor at room temperature is passed through an
ILPF of bandwidth 2.5 MHz. The filtered noise is sampled every microsecond.
Denoting the random variables corresponding to the two adjacent samples as
1
Y and
2
Y , obtain the expression for the joint PDF of
1
Y and
2
Y .

3.59
essentially uncorrelated. If ( ) N t happens to be a Gaussian process, then ( )
1
N t
and ( )
2
N t will be, for all practical purposes, independent.

In example 3.15, we found that the average output noise power is equal to
0
N B where B is the bandwidth of the ILPF. Similarly, in the case of example
3.16, we find the average output noise power being equal to
0
4
N
RC

where
1
2 RC
is the 3-dB bandwidth of the RC-LPF. That is, in both the cases, we find
that the average output noise power is proportional to some measure of the
bandwidth. We may generalize this statement to include all types of low pass
filters by defining the noise equivalent bandwidth as follows.

Suppose we have a source of white noise with the PSD
( )
0
2
W
N
S f =
connected to an arbitrary low pass filter of transfer function
( ) H f . The resulting
average output noise power N is given by,
( )
2
0
2
N
N H f d f

=

( )
2
0
0
N H f d f
=

Consider next the same source of white noise connected to the input of an
ILPF of zero frequency response
( ) 0 H and bandwidth B . In this case, the
average output noise power is,
( )
2
0
' 0 N N BH = .
Def. 3.10:
The noise equivalent bandwidth of an arbitrary filter is defined as the
bandwidth of an ideal rectangular filter that would pass as much white noise

3.60
power as the filter in question and has the same maximum gain as the filter
under consideration.

Let
N
B be the value of B such that ' N N = . Then
N
B , the noise
equivalent bandwidth of
( ) H f , is given by
( )
( )
2
0
2
0
N
H f d f
B
H
=

(3.32)
The notion of equivalent noise bandwidth is illustrated in Fig. 3.19.

Fig. 3.19: Pictorial representation of noise equivalent bandwidth

The advantage of the noise equivalent bandwidth is that if
N
B is known,
we can simply calculate the output noise power without worrying about the actual
shape of ( ) H f . The definition of noise equivalent bandwidth can be extended to
a band pass filter.

Example 3.17
Compute the noise equivalent bandwidth of the RC-LPF.

We have ( ) 0 1 H = . Hence,

( )
2
0
1
4
1 2
N
d f
B
RC
f RC
= =
+


3.61
3.10 Narrowband Noise
The communication signals that are of interest to us are generally
NarrowBand BandPass (NBBP) signals. That is, their spectrum is concentrated
around some (nominal) centre frequency, say
c
f , with the signal bandwidth
usually being much smaller compared to
c
f . Hence the receiver meant for such
signals usually consists of a cascade of narrowband filters; this means that even
if the noise process at the input to the receiver is broad-band (may even
considered to be white) the noise that may be present at the various stages of a
receiver is essentially narrowband in nature. We shall now develop the statistical
characterization of such NarrowBand Noise (NBN) processes.

Let ( ) N t denote the noise process at the output of a narrowband filter
produced in response to a white noise process, ( ) W t , at the input. ( )
W
S f is
taken as
0
2
N
. If ( ) H f denotes the filter transfer function, we have
( ) ( )
2
0
2
N
N
S f H f = (3.33)
In fact, any narrowband noise encountered in practice could be modeled as the
output of a suitable filter. In Fig. 3.20, we show the waveforms of experimentally
generated NBBP noise. This noise process is generated by passing the BLWN
(with a flat spectrum upto 110 kHz) through a NBBP filter, whose magnitude
characteristic is shown in Fig. 3.21. This filter has a centre frequency of 101 kHz
and a 3-dB bandwidth of less than 3 kHz.


3.62

Fig. 3.20: Some oscilloscope displays of narrowband noise
X-axis: time
Y-axis: voltage

3.63
The magnitude response of the filter is obtained by sweeping the filter input from
87 kHz to 112 kHz.

Fig. 3.21: Frequency response of the filter used to generate NBN

Plot in Fig. 3.20(a) gives us the impression that it is a 101 kHz sinusoid with
slowly changing envelope. This is only partially correct. Waveforms at (b) and (c)
are expanded versions of a part of the waveform at (a). From (b) and (c), it is
clear that zero crossings are not uniform. In Fig. 3.20(c), the cycle of the
waveform shown in green fits almost into the space between two adjacent
vertical lines of the graticule, whereas for the cycle shown in red, there is a clear
offset. Hence the proper time-domain description of a NBBP signal would be: it is
a sinusoid undergoing slow amplitude and phase variations. (This will be justified
later on by expressing the NBBP noise signal in terms of its envelope and
phase.)

We shall now develop two representations for the NBBP noise signal.
These are (a) canonical (also called in-phase and quadrature component)
representation and (b) Envelope and Phase representation.


3.64
3.10.1 Representation of narrowband noise
a) Canonical representation
Let ( ) n t represent a sample function of a NBBP noise process ( ) N t , and
( )
pe
n t and ( )
c e
n t , its pre-envelope and complex envelope respectively. We will
assume
c
f to be the (nominal) centre frequency of the noise process. Then, we
can write:
( ) ( )

( )
pe
n t n t j n t = + and (3.34)
( ) ( ) ( ) exp 2
c e pe c
n t n t j f t = (3.35)
where
( ) n t is the Hilbert transform of ( ) n t .

The complex envelope ( )
c e
n t itself can be expressed as
( ) ( ) ( )
c e c s
n t n t j n t = + (3.36)
With the help of Eq. 3.34 to 3.36, it is easy to show,

( ) ( ) ( )

( ) ( ) cos 2 sin 2
c c c
n t n t f t n t f t = + (3.37)
( )

( ) ( ) ( ) ( ) cos 2 sin 2
s c c
n t n t f t n t f t = (3.38)
As ( ) ( )
2
Re
c
j f t
c e
n t n t e

=

( ) ( )
{ }
2
Re
c
j f t
c s
n t j n t e

= +

We have,
( ) ( ) ( ) ( ) ( ) cos 2 sin 2
c c s c
n t n t f t n t f t = (3.39)
As in the deterministic case, we call ( )
c
n t the in-phase component and ( )
s
n t the
quadrature component of ( ) n t .
As Eq. 3.39 is valid for any sample function ( ) n t of ( ) N t , we shall write
( ) ( ) ( ) ( ) ( ) cos 2 sin 2
c c s c
N t N t f t N t f t = (3.40a)

3.65
Eq. 3.40(a) is referred to as the canonical representation of ( ) N t . ( )
c
N t and
( )
s
N t are low pass random processes; ( )
c
n t and ( )
s
n t are sample functions of
( )
c
N t and ( )
s
N t respectively.

b) Envelope and phase representation
Let us write ( ) n t as
( ) ( ) ( ) cos 2
c
n t r t f t t = +

(3.41a)
( ) ( ) ( ) ( ) ( ) cos cos 2 sin sin 2
c c
r t t f t t f t =

(3.41b)
Comparing Eq. 3.41(b) with Eq. 3.39, we find that
( ) ( ) ( ) cos
c
n t r t t =
( ) ( ) ( ) sin
s
n t r t t =
or ( ) ( ) ( )
1
2 2
2
c s
r t n t n t = +

(3.42a)
and ( )
( )
( )
tan
s
c
n t
t arc
n t

=

(3.42b)
( ) r t is the envelope of ( ) n t and ( ) t its phase. Generalizing, we have
( ) ( ) ( ) ( )
cos 2
c
N t R t f t t

= +

(3.43)
where ( ) R t is the envelope process and ( ) t is the phase process. Eq. 3.43
justifies the statement that the NBN waveform exhibits both amplitude and phase
variations.

Exercise 3.15
Show that
( ) ( ) ( ) ( ) ( ) sin 2 cos 2
c c s c
N t N t f t N t f t = + (3.40b)

3.66
3.10.2 Properties of narrowband noise
We shall now state some of the important properties of NBN. (For proofs,
refer to Appendix A3.1.)
P1) If ( ) N t is zero mean, then so are ( )
c
N t and ( )
s
N t .
P2) If ( ) N t is a Gaussian process, then ( )
c
N t and ( )
s
N t are jointly Gaussian.
P3) If ( ) N t is WSS, then ( )
c
N t and ( )
s
N t are WSS.
P4) Both ( )
c
N t and ( )
s
N t have the same PSD which is related to ( )
N
S f of the
original narrowband noise as follows:
( ) ( )
( ) ( ) ,
0 ,
c s
N c N c
N N
S f f S f f B f B
S f S f
elsewhere
+ +
= =

where it is assumed that ( )
N
S f occupies the frequency interval
c c
f B f f B + and
c
f B > .
P5) If the narrowband noise ( ) N t is zero mean, then ( )
c
N t and ( )
s
N t have
the same variance as the variance of ( ) N t .
P6) The cross-spectral densities of the quadrature components of narrowband
noise are purely imaginary, as shown by
( ) ( )
( ) ( ) ,
0 ,
c s s c
N c N c
N N N N
j S f f S f f B f B
S f S f
elsewhere
+

= =

P7) If ( ) N t is zero-mean Gaussian and its PSD, ( )
N
S f is locally symmetric
about
c
f , then ( )
c
N t and ( )
s
N t are statistically independent.

Property P7) implies that if ( )
N
S f is locally symmetric about
c
f , then
( )
c s
N N
R is zero for all . Since ( )
c
N t and ( )
s
N t are jointly Gaussian, they
become independent. However, it is easy to see that ( ) 0 0
c s
N N
R = whether or
not there is local symmetry. In other words the variables ( )
1 c
N t and ( )
1 s
N t , for
any sampling instant
1
t , are always independent.

3.67
Note: Given the spectrum of an arbitrary band-pass signal, we are free to choose
any frequency
c
f as the (nominal) centre frequency. The spectral shape ( )
c
N
S f
and ( )
s
N
S f will depend on the
c
f chosen. As such, the canonical representation
of a narrowband signal is not unique. For example, for the narrowband spectra
shown in Fig. 3.22, if
1
f is chosen as the representative carrier frequency, then
the noise spectrum is
1
2 B wide, whereas if
2
f (which is actually the mid-
frequency of the given band) is chosen as the representative carrier frequency,
then the width of the spectrum is
2
2 B . Note for the ( )
N
S f of Fig. 3.22, it is not
possible for us to choose an
c
f such that ( )
N
S f exhibits local symmetry with
respect to it.

Fig. 3.22: Narrowband noise spectrum with two different centre frequencies

Example 3.18
For the narrowband noise spectrum ( )
N
S f shown in Fig. 3.23, sketch
( )
c
N
S f for the two cases, namely a) 10
c
f k Hz = and b) 11
c
f k Hz = . c) What
is the variance ( )
c
N t ?


3.68

Fig. 3.23: ( )
N
S f for the example 3.18

a) From P4), we have,
( )
( ) ( ) ,
0 ,
c
N c N c
N
S f f S f f B f B
S f
elsewhere
+ +

Using 10
c
f k Hz = , and plotting ( )
N c
S f f and ( )
N c
S f f + , we obtain Fig.
3.24.

Fig. 3.24: Shifted spectra: (a) ( )
N c
S f f + and (b) ( )
N c
S f f

Taking 2 B k Hz = and extracting the relevant part of ( )
N c
S f f + and
( )
N c
S f f , we have ( )
c
N
S f as shown in Fig. 3.25.

3.69

Fig. 3.25: ( )
c
N
S f with 10
c
f k Hz =

b) By repeating the above procedure, we obtain ( )
c
N
S f (the solid line) shown
in Fig. 3.26.

Fig. 3.26: ( )
c
N
S f with 11
c
f k Hz =

c) ( )
2 2
1.0
c
N N N
S f d f Watt

= = =


3.70

Exercise 3.16
Assuming ( ) N t to be WSS and using Eq. 3.40(a), establish
( ) ( ) ( ) ( ) ( ) =
c s c
N N c N N c
R R f R f cos 2 sin 2 (3.44)

Exercise 3.17
Let ( ) N t represent a narrowband, zero-mean Gaussian process with
( )
0
,
2
0 ,
c c
N
N
f B f f B
S f
otherwise

Let X and Y be two random variables obtained by sampling ( )
c
N t and
( )
c
d N t
d t
at
1
t t = , where ( )
c
N t is the in phase component of ( ) N t .
a) Show that X and Y are independent.
b) Develop the expression for ( )
,
,
X Y
f x y .

Exercise 3.18
Let ( ) N t represent a NBN process with the PSD shown in Fig. 3.27
below.

Fig. 3.27: Figure for exercise 3.18

Let ( ) ( ) ( ) ( ) ( ) cos sin
c c s c
N t N t t N t t = with 50
c
f = kHz.
Show that ( ) ( ) ( )
3 3
2
sin 10 sin 3 10
3
s c
N N
R c = .

3.71

We shall now establish a relation between ( )
N
R and ( )
c e
N
R , where
( )
c e
N
R evidently stands for the ACF of the complex envelope of the narrowband
noise. For complex signals, such as ( )
c e
N t , we define its ACF as
( ) ( ) ( ) ,
c e
N c e c e
R t t E N t N t
+ = +

( ) ( ) ( ) ( ) ( ) ( )
c s c s
E N t j N t N t j N t

= + + +

( ) ( ) ( ) ( )
c s s c c s
N N N N N N
R R j R j R = + +
But ( ) ( )
c s
N N
R R = (Eq. A3.1.7)
and ( ) ( )
s c c s
N N N N
R R = (Property of cross correlation)
( )
c s
N N
R =

The last equality follows from Eq. A3.1.8. Hence,
( ) ( ) ( ) ( ) ( )
, 2
c e ce c s c
N N N N N
R t t R R j R + = = + . From Eq. 3.44, we have
( ) ( )
2
1
Re
2
c
c e
j f
N N
R R e
+

=

( ) ( )
2 2
1
4
c c
c e c e
j f j f
N N
R e R e

= +

(3.45)
Taking the Fourier transform of Eq. 3.45, we obtain
( ) ( ) ( )

= +

1
4
c e c e
N N c N c
S f S f f S f f (3.46)

Note: For complex signals, the ACF is conjugate symmetric; that is, if ( ) X t
represents a complex random process that is WSS, then ( ) ( )
X X
R R
= . The
PSD ( ) ( )
X X
S f F R =

is real, nonnegative but not an even function of
frequency.


3.72
3.10.3 PDF of the envelope of narrowband noise
From Eq. 3.43, we have,
( ) ( ) ( ) ( )
cos 2
c
N t R t f t t

= +

with ( ) ( ) ( ) ( )
cos
c
N t R t t = (3.47a)
and ( ) ( ) ( ) ( )
sin
s
N t R t t = (3.47b)
That is,
( ) ( ) ( )
1
2 2
2
c s
R t N t N t = +

(3.48a)
( )
( )
( )
1
tan
s
c
N t
t
N t

=

(3.48b)

Our interest is the PDF of the random variable ( )
1
R t , for any arbitrary
sampling instant
1
t t = .
Let
c
N and
s
N represent the random variables obtained from sampling
( )
c
N t and ( )
s
N t at any time
1
t t = . Assuming N(t) to be a Gaussian process,
we have
c
N and
s
N as zero mean, independent Gaussian variables with
variance
2
where ( )
2
N
S f d f

=

.
Hence,
( )
2 2
2 2
1
, exp
2 2
c s
c s
N N c s
n n
f n n
+
=

For convenience, let ( )
1
R t R = and ( )
1
t = . Then,
cos
c
N R = and sin
s
N R =
and the J acobian of the transformation is,

= =

1
cos sin
sin cos
J r
r r


3.73
Therefore,
( )
2
2 2
,
exp , 0, 0 2
, 2 2
0 ,
R
r r
r
f r
otherwise

<

=

( ) ( )
2
,
0
,
R R
f r f r d

It is easy to verify that
( )
2
2 2
exp , 0
2
0 ,
R
r r
r
f r
otherwise

=
(3.49)
Similarly, it can be shown that
( )
1
, 0 2
2
0 ,
f
otherwise
<
(3.50)
As ( ) ( ) ( )
,
,
R R
f r f r f

= , we have R and as independent variables.
The PDF given by Eq. 3.49 is the Rayleigh density which was introduced
in Chapter 2. From the above discussion, we have the useful result, namely, the
envelope of narrowband Gaussian noise is Rayleigh distributed.

Let us make a normalized plot of the Rayleigh PDF by defining a new
random variable V as
R
V =
(transformation by a multiplicative constant).

Then,
( )
2
exp , 0
2
0 ,
V
v
v v
f v
otherwise

=

( )
V
f v is plotted in Fig. 3.28.

3.74

Fig. 3.28: Normalized Rayleigh PDF

The peak value of the PDF occurs at 1 v = and ( ) 1 0.607
V
f = .

3.10.4 Sine wave plus narrowband noise
Let a random process ( ) X t be given by
( ) ( ) ( ) cos 2
c
X t A f t N t = + (3.51)
where A and
c
f are constants, ( ) N t represents the narrowband noise process
whose centre frequency is taken as
c
f . Our interest is to develop an expression
for the PDF of the envelope of ( ) X t .

Using the canonical form for ( ) N t , Eq. 3.51 can be written as
( ) ( ) ( ) ( ) ( ) ( ) cos 2 cos 2 sin 2
c c c s c
X t A f t N t f t N t f t = +
Let ( ) ( )
c c
'
N t A N t = + . Assume ( ) N t to be a Gaussian process with zero mean
and variance
2
. Then, for any sampling instant
1
t t = , let
c
'
N denote the
random variable ( )
1 c
'
N t and let
s
N denote the random variable ( )
1 s
N t . From our
earlier discussion, we find that
c
'
N is
( )
2
, N A ,
s
N is
( )
2
0, N and
c
'
N is
independent of
s
N . Hence,

3.75

( )
( )
2
2
2 2
,
1
, exp
2 2
c s
c s
c s
N N
'
'
'
n A n
f n n

+

=

(3.52)
Let ( ) ( ) ( )
c s
'
R t N t N t
1
2
2
2

= +

( )
( )
( )
1
tan
s
c
'
N t
t
N t

=

where ( ) R t and ( ) t are the envelope and phase, respectively of ( ) X t . By a
procedure similar to that used in the computation of the PDF of the envelope of
narrowband noise, we find that

( )
2 2
, 2 2
2 cos
, exp , 0, 0 2
2 2
R
r r A Ar
f r r
+
=

where ( )
1
R R t = and ( )
1
t =
The quantity of interest is ( )
R
f r , where

( ) ( )
2
,
0
,
R R
f r f r d

That is,
( )
2
2 2
2 2 2
0
exp exp cos
2 2
R
r r A Ar
f r d
+
=

(3.53)

The integral on the RHS of Eq. 3.53 is identified in terms of the defining
integral for the modified Bessel function of the first kind and zero order.

Let ( ) ( )
2
0
0
1
exp cos
2
I y y d

(3.54)
A plot of ( )
0
I y is shown in Fig. 3.29. In Eq. 3.54, if we let
2
Ar
y =
, Eq. 3.53 can

be rewritten as

3.76
( )
2 2
0 2 2 2
exp
2
R
r r A Ar
f r I
+
=

(3.55)
The PDF given by Eq. 3.55 is referred to as Rician distribution.

Fig. 3.29: Plot of ( )
0
I y

The graphical presentation of the Rician PDF can be simplified by introducing
two new variables, namely,
R
V =
and
A
=
. Then, the Rician density of Eq.

3.55 can be written in a normalized form,
( ) ( )
2 2
0
exp
2
V
v
f v v I v
+
=

(3.56)
which is plotted in Fig. 3.30 for various values of .

3.77

Fig. 3.30: Normalized Rician PDF for various values of

Based on these curves, we make the following observations:
i) From Fig. 3.28, we find ( )
0
0 1 I = . If 0 A = , then 0 = and
( )
2
exp , 0
2
0 ,
V
v
v v
f v
otherwise

=

which is the normalized Rayleigh PDF shown earlier in Fig. 3.27. This is
justified because if 0 A = , then ( ) R t is the envelope of only the
narrowband noise.
ii) For 1 y >> , ( )
0
2
y
e
I y
y
. Using this approximation, it can be shown that
( )
V
f v is approximately Gaussian in the vicinity of v = , when is
sufficiently large. That is, when the sine-wave amplitude A is large
compared with (which is the square root of the average power in ( ) N t ),
( )
V
f v can be approximated by a Gaussian PDF over a certain range. This
can be seen from Fig. 3.29 for the cases of 3 and 5 = .


3.78
Appendix A3.1
Properties of Narrowband Noise: Some Proofs.
We shall now give proofs to some of the properties of the NBN, mentioned
in sec.3.10.2.
P1) If ( ) N t is zero mean Gaussian, then so are ( )
c
N t and ( )
s
N t .

Proof: Generalizing Eq. 3.37 and 3.38, we have
( ) ( ) ( )

( ) ( ) cos 2 sin 2
c c c
N t N t f t N t f t = + (A3.1.1)
( )

( ) ( ) ( ) ( ) cos 2 sin 2
s c c
N t N t f t N t f t = (A3.1.2)
( ) N t is obtained as the output of an LTI system with ( )

1
h t
t
=
, with the
input ( ) N t . This implies that if ( ) N t is zero mean, then so is
( ) N t . Taking
the expectation of A3.1.1, we have
( ) ( ) ( )

( ) ( ) cos 2 sin 2
c c c
N t N t f t N t f t = +
As ( )

( ) 0 N t N t = = , we obtain ( ) 0
c
N t = . Taking the expectation of
A3.1.2, we obtain ( ) 0
s
N t = .

P2) If ( ) N t is a Gaussian process, then ( )
c
N t and ( )
s

Proof: This property follows from Eq. 3.40(a) because, ( ) N t is guaranteed
to be Gaussian only if ( )
c
N t and ( )
s

P3) If ( ) N t is WSS, then ( )
c
N t and ( )
s
N t are WSS.

Proof: We will first establish that
( ) ( ) ( ) ( )

( ) ( ) , cos 2 sin 2
c c
N
N N N c c
R t t R R f R f + = = + (A3.1.3)
(In Eq. A3.1.3,
( ) N R is the Hilbert transform of ( )

N
R .)

3.79
Consider the scheme shown in Fig A3.1.1.

Fig. A3.1.1: Scheme to obtain
( ) N t from ( ) N t

Eq. 3.21(a) gives us
( ) ( )
1
N
NN
R R =
(A3.1.4a)

( ) N R = (A3.1.4b)
( )

( )
NN NN
R R = ,
( )
( )
1
N
R =

( )
1
N
R =
( ) N R =
That is,
( )

( )
NN NN
R R = (A3.1.5)
( ) ( ) ( ) ,
c
N c c
R t t E N t N t + = +

Expressing ( )
c
N t + and ( )
c
N t in terms of the RHS quantities of Eq.
A3.1.1 and after some routine manipulations, we will have

( ) ( )

( ) ( )
( )

( ) ( ) ( )
( )

( ) ( )
( )

( ) ( ) ( )
1
, cos 2
2
1
cos 2 2
2
1
sin 2
2
1
sin 2 2
2
c
N N c
N
N c
N
c
NN NN
c
NN NN
R t t R R f
R R f t
R R f
R R f t
+ = +

+ +

+

+ + +

(A3.1.6)

3.80
( )
N
R is the autocorrelation of the Hilbert transform of ( ) N t . As ( ) N t and
( ) N t have the same PSD, we have ( )

( )
N
N
R R = and from Eq. A3.1.5,
we have
( )

( )
NN NN
R R T = . Using these results in Eq. A3.1.6, we obtain
Eq. A3.1.3.

P4) Both ( )
c
N t and ( )
s
N t have the same PSD which is related to ( )
N
S f of the
original narrowband noise as follows:
( ) ( )
( ) ( ) ,
0 ,
c s
N c N c
N N
S f f S f f B f B
S f S f
elsewhere
+ +
= =

where it is assumed that ( )
N
S f occupies the frequency interval
c c
f B f f B + and
c
f B > .

Proof: Taking the Fourier transform of Eq. A3.1.3, we have
( ) ( ) ( )
( ) ( ) ( ) ( )
1
2
1
sgn sgn
2
c
N N c N c
N c c N c c
S f S f f S f f
S f f f f S f f f f
= + +

+ +

( ) ( ) ( ) ( )
1 1
1 sgn 1 sgn
2 2
N c c N c c
S f f f f S f f f f = + + + + +

(A3.1.9)
But ( )
2,
1 sgn
0,
c
c
f f
f f
outside
<
=

(A3.1.10a)
( )
2,
1 sgn
0,
c
c
f f
f f
outside
>
+ + =

(A3.1.10b)
Exercise A3.1.1
Show that ( ) ( )
s c
N N
R t R = (A3.1.7)
and ( ) ( ) ( )

( ) ( ) sin 2 cos 2
c s
N
N N N c c
R R f R f = (A3.1.8)

3.81
Using Eq. A3.1.10 in Eq. A3.1.9 completes the proof.

Exercise A3.1.2
Establish properties P5 to P7.

3.82
References
1) A. Papoulis, Probability, Random variables and Stochastic processes (3
rd

ed), McGraw Hill, 1991
2) Henry Stark and J ohn W. Woods, Probability and Random Processes with
Applications to Signal processing (3
rd
ed), Pearson Education Asia, 2002
3) K. Sam Shanmugam and A. M. Breiphol, Random Signals: Detection,
Estimation and data analysis, J ohn Wiley, 1988

4.1

4 CHAPTER 4

Linear Modulation

4.1 Introduction
We use the word modulation to mean the systematic alteration of one
waveform, called the carrier, according to the characteristic of another waveform,
the modulating signal or the message. In Continuous Wave (CW) modulation
schemes, the carrier is a sinusoid. We use ( ) c t and ( ), m t to denote the carrier
and the message waveforms respectively.

The three parameters of a sinusoidal carrier that can be varied are:
amplitude, phase and frequency. A given modulation scheme can result in the
variation of one or more of these parameters. Before we look into the details of
various linear modulation schemes, let us understand the need for modulation.
Three basic blocks in any communication system are: 1) transmitter 2) Channel
and 3) Receiver (Fig. 4.1).

Fig. 4.1: A basic communication system

The transmitter puts the information from the source (meant for the
receiver) onto the channel. The channel is the medium connecting the transmitter
and the receiver and the transmitted information travels on this channel until it
reaches the destination. Channels can be of two types: i) wired channels or ii)
wireless channels. Examples of the first type include: twisted pair telephone

4.2

channels, coaxial cables, fiber optic cable etc. Under the wireless category, we
have the following examples: earths atmosphere (enabling the propagation of
ground wave and sky wave), satellite channel, sea water etc.

The main disadvantage of wired channels is that they require a man-made
medium to be present between the transmitter and the receiver. Though wired
channels have been put to extensive use, wireless channels are equally (if not
more) important and have found a large number of applications.

In order to make use of the wireless channels, the information is to be
converted into a suitable form, say electromagnetic waves. This is accomplished
with the help of a transmitting antenna. The antenna at the receiver (called the
receiving antenna) converts the received electromagnetic energy to an electrical
signal which is processed by the receiver.

The question is: can we radiate the baseband
1
information bearing signal
directly on to the channel?

For efficient radiation, the size of the antenna should be 10 or more
(preferably around 4 ), where is the wavelength of the signal to be radiated.
Take the case of audio, which has spectral components almost from DC upto 20
kHz. Assume that we are designing the antenna for the mid frequency; that is,10
kHz. Then the length of the antenna that is required, even for the 10 situation
is,
8
3
4
3 10
3 10
10 10 10
c
f

= =

meters, c being the velocity of light.

1
Baseband signals have significant spectral content around DC. Some of the baseband signals
that are of interest to us are: a) Speech b) music and c) video (TV signals).
Approximate spectral widths of these signals are: Speech: 5 kHz, Audio : 20 kHz, Video : 5 MHz

4.3

Even an antenna of the size of 3 km, will not be able to take care of the entire
spectrum of the signal because for the frequency components around 1 kHz, the
length of the antenna would be 100 . Hence, what is required from the point of
view of efficient radiation is the conversion of the baseband signal into a
narrowband, bandpass signal. Modulation process helps us to accomplish this;
besides, modulation gives rise to some other features which can be exploited for
the purpose of efficient communication. We describe below the advantages of
modulation.

1. Modulation for ease of radiation
Consider again transmission of good quality audio. Assume we choose the
carrier frequency to be 1 MHz. The linear modulation schemes that would be
discussed shortly give rise to a maximum frequency spread (of the modulated
signal) of 40 kHz, the spectrum of the modulated signal extending from (1000 -
20) = 980 kHz to (1000 + 20) = 1020 kHz. If the antenna is designed for 1000
kHz, it can easily take care of the entire range of frequencies involved because
modulation process has rendered the signal into a NBBP signal.

2. Modulation for efficient transmission
Quite a few wireless channels have their own appropriate passbands. For
efficient transmission, it would be necessary to shift the message spectrum into
the passband of the channel intended. Ground wave propagation (from the lower
atmosphere) is possible only up to about 2 MHz. Long distance ionospheric
propagation is possible for frequencies in the range 2 to 30 MHz. Beyond 30
MHz, the propagation is line of sight. Preferred frequencies for satellite
communication are around 3 to 6 GHz. By choosing an appropriate carrier
frequency and modulation technique, it is possible for us to translate the
baseband message spectrum into a suitable slot in the passband of the channel
intended. That is, modulation results in frequency translation.


4.4

3. Modulation for multiplexing
Several message signals can be transmitted on a given channel, by assigning to
each message signal an appropriate slot in the passband of the channel. Take
the example of AM broadcast, used for voice and medium quality music
broadcast. The passband of the channel used to 550 kHz to 1650 kHz. That is,
the width of the passband of the channel that is being used is 1100 kHz. If the
required transmission bandwidth is taken as 10 kHz, then it is possible for us to
multiplex, atleast theoretically, 110 distinct message signals on the channel and
still be able to separate them individually as and when we desire because the
identity of each message is preserved in the frequency domain.

4. Modulation for frequency assignment
Continuing on the broadcast situation, let us assume that each one of the
message signals is being broadcast by a different station. Each station can be
assigned a suitable carrier so that the corresponding program material can be
received by tuning to the station desired.

5. Modulation to improve the signal-to-noise ratio
Certain modulation schemes (notably frequency modulation and phase
modulation) have the feature that they will permit improved signal-to-noise ratio
at the receiver output, provided we are willing to pay the price in terms of
increased transmission bandwidth (Note that the transmitted power need not be
increased). This feature can be taken advantage of when the quality of the
receiver output is very important.

Having understood the need and the potential benefits due to modulation,
let us now get into the details of various linear modulation schemes. The four
important types of linear modulation schemes are
1) Double SideBand, Suppressed Carrier (DSB-SC)
2) Double SideBand, Large Carrier (DSB-LC) (also called conventional AM or
simply AM)

4.5

3) Single SideBand (SSB)
4) Vestigial SideBand (VSB)
We shall begin our discussion with DSB-SC.

4.2 DSB-SC Modulation
4.2.1. Modulation
The DSB-SC is the simplest of the four linear modulation schemes listed
above (simplest in terms of the mathematical description of modulation and
demodulation operations). Consider the scheme shown in Fig. 4.2

Fig. 4.2: DSB-SC modulation scheme

( ) m t is a baseband message signal with = M f ( ) 0 for ( ) , f W c t > is a high
frequency carrier, usually with >>
c
f W .

DSB-SC modulator is basically a multiplier. Let
m
g denotes the amplitude
sensitivity (or gain constant) of the modulator, with the units per volt (we assume
that ( ) m t and
c
A are in volts). Then the modulator output ( ) s t is,
( ) ( ) ( ) ( )
cos
m c c
s t g m t A t = (4.1a)
For convenience, let 1
m
g = . Then,
( ) ( ) ( ) cos
c c
s t A m t t = (4.1b)


4.6

As DSB-SC modulation involves just the multiplication of the message
signal and the carrier, this scheme is also known as product modulation and can
be shown as in Fig. 4.3.

Fig. 4.3: Product Modulation scheme

The time domain behavior of the DSB-SC signal (with 1
c
A = ) is shown in
Fig. 4.4(b), for the ( ) m t shown in Fig. 4.4(a).


4.7

Fig. 4.4: (a) The message signal
(b) The DSB-SC signal

Note that the carrier undergoes a 180

phase reversal at the zero crossings of
( ) m t . This is brought out more clearly in the oscillograms, shown in Fig. 4.5 and
Fig. 4.6, where ( ) m t is a sinusoidal signal.

With reference to Fig. 4.5, between the points a and b, the carrier in the
DSB-SC signal and the actual carrier (bottom picture) are in phase whereas
between the points b and c, they are 180
0
out of phase. Fig. 4.6 is an
expanded version of the central part of the waveforms in Fig. 4.5. Here, we can

4.8

very clearly observe that to the left of b, both the carriers are in phase whereas
to the right, they are 180
0
out of phase.

Fig. 4.5: (top) DSB-SC signal with tone modulation
(bottom) The carrier

Fig. 4.6: Expanded versions of a part of the waveforms in Fig. 4.5

Consider waveforms shown in Fig. 4.7. We have on the top, modulating tone
signal and at the bottom, the corresponding DSB-SC. What do we observe on
the oscilloscope, if we feed the X-plates the tone signal and the Y-plates, the
DSB-SC signal? The result is shown in Fig. 4.8, which can be explained as
follows.


4.9

At the point a in Fig. 4.7, the modulating tone is at its maximum and
hence the DSB-SC signal has the maximum value. Point a in Fig. 4.8
corresponds to the point a in Fig. 4.7. Between the points a and b in Fig. 4.7,
the tone amplitude decreases (reaching the value zero at point b); hence the
maximum value reached by the DSB-SC signal during each carrier cycle keeps
decreasing. As the X-plates are being fed with the same tone signal, this
decrease will be linear and this corresponds to segment a to b in Fig. 4.8.
(Note that DSB-SC signal is zero at point b). In the time interval between b and
c of Fig. 4.7, the DSB signal increases and this increase is seen as a straight
line between the points b and c in Fig. 4.8. Between the points c and e in Fig.
4.7, the tone amplitude varies from the most negative value to the most positive
value. Correspondingly, the display on the oscilloscope will follow the trace
c d e shown in Fig. 4.8.

Fig. 4.7: (top) modulating signal
(bottom) DSB-SC signal


4.10

Fig. 4.8 Display on the oscilloscope with the following inputs:
X-plates: Tone signal
Y-plates: DSB-SC signal

Taking the Fourier transform of Eq. 4.1(b), we have
( ) ( ) ( )
2
c
c c
A
S f M f f M f f = + +

(4.2)
If we ignore the constant
2
c
A
on the R.H.S of Eq. (4.2), we see that the
modulation process has simply shifted the message spectrum by
c
f . As the
frequency translation of a given spectrum occurs quite often in the study of
modulation and demodulation operations, let us take a closer look at this.
i) Let ( ) m t be a real signal with the spectrum ( ) M f shown below (Fig.
4.9(a)). Let
c
f be 100 kHz. Assuming = 1
2
c
A
, we have ( ) S f as shown in
Fig. 4.9(b).

4.11

Fig. 4.9: Frequency translation (a) baseband spectrum (real signal)
(b) Shifted spectrum.

Note that ( ) ( ) ( )
102 kHz
2 kHz 202 kHz
f
S f M M
=
= +
1 0 = + 1 =
and is the point a in Fig. 4.9
ii) Let ( ) m t be a complex signal with ( ) M f as shown in Fig. 4.10(a). The
corresponding shifted spectrum (with 100
c
f = kHz) is shown in Fig.
4.10(b)


4.12

Fig. 4.10: Frequency translation (a) Baseband spectrum (complex signal)
(b) Shifted spectrum.

In figures 4.9(b) and 4.10(b), the part that is hatched in red is called the
Upper Sideband (USB) and the one hatched in blue is called the Lower Sideband
(LSB). Any one of these two sidebands has the complete information about the
message signal. As we shall see later, SSB modulation conserves the bandwidth
by transmitting only one sideband and recovering the ( ) m t with appropriate
demodulation.

Example 4.1
Consider the scheme shown in Fig. 4.11(a). The ideal HPF has the cutoff
frequency at 10 kHz. Given that f
1
10 = kHz and f
2
15 = kHz, let us sketch
( ) Y f for the ( ) X f given at (b).


4.13

Fig. 4.11: (a) The scheme of example 4.1
(b) The input spectrum, ( ) X f

We have ( ) ( ) ( ) V f X f f X f f
1 1
= + + , which is as shown in Fig. 4.12(a).
The HPF eliminates the spectral components for f 10 kHz. Hence ( ) W f is
as shown in Fig. 4.12(b).
( ) ( ) ( ) Y f W f f W f f
2 2
= + + . This is shown in Fig. 4.12(c).


4.14

Fig. 4.12: Spectra at various points in the scheme of Fig. 4.11

4.2.2. Coherent demodulation
The process of demodulation of a DSB-SC signal, at least theoretically, is
quite simple. Let us assume that the transmitted signal ( ) s t has been received
without any kind of distortion and is one of the inputs to the demodulator as
shown in Fig. 4.13. That is, the received signal ( ) ( ) r t s t = . Also, let us assume
that we are able to generate at the receiving end a replica of the transmitted
carrier (denoted ( ) ( )
'
cos
r c c
c t A t = in Fig. 4.13) which is the other input to the
demodulator.

4.15

Fig. 4.13: Coherent demodulation of DSB-SC

The demodulation process consists of multiplying these two inputs and lowpass
filtering the product quantity ( ) v t .
From Fig. 4.13, we have
( ) ( ) ( ) ( ) ( )
( )
'
cos cos
g c c c c
v t d A m t t A t =
where
g
d is the gain constant of the multiplier, called the detector gain constant,
in the context of demodulation. For convenience, let us take = 1
g
d
( ) ( ) ( )
2 '
cos
c c c
v t A A m t t =
( )
( ) 1 cos 2
'
2
c
c c
t
A A m t
+

=
Assuming that =
'
2
c c
A A we have
( ) ( ) ( ) ( ) cos 4
c
v t m t m t f t = + (4.3)
The second term on the R.H.S of Eq. 4.3 has the spectrum centered at 2
c
f
and would be eliminated by the lowpass filter following ( ) v t . Hence ( )
0
v t , the
output of the demodulation scheme of Fig. 4.13 is the desired quantity, namely,
( ) m t .

Let us illustrate the operation of the detector in the frequency domain. Let
( ) m t be real with the spectrum shown in Fig. 4.14(a). Let

4.16

( ) ( ) ( ) ( ) 2 cos
c
r t s t m t t = = . Then ( ) ( ) ( )
c c
S f M f f M f f = + + , shown in
Fig. 4.14(b).

Fig. 4.14: Spectra at various points in the demodulation scheme of Fig. 4.13

(Note that the positive frequency part of ( ) S f is shown in red and the negative
frequency part in blue). Assuming ( ) ( ) ( ) cos
c
v t s t t = (Fig. 4.13 with
c
A
'
1 = ),
then ( ) ( ) ( )
1
2
c c
V f S f f S f f = + +

. ( )
1
2
c
S f f and ( )
1
2
c
S f f + are shown in

4.17

Fig. 4.14(c) and (d) respectively. ( )
0
V f is the sum of the outputs of the lowpass
filters shown in Fig. 4.14(c) and (d) which is the desired message spectrum,
( ) M f .

From the discussion of the demodulation process so far, it appears that
demodulation of DSB-SC is quite simple. In practice, it is not. In the scheme of
Fig. 4.13, we have assumed that, we have available at the receiver, a carrier
term that is coherent (of the same frequency and phase) with the carrier used to
generate the DSB-SC signal at the transmitter. Hence this demodulation scheme
is known as coherent (or synchronous) demodulation. As the receiver and the
transmitter are, in general, not collocated, the carrier source at the receiver is
different from that used at the transmitter and it is almost impossible to
synchronize two independent sources. Fairly sophisticated circuitry has to be
used at the receiver in order to generate the coherent carrier signal, from an
( ) r t that has no carrier component in it. Before we discuss at the generation of
the coherent carrier at the receiver, let us look at the degradation caused to the
demodulated message due to a local carrier that has phase and frequency
differences with the transmitted one.

Case i): Constant phase difference between ( ) c t and ( )
r
c t
Let ( ) ( ) cos 2
c
c t f t = and ( ) [ ]
cos 2
r c
c t f t = + (the amplitude quantities,
c
A and
'
c
A can be treated as 1)
( ) ( ) ( ) ( ) cos cos
c c
v t m t t t = +
( ) ( ) ( ) ( ) cos cos cos sin sin
c c c
m t t t t =

( ) ( ) ( ) ( )
2
cos cos sin cos sin
c c c
m t t t t

=

( )
( ) ( ) ( ) 1 cos 2 sin 2
cos sin
2 2
c c
t m t t
m t
+
=


4.18

At the output of the LPF, we will have only the term
( ) cos
2
m t
. That is, the
output of the demodulator, ( )
0
v t , is proportional to ( ) cos m t . As long as
remains a constant, the demodulator output is a scaled version of the actual
message signal. But values of close to 2 will force the output to near about
zero. When 2 = we have zero output from the demodulator. This is called
the quadrature null effect of the coherent detector.

Case ii): Constant frequency difference between ( ) c t and ( )
r
c t
Let ( ) ( ) cos 2
c
c t f t = and ( ) ( ) cos 2
r c
c t f f t = +

. Then,
( ) ( ) ( ) ( ) cos 2 cos 2
c c
v t m t f t f f t = +

By carrying out the analysis similar to case (i) above, we find that
( ) ( ) ( )
0
cos 2 v t m t f t

(4.4a)

Let us look in some detail the implications of Eq. 4.4(a). For convenience,
let ( ) ( ) ( )
0
cos 2 v t m t f t =

(4.4b)
Assume 100 f = Hz and consider the spectral component at 1 kHz in ( ) M f .
After demodulation, this gives rise to two spectral components, one at 900 Hz
and the other at 1100 Hz, because

( )
( ) ( ) ( )
3
1
cos 2 10 cos 2 100 cos 2 1100 cos 2 900
2
t t t t

= +

The behavior of the sum of these two components is shown in Fig. 4.15.
As can be seen from this figure, the envelope of sum signal (broken red line)
attains the peak value twice in a cycle of the beat frequency f . Also, it goes
through zero twice in a cycle of the beat frequency.


4.19

Fig. 4.15: Time-domain behavior of
( )
( )
3
cos 2 10 cos 2 100 t t

Let us examine the effect of frequency offset in the frequency domain. Let ( ) M f
be as shown in Fig. 4.16(a). Assume 300 f = Hz. Then,
( ) ( ) ( )
0
1
2
V f M f f M f f = + +

will be as shown in Fig. 4.16(d), which is
one-half the sum of the spectra shown at (b) and (c). Comparing Fig. 4.16(a) and
(d), we are tempted to surmise that the output of the demodulator is a fairly
distorted version of the actual message signal. A qualitative feeling for this
distortion can be obtained by listening to the speech files that follow.

Introduction
Output 1
Output 2
Output 3
Output 4


4.20

Fig. 4.16: The effect of frequency offset in the demodulation of DSB-SC:
(a) Typical message spectrum, ( ) M f
(b) ( ) 300 M f +
(c) ( ) 300 M f
(d) ( ) ( ) M f M f
1
300 300
2
+ +


4.21

Example 4.2
In this example, we will show that the DSB-SC signal can be demodulated
with the help of any periodic function ( ) p t , as long as ( ) p t has a spectral
component at
c
f , the carrier frequency. This component could be due to the
fundamental or some harmonic of the fundamental.
a) Let ( ) ( ) ( )
c c
s t A m t t cos = . Consider the product ( ) ( )
p
s t x t where ( )
p
x t
is any periodic signal with the period
c
T
f
0
1
= . That is,
( )
c
j nf t
p n
n
x t x e
2
=
=

where
n
x is the n
th
Fourier coefficient. We will show that if x
1
0 , then it is
possible to extract ( ) m t from the product ( ) ( )
p
s t x t .
b) Let ( )
p
y t be another periodic signal with the period T NT
0 0
'
= . We will
show that, appropriate filtering of the product ( ) ( )
p
s t y t , will result in ( ) m t .

a) As ( )

= +

2 2
1
cos 2
2
c c
j f t j f t
c
f t e e , we have
( ) ( )
( )
( ) ( )
c c
j n t j n t c
p n n
n n
A m t
s t x t x e x e
1 1
2
+

= +

( )
( ) ( )
c c
j n t j n t c
n n
n n
n n
A m t
x x e x x e
1 1
1 1
, ,
1 1
2
+

= + + +

as
[ ]
x x
x
1 1
1
Re
2
+
= , the output, after lowpass filtering would be,

[ ] ( )
c
x A m t
1
Re . (We assume that the LPF will reject all the other
spectral components)
b) The product ( ) ( )
p
s t y t can be written as

4.22

( ) ( )
( )
c c
n n
j f t j f t
c N N
p n n
n n
A m t
s t y t y e y e
2 1 2 1
2

+

= +

( )
c c
n n
j f t j f t
c N N
N n N n
n n
n N n N
A m t
y y e y y e
2 1 2 1
, ,
2

+

= + + +

We assume that
N
y 0 . Then, the output of the LPF would be
[ ] ( )
N c
y A m t Re . (Note that ( ) ( )
p
y t s t has spectral lobes at
c c
f f
N N
2
0, , , etc. We assume that the LPF will extract the lobe at f 0 =
and reject others).

Example 4.3
Consider the scheme shown in Fig. 4.17. ( ) s t is the DSB-SC signal
( ) ( )
c
m t t cos with
( )
f
S f
outside
1 , 99 kHz 101kHz
0 ,

Let ( ) g t be another bandpass signal with
( )
f
G f
outside
1 , 98 kHz 102 kHz
0 ,

a) We will show that the output ( ) ( ) y t m t .
b) We will show that it would not be possible to recover ( ) m t from ( ) v t if
( )
<
1 , 98.5 kHz 101.5 kHz

0 ,
f
G f
outside


4.23

Fig. 4.17: Scheme of DSB-SC demodulation (example 4.3)

a) Let ( ) ( ) ( )
c
g t m t f t
1
cos 2 = where
c
f 100 kHz = and
( )
f
M f
outside
1
2 , 2 kHz
0 ,

From ( ) S f , we see that
( )
f
M f
outside
2 , 1kHz
0 ,

We have,
( ) ( ) ( ) ( )
c
v t m t m t t
2
1
cos =
( ) ( )
( )
c
t
m t m t
1
1 cos 2
2
+
=

( ) ( ) ( ) ( )
( )
c
m t m t m t m t
t
1 1
cos 2
2 2
= +
We will assume that the LPF rejects the spectrum around
c
f 2 ,

( ) ( ) ( ) ( ) m t m t M f M f
1 1
2 2

( ) ( ) M f M f
1
will have a flat spectrum for f 1kHz . By using an ILPF
with cutoff at 1 kHz, we can recover ( ) m t from ( ) v t .
b) For this case ( ) M f
1
would be
( )
f
M f
outside
1
2 , 1.5 kHz
0 ,

( ) ( ) M f M f
1
will be flat only for f 0.5 kHz . Hence ( ) m t cannot be
recovered.


4.24

Exercise 4.1
A signal ( ) m t whose spectrum is shown in Fig. 4.18(a) is generated
using the signals ( ) m t
1
and ( ) m t
2
. ( ) M f
1
and ( ) M f
2
are shown at (b) and
(c) respectively in Fig. 4.18. The signal ( ) ( )
( )
s t m t t
5
2 cos 10 = is
transmitted on the channel.
a) Suggest a scheme to obtain ( ) m t from ( ) m t
1
and ( ) m t
2
.
b) ( ) m t
1
and ( ) m t
2
are to be recovered from the received signal
( ) ( ) r t s t = . A part of this receiver is shown in Fig. 4.18(d). Complete
the receiver structure by indicating the operations to be performed by the
boxes with the question mark inside.

(a)

(b) (c)

(d)
Fig. 4.18: Proposed receiver structure for the exercise 4.1

4.25

4.2.3 Carrier recovery for coherent demodulation
As explained in detail in sec. 4.2.2, coherent demodulation requires a
carrier at the receiving end that is phase coherent with the transmitted carrier.
Had there been a carrier component in the transmitted signal, it would have been
possible to extract it at the receiving end and use it for demodulation. But the
DSB-SC signal has no such component and other methods have to be devised to
generate a coherent carrier at the receiver. Two methods are in common use for
the carrier recovery (and hence demodulation) from the suppressed carrier
modulation schemes, namely (a) Costas loop and (b) squaring loop.

a) Costas loop: This scheme is shown in Fig. 4.19.

Fig. 4.19: Costas loop

The VCO (Voltage Controlled Oscillator) is a source that produces a periodic
waveform
1
whose frequency is controlled by the input voltage ( )
c
e t . The output
frequency
o
f of the VCO when ( ) 0
c
e t = is called the free running frequency of
the VCO. The frequency put out by the VCO at any instant depends on the sign
and magnitude of the control voltage, ( )
c
e t .

1
Here, we shall assume that the VCO output is sinusoidal.

4.26

To understand the loop operation, let us assume that the frequency and
phase of the VCO output are the same as that of the incoming carrier. Then,
( ) ( ) ( ) ( ) ( ) ( )
1 0
cos cos
c c c
v t A m t t A t =
( ) ( )
2
0
cos
c c
A A m t t =
( )
( )
0
1 2cos
2
c
c
t
A A m t
+

=
The output of the LPF1 is
( )
( )
0
2
2
c
A A m t
v t = ; that is, ( ) ( )
2
v t m t , the desired signal.
Similar analysis shows
( )
4
0 v t =
Now suppose that VCO develops a small phase offset of radians. The I-
channel output will remain essentially unchanged but a small voltage will develop
at the output of the Q-channel which will be proportional to sin (If the phase
shift is rad, then the Q channel output is proportional to sin ). Because of
this, ( ) e t is a non-zero quantity given by
( ) ( ) ( ) ( )
2
2 4 0
1
cos sin
4
c
e t v t v t A A m t = =

( )
2
0
1
sin2
8
c
A A m t =

( ) e t is input to LPF3, which has very narrow passband (Note that LPF1 and
LPF2 should have a bandwidth of at least W Hz). Hence ( )
0
sin2
c
e t C =
where
0
C is the DC value of ( )
2
0
1
8
c
A A m t

. This DC control voltage ensures
that the VCO output is coherent with the carrier used for modulation.

b) Squaring loop
The operation of the squaring loop can be explained with the help of Fig.
4.20.

4.27

Fig. 4.20: Demodulation of DSB-SC using a squaring loop

Let ( ) ( ) ( ) ( ) cos
c c
r t s t A m t t = = . Then,
( ) ( ) ( ) ( )
2
2 2
1 cos 2
2
c
c
A
v t r t m t t = = +

( )
2
m t will have nonzero DC value which implies its spectrum has an impulse at
= 0 f . Because of this, ( ) V f will have a discrete spectral component at 2
c
f .
( ) v t is the input to a very narrowband bandpass filter, with the centre frequency
2
c
f . By making the bandwidth of LPF1 very narrow, it is possible to make the
VCO to lock on to the discrete component at 2
c
f , present in ( ) w t . (The dotted
box enclosing a multiplier, LPF and a VCO, connected in the feedback
configuration shown is called the Phase Locked Loop (PLL)). The VCO output
goes through a factor of two frequency divider, yielding a coherent carrier at its
output. This carrier is used to demodulate the DSB-SC signal. Note that LPF2
must have adequate bandwidth to pass the highest frequency component
present in ( ) m t .

Both the Costas loop and squaring loop have one disadvantage, namely,
an 180
0
phase ambiguity. Consider the Costas loop; if the input to the loop were

4.28

to be ( ) ( ) cos
c c
A m t t , output of the LPF1 would be ( )
0
1
cos
2
c
A A m t and
that of LPF2 would be ( )
0
1
sin
2
c
A A m t with the result that, ( ) e t would be the
same as in the case discussed earlier. Similarly, for the squaring loop, ( ) v t would
be the same whether ( ) ( ) ( ) cos
c c
r t A m t t = or ( ) ( ) cos
c c
A m t t . Hence
the demodulated output could be either ( ) m t or ( ) m t . However, this will not
cause any problem for audio transmission because ( ) m t and ( ) m t , would
sound the same to our ears.

Though DSB-SC modulation schemes place the entire transmitted power
into the useful sidebands, the demodulation has to be coherent. The circuit
required to generate a coherent carrier increases the cost of the receiver. If only
a few receivers are to be built for a specific communication need, the cost may
not be a major factor. But in a broadcast situation, there would be a large number
of receivers tuned to a given station and in that scenario, it is better make the
receiver fairly cheap and push the cost up of the transmitter, if required. As will
seen later, the Envelope Detector(ED) is fairly cheap to implement as compared
to a coherent detector. But to make use of ED, the modulated carrier should
carry ( ) m t in its envelope. Evidently, DSB-SC does not satisfy this property as
explained below.
Let ( ) ( ) ( ) cos
c c
s t A m t t =
Pre-envelope of ( ) ( ) ( )
c
i t
c
pe
s t s t A m t e

= =

Complex envelope of ( ) ( ) ( )
c
ce
s t s t A m t

= =

Hence the envelope of DSB-SC ( ) ( )
pe ce
s t s t = =
( ) m t

4.29

We shall now describe a modulation scheme that has ( ) m t as its envelope
which can easily be extracted.

4.3 DSB-LC Modulation (or AM)
By adding a large carrier component to the DSB-SC signal, we will have
DSB-LC, which, for convenience, we shall call simply as AM. By choosing the
carrier component properly, it is possible for us to generate the AM signal such
that it preserves ( ) m t in its envelope. Consider the scheme shown in Fig. 4.21.

Fig. 4.21: Generation of an AM signal from a DSB-SC signal

Let ( ) ( ) ( ) cos
c m c
v t A g m t t = . Then,
( ) ( ) ( ) cos
c c
AM
s t A t v t = +

( ) ( ) 1 cos
c m c
A g m t t = +

(4.5)
In this section, unless there is confusion, we use ( ) s t in place of
[ ]
( )
AM
s t . We
shall assume that ( ) m t has no DC component and ( ) ( ) ( ) ( )
max min
m t m t = . Let
( )
m
g m t be such that ( ) 1
m
g m t for all t. Then ( ) 1 0
m
g m t +

and
( )
AM
s t

preserves ( ) m t in its envelope because
( ) ( ) 1
c
j t
c m
pe
s t A g m t e

= +

( ) ( ) 1
c m
ce
s t A g m t = +


4.30

As ( ) 1 0
m
g m t +

, we have
Envelope of
[ ]
= = + ( ) ( ) 1 ( )
ce c m
s t s t A g m t
The quantity after the DC block is proportional to ( ) m t .
If ( ) 1
m
g m t +

is not nonnegative for all t , then the envelope would be different
from ( ) m t . This would be illustrated later with a few time domain waveforms of
the AM signal. Fig. 4.22(b) illustrates the AM waveform for the case
( ) 1 0
m
g m t +

for all t .

Fig. 4.22: (a) An arbitrary message waveform ( ) m t
(b) Corresponding AM waveform


4.31

A few time instants have been marked in both the figures (a) and (b). At
the time instants when ( ) 0 m t = , carrier level would be
c
A which we have
assumed to be 1. Maximum value of the envelope (shown in red broken line)
occurs when ( ) m t has the maximum positive value. Similarly, the envelope will
reach its minimum value when ( ) m t is the most negative. As can be seen from
the figure, the envelope of ( ) s t follows ( ) m t in a one-to-one fashion.
Let ( )
max
1
m
g m t x = . Then ( ) s t is said to have (100x) percentage
modulation. For the case of 100% modulation,
( ) ( )
max min
1
m m
g m t g m t = =

. If ( )
max
1
m
g m t > , then we have over
modulation which results in the envelope distortion. This will be illustrated in the
context of tone modulation, discussed next.

Exercise 4.2
For the waveform ( ) m t shown in Fig. 4.23, sketch the AM signal with
the percentage modulation of 40. Assume
c
A 1 = (the figure has to be shown
with reference to ( ) m t )

Fig. 4.23: Baseband signal for the exercise 4.2

4.32

4.3.1. Tone Modulation
Let ( ) m t to be a tone signal; that is, ( ) ( ) cos 2
m m
m t A f t = where
m c
f f << . Then Eq. 4.5 becomes
( ) ( ) ( ) 1 cos cos
c m m m c
s t A g A t t = +

(4.6a)
( ) ( ) cos
c
A t t = (4.6b)
Let
m m
g A = . Then for tone modulation,
( ) ( ) ( ) 1 cos cos
c m c
s t A t t = +

(4.7)
is called the modulation index or modulation factor. 100 is the
percentage modulation. To avoid envelope distortion, we require, 1.
As ( ) ( ) 1 cos
c m
A t A t = +

, we have
( ) [ ]
max
1
c
A t A = +

( ) [ ]
min
1
c
A t A =

( )
( )
max
min
1
1
A t
A t

+

=

or
( ) ( )
( ) ( )
max min
max min
A t A t
A t A t

=
+

Fig. 4.24 to 4.26 illustrate the experimentally generated AM waveforms

for
= 0.5, 1and 1.5 respectively (with 1 > , we have overmodulation).


4.33

Fig. 4.24: AM with tone modulation ( 0.5 = )

Fig. 4.25: AM with tone modulation ( 1 = )


4.34

Fig. 4.26: AM with tone modulation ( 1.5 = )

Fig. 4.27: Envelope of the AM signal of Fig. 4.26

As can be seen from 4.24 and 4.25, the envelope (shown with a red
broken line) is one-to-one related to the message sinusoid. Note that, for = 1,
the carrier amplitude (and hence the envelope) goes to zero, corresponding to
the time-instant when the sinusoid is going through the negative peak. However,
when > 1, the one-to-one relationship between the envelope of the modulated
carrier and the modulating tone is no longer maintained. This can be more clearly
seen in Fig. 4.27 which shows the output of the envelope detector when the input

4.35

is the modulated carrier of Fig. 4.26. Notice that the tone signal between ( ) t t
1 2
,
and to the right of t
3
of Fig. 4.26 is getting inverted; in other words, the output of
the ED is proportional to ( ) + 1 cos
m
t which is not equal to ( ) + 1 cos
m
t ,
when > 1.

Fig. 4.28: Oscillogram when the CRO inputs are:
X-plates: tone signal
Y-plates: AM signal with
1
2
=

Fig. 4.29: Oscillogram when the CRO inputs are:
X-plates: tone signal
Y-plates: AM signal with ( 1 = )

4.36

Fig. 4.28 and Fig. 4.29 illustrate the oscillograms when the X-plates of the
CRO are fed with the modulating tone and the Y-plates with the AM signal with
= 0.5 and = 1 respectively. In Fig. 4.28, A represents the peak-to-peak
value of the carrier at its minimum (that is,
[ ]
= 2 1
c
A A ) where as B is the
peak-to-peak value of the carrier at its maximum (that is,
[ ]
= + 2 1
c
B A ).
Hence can be calculated as
B A
B A
=
+

In Fig. 4.29, as A = 0 we have = 1

4.3.2. Spectra of AM signals
Taking the Fourier transform of Eq. 4.5,
( ) ( ) ( ) ( ) ( )
2 2
c c
c c m c c
AM
A A
S f f f f f g M f f M f f = + + + + +

(4.8)
The plot of [ ( )]
AM
S f is given in Fig. 4.30.

Exercise 4.3
Picture the oscillogram when the X-plates of the CRO are fed with the
modulating tone and the Y-plates with the AM signal with = 1.5 .

4.37

Fig. 4.30: (a) Baseband message spectrum ( ) M f
(b) Spectrum of the AM signal

Based on Fig. 4.30, we make the following observations:
1) The spectrum has two sidebands, the USB [between
c
f to
c
f W + , and
( )
c
f W to
c
f , hatched in red] and the LSB (
c
f W to
c
f and
c
f to
( )
c
f W + , hatched in blue).
2) If the baseband signal has bandwidth W , then the AM signal has
bandwidth 2 W . That is, the transmission bandwidth
T
B , required for the
AM signal is 2 W .
3) Spectrum has discrete components at
c
f f = , indicated by impulses of
area
2
c
A

4) In order to avoid the overlap between the positive part and the negative part
of ( ) S f ,
c
f W > (In practice,
c
f W >> , so that ( ) s t is a narrowband
signal)

4.38

The discrete components at
c
f f = , do not carry any information and as
such AM does not make efficient use of the transmitted power. Let us illustrate
this taking the example of tone modulation.

Example 4.4
For AM with tone modulation, let us find
Total sideband power
=
Total power
, as a
function of modulation index .

For tone modulation, we have
( ) ( ) ( ) 1 cos cos
c m c
s t A t t = +

Carrier term = ( ) cos
c c
A t
Carrier Power =
2
2
c
A

USB term = ( ) cos
2
c
c m
A
t
+
Power in USB =
2
2
2
2 8
c
2
c
A
A

=
Power in LSB = Power in USB
Total sideband Power =
2
2 c
c
A
A
2 2
2
8 4

=
Total Power=
( )
2
2 2 2
c
c c c
A
A A A
2
2 2
2
1
2 4 2 2 4
+

+ = + =

Hence,
( )
c
c
A
A
2 2
2
2 2 2
4
2 2
4
= =
+ +

Calculating the value of for a few value of , we have

4.39

0.25 0.03
0.50 0.11
0.75 0.22
1.0 0.33

As can be seen from the above tabulation, increases as 1 ;
however, even at 1 = , only 1/3 of the total power is in the sidebands (or side
frequencies), the remaining 2/3 being in the carrier. From this example, we see
that AM is not an efficient modulation scheme, in terms of the utilization of the
transmitted power.

The complex envelope behavior of ( ) s t for tone modulation is quite
illustrative.
( ) ( ) ( )
c m c
s t A t t 1 cos cos = +

( ) ( )
c m c m c
j j t j t t c
c
A
A e e e Re
2
+

= + +

( )
m m
j t j t c c
c
ce
A A
s t A e e
2 2

= + +

(4.9)

Let us draw a phasor diagram, using the carrier quantity as the reference.
The term
m
j t c
A
e
2

can be represented as a rotating vector with a magnitude

of
c
A
2
, rotating counterclockwise at the rate of

m
f rev/sec. Similarly,
m
j t c
A
e
2

can be shown as a vector with clockwise rotational speed of

m
f
rev/sec.

4.40

Fig. 4.31: Phasor diagram for AM with tone modulation

Fig. 4.31 depicts the behavior of all the quantities on the RHS of Eq. 4.9.
From Eq.4.9, we find that the complex envelope is real and is given by
( )
c m
A t 1 cos +

. This can also be seen from the phasor diagram, because
at any given time, the quadrature components of the sideband phasors cancel
out where as the in-phase components add up; the resultant of the sideband
components is collinear with the carrier.

The length of the in-phase component of ( )
ce
s t

depends on the sign of
the resultant of the sideband phasors. As can be seen from Fig. 4.31 this varies
between the limits
[ ]
c
A 1 to
[ ]
c
A 1 + . If the modulation index is less than
1,
[ ]
c
A 1 0 > and envelope of ( ) s t is
( ) ( ) ( )
= + = +

1 cos 1 cos
c m c m
ce
s t A t A t

Phasor diagrams such as the one shown in Fig. 4.31 are helpful in the
study of unequal attenuation of the sideband components. We shall illustrate this
with an example.

Example 4.5
Let 1
c
A = , =
1
2
and let the upper sideband be attenuated by a factor of
2. Let us find the expression for the resulting envelope, ( ) A t .

4.41

The phasor diagram for this case is shown in Fig. 4.32.

Fig. 4.32: Phasor diagram for an AM signal with unequal sidebands

As can be seen from the figure, the resultant of the sidebands is no longer
collinear with the carrier.
( ) ( ) ( ) ( ) ( ) ( ) ( )
m m m m
ce
s t t j t t j t
1 1
1 cos sin cos sin
8 4
= + + +

( ) ( )
m m
t j t
3 1
1 cos sin
8 8
= +
( ) ( ) ( )
m m
A t t t
1
2 2
2
3 1
1 cos sin
8 8

= + +

Evidently, it is not possible for us to recover the message from the above ( ) A t .

4.4 Generation of AM and DSB-SC signals
Let ( ) x t be the input to an LTI system with the impulse response ( ) h t
and let ( ) y t be the output. Then,
( ) ( ) ( ) y t x t h t =
( ) ( ) ( ) Y f X f H f =
That is, an LTI system can only alter a frequency component (either boost or
attenuate), that is present in the input signal. In other words, an LTI system
cannot generate at its output frequency components that are not present in
( ) X f . We have already seen that the spectrum of a DSB or AM signal is

4.42

different from that of the carrier and the baseband signal. That is, to generate a
DSB signal or an AM signal, we have to make use of nonlinear or time-varying
systems.

4.4.1 Generation of AM
We shall discuss two methods of generating AM signals, one using a
nonlinear element and the other using an element with time-varying
characteristic.

a) Square law modulator
Consider the scheme shown in Fig. 4.33(a).

Fig. 4.33 (a): A circuit with a nonlinear element
(b): v i characteristic of the diode in Fig. 4.28(a)

A semiconductor diode, when properly biased has a v i characteristic that
nonlinear, as shown in Fig. 4.33(b).

4.43

For fairly small variations of v around a suitable operating point, ( ) v t
2
can be
written as
( ) ( ) ( ) v t v t v t
2
2 1 1 2 1
= + (4.10)
where
1
and
2
are constants.
Let ( ) ( ) ( )
c c
v t A f t m t
1
'
cos 2 = + . Then,
( ) ( ) ( ) ( ) ( )
c c c c
v A m t f t m t m t A f t t
2 2 2 2
2 1 1 2 2
1
2
'
1 cos 2 ( ) cos 2
'
= + + + +

(4.11)
The first term (on the RHS of Eq. 4.11) is ( )

AM
s t , with the carrier amplitude
c c
A A
1
'
= and
m
g
2
1
2
=
.

Now the question is: can we extract ( )

AM
s t from the sum of terms on
the RHS of Eq. 4.11? This can be answered by looking at the spectra of the
various terms constituting v t
2
( ) . Fig. 4.34 illustrates these spectra (quantities
marked A to E) for the ( ) M f of Fig. 4.14(a). The time domain quantities
corresponding to A to E are listed below.

Fig. 4.34: Spectra of the components of ( ) v t
2
of Eq. 4.11


4.44

Component Spectrum indicated by
(i) ( )
c c
A f t cos 2
A
(ii) ( ) ( )
c c
A m t f t
2
'
2 cos 2
B
(iii) ( ) m t
1

C
(iv) ( ) m t
2
2

D
(v)
( ) c c
A f t
2
2
2
'
cos (2 )
E

( )

AM
s t consists of the components (i) and (ii) of the above list. If it is possible
for us to filter out the components (iii), (iv) and (v), then the required AM signal
would be available at the output of the filter. This is possible by placing a BPF
with centre at
c
f and bandwidth 2W provided ( )
c
f W W 2 > or
c
f W 3 > .
Usually, this is not a very stringent requirement. However, this scheme
suffers from a few disadvantages.
i) The required square-law nonlinearity of a given device would be available
only over a small part of the ( ) v i characteristic. Hence, it is possible to
generate only low levels of the desired output.
ii) If
c
f is of the order of 3W, then we require a BPF with very sharp cut off
characteristics.

b) Switching modulator
In the first method of generation of the AM signals, we have made use of
the nonlinearity of a diode. In the second method discussed below, diode will be
used as a switching element. As such, it acts as a device with time-varying
characteristic, generating the desired AM signals when it is used in the circuit
configuration shown in Fig. 4.35.


4.45

Fig. 4.35: (a) Switching modulator
(b) Switching characteristic of the diode-load combination.

The (ideal) transfer characteristics of the diode-load combination is shown at (b)
in Fig. 4.35. This is explained as follows. We have,
( ) ( ) ( ) v t c t m t
1
= +
( ) ( )
c c
A f t m t cos 2 = +
If we assume that ( )
c
m t A
max
<< , then the behavior of the diode is governed by
( ) c t and can be approximated as
( )
( ) ( )
( )
v t c t
v t
c t
1
2
, 0
0 , 0
>

(That is, the diode offers infinite impedance when reverse biased and has zero
impedance, when forward biased. Hence, whether ( ) v t
1
is switched to the output
or not depends on the carrier cycle)
We can express ( ) v t
2
as
( ) ( ) ( )
p
v t v t x t
2 1
= (4.12)
where ( )
p
x t is the periodic rectangular pulse train of example 1.1. That is,
( )
( ) ( ) ( )
( ) ( ) ( )
p
c t c t
x t
c t c t
if
if
1 , 0 positivehalf cycleof
0 , 0 negativehalf cycleof
>
=

<

But from example 1.1 ( with =
0
),
c
f f

4.46

( )

=
=

2
c
j nf t
p n
n
x t x e
where
n
n
x c
1
sin
2 2

=

(4.13a)

From Eq. 4.13(a), we find that 0
n
x = , for n 2, 4 = etc. Combining the terms
n
x and
n
x , we obtain the trigonometric Fourier series, namely,
( ) ( )
( )
( )
n
p c c
n
x t f t n f t
n
1
2
1
1 2
cos 2 cos 2 2 1
2 2 1
= + +

(4.13b)
From Eq. 4.12 and 4.13(b), we see that ( ) v t
2
is composed of two components,
namely,
a) The desired quantity: ( ) ( )

+

4
1 cos 2
2
c
c
c
A
m t f t
A

b) The undesired terms with
i) Impulses in spectra at
c c
f f f 0, 2 , 4 = etc.
ii) Spectral lobes (same in shape as M(f)) of width 2W, centered at
c c
f f 0, 3 , 5 etc.
As compared to the square law modulator, switching modulator has the following
advantages:
a) Generated AM signals can have larger power levels.
b) Filtering requirements are less stringent because we can separate the
desired AM signal if
c
f W 2 > .
However, the disadvantage of the method is that percentage modulation has to
be low in order that the switching characteristics of the diode are controlled only
by the carrier.


4.47

4.4.2. Generation of DSB-SC
a) Product modulator
Generation of a DSB-SC signal involves the multiplication of ( ) m t with
( )
c c
A t cos . To generate this signal in the laboratory, any of the commercially
available multipliers can be used. Some of them are:
National: LM 1496
Motorola: MC 1496
Analog Devices: AD 486, AD 632 etc
Signetics: 5596
The power levels that can be generated, the carrier frequencies that can be used
will depend on the IC used. The details can be obtained from the respective
manuals. Generally, only low power levels are possible and that too over a
limited carrier frequency range.

b) Ring modulator
Consider the scheme shown in Fig. 4.36. We assume that the carrier

Fig. 4.36: Ring modulator

signal ( ) c t is much larger than ( ) m t . Thus ( ) c t controls the behavior of diodes
which would be acting as ON-OFF devices. Consider the carrier cycle where the

4.48

terminal 1 is positive and terminal 2 is negative. T
1
is an audio frequency
transformer which is essentially an open circuit at the frequencies near about the
carrier. With the polarities assumed for ( ) c t , D
1
, D
4
are forward biased, where as
D
2
, D
3
are reverse biased. As a consequence, the voltage at point a gets
switched to a
'
and voltage at point b to b
'
. During the other half cycle of ( ) c t ,
D
2
and D
3
are forward biased where as D
1
and D
4
are reverse biased. As a
result, the voltage at a gets transferred to b
'
and that at point b to a
'
. This
implies, during, say the positive half cycle of ( ) c t , ( ) m t is switched to the output
where as, during the negative half cycle, ( ) m t is switched. In other words,
( ) v t can be taken as
( ) ( ) ( )
p
v t m t x t = (4.14)
where ( )
p
x t is square wave as shown in Fig. 4.37.

Fig. 4.37: ( )
p
x t of Eq. 4.14


4.49

Fig. 4.38: (a) A message waveform ( ) m t
(b) ( ) v t of the ring modulator

Fig. 4.38(b) illustrates the product quantity ( ) ( )
p
m t x t , for the ( ) m t shown in
Fig. 4.38(a). The Fourier series expansion of ( )
p
x t can be written as
( )
( )
( )
n
p c
n
x t n t
n
1
2
1, 3, 5, ...
1
4
cos

.
When ( ) v t is passed through a BPF tuned to
c
f , the output is the desired DSB-
SC signal, namely, ( ) ( ) ( )
c
s t m t t
4
cos =
.

Example 4.6: Generation of DSB-SC
Consider the scheme shown in Fig. 4.39. The non-linear device has the
input-output characteristic given by
( ) ( ) ( ) y t a x t a x t
3
0 1
= +


4.50

Fig. 4.39: The scheme for the example 4.6

Let ( ) ( ) ( ) x t A f t m t
1
cos 2 = + where ( ) m t is the message signal. If the
required output ( ) s t is a DSB-SC signal with a carrier frequency of 1 MHz, let us
find the value of f
1
, assuming that a suitable BPF is available.

( ) ( ) ( ) ( ) ( ) y t a A f t m t a A f t m t
3
0 1 1 1
cos 2 cos 2 = + + +

1 2

( ) ( ) ( ) ( ) ( ) ( ) a A f t m t A f t m t A f t m t
3 3 3 2 2 2
1 1 1 1
cos 2 3 cos 2 3 cos 2 = + + +

2

In the equation for the quantity 2 above, the only term on the RHS that
can give rise to the DSB-SC signal is ( ) ( ) a A m t f t
2 2
1 1
3 cos 2 .
( ) ( ) ( )
( )
{ }
f t
a A m t f t a A m t
1
2 2 2
1 1 1
1 cos 2 2
3 cos 2 3
2
+

=
Assume that the BPF will pass only the components centered around f
1
2 . Then,
choosing f
1
500 = kHz, we will have
( ) ( ) ( )
c c
s t A m t f t cos 2 =
where
c
A a A
2
1
3 = and
c
f 1 = MHz.


4.51

Exercise 4.4
Consider the circuit configuration (called Cowan modulator) shown in Fig.
4.40. Show that the circuit can produce at its output the DSB-SC signal. T
1
is the
audio frequency transformer where as T
2
and T
3
are designed to operate around
the carrier frequency.

Fig. 4.40: Cowan modulator

Exercise 4.5: Balanced Modulator (BM)
Consider the scheme shown in Fig. 4.41. This configuration is usually
called a balanced modulator. Show that the output ( ) s t is a DSB-SC signal,
thereby establishing that BM is essentially a multiplier.

Fig. 4.41: Balanced modulator (BM)

4.52

4.5 Envelope Detector
As mentioned earlier, the AM signal, when not over modulated allows the
recovery of ( ) m t from its envelope. A good approximation to the ideal envelope
detector can be realized with a fairly simple electronic circuit. This makes the
receiver for AM somewhat simple, there by making AM suitable for broadcast
applications. We shall briefly discuss the operation of the envelope detector,
which is to be found in almost all the AM receivers.

Consider the circuit shown in Fig. 4.42.

Fig. 4.42: The envelope detector circuit

We assume the diode D to be ideal. When it is forward biased, it acts as a short
circuit and thereby, making the capacitor C charge through the source
resistance
s
R . When D is reverse biased, it acts as an open circuit and C
discharges through the load resistance
L
R .

As the operation of the detector circuit depends on the charge and
discharge of the capacitor C, we shall explain this operation with the help of Fig.
4.43.


4.53

Fig. 4.43: Envelope detector waveforms
(a) ( ) v t
1
(before DC block)
(b) ( )
out
v t (after DC block)

If the time constants
s
R C and
L
R C are properly chosen, ( ) v t
1
follows the
envelope of ( ) s t fairly closely. During the conduction cycle of D, C quickly
charges to the peak value of the carrier at that time instant. It will discharge a
little during the next off cycle of the diode. The time constants of the circuit will
control the ripple about the actual envelope.
B
C is a blocking capacitor and the
final ( )
out
v t will be proportional to ( ) m t , as shown in Fig. 4.43(b). (Note that a
small high frequency ripple, at the carrier frequency could be present on ( )
out
v t .
For audio transmission, this would not cause any problem, as
c
f is generally
much higher than the upper limit of the audio frequency range).


4.54

How do we choose the time constants?
s
R , though not under our control
can be assumed to be fairly small. Values for
L
R and C can be assigned by us.
During the charging cycle, we want the capacitor to charge to the peak value of
the carrier in as short a time as possible. That is,
1
s
c
R C
f
<< (4.15a)
Discharge time constant should be large enough so that C does not discharge
too much between the positive peaks of the carrier but small enough to be able
follow the maximum rate of change of m(t). This maximum rate depends on W,
the highest frequency in M(f). That is
L
c
R C
f W
1 1
<< << (4.15b)

Too small a value for
L
R C will make ( )
1
V t somewhat ragged (sort of saw
tooth ripple on the top) where as, with too large value for
L
R C, ED fails to follow
the envelope during the periods when ( ) m t is decreasing. A more accurate
analysis of the behavior of ED with ( ) m t as a tone signal is given in appendix
A4.1.

Example 4.7
Consider the scheme shown in Fig. 4.44. ( ) x t is a tone signal given by
( )
( )
x t t
4
cos 2 10

=

and ( ) ( )
c
c t f t cos 2 = with
c
f 10 = MHz.
( ) c t is the
HT of ( ) c t . ( ) v t , the output of the Balanced Modulator (BM), is applied as input
to an ideal HPF, with cutoff at 10 MHz. We shall find the expression for ( ) y t , the
output of an ideal envelope detector.


4.55


It is not too difficult to see that
( ) ( ) w t f t
6
1
cos 2 10 10
2

= +

where f 0.01 = Hz. That is,
( ) ( ) ( ) ( ) ( )
c c
w t f t t f t t
1
cos 2 cos sin 2 sin
2
=

As ( ) ( ) ( )
c
z t w t f t sin 2 = +

, we have
( ) ( ) ( ) ( ) ( )
c c
z t f t t f t t
1 1
cos 2 cos sin 2 1 sin
2 2

=

( ) z t represents a narrowband signal with the in phase component
( ) f t
1
cos 2
2

and the quadrature component ( ) f t
1
sin 2 1
2

. Hence,
( ) ( ) ( ) y t f t f t
1
2
2
2
1 1
cos 2 sin 2 1
4 2

= +

( )
t
1
2
4
5
sin 2 10
4

=

Example 4.8
Consider the scheme shown in Fig.4.45.


Let us find the output ( ) y t when,

4.56

a) ( ) ( ) ( )
m c
x t g m t t 1 cos = +

. ( )
m
g m t 1 < and ( ) m t is band-limited to
W Hz and the LPF has a bandwidth of W 2 . Assume that
c
f W 2 >> .
b) ( ) x t is a DSB-SC signal; that is ( ) ( ) ( )
c
x t m t t cos = .

a) ( ) ( ) ( ) ( )
m c
v t x t g m t t
2
2 2
1 cos = = +

( )
( )
c
m
t
g m t
2
1 cos 2
1
2
+

= +

( ) ( )
( )
m m
c
g m t g m t
t
2 2
1 1
cos 2
2 2
+ +

= + .
The second term on the RHS will be eliminated by the LPF. Hence,
( )
( )
m
g m t
w t
2
1
2
+

= . As ( )
m
g m t 1 0 +

, we have
( )
( )
m
g m t
y t
1
2
+

= .
b) When ( ) ( ) ( )
c
x t m t t cos = , we have
( ) ( ) ( ) ( )
( )
c
c
t
v t m t t m t
2 2 2
1 cos 2
cos
2
+

= =
The output of the LPF would be
( )
( ) m t
w t
2
2
=
As the squaring operation removes the information about the sign of the
signal, the output of ( ) y t is
( )
( ) m t
y t
2
=


4.57

4.6 Theory of Single-Sideband
Assume that from a DSB-SC signal, we have completely suppressed one
of the sidebands, say, the LSB. Let ( ) ( ) ( )
2
c
c c
DSB
A
S f M f f M f f = + +

where ( ) M f is as shown in Fig. 4.30(a). The resulting spectrum ( )

USB
S f will
be as shown in Fig. 4.47. Can we get back ( ) m t from the above signal? The
answer is YES. Let
( ) ( ) ( )
=

cos 2
c
USB
v t s t f t
If ( ) ( )
=

'
USB
S f S f , then ( ) ( ) ( )
1
' '
2
c c
V f S f f S f f

= + +

Exercise 4.6
Consider the waveform ( ) m t shown in Fig. 4.46. A DSB-SC is
generated using ( ) m t and a suitable high frequency carrier. Sketch the output
of an ideal envelope detector when the input to the detector is the DSB-SC
signal.

Fig. 4.46: ( ) m t for the exercise 4.6

4.58

Fig 4.47: Spectrum of the upper sideband signal

By plotting the spectrum of ( ) v t and extracting the spectrum for f W , we see
that it is ( )
1
2
c
A M f . A similar analysis will show that it is possible to extract
( ) m t from ( )

LSB
S f . In other words, with coherent demodulation, it is possible
for us to recover the message signal either from USB or LSB and the
transmission of both the sidebands is not a must. Hence it is possible for us to
conserve transmission bandwidth, provided we are willing to go for the
appropriate demodulation.

Let us now derive the time domain equation for an SSB signal. Let us start
with the two-sided spectrum and then eliminate the unwanted sideband. We shall
retain the upper sideband and try to eliminate the lower sideband. Consider
( ) ( ) ( ) sgn
2
c
c c c
A
M f f f f M f f +

But ( ) ( )
( )
( )
,
sgn
,
c c
c c
c c
M f f f f
f f M f f
M f f f f
>
=

<

Hence ( ) ( )
( )
>
+ =

<
,
1 sgn
2 0 ,
c c c
c
c c
c
A M f f f f
A
M f f f f
f f

That is, the lower sideband has been eliminated from the positive part of the
spectrum.
( ) ( )
2
c
j f t
c
M f f m t e


4.59

( ) ( )

( )
2
1
sgn
c
j f t
c c
f f M f f m t e
j

,
where
( ) m t is the HT of ( ) m t .
That is, ( ) ( ) ( )

( )
2
1
1 sgn
2 2
c
j f t
c c
c c
A A
M f f f f m t m t e
j

+

(4.16a)
Similarly, ( ) ( )
( )
+ <
+ + =

>
,
1
1 sgn
2 0 ,
c c
c c
c
M f f f f
M f f f f
f f

( ) ( ) ( )

( )
2 2
1 sgn
2 2 2
c c
j f t j f t
c c c
c c
A A A
M f f f f m t e m t e
j

+ + +

(4.16b)
Combining Eq. 4.16(a) and Eq. 4.16(b), we have the time domain equation for
the upper single sideband signal, namely,
( ) ( ) ( )

( ) ( ) cos sin
c c c
USB
s t A m t t m t t

=

(4.17)
Assume that the USSB signal is obtained from the DSB-SC signal,
( ) ( ) cos
c c
A m t t , by filtering out the LSB part. Then,
( ) ( ) ( )

( ) ( ) cos sin
2
c
c c
USB
A
s t m t t m t t

=

(4.18)

A few authors take Eq. 4.18 as representative of the SSB signal. Eq. 4.18
has the feature that the average power of the SSB signal is one-half the average
power of corresponding DSB signal. We shall make use of both Eq. 4.17 and
Eq. 4.18 in our further studies.
By a procedure similar to that outlined above, we can derive a time
domain expression for the LSB signal. The result would be
( ) ( ) ( )

( ) ( )

= +

cos sin
c c c
LSB
s t A m t t m t t or (4.19a)
( ) ( ) ( )

( ) ( ) cos sin
2
c
c c
LSB
A
s t m t t m t t

= +

(4.19b)
An SSB signal, whether upper or lower, is also a narrowband bandpass signal.
Eq. 4.18 can be treated as the canonical representation of USB signal with ( ) m t
as the in-phase component and
( ) m t as the quadrature component. Similarly


4.60

Eq. 4.19 provides the canonical representation of the LSB signal where ( ) m t is
the in-phase component and
( ) m t is the quadrature component.

We have already seen that a narrowband signal can also be expressed in
the envelope and phase form. Taking the USB signal, we have for the complex
envelope the quantity ( )

( )
( )
2
c
A
m t j m t + . Hence the envelope ( ) A t of the USB
signal is
( ) ( )

( )
2
2
2
c
USB
A
A t m t m t = +

(4.20a)
Similarly for the phase ( ) t , we have
( )

( )
( )
arc tan
m t
t
m t

=

(4.20b)
Expressing the USB signal with the envelope-phase form, we have
( ) ( ) ( ) ( ) cos
c
USB
s t A t t t = +

(4.21)
where ( ) A t and ( ) t are given by Eqs. 4.20(a) and 4.20(b) respectively. The
expression for ( )
LSB
s t

is identical to Eq. 4.21 but with ( ) t given by
( )

( )
( )
tan
m t
t arc
m t

=

(4.22)
That is, ( ) ( ) ( ) ( )
cos
c
SSB
s t A t t t = +

(4.23)
where ( ) t is given either by Eq. 4.20(b) or Eq. 4.22. Eq. 4.23 indicates that an
SSB signal has both amplitude and phase variations. (AM and DSB-SC signals
have only the amplitude of the carrier being changed by the message signal.
Note that AM or DSB-SC signals do not have quadrature components.) As such,
SSB signals belong to the category of hybrid amplitude and phase modulation.


4.61

Example 4.9: SSB with tone modulation
As a simple application of the Eqs. 4.18 and 4.19, let ( ) m t be ( ) cos
m
t .
Let us find the SSB signals.

( ) ( )

( ) ( )
m m
m t t m t t cos sin = = . Therefore,
( ) ( ) ( ) ( ) ( ) cos cos sin sin
2
c
m c m c
USB
A
s t t t t t =

( ) cos
2
c
c m
A
t = +

( ) ( ) ( ) ( ) ( ) cos cos sin sin
2
c
m c m c
LSB
A
s t t t t t = +

( ) cos
2
c
c m
A
t =

Alternatively,
( ) ( ) ( ) cos cos
c m c
DSB SC
s t A t t
=

( ) ( ) cos cos
2
c
c m c m
A
t t = + +

Extracting the USB, we have
( ) ( ) cos
2
c
c m
USB
A
s t t = +

If we eliminate the USB, then
( ) ( ) cos
2
c
c m
LSB
A
s t t =

Example 4.10
Let ( ) ( ) ( ) m t x t y t = where ( ) X f and ( ) Y f are as shown in Fig. 4.48.
An LSB signal is generated using ( ) m t as the message signal. Let us develop
the expression for the SSB signal.


4.62

Fig. 4.48: ( ) X f and ( ) Y f of example 4.10

Let us take the SSB signal as
( ) ( ) ( )

( ) ( )
c c
LSB
s t m t t m t t cos sin = +

(4.24)
We have
( )
( )
x t c t
3 3
2 10 sin 2 10

=

( ) ( )
( )
y t c t t
3 2 3
10 sin 500 cos 4 10

=

( ) ( ) ( ) m t x t y t = (4.25a)
What is required is
( ) m t , the HT of ( ) m t . ( ) m t is the product of a lowpass and

a bandpass signal. Hence
( ) ( )

( ) m t x t y t = . (See the note, after example 1.25)

But
( ) y t , from the result of example 1.25, is

( ) ( )
( )
y t c t t
3 2 3
10 sin 500 sin 4 10

=

That is,
( )
( )
( )
( )
m t c t c t t
6 3 2 3
2 10 sin 2 10 sin 500 sin 4 10

=

(4.25b)
( )
LSB
s t

is obtained by using Eq. 4.25 in Eq. 4.24.


4.63

Exercise 4.7
Let ( ) M f be as shown in Fig. 4.49. An upper sideband signal is
generated using the signal with this ( ) M f . Compute and sketch the spectrum
of the quadrature component of the SSB signal.

Fig. 4.49: Baseband spectrum for the Exercise 4.7

Exercise 4.8
Let ( ) ( ) m t c t sin = . Develop the expression for the following:
a) USB signal, in the canonical form.
b) USB signal, in the envelope and phase form
Ans. (b):
c
t
c f t
1
sin cos 2
2 4

+

Exercise 4.9
Let ( ) ( ) ( )

( ) ( )
c c
s t m t t m t t cos sin =
where ( ) M f is as shown in Fig. 4.49. Let W 5 = kHz and ( ) M 0 1 = .
a) Sketch the spectrum of (i) ( ) ( )
c
s t t cos and (ii)
( ) ( )
c
s t t sin
b) Show that sum of the spectra of part (a) is proportional to ( ) M f
c) Sketch the spectrum of
( ) ( ) ( ) ( )
c c
s t t s t t cos sin . Is this related to
( ) M f ? Explain.

4.64

4.7 Generation of SSB Signals
We shall consider two broad categories of SSB generation, namely, (i)
frequency discrimination method and (ii) phase discrimination method. The
former is based on the frequency domain description of SSB, whereas the latter
in based on the time-domain description of an SSB signal.

4.7.1 Frequency discrimination method
Conceptually, it is a very simple scheme. First generate a DSB signal and
then filter out the unwanted sideband. This method is depicted in Fig. 4.50.

Fig. 4.50: Frequency discrimination method of SSB generation

( ) v t is the DSB-SC signal generated by the product modulator. The BPF is
designed to suppress the unwanted sideband in ( ) V f , thereby producing the
desired SSB signal.

As we have already looked at the generation of DSB-SC signals, let us
now look at the filtering problems involved in SSB generation. BPFs with abrupt
pass and stopbands cannot be built. Hence, a practical BPF will have the
magnitude characteristic ( ) H f , as shown in Fig. 4.51. As can be seen from the
figure, ( ( ) H f is shown only for positive frequencies) a practical filter, besides the
PassBand (PB) and StopBand (SB), also has a TransitionBand (TB), during
which the filter transits from passband to stopband. (The edges of the PB and SB
depend on the attenuation levels used to define these bands. It is a common
practice to define the passband as the frequency interval between the 3-dB

4.65

points. Attenuation requirements for the SB depend on the application. Minimum
attenuation for the SB might be in the range 30 to 50 dB.)

Fig. 4.51: Magnitude characteristic of a practical BPF. Centre frequency
0
f =

Because of TB (where the attenuation is not to the desired level), a part of
the undesired sideband may get through the filter. As a rule of the thumb, it is
possible to design a filter if the permitted transitionband is not less than 1% of
center frequency of a bandpass filter. Fortunately, quite a few signals have a
spectral null around DC and if it is possible for us to fit in the transitionband into
this gap, then the desired SSB signal could be generated. In order to accomplish
this, it might become necessary to perform the modulation in more than one
stage. We shall illustrate this with the help of an example.

Example 4.11
Telephone quality speech signal has a spectrum in the range 0.3 to 3.4
kHz. We will suggest a scheme to generate upper sideband signal with a carrier
frequency of 5 MHz. Assume that bandpass filters are available, providing an
attenuation of more than 40 dB in a TB of width 0.01
0
f , where
0
f is the centre
frequency of the BPF.

Let us look at the generation of the SSB signal in one stage using a carrier
of 5 MHz. When a DSB signal is generated, it will have a spectral null of 600 Hz
centered at 5 MHz. That is, the transitionband is about 0.01 percent of the carrier

4.66

and hence it would not be possible to design such a sideband filter. However, it
would be possible to generate the desired SSB signal using two stages of
modulation.


Fig. 4.52: Two stage generation of SSB signal

( )
1
v t is a DSB-SC signal with the USB occupying the (positive frequency) range
( )
1
300
c
f + Hz to ( )
1
3400
c
f + Hz. The frequency range of the LSB is
( )
1
3400
c
f Hz to ( )
1
300
c
f Hz. Let us extract the upper sideband from ( )
1
v t
with the help of BPF1. Then the centre frequency of BPF1, ( )
0
1
f , is
( )
( ) ( )
1 1
0
1
300 3400
2
c c
f f
f
+ + +
=
( )
1
1850
c
f = +
width of the spectral null around
1
600
c
f = Hz.
Hence
1
1850
600
100
c
f +

or
1
60,000 1850
c
f
58.1 kHz
Let us take
1 c
f as 50 kHz. Let ( ) M f be as shown in Fig. 4.53(a). Then ( )
2
V f will
be as shown in Fig. 4.53(b).

4.67

Fig. 4.53: (a) Message spectrum
(b) The spectrum of the USB signal with
1
50
c
f = kHz

( )
3
v t is a DSB-SC signal with the carrier
2 c
f , ( )
2
v t being the modulating signal.
Then ( )
3
V f will be as shown in Fig. 4.54. In this figure,

Fig. 4.54: Spectrum of ( )
3
v t of Fig. 4.52

( )
1 2
53,400
c
f f = Hz
( )
2 2
50,300
c
f f = Hz
( )
3 2
50,300
c
f f = + Hz
( )
4 2
53,400
c
f f = + Hz


4.68

Hence the transition band available to design the sideband filter is
( )
3 2
100.6 f f = kHz. With this TB, we can choose centre frequency of BPF2
less than or equal to 10.06 MHz. If we choose
2 c
f as 4.95 MHz then we will have
the upper sideband occupying frequency range (4.95 + 0.0503) = 5.0003 MHz to
(4.95 + 0.0534) = 5.0034 MHz. This is exactly what would have happened if the
modulation scheme was attempted in one step with 5 MHz as the carrier.
Note: As the spectral occupancy of the USB signal is from 5.0003 MHz to 5.0034
MHz, theoretical centre frequency of the BPF2 is 5.00185. With respect to this
frequency, we have

TB width 100.6
100 2.01 percent
centre freq. 5001.85
= =
which is about twice the permitted ratio. Hence, it is possible to choose for
1 c
f a
value lower than 50 kHz.

4.7.2 Phase discrimination method
This method implements Eq. 4.17 or Eq. 4.19(a), to generate the SSB
signal. Consider the scheme shown in Fig. 4.55. This scheme requires two
product modulators, two
2
phase shifters and an adder. One of the phase shifter

is actually a Hilbert transformer (HT); it should provide a
2
phase shift for all the

components in ( ) M f . This is not such an easy circuit to realize. Assuming it is
possible to build the HT, the SSB can be generated for any
c
f , provided the
product modulators (multipliers) can work at these frequencies.

4.69

Fig. 4.55: SSB generation: Phase discrimination method

Instead of a single wide band phase shifter acting as the HT, it is possible
to have an SSB generator with two Phase Shifting Networks, (PSN), one in each
branch as shown in Fig. 4.56.

Fig. 4.56: An alternate configuration for the phase discrimination scheme

( )
1
H f and ( )
2
H f are the phase shifting networks. Let ( )
( )
1
1
j f
H f e

= and
( )
( )
2
2
j f
H f e

= . ( )
1
f and ( )
2
f are such that ( ) ( )
1 2
2
f f

=

for the
frequency range of interest. That is, PSN1 and PSN2 maintain a constant
difference of
2
.


4.70

Let us explain the operation of the scheme shown in Fig. 4.56 by taking
( ) m t to be a tone signal; that is, ( ) ( ) cos
m m
m t A t = . Let
( )
1 1
m
f f
f
=
= and ( )
2 2
m
f f
f
=
=
where
2 1
2
= + . Then,
( ) ( )
1 1
cos
m m
v t A t = + and ( ) ( )
2 2
cos
m m
v t A t = + .
( ) ( ) ( )
3 1
cos cos
m c m c
v t A A t t = +
( ) ( ) ( )
4 2
cos sin
m c m c
v t A A t t = +
( )
1
cos sin
2
m c m c
A A t t

= + +

( ) ( )
1
sin sin
m c m c
A A t t = +
( ) ( ) ( )
3 4 1
cos
m c c m
v t v t A A t + = + +

After coherent demodulation, we will have ( )
1
cos
m
t + . We shall assume that
the additional phase shift
1
which is actually frequency dependent will not cause
any problem after demodulation.

As it is not too difficult to design a Hilbert transformer using digital filter
design techniques, phase shift method of SSB generation is better suited for
digital implementation. For a brief discussion on SSB generation using digital
signal processing, the reader is referred to [1].


4.71

4.8 Demodulation of SSB
SSB signals can be demodulated using coherent demodulation as shown
in Fig. 4.58.

Exercise 4.10
There is a third method of generating the SSB signal, known as Weavers
method. This scheme is shown in Fig. 4.57.

Fig. 4.57: Weavers method of SSB generation
Let ( ) M f be non-zero only in the interval
l u
f f f ; and let
1
2
l u
f f
f
+
= .
The lowpass filters in the I and Q channels are identical and have a cutoff
frequency of
0
2
u l
c
f f
f

= . Assume that
2 0 c
f f >> . By sketching the spectra at
various points in the above scheme, show that ( ) s t is an SSB signal. What is
the actual carrier frequency with respect to which, ( ) s t would be an SSB
signal?

4.72

Fig. 4.58: Coherent demodulation of SSB

The received SSB signal is multiplied by the local carrier which is of the same
frequency and phase as the carrier used at the transmitter. From Fig, 4.58, we
have
( ) ( ) ( )
'
cos
c c
SSB
v t s t A t =

( ) ( )

( ) ( ) ( )
'
cos sin cos
2
c
c c c c
A
m t t m t t A t

=

( ) ( )

( ) ( ) ( )
2
'
cos cos sin
2
c c
c c c
A A
m t t m t t t

=

(4.26)
The second term on the RHS of Eq. 4.26 has the spectrum centered at 2
c
f
which will be eliminated by the LPF following ( ) v t . The first term of Eq. 4.26 can
be written as,
( )
( )
'
1 cos2
2 2
c c c
t A A
m t
+

.

As ( ) ( )
c
m t t cos 2 has the spectrum centered at 2
c
f , even this will be
eliminated by the LPF. Hence ( ) ( )
0
'
4
c c
A A
v t m t = . That is,
( ) ( )
0
v t m t
The difficulty in demodulation is to have a coherent carrier at the receiver. Can
we use a squaring loop or Costas loop to recover the carrier from ( )
SSB
s t

?
The answer is NO. Let us look at the squaring loop. From Eq. 4.23,
( ) ( ) ( ) ( ) cos
c
SSB
s t s t A t t t = = +

After squaring, we obtain,

4.73

( ) ( ) ( )
2 2 2
cos
c
s t A t t t = +

( ) ( ) ( )
{ }
2 2
1 cos 2
c
A t t t

= + +

As ( ) t is a function of time, we do not have a discrete component at 2
c
f f =
and hence, carrier acquisition is not possible. It is left as an exercise to show that
Costas loop will not be able to put out ( ) m t when the input to the loop is the SSB
signal. Hence, when SSB is used in a communication system, highly stable
crystal oscillators are used both at the transmitter and receiver. If this scheme
does not work (especially, at very high frequencies) a small pilot carrier can be
sent along with the SSB signal. This pilot carrier is recovered at the receiver and
is used for demodulation.

Let us now look at the effects of frequency and phase offset in the carrier
used for demodulation. Let the carrier term at the receiver be
( )
'
cos 2
c c
A f f t +

. Let the received input to the demodulator (Fig. 4.58) be
( ) ( )

( ) ( )
1
cos sin
2
c c c
A m t t m t t

Then, ( ) ( ) ( )

( ) ( ) ( )

= +

1
'
cos sin cos
2
c c c c c
v t A A m t t m t t t
( )
0
v t , the output of the LPF would be
( ) ( ) ( )

( ) ( )
0
1
'
cos 2 sin 2
4
c c
v t A A m t f t m t f t

+

(4.27)
Assume that
( ) ( ) ( )

( ) ( )
0
cos 2 sin 2 v t m t f t m t f t = + (4.28)
Consider a special case, namely, a frequency component at 1 f = kHz in ( ) M f
and 100 f = Hz. With these values, Eq. 4.28 becomes
( )
( )
( )
( )
( )
3 3
0
cos 2 10 cos 2 100 sin 2 10 sin 2 100 v t t t t t = +
( )
2 900 2 900
cos 2 900
2
j t j t
e e
t

+
= = (4.29)

4.74

As this is true of every frequency component in ( ) M f , we have the result that
when f is positive and the input is a USB signal, there is an inward shift of
( ) M f by f . (We see from Eq. 4.28,
2 1000 j t
e

is converted to
2 900 j t
e

and
2 1000 j t
e

is converted to
2 900 j t
e

. That is, both the spectral components
have been shifted inward by 100 Hz.) By a similar analysis, we can easily see
that if f is negative and the input is a USB signal, then, after demodulation, the
spectral components in ( ) M f would undergo an outward shift by f . In all, we
have four cases to be taken into account and the effects of non-zero f on the
resulting output after demodulation are summarized below.
Case i) 0 f > and the input signal is USB: Spectral components in ( ) M f will
undergo an inward shift by f
Case ii) 0 f > and the input signal is LSB: Spectral components in ( ) M f will
undergo an outward shift by f
Case iii) 0 f < and the input signal is USB: Spectral components in ( ) M f will
undergo an outward shift by f
Case iv) 0 f < and the input signal is LSB: Spectral components in ( ) M f will
undergo an inward shift by f

Let ( ) M f be as shown in Fig. 4.59(a). Let 300 f = Hz. Then, if the input
is a USB signal, the spectrum of the demodulated output, ( )
0
V f , would be as
shown in Fig. 4.59(b).


4.75

Fig. 4.59: (a) Baseband message spectrum
(b) Inward spectral shift (after demodulation) of a
1 USB signal. Frequency offset, 300 f = Hz


4.76

Example 4.12
Let ( ) ( ) ( ) ( ) m t A t A t A t
1 2 3
cos 200 cos 1100 cos 4000 = + + . An
upper sideband signal is generated using this ( ) m t . The carrier used for
demodulation had a positive offset of 150 Hz. Let us find the frequencies of the
spectral components after demodulation.

As the received signal is USB and f 0 > , there would be an inward shift
of the frequency components by 150 Hz. Spectral components in ( ) M f are
shown in Fig. 4.61(a).

Exercise 4.11: Effect of phase error on SSB
Consider the scheme shown in Fig. 4.60. Let
( ) ( ) ( )

( ) ( ) cos sin
c c
s t m t t m t t

= +

. The carrier used to demodulate has
a phase difference of with respect to the carrier used for modulation. Show
that ( )
0
v t has phase distortion (when compared to ( ) m t ) by establishing
( )
( )
( )
0
, 0
, 0
j
j
M f e f
V f
M f e f
>
=

<

Fig. 4.60: SSB demodulation with carrier phase difference

Hint: Show that ( ) ( )

( ) v t m t m t
0
cos sin

+


4.77

Fig. 4.61: (a) message spectrum
(b) Spectrum after demodulation with frequency error

Note that the negative frequency components have been shown with a broken
line. Let ( ) m t be the demodulated output. After demodulation, there would be an
inward shift by 150 Hz and this is shown in Fig. 4.61(b). From this spectrum, we
see that ( ) m t
'
is consisting of components at 50 Hz, 350 Hz and 1850 Hz.

The speech files that follow provide a qualitative measure of the distortion
caused by the frequency offset of the local carrier in the demodulation of SSB
signals. Input to the demodulator is a USB signal.
Introduction
Output 1
Output 2a
Output 2b
Output 2c

4.78

Output 3a
Output 3b
Output 3c
Output 4a
Output 4b
Output 4c

After listening to these speech outputs, one gets the feeling that, for a
given frequency offset, SSB performs better than the DSB. Even in SSB, outward
frequency shift of the original message spectrum ( f negative for USB) has
better clarity than the corresponding inward shift. Of course, voice tends to
become somewhat shrill, which, of course, is expected.


4.79

Exercise 4.12: Quadrature Carrier Multiplexing (QCM)
It is possible to transmit two DSB-SC signals with in a bandwidth of 2 W ,
by using a scheme called QCM, as shown in Fig. 4.62. (QCM is also referred to
as Quadrature Amplitude Modulation (QAM) or simply quadrature modulation.
Using QAM, we are able to achieve the BW efficiency of SSB modulation.)

Fig. 4.62: Quadrature carrier multiplexing scheme

Two message signals ( )
1
m t and ( )
2
m t are used to generate two DSB-SC
signals, ( )
1
v t and ( )
2
v t respectively. The carriers used in generating ( )
1
v t and
( )
2
v t are in phase quadrature. The transmitted signal ( ) ( ) ( )
1 2
s t v t v t = + . At
the receiver, coherent demodulation is used to recover the two messages ( )
1
m t
and ( )
2
m t .
a) Show that ( )
1
m t and ( )
2
m t will be recovered by the receiver shown.
b) Let the local carrier have some frequency and phase offset; that is, instead
of ( ) 2 cos 2
c
f t , let it be ( ) 2 cos 2
c
f f t + +

. Then show that the
output of the upper branch is
( ) ( ) ( ) ( )
{ }
1 2
cos 2 sin 2
c
A m t f t m t f t + +

where as the output of the lower branch is
( ) ( ) ( ) ( )
{ }
2 1
cos 2 sin 2
c
A m t f t m t f t + + +

Note: We see from the above result that the carrier phase and frequency have to
be fairly accurate to have proper demodulation; otherwise ( )
1
m t will interfere with
( )
2
m t and vice versa. This is called cochannel interference. QAM is used in
color TV for multiplexing the chrominance signals.

4.80

4.9 Vestigial SideBand (VSB) Modulation
One of the widespread applications of VSB has been in the transmission
of picture signals (video signals) in TV broadcast. The video signal has the
characteristic that it has a fairly wide bandwidth (about 5 MHz) with almost no
spectral hole around DC. DSB modulation, though somewhat easy to generate,
requires too much bandwidth (about 10 MHz) where SSB, though bandwidth
efficient, is extremely difficult to generate, as explained below.

With analog circuitry it is very difficult to build the
2
phase shifter over a 5

MHz bandwidth; as such phase shift discrimination method is not feasible. To
make use of the frequency discrimination method, we require very sharp cutoff
filters. Such filters have a highly non-linear phase characteristic at the band
edges and spectral components around the cut-off frequencies suffer from phase
distortion (also called group delay distortion). The human eye (unlike the ear)
being fairly sensitive to phase distortion, the quality of the picture would not be
acceptable.

VSB refers to a modulation scheme where in the wanted sideband (either
USB or LSB) is retained almost completely; in addition, a vestige (or a trace) of
the unwanted sideband is added to the wanted sideband. This composite signal
is used for transmitting the information. This vestige of the wanted sideband
makes it possible to come up with a sideband filter that can be implemented in
practice.

4.9.1 Frequency domain description of VSB
Figure 4.63 depicts the scheme of VSB generation. In this figure, ( ) v t is a
DSC-SC signal, which is applied as input to a Sideband Filter (SBF), ( )
v
H f , that
shapes ( ) V f so that ( ) s t is a VSB signal.


4.81

Fig. 4.63: Generation of VSB using the filtering method

Now the questions that arise are: what is the shape of the SBF and how
do we demodulate such a signal. As coherent demodulation is fairly general, let
us try to demodulate the VSB signal as well, using this method. In the process of
demodulation, we shift the modulated carrier spectrum (bandpass spectrum) up
and down by
c
f and then extract the relevant baseband. Because of the vestige
of the unwanted sideband, we expect some overlap (in the baseband) of the
shifted spectra. In such a situation, overlap should be such that, ( ) M f is
undistorted for f W . In other words, ( ) ( )
v c v c
H f f H f f + + should result in
a filter with a rectangular passband within the frequency range ( ) to W W .
With a little intuition, it is not too difficult to think of one such ( )
v
H f .

Assume that we are retaining the USB almost completely and permitting
the vestige of the LSB. Consider the ( ) ( )
( )
v
j f
v v
H f H f e

= shown in Fig. 4.64.
We assume the phase characteristic ( )
v
f to be linear over the frequency range
l c
f f f W + with ( ) ( )
v c v c
f f m 2 = = , where m is an integer.

If a DSB signal is given as input to the above ( )
v
H f , it will partially
suppress USB (in the frequency range
c u
f f f ) and allow the vestige of the
LSB (from
l c
f f f ). The demodulation scheme is the same as shown in Fig.
4.13, with the input to the detector being the VSB signal. ( )
0
V f , the FT of the
output of the detector, is

4.82

( ) ( ) ( ) ( )
v c v c
V f K M f H f f H f f
0 1
= + +

, for f W .
where K
1
is the constant of proportionality.

Fig. 4.64: An example of a SBF generating VSB

If ( ) ( )
d
j f t
v c v c
H f f H f f K e
2
2

+ + =

, f W ,
where
d
t determines the slope of the phase characteristic and K
2
is a constant,
then ( ) ( )
0 d
v t K m t t = , where K K K
1 2
= .

Let ( ) ( )
( )
( )
v
j f
v v v c v
H f H f e H f f W f f
1,
1, 1,
,
= =
where
v c l
f f f = , is the width of the vestige in LSB.

For the ( )
v
H f shown in Fig. 4.64, ( )
1, v
H f is as shown in Fig. 4.65.

4.83

Fig. 4.65: ( )
v c
H f f for f W
(a) magnitude characteristics
(b) phase characteristics

Let ( ) ( )
( )
( )
2,
2, 2,
v
j f
v v v c
H f H f e H f f
= = + for
v
f f W . ( )
2, v
H f is
shown in figure 4.66.


4.84

Fig. 4.66: ( )
v c
H f f + for f W
(a) magnitude characteristics
(b) phase characteristics
As ( )
v
H f
1,
0 = for
v
f f W and
j m
e
2
1

= , we can write
( ) ( )
2
1,
,
d
j f t
v c v
H f f H f e f W

= (4.30a)
By a similar reasoning,
( ) ( )
2
2,
,
d
j f t
v c v
H f f H f e f W

+ = (4.30b)
Therefore,
( ) ( ) ( ) ( )
2
1, 2,
,
d
j f t
v c v c v v
H f f H f f H f H f e f W

+ + = +

(4.30c)
Summing ( )
v
H f
1,
and ( )
v
H f
2,
, we have the ideal LPF with unity gain for
f W .
Let us take a closer look at the sideband filter for VSB. The magnitude
characteristic shown in Fig. 4.64(a) can be taken as the sum of
( )
( )
( )
( )
1 2
v v
H f H f +


4.85

where
( )
( )
1
v
H f and
( )
( )
2
v
H f are shown in Fig. 4.67(a) and (b) respectively.

Fig. 4.67: Decomposition of the ( )
v
H f of Fig. 4.64

Note that with
( )
( )
v
H f
1
alone for the sideband filter, we would have generated the
SSB signal. It is the addition of
( )
( )
v
H f
2
to
( )
( )
v
H f
1
that gives rise to VSB output.
Consider f 0 > ;
( )
( )
v
H f
2
consists of two straight line segments, one between the
points ( ) a b , and the other between the points ( ) c d , . Similar is the case for
f 0 < . Now the question is: should
( )
( )
v
H f
2
consist of only straight line
segments? The answer is NO. What is required is that
( )
( )
v
H f
2
exhibit odd
symmetry about
c
f . Two more such characteristics have been shown in fig.
4.68, for f 0 > .


4.86

Fig. 4.68: Two more examples of
( )
( )
v
H f
2

As such, we have a great deal of flexibility in designing ( )
v
H f . (Subject to some
upper limit, there could be choice in fixing
v
f , the width of the vestige.) These
features facilitate the design and implementation of a practical sideband filter.

4.9.2 Time domain description of VSB
Let ( )
v
h t denote the impulse response of the sideband filter, ( )
v
H f . From
Fig. 4.63, we have
( ) ( ) ( ) ( )
v
VSB
s t s t h v t d

= =

But ( ) ( ) ( )
c c
v t A m t t cos = . For convenience, let
c
A 1 = .
Then, ( ) ( ) ( ) ( )
v c
s t h m t t d cos

=


4.87

( ) ( ) ( ) ( ) ( ) ( )
v c c c c
h m t t t d cos cos sin sin

= +

( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )
v c c
v c c
h m t d t
h m t d t
cos cos
sin sin 4.31

Eq. 4.31 is in the canonical form for the representation of narrowband
signal. Let ( )
c
m t denote the in-phase component of ( ) s t and ( )
s
m t , the
quadrature component. Then,
( ) ( ) ( ) ( )
c v c
m t h m t d cos
(4.32a)
and ( ) ( ) ( ) ( )
s v c
m t h m t d sin
(4.32b)
Then, ( ) ( ) ( ) ( )
c c s c
VSB
s t m t t m t cos sin =

(4.33)
Eq. 4.33 is the canonical representation of the VSB signal.

Let ( ) ( ) ( )
i v c
h t h t t cos = and ( ) ( ) ( )
q v c
h t h t t sin = . Then,
( ) ( ) ( )
c i
m t m t h t = (4.34a)
( ) ( ) ( )
s q
m t m t h t = (4.34b)
Taking the FT of Eq. 4.34, we have
( ) ( ) ( )
c i
M f M f H f = and ( ) ( ) ( )
s q
M f M f H f =
If ( ) M f 0 = for f W > , then ( )
c
M f and ( )
s
M f are bandlimited to atmost W .
This implies that ( )
c
m t and ( )
s
m t are lowpass signals, as required.
As ( ) ( ) ( )
i v c
h t h t t cos = we have,
( )
( ) ( )
v c v c
i
H f f H f f
H f
2
+ +
= (4.35a)

4.88

Similarly, ( )
( ) ( )
v c v c
q
H f f H f f
H f
j 2
+
= (4.35b)
But ( ) ( )
v c v c
H f f H f f 1 + + =

, for f W . Hence, ( )
( )
c
m t
m t
2
= .
Let us look at ( )
q
H f .
From Eq. 4.35(b),
( )
( ) ( )
q
v c v c
H f
H f f H f f
j
2
= +
Let ( )
v
H f be as shown in Fig. 4.69.

Fig. 4.69: ( )
v
H f with vestige in LSB

Fig. 4.70:
( )
q
H f
j
2
(solid line) for the ( )
v
H f of Fig. 4.69

Then,
( )
q
H f
j
2
for f W will be as shown in Fig. 4.70.

4.89

Based on Eq. 4.33, we have another implementation for the generation of
the VSB signal called the phase discrimination method of generating VSB. This
scheme is shown in Fig. 4.71. Plus sign at the last summer will result in a VSB
signal with a vestige in LSB whereas the minus sign will result in a VSB signal
with the vestige in USB for the ( )
q
H f shown in Fig. 4.70.

Fig. 4.71: Phase discrimination method of VSB generation

Comparing the scheme shown in Fig. 4.71 with that shown in Fig. 4.55 for the
generation of SSB, we find a close resemblance between them. When ( )
q
h t is a
HT (with the magnitude response equal to 1/2), we have ( ) ( )
VSB SSB
s t s t =

.
If ( )
q
h t 0 = , we have ( ) ( )
VSB DSB
s t s t =

. In other terms, DSB and SSB can
be treated as special cases of VSB. Transmission bandwidth of VSB is,

[ ]
T v
VSB
B W f = +
With
v
f 0 = , we have the SSB situation where as,
v
f W = leads to DSB.

Example 4.13
A VSB signal is generated from the DSB-SC signal, ( ) ( )
c
m t t 2 cos .
( ) M f is as shown in Fig. 4.72(a). If the vestige is as shown in Fig. 4.72(b), lets

4.90

find the values of all spectral components in the VSB signal for
c
f f > , assuming
that demodulation in done by multiplying the received VSB signal with
( )
c
t 2cos . ( ) M f is to be restored to its original values.

Fig. 4.72: (a) Baseband message spectrum used to generate the VSB
(b) Spectral components in the vestige in LSB

Let
i
be the magnitude of the spectral component at
c i
f f i , 1, 2, ..., 5 + = in
the VSB spectrum. When the VSB spectrum is shifted to the left by
c
f , we have
the spectrum shown in Fig. 4.73(a). Similarly, when the VSB spectrum is shifted
to the right by
c
f , we have the spectrum shown in Fig. 4.73(b).

4.91

Fig. 4.73: Shifted VSB spectra: (a) left shift (b) right shift

From the spectral plots in Fig. 4.73(a) and (b), we have

1
0.1 0.3 + = ; That is
1
0.2 =

2
0.2 0.8 + = ; That is
2
0.6 =

3 4 5
1, 0.65 and 0.4 = = =
Hence the VSB spectrum for f 0 > is shown in Fig. 4.74.


4.92

Fig. 4.74: VSB spectrum of example 4.13

4.10 Envelope Detection of VSB+C
As mentioned earlier, one important application of VSB has been in TV
broadcast. As a large number of receivers are involved, it is preferable to make
the detector circuit fairly simple and inexpensive. The envelope of a VSB signal is
not one-to-one related to the message ( ) m t . Hence direct envelope detection of
a VSB signal would be of no use. However, if there is a carrier component along
with the VSB signal, that is, VSB+C, then ED might work. We shall now look into
this.

Exercise 4.13
Let ( ) ( ) ( ) ( ) m t t t t 2 4 cos 200 6 cos 300 5 cos 400 = + + + .
Specify the frequency response of a VSB filter that passes the LSB almost
completely and leaves a vestige of the first frequency component in USB. It is
given that the magnitude of the filter at the vestige frequency is
1
8
. Sketch the
spectrum of the VSB signal.

4.93

Let the input to the ED be
( ) ( ) ( ) ( ) ( )
c c c c s c
A t A m t t m t t cos cos sin +

where is an adjustable scale factor. Then ( ) A t , the output of the ED is
( ) ( ) ( ) ( )
{ } c s
A t A m t m t
1
2 2
2
1 = + +

( )
( )
( )
s
c
m t
A m t
m t
1
2
2
1 1
1

= + +

+

(4.36)
If the distortion component
( )
( )
s
m t
m t
1
1
<<
+
for all t , then the output of the ED
will be ( )
c
A m t 1 +

, which, after the DC block, will provide the message
output. This level of the distortion component can be reduced by
i) increasing
v
f . Note that as
v
f W , we will have the DSB signal; that is,
( )
s
m t 0 = .
ii) decreasing the value of .

In commercial TV broadcast,
v
f is about 75 kHz which is about 1/6 of the
width of a full sideband which is about 5 MHz. It has been found that with a
vestige of about 75 kHz, the distortion component is not too bothersome.

It would be instructive to compare the envelope detection of SSB+C with
that of VSB+C. Let the input to the ED be SSC+C, namely,
( ) ( ) ( )

( ) ( )
c c c c
A t m t t m t t cos cos sin

+

then, ( ) ( )

( )
c
A t A m t m t
1
2
2 2

= + +


4.94

( ) ( )

( )
c
c c
c
m t m t m t
A
A A
A
1
2
2
2
2
2
1

= + + +

(4.37)
If
( )
c
m t
A
and
( )
c
m t
A
are very much less than unity, then
( )
c
m t
A
2

and
( )
c
m t
A
2

can
be dropped from Eq. 4.37, giving us
( )
( )
c
c
m t
A t A
A
1
2 2
1

+

Retaining only first order term of the binomial expansion,
( )
( )
( )
c c
c
m t
A t A A m t
A
1

+ = +

Of course, if we ensure ( ) ( )
c
m t m t A
max min
= << , then
( )
c
m t
A
can be
neglected in comparison with unity. This may be adequate to make
( )
c
m t
A
much
smaller than unity, most of the time. In any case, it is obvious that ED of SSB+C
results in excessive wastage of the transmitted power. In contrast, ED of AM is
reasonably efficient as the requirement to avoid envelope distortion is
( )
c
m t A
max
.

Example 4.14
Consider the SSB+C signal,

( ) ( ) ( ) ( )

( ) ( )
c c c c
s t A t m t t m t t cos cos sin = + +
where ( ) m t
t
2
1
1
=
+
. We will show that, if
c
A 1 >> , then the output of the ED
can be taken as ( ) +
c
A m t .


4.95

As ( ) m t
t
2
1
1
=
+
, we have
( )
t
m t
t
2
1
=
+
. Hence ( ) A t , the envelope of
( ) s t is,
( )
c
t
A t A
t t
1
2 2
2
2 2
1
1 1

= + +

+ +

( ) ( )
c
c
A t
A
t
t t
1
2
2
2
2 2 2
2 2
2 1
1
1 1

= + + +

+
+ +

c
c
A
A
t t
1
2
2
2 2
2 1
1 1

= + +

+ +

( ) ( )
c
c c
A
A t A t
1
2
2 2 2
2 1
1
1 1

= + +

+ +

Neglecting the term
( )
+
2 2
1
1
c
A t
, we have
( )
( )
c
c
A t A
A t
1
2
2
2
1
1

+

+

Using the binomial expansion upto the second term, we have
( )
( )
c
c
A t A
A t
2
1 2
1
2
1

+

+

c
A
t
2
1
1
+
+

which is the desired result.


4.96

A note on the linearity of AM, DSB-SC, SSB and VSB: Having discussed
these modulation schemes, let us look at the linearity aspect of these schemes.
By linearity, we imply that the schemes obey the superposition property. This can
be easily verified in the case of DSB-SC, SSB and VSB. Consider DSB-SC.
When message signals ( ) m t
1
and ( ) m t
2
are applied separately, the resulting
modulated waveforms are ( ) ( )
c c
A m t t
1
cos and ( ) ( )
c c
A m t t
2
cos . Let
( ) ( ) ( ) m t m t m t
1 1 2 2
= +
where
1
and
2
are constants. Then the modulated carrier is
( ) ( ) ( ) ( ) ( )
c c c c
A m t t A m t m t t
1 1 2 2
cos cos = +

,
Exercise 4.14
Consider SSB+C with tone modulation. Let
( ) ( ) ( )
c c c m
s t A t t cos cos = + +

Assume that
c m
f f >> and 1 < .
a) Construct the phasor diagram and develop the expression for the
envelope ( ) A t .
b) Let be such that terms involving
m
m , 3 can be neglected. Show
that
( ) ( ) ( )
c m m
A t A t t
2 2
1 cos cos 2
4 4

+ +

c) Find the value of so that second harmonic envelope distortion, that is,
the ratio,
( )
( )
A t
A t
Amplitude of the second harmonic in
Amplitude of the fundamental in
is less than five
percent.

4.97

which establishes the linearity property. Similarly, SSB and VSB can be shown to
be linear. AM is not linear in its strict sense, because if ( ) m t is applied as input
to an AM modulator, the output is
( ) ( ) ( ) ( ) ( )
{ }
c m c c m c
A g m t t A g m t m t t
1 1 2 2
1 cos 1 cos + = + +

and this is not equal to
( ) ( ) ( ) ( )
c m c c m c
A g m t t A g m t t
1 1 2 2
1 cos 1 cos + + +

That is, superposition does not apply to the carrier component. As this is only a
minor deviation, all the four modulation types are put under the category of linear
modulation.

4.11 Superheterodyne Receiver
The important function of a receiver is demodulation; that is, to recover the
message signal from the received modulated waveform. The other functions
which become necessary for the proper reception of a signal are: amplification
and tuning or selective filtering. Amplification becomes necessary because, most
often, the received signal is quite weak and without sufficient amplification, it may
not even be able to drive the receiver circuitry. Tuning becomes important
especially in a broadcast situation because there is more than one station
broadcasting at the same time and the receiver must pick the required station
and reject the inputs from the other (unwanted) stations; tuning also ensures that
out of band noise components do not affect the receiver's performance. Besides
these operations, most of the receivers also incorporate certain other features
such as frequency conversion, automatic gain control etc.

Two of the demodulation methods which we have already discussed are
coherent or synchronous detection and envelope detection. Coherent detection
can be used to demodulate any linear modulation scheme: DSB-SC, DSB-LC,
SSB or VSB. In practice, it is used to demodulate only suppressed carrier

4.98

signals. Envelope detection is mainly used in the demodulation of DSB-LC and
VSB+C signals. We shall now describe the receiver used in AM broadcast.
The receiver for the broadcast AM is of the superheterodyne (or superhet)
variety
1
. This is shown schematically in the Fig. 4.75.

Fig. 4.75: Superheterodyne Receiver

The wanted signal ( ) s t , along with other signals and noise, is input to the Radio
Frequency (RF) stage of the receiver. The RF section is tuned to
c
f , the carrier
frequency of the desired signal ( ) s t . The bandwidth of the RF stage,
RF
B , is
relatively broad; hence along with ( ) s t , a few adjacent signals are also passed
by it. The next stage in the receiver is the frequency conversion stage consisting
of a mixer and a local oscillator. The local oscillator frequency,
LO
f tracks the
carrier frequency
c
f , (with the help of a ganged capacitor) and is usually
( )
c IF
f f + , where
IF
f denotes the Intermediate Frequency (IF). The mixer output
consists of, among others, the frequency components at
c IF
f f 2 + and
IF
f . The
following stage, called the IF stage, is a tuned amplifier, which rejects all the
other components and produces an output that is centered at
IF
f . The bandwidth
of the IF stage,
IF
B , is approximately equal to the transmission bandwidth,
T
B , of

1
Some of the other applications of superhet are: reception of FM and TV broadcast signals and
RADAR

4.99

the modulation scheme under consideration. For example, if the input signal is of
the double sideband variety, then
IF
B W 2

The IF stage constitutes a very important stage in a superheterodyne
receiver. It is a fixed frequency amplifier (it could consist of one or more stages of
amplification) and provides most of the gain of the superhet. Also, as
IF T
B B = ,
it also rejects the adjacent channels (carrier frequency spacing ensures this).
Next to IF, we have the detector or demodulation stage which removes the IF
carrier and produces the baseband message signal at its output. Finally, the
demodulator output goes through a baseband amplification stage (audio or video,
depending upon the type of the signal) before being applied to the final
transducer (speaker, picture-tube etc.)

The spectral drawings shown in Fig. 4.76 and 4.77 help clarify the action
of a superhet receiver. We shall assume the input to the receiver is a signal with
symmetric sidebands.

Fig. 4.76: Typical spectrum at the input to the RF stage of a superhet

Fig. 4.77: Spectrum at the input of the IF stage of a stage of a superhet


4.100

Fig. 4.76 shows the spectrum of the input signal to the RF stage. As can be
seen, it has the desired signal, plus a few adjacent channels and possibly a
signal with the carrier frequency
c c IF LO IF c
f f f f f f
' '
2 = + = + is called the image
frequency corresponding to
c
f . If
c
f
'
is allowed to pass through the RF stage,
then it will also give rise to an output at the IF stage which would interfere with
the wanted signal, namely the signal whose carrier frequency is
c
f . Hence the
main task of the RF section is to pass the frequency components in the range
c T
f B
2
while rejecting the signal with the spectrum centered at
c
f
'
(image
frequency rejection). If the 3-dB bandwidth,
RF
B is such that

T RF F
B B f
1
2 < <
then, ( )
RF
H f , should be able to provide sufficient attenuation at the image
frequency.

Fig. 4.77 depicts the frequency response characteristic of the IF stage. As
seen from this figure, the IF filter takes care of adjacent channel rejection.

The superheterodyne
1
structure results in several practical benefits:
i) Tuning takes place entirely in the front end (RF and mixer stage) so that the
rest of the circuitry (IF stage, detector and the final power amplifier stage)
requires no adjustments to changes in
c
f .
ii) Separation between
c
f and
IF
f eliminates potential instability due to stray
feedback from the amplified output to the receiver's input.
iii) Most of the gain and selectivity is concentrated in the fixed IF stage.
IF
f is
so selected so that
IF IF
B f results in a reasonable fractional bandwidth (for

1
The word superheterodyne refers to the operation of the receiver, namely, the incoming signal
at the carrier frequency is heterodyned or mixed with the LO signal whose frequency is higher
than ( )
c LO c IF
f f f f = + .

4.101

AM broadcast various frequency parameters are given in table 4.1). It has
been possible to build superhets with about 70 dB gain at the IF stage itself.

Table 4.1 Parameters of AM radio

Carrier frequency range 540-1600 kHz
Carrier spacing 10 kHz
Intermediate frequency 455 kHz
IF bandwidth 6-10 kHz
Audio bandwidth 3-5 kHz

An IF of 455 kHz has been arrived at by taking the following points into
consideration.
1) IF must not fall within the tuning range of the receiver. Assume that there is
a station broadcasting with the carrier frequency equal to
IF
f . This signal,
could directly be picked off by the IF stage (every piece of wire can act as
an antenna). Interference would then result between the desired station and
the station broadcasting at
c IF
f f = .
2) Too high an IF would result in poor selectivity which implies poor adjacent
channel rejection. Assume that IF was selected to be 2 MHz. With the
required bandwidth of less than 10 kHz, we require very sharp cutoff filters,
which would push up the cost of the receiver.
3) As IF is lowered, image frequency rejection would become poorer. Also,
selectivity of the IF stage may increase; thereby a part of the sidebands
could be lost.

We had mentioned earlier, that
LO c IF
f f f = + . If we have to obtain the
IF
f
component after mixing, this is possible even if
LO c IF
f f f = . But this causes
the following practical difficulty. Consider the AM situation. If we select
LO c IF
f f f = , then the required range of variation of
LO
f so as to cover the

4.102

entire AM band of 540-1600 kHz is (540 - 455) = 85 kHz to (1600 - 455) = 1145
kHz. Hence the tuning ratio required is 85 : 1145 1: 13 . If
LO c IF
f f f = + , then
the tuning ratio required is 995 : 2055 1: 2 . This is much easier to obtain than
the ratio 1:13. With the exception of tuning coils, capacitors and potentiometers,
all the circuitry required for proper reception of AM signals is available in IC chips
(for example, BEL 700).

Example 4.15: Double conversion superhet receiver
In receivers operating in the VHF (30 - 300 MHz) range and meant for
receiving fairly narrowband signals (say telemetry signals), to achieve good
image frequency rejection and selectivity using a single mixer stage is quite
difficult. Hence, receivers have been developed with more than one frequency
conversion stage and more than one IF. We shall now look at the image
frequency problem of a receiver with two mixer stages, usually called double or
dual conversion receiver.


Fig. 4.78: Block diagram of a double conversion receiver

In the scheme shown, the first mixer stage has a tunable LO, its output
being 30 MHz below the incoming signal frequency. The second mixer stage has
an LO producing a fixed frequency output at 40 MHz. If the RF stage is tuned to
200 MHz, let us find the possible image frequencies of the incoming signal. We
assume that none of the filters in the cascade are ideal.

4.103

First local oscillator frequency = (200 - 30) = 170 MHz. Hence (170 - 30) =
140 MHz will be an image frequency for the incoming signal w.r.t the first IF. As
the second IF is 10 MHz, a component at 50 MHz would be an image of 30 MHz.
Hence if a frequency component at ( ) 170 50 220 = or 120 MHz get through
the RF stage, it would interfere with the reception of the wanted carrier at 200
MHz. In other words, the image frequencies are at 220 MHz, 140 MHz and 120
MHz.


4.104

Exercise 4.15
a) For the tuned circuit shown in Fig. 4.79,
( )
( )
( )
V f
H f
I f
0
= is,
( ) H f
j f C
R f L
1
1 1
2
2
=

+

( )
R
j R LC
L
2
1
1 1
=
+

Let f
0
be the resonant frequency; then f
LC
0
1
2
=
and Q-factor of
filter is
C
Q R
L
=

Fig. 4.79: A tuned circuit with L, C and R

Show that
( )
( )
( )
( )
V f
R R
H f
I f f f
f f
j Q
j Q
f f
f f
0
2 2
0
0
0
0
1
1
= = =

+

+

and hence
( )
( )
R
H f
Q f
2 2
1
=
+

where ( )
f f
f
f f
0
0

=


4.105

b) Let the RF stage in a superhet consist of a simple tuned circuit (TC1)
whose output is input to the mixer stage as shown in Fig. 4.80.

Fig. 4.80: Superhet without RF amplifier
Let denote the image frequency rejection, where
( )
( )
c
c
H f
H f
'
= ,
c
f
'
being
the image frequency of
c
f .
That is,
( ) c
Q f
2 2 '
1 = + . Calculate the value of when the receiver
is tuned to (i) 1.0 MHz and (ii) 20.00 MHz. Assume Q of the resonant
circuit TC
1
to be 100.
(Ans.: (i) 138.6 = (ii) 70.6 = )
c) It is required to have the value of at 20 MHz the same as at 1.00 =
MHz. For this purpose, the RF stage has been modified to include a
tuned amplifier stage as shown in Fig. 4.81. Calculate the required Q of
the tuned circuit, TC
2

Answer: Q of TC
2
5.17 =

Fig. 4.81: Superhet with RF amplifier

4.106

Appendix A4.1: Analysis of ED with Tone Modulation
Consider ED circuit shown in Fig. 4.42 which is redrawn in Fig. A4.1.

Fig. A4.1: Envelope detector

Let ( ) ( ) ( )
m c
AM
s t f t f t 1 cos 2 cos 2 = +

We assume 1 < so that ( )
m
f t 1 cos 2 +

is positive for all t . Then, the
envelope quantity is, ( ) ( )
m
A t t 1 cos = +

. We shall derive on upper bound
on in terms of
L
R C and
m
so that the detector is able to follow the envelope
for all t .

Consider the waveform ( ) v t
1
shown in Fig. A4.2. The value of the
envelope ( ) A t at t t
0
= ,
( ) ( )
m
A t t
0 0
1 cos = +


4.107

Fig. A4.2: Waveforms of ED of Fig. A4.1

Let ( ) ( ) v t A t
1 0 0
= . Note that ( ) v t
1 0
is the voltage across the capacitor at
t t
0
= . Assuming that the capacitor discharges until the next positive peak in the
carrier cycle, we have,
( )
L c
R C f
c
v t v t e
f
1 1
1 0 1 0
1

+ =

L c
R C f 1 >> . Hence
( ) ( )
m
c L c
v t t
f R C f
1 0 0
1 1
1 cos 1

+ +

m
c c
A t t
f f
0 0
1 1
1 cos

+ = + +

( ) ( )
m m
m m
c c
t t
f f
0 0
1 cos cos sin sin

+

Assuming
m c
f f << , we approximate

m
c
f
f
cos 2 1

and
m m
c c
f f
f f
2
sin 2

Hence,
( ) ( ) ( ) ( )
m
m m m
L c c
t t t
R C f f
0 0 0
1
1 cos 1 1 cos sin

+ +

That is,

4.108

( )
( )
m
L
m m
t
R C
t
0
0
1 cos
sin
+

(A4.1a)
or
( )
( )

+
0
0
sin
1
1 cos
m m
L m
t
R C t
(A4.1b)
Rearranging Eq. A4.1(b), we have
( ) ( )
m m m
L L
t t
R C R C
0 0
1
sin cos

(A4.2)
Let
m L
R C
1
1
tan
as shown
Then,
m
m
L
R C
2
2
cos
1
=

+

and,
L
m
L
R C
R C
2
2
1
sin
1
=

+

The inequality A4.2 can be written as
( ) ( )
m m
L
D t D t
R C
0 0
1
cos sin sin cos
where
m
L
D
R C
2
2
1
1
=

+

. That is,
( )
L m
R C D t
0
sin 1 (A4.3)
The inequality should hold even when ( )
m
t
0
sin 1 = . That is,

L
R C D 1

( )
L
m L
R CD
R C
2
1 1
1
=
+

or ( )
L
m
R C
2
2
1 1


4.109

Evidently, we will not be able to demodulate, using the ED circuit, a tone
modulated AM signal with 1 = .

Fig. A4.3 displays the experimentally generated demodulator output (top
waveform) when the time constant
L
R C is of proper value so as to follow the
envelope for all t (modulating tone is shown at the bottom). Fig. A4.4 is the ED
output (top waveform) when the time constant is too large. As can be seen from
the figure, ED is not able to follow the negative half cycle of the tone fully,
resulting in the clipping of a part of their cycle. (These waveforms are from
Shreyas experimentor kit.)

Fig. A4.3

4.110

Fig. A4.4

Exercise A4.1
A 1 kHz square wave, switching between the levels 1 V, amplitude
modulates a carrier to a depth of 50 %. The parameters of the carrier are:
c
A 1 = volt and
c
f 1 = MHz.
a) Sketch the resulting AM signal.
b) Let the signal of (a) be the input of the ED of Fig. A4.1. Sketch the
voltage across the capacitor C for the following cases:
i)
L
R C 25 sec =
ii)
L
R C m 2 sec =

4.111

References
1) Charles Schuler and Mahesh Chugani, Digital Signal Processing: a hands
on approach, Tata McGraw Hill, 2005

Some suggested references
1) A. Bruce Carlson, Paul B. Crilly and Janet C. Rutledge, Communication
systems (4
th
ed.), Mc Graw Hill international edition, 2002
2) B. P. Lathi, Modern Digital and Analog Communication Systems, (3
rd
ed.)
Oxford University press, 1998
3) John G. Proakis and Masoud Salehi, Communication Systems Engineering,
Prentice Hall international edition, 1994

5.1
CHAPTER 5
CHAPTER 5
Angle Modulation

5.1 Introduction
Consider a sinusoid, ( )
c c
A f t
0
cos 2 + , where
c
A is the (constant)
amplitude,
c
f is the (constant) frequency in Hz and
0
is the initial phase angle.
Let the sinusoid be written as ( )
c
A t cos

where ( )
c
t f t
0
2 = + . In chapter
4, we have seen that relaxing the condition that
c
A be a constant and making it a
function of the message signal ( ) m t , gives rise to amplitude modulation. We
shall now examine the case where
c
A is a constant but ( ) t , instead of being
equal to
c
f t
0
2 + , is a function of ( ) m t . This leads to what is known as the
angle modulated signal. Two important cases of angle modulation are Frequency
Modulation (FM) and Phase modulation (PM). Our objective in this chapter is to
make a detailed study of FM and PM.

An important feature of FM and PM is that they can provide much better
protection to the message against the channel noise as compared to the linear
(amplitude) modulation schemes. Also, because of their constant amplitude
nature, they can withstand nonlinear distortion and amplitude fading. The price
paid to achieve these benefits is the increased bandwidth requirement; that is,
the transmission bandwidth of the FM or PM signal with constant amplitude and
which can provide noise immunity is much larger than W 2 , where W is the
highest frequency component present in the message spectrum.


5.2
Now let us define PM and FM. Consider a signal ( ) s t given by
( ) ( )
c i
s t A t cos =

where ( )
i
t , the instantaneous angle quantity, is a
function of ( ) m t . We define the instantaneous frequency of the angle modulated
wave ( ) s t , as
( )
( )
i
i
d t
f t
d t
1
2

=
(5.1)
(The subscript i in ( )
i
t or ( )
i
f t is indicative of our interest in the instantaneous
behavior of these quantities). If ( )
i c
t f t
0
2 = + , then ( )
i
f t reduces to the
constant
c
f , which is in perfect agreement with our established notion of
frequency of a sinusoid. This is illustrated in Fig. 5.1.

Fig. 5.1: Illustration of instantaneous phase and frequency

Curve 1 in Fig. 5.1 depicts the phase behavior of a constant frequency sinusoid
with
0
0 = . Hence, its phase, as a function of time is a straight line; that is
( )
i c
t f t 2 = . Slope of this line is a constant and is equal to the frequency of
the sinusoid. Curve 2 depicts an arbitrary phase behavior; its slope changes
with time. The instantaneous frequency (in radians per second) of this signal at
t t
1
= is given by the slope of the tangent (green line) at that time.


5.3
a) Phase modulation
For PM, ( )
i
t is given by
( ) ( )
i c p
t f t k m t 2 = + (5.2)
The term
c
f t 2 is the angle of the unmodulated carrier and the constant
p
k is
the phase sensitivity of the modulator with the units, radians per volt. (For
convenience, the initial phase angle of the unmodulated carrier is assumed to be
zero). Using Eq. 5.2, the phase modulated wave ( ) s t can be written as
( ) ( )
c c p
PM
s t A f t k m t cos 2 = +

(5.3)
From Eq. 5.2 and 5.3, it is evident that for PM, the phase deviation of ( ) s t from
that of the unmodulated carrier phase is a linear function of the base-band
message signal, ( ) m t . The instantaneous frequency of a phase modulated
signal depends on
( )
( )
d m t
m t
d t
'
= because
( )
( )
p i
c
k d t
f m t
d t
1
'
2 2
= +

.
b) Frequency Modulation
Let us now consider the case where ( )
i
f t is a function of ( ) m t ; that is,
( ) ( )
i c f
f t f k m t = + (5.4)
or ( ) ( )
t
i i
t f d 2

=
(5.5)
( )
t
c f
f t k m d 2 2

= +
(5.6)
f
k is a constant, which we will identify shortly. A frequency modulated signal
( ) s t is described in the time domain by
( ) ( )
t
c c f
FM
s t A f t k m d cos 2 2

= +

(5.7)
f
k is termed as the frequency sensitivity of the modulator with the units Hz/volt.
From Eq. 5.4 we infer that for an FM signal, the instantaneous frequency

5.4
deviation of ( ) s t from the (unmodulated) carrier frequency
c
f is a linear function
of ( ) m t . Fig. 5.2 to 5.5 illustrate the experimentally generated AM, FM and PM
waveforms for three different base-band signals. From these illustrations, we
observe the following:
i) Unlike AM, the zero crossings of PM and FM waves are not uniform (zero
crossings refer to the time instants at which a waveform changes from
negative to positive and vice versa).
ii) Unlike AM, the envelope of PM or FM wave is a constant.
iii) From Fig. 5.2(b) and 5.3(b), we find that the minimum instantaneous
frequency of the FM occurs (as expected) at those instants when ( ) m t is
most negative (such as t t
1
= ) and maximum instantaneous frequency
occurs at those time instants when ( ) m t attains its positive peak value,
p
m
(such as t t
2
= ). When ( ) m t is a square wave (Fig. 5.4), it can assume
only two possible values. Correspondingly, the instantaneous frequency
has only two possibilities. This is quite evident in Fig. 5.4(b).


5.5

Fig 5.2: AM and FM with tone modulation

5.6

Fig. 5.3: AM and FM with the triangular wave shown as ( ) m t


5.7

Fig. 5.4: AM and FM with square wave shown as ( ) m t

iv) A triangular wave has only two possibilities for its slope. In Fig. 5.5(b), it has
a constant positive slope between t
1
and t
2
, and constant negative slope to
the right of t
2
for the remaining part of the cycle shown. Correspondingly,
the PM wave has only two values for its ( )
i
f t , which is evident from the
figure.
v) The modulating waveform of Fig. 5.5(c) is a square wave. Except at time
instants such as t t
1
= , it has zero slope and
( )
t t
d m t
d t
1
=
is an impulse.

5.8
Therefore, the modulated carrier is simply a sinusoid of frequency
c
f ,
except at the time instants when ( ) m t changes its polarity. At t t
1
= , ( )
i
f t
has to be infinity. This is justified by the fact that at t t
1
= , ( )
i
t undergoes
sudden phase change (as can be seen in the modulated waveform) which
implies
( )
i
d t
d t
tends to become an impulse.

Eq. 5.3 and 5.7 reveal a close relationship between PM and FM. Let
( ) ( )
t
I
m t m d

=
. If ( )
I
m t phase modulates a carrier with modulator
sensitivity
p f
k k 2 = , then the resulting signal is actually an FM signal as given
by Eq. 5.7. Similarly a PM signal can be obtained using frequency modulator by
differentiating ( ) m t before applying it to the frequency modulator. (Because of
differentiation, ( ) m t should not have any discontinuities.)

As both PM and FM have constant amplitude
c
A , the average power of a
PM or FM signal is,

c
av
A
P
2
2
= ,
regardless of the value of
p
k or
f
k .


5.9

Fig 5.5: PM with ( ) m t (a) a sine wave
(b) a triangular wave
(c) a square wave

5.10
Example 5.1
An angle modulated signal is given by
( ) ( ) ( )
( )
s t t t t
6
cos 2 2 10 30sin 150 40cos 150

= + +

Let us find the maximum phase and frequency derivations.

The terms ( ) ( ) t t 30sin 150 40cos 150 +

can be expressed in the form
( ) ( ) t t cos sin 150 sin cos 150 +

.
As
2 2
30 40 50 + = , we have
( ) ( ) ( ) ( ) t t t t
3 4
30sin 150 40cos 150 50 sin 150 cos 150
5 5

+ = +

( ) t 50sin 150 = + where
1
4
tan
3
=
Hence ( ) ( ) s t t t
6
cos 4 10 100 sin 150

= + +

Evidently, the maximum phase deviation is ( ) 100 .
Let ( ) ( ) t t 100 sin 150 = +

( )
( )
d t
t
d t
1
50cos 150 150
2

= +

( ) t 7500 cos 150 = +
Hence maximum frequency deviation 7500 = Hz.

Example 5.2
Let ( ) m t be a periodic triangular wave with period
3
10
sec. with
( ) ( ) m t m t
max min
1 = = volt. We shall find the maximum and minimum values
of the instantaneous frequency for
a) FM with
f
k
4
10 = Hz/volt
b) PM with
p
k = rad/volt

5.11
Assume the carrier frequency to be 100 kHz.

a) For FM, ( ) ( )
i c f
f t f k m t = +
( ) ( )
i
f t
3 4
min
100 10 10 90 = = kHz
( ) ( )
i
f t
3 4
max
100 10 10 110 = + = kHz
b) For PM, ( ) ( )
p
i c
k
f t f m t
'
2
= +

Note that ( ) m t
'
is a square wave with maximum and minimum values as
4,000 .
Hence ( ) ( )
i
f t
3
min
1
100 10 4,000
2
=
98 = kHz
Similarly, ( ) ( )
i
f t
max
102 = kHz

Example 5.3
Let ( ) s t be a general angle modulated signal given by
( ) ( ) ( )
c i c c
s t A t A t t cos cos = = +

It is given that when ( )
m
m t t cos = , ( ) s t has the instantaneous frequency
given by ( ) ( )
i c m m
f t f k f t
2
2 cos = + , where k is a suitable constant. Let us
find the expression for ( )
i
t . If ( ) m t is different from
m
t cos , what could be
the expression for ( )
i
t and ( ) s t .

( )
( ) ( )
i
i c m m
d t
f t f k f t
d t
2 1
2 cos
2

= = +

( )
( )
i
c m m
d t
f k t
d t
2
2 cos
= +

5.12
or ( ) ( )
i c m m
t f t k t 2 sin = +
( )
c m
d
f t k t
d t
2 cos =

Generalizing,
( ) ( )
i c
d
t f t k m t
d t
2 =

and ( )
( )
c c
d m t
s t A f t k
d t
cos 2

=

Exercise 5.1
A periodic signal ( ) m t angle modulates a very high frequency carrier.
Let the modulated signal be given by, ( ) ( )
c p
s t A t k m t
8
cos 2 10

= +

,
where ( ) m t is as shown in Fig. 5.6.

Fig. 5.6: Modulating signal for the exercise 5.1

If ( ) ( )
i i
f f
max min
is to be 100 kHz show that
p
k 10 = rad/volt.

5.13
5.2 Bandwidth of FM
In this section, we shall make a detailed analysis of the bandwidth
requirements of FM. (PM is considered in section 5.4.) Let
f f
c k 2 = . For
convenience, let ( ) s t denote the FM signal. With ( )
t
I
m t m d ( ) , the pre-
envelope of ( ) s t is
( ) ( ) ( )
c c f I
pe
s t A j t c m t exp
= +

( ) ( ) ( ) ( ) ( )
c
n
n
j t n f f
c f I I I
c c
A j c m t m t j m t e
n
2
2
1
2! !

= + + + +

(5.8)
As ( ) ( )
{ }
pe
s t s t Re =

, we have
( ) ( ) ( ) ( ) ( ) ( ) ( )
f
c c f I c I c
c
s t A t c m t t m t t
2
2
cos sin cos
2!

= +

(5.9)
If ( ) m t is band-limited to W , then so is ( )
I
m t , as ( )
( )
( )
I
M f
M f
j f 2
=
. But ( ) s t
consists of terms ( ) ( ) ( ) ( ) ( ) ( )
I c I c
m t t m t t
2 3
cos , sin etc. The spectrum of
( ) ( )
I
m t
2
is of width W 2 , and that of ( ) ( )
I
m t
3
is W 3 etc. The spectrum of
( ) ( ) ( )
I c
m t t
2
cos occupies the interval
c
f W 2 and that of ( ) ( ) ( )
n
I c
m t t cos
occupies the interval
c
f nW . Clearly, the spectrum of the FM signal is not
band-limited. It appears that, at least theoretically, it has infinite bandwidth and
( ) S f is not simply related to ( ) M f . A similar situation prevails even in the case
of a PM signal. (Recall that the bandwidth of any linear modulated signal is less
than or equal to W 2 , where W is the message band-width). Let us take a closer
look at this issue of FM bandwidth.


5.14

5.2.1 NarrowBand FM (NBFM)
Assume that ( )
f I
c m t
max
1 << . Then ( ) s t of Eq. 5.9 can be well
approximated by the first two terms; that is,
( ) ( ) ( ) ( )
c c f I c
s t A t c m t t cos sin

(5.10)
Eq. 5.10 is similar to that of an AM wave. Now ( ) S f is band-limited to W 2 as in
the case of an AM signal
1
. For this reason, the case ( )
f I
c m t
max
1 << is called
narrowband FM (NBFM). Similarly, the narrowband PM (NBPM) signal can be
written as
( ) ( ) ( ) ( )
c c p c
s t A t k m t t cos sin

(5.11)
Though the expressions for NBFM or NBPM resemble fairly closely that of an AM
signal, there is one important difference; namely, the sideband spectrum of PM
or FM has a phase shift of
2
with respect to the carrier. This makes the time

domain signal of PM or FM very much different from that of an AM signal. A
NBFM signal has phase variations with very little amplitude variations whereas
the AM signal has amplitude variations with no phase variations.

Note that from Eq. 5.10, we have the envelope of the NBFM signal given by
( ) ( )
c f I
A t A c m t
2 2
1 = +
If ( )
f I
c m t 1 << , then ( )
c
A t A .

Example 5.4

1
In general, FM is nonlinear as
( ) ( ) [ ] { } ( ) { } ( ) { }
c c f I I c c f I c c f I
A t c m t m t A t c m t A t c m t
1 2 1 2
cos cos cos + + + + +

5.15
A NBFM signal, ( ) s t , is generated by using ( )
( )
m t c t
3 2 3
10 sin 10 = . We
shall find ( ) S f .
From Eq. 5.10, we have
( ) ( ) ( ) ( )
c c f I c
s t A t c m t t cos sin =

But ( )
( )
I
M f
m t
j f 2
and
f f
c k 2 = .
Hence
( ) ( ) ( )
( )
( )
( )
( )
c c c c f
c c
c c
M f f M f f A A K
S f f f f f
f f f f 2 2
+
= + + +

+

where ( )
f
M f tri
3
10

=

5.2.2 WideBand FM (WBFM)
If the condition ( )
f I
c m t
max
1 << is not satisfied, then we have the
wideband FM, which, as mentioned earlier has, at least theoretically, infinite
bandwidth. However, as will be seen a little later, most of the power of the FM
signal resides in a finite bandwidth, called the transmission bandwidth. In order
to estimate this bandwidth, we observe that
( ) ( )
i c f
f t f k m t = +
Let ( ) ( )
p
m m t m t
max min
= = . Then the instantaneous frequency varies in the
Exercise 5.2
Let ( )
f
M f ga
3 3
1
10 10

=

and
c
f
6
10 = Hz. It is given that
f
k 250 =
Hz/volt, and
c
A 4 = V. Sketch ( ) S f for a NBFM signal.

5.16
range
( )
c f p
f k m to
( )
c f p
f k m + . As the total range of frequency deviation (with
centre at
c
f ) is
f p
k m 2 , can we assume that, for all practical purposes, the
transmission bandwidth of an FM signal, ( )
T
FM
B is
( )
T f p
FM
B k m
?
2 =
Let f denote the maximum deviation of the carrier frequency
c
f . That is,
f p
f k m = . Then, can we assume that, to a very good approximation,
( )
T
FM
B f 2 = ?
The above expression for the transmission bandwidth is valid only when
f W >> . In the case f W << , ( )
T
FM
B f 2 but W 2 , as will be shown
later. This is indeed the fallacy that gave birth to FM in the first place. In the
1920s radio engineers, thinking that the actual bandwidth of an FM signal is
f 2 , felt that bandwidth requirement of FM can be made less than that of AM
(that is, less than W 2 ) by choosing f appropriately! The fallacy here lies in
equating the instantaneous frequency to the spectral frequency. Although ( )
i
f t is
measured in Hz, it should not be equated with spectral frequency. Spectral
frequency f is the independent variable of the frequency domain where as ( )
i
f t
is a time dependent quantity indicating the time behavior of a signal with angle
modulation. When we refer to the spectrum ( ) X f of a signal ( ) x t , we imply that
( ) x t is composed of the complex exponentials with the magnitude and phase
specified by ( ) X f , and each one of these exponentials exists for all t with the
given frequency. ( )
i
f t , on the other hand represents the frequency of a cosine
signal that can be treated to have a constant frequency for a very short duration,
maybe only for a few cycles. We shall now give some justification that bandwidth
of an FM signal is never less than W 2 .


5.17
Let the base-band signal ( ) m t be approximated by a staircase signal, as
shown in Fig. 5.7(a). From the sampling theorem (to be discussed in chapter 6),
this staircase approximation is justified as long as the width of each rectangle
(separation between two adjacent samples) is less than or equal to
W
1
2
. For
convenience, we shall treat the adjacent sample separation to be equal to
W
1
2
.

The FM wave for the staircase-approximated signal will consist of a
sequence of sinusoidal pulses, each of a constant frequency and duration of
W
1
2
sec. A typical pulse and its spectrum are shown in Fig. 5.7(b) and (c)
respectively.
The frequency of the RF pulse in the interval
k k
t t
W
1
,
2

+

is
( ) ( )
i k c f k
f t f k m t = + . Hence, its spectral range (between the first nulls) as
shown in Fig. 5.7(c), is from ( )
c f k
f k m t W 2 +

to ( )
c f k
f k m t W 2 + +

.
Clearly the significant part of the spectrum of the FM signal will lie in the interval
( )
c f p
f k m W 2 to
( )
c f p
f k m W 2 + + . (Note that about 92% of the energy of
a rectangular pulse of duration T sec, lies in the frequency range f
T
1
).
Hence, we can take as one possible value of the transmission bandwidth of an
FM signal, as
( ) ( )
T f p
FM
B k m W f W f W 2 4 2 4 2 2 = + = + = + (5.12)
For the wideband FM case, where f W >> , ( )
T
FM
B can be well approximated
by f 2 .
In the literature, other rules of thumb for FM bandwidths are to be found.
The most commonly used rule, called the Carsons rule, is
( ) ( )
T
FM
B f W 2 = + (5.13)

5.18

Fig. 5.7: Estimation of FM spectral width

Carsons rule gives a better bandwidth estimate than does Eq. 5.12 for the NBFM
( ) f W << case. This is in agreement with our result on NBFM, namely, its
bandwidth is approximately W 2 . In other cases, where f W << is not satisfied
(wideband and intermediate cases) Eq. 5.12 gives a better estimate than does
Eq. 5.13.


5.19
We now define the deviation ratio D, as
f
D
W
= (5.14)
Then Eq. 5.12 and 5.13 can be combined into
( ) ( )
T
FM
B W D k 2 = + (5.15)
where k varies between 1 and 2. As can be seen, Eq. 5.12 and Eq. 5.13 are
special cases of Eq. 5.15.

5.3 Tone Modulation
Let ( ) ( )
m m
m t A f t cos 2 =
Then ( ) ( )
m
I m
m
A
m t t sin =
, assuming ( )
I
m 0 =
( ) ( )
m
pe c c f m
m
A
s t A j t c t exp sin

= +

( )
f m
c c m
m
k A
A j t t
f
exp sin

= +

For tone modulation,
f m
f k A = and
m
W f = . In this case, the deviation ratio is
referred to as the modulation index and is usually denoted by the symbol .
That is, for tone modulation,
f m
m
k A
f
= and
( )
( )
m c
j t j t
pe c
s t A e e
sin

=

(5.16a)
( ) ( ) ( ) ( )
pe c c m
s t s t A t t Re cos sin = = +

(5.16b)

5.3.1 NBFM
( ) ( )
c c m
s t A t t cos sin = +

( ) ( ) ( ) ( ) ( ) ( ) { }
c c m c m
A t t t t cos cos sin sin sin sin =


5.20
NBFM: small .
Let be small enough so that we can use the approximations
( ) ( )
m
t cos sin 1 , and
( ) ( ) ( )
m m
t t sin sin sin .
Then, ( ) ( ) ( ) ( )
c c m c
s t A t t t cos sin sin =

(5.17a)
( ) ( ) ( )
c c c m c m
A t t t cos cos cos
2

= + +

(5.17b)
Corresponding expression for AM (Eq 4.7) is
( ) ( ) ( ) ( )
c c c m c m
AM
s t A t t t cos cos cos
2

= + + +

Eq. 5.17(b) can be written as
( )
( ) ( )
c m c m c
j t j t j t
c
c
A
s t A e e e Re
2
+

= +

(5.17c)

Fig. 5.8: Phasor diagram for NBFM with tone modulation

Using Eq. 5.17(c), we construct the phasor diagram for NBFM, shown in Fig. 5.8.
(We have taken the carrier phasor as the reference.) Comparing the phasor
diagram for the NBFM with that of the AM signal (Fig. 4.26), we make the
following observation.

In the case of AM, the resultant of the side-band phasors is collinear with
the carrier phasor whereas, it is perpendicular to the carrier phasor in NBFM.


5.21
It is this quadrature relationship between the two phasors of NBFM that
produces angle variations resulting in corresponding changes in ( )
i
f t .

5.3.2 WBFM
The exponential term in the brackets in Eq. 5.16(a) is a periodic signal
with period
m
f
1
. Expressing this term in Fourier series, we have

( )
m m
j t j n t
n
n
e x e
sin

=
=

where
( )
m
m m
m
f
j t j n t
n m
f
x f e e d t
1
2
sin
1
2

=

Let
m
t = . Then,

( ) j n
n
x e d
sin
1
2

(5.18)
The integral on the RHS of Eq. 5.18 is recognized as the nth order Bessel
function of the first kind and argument and is commonly denoted by the symbol
( )
n
J ; that is,
( )
( ) j n
n n
x J e d
sin
1
2

= =

(5.19a)
That is,

( )
( )
m m
j t j n t
n
n
e J e
sin

=
=
(5.19b)
Hence,

( )
( ) ( )
m
j t
n m
n
e J f nf
sin

=

(5.19c)
Using Eq. 5.19, we can express ( )
pe
s t as

5.22
( ) ( )
( )
c m
j f nf t
pe c n
n
s t A J e
2
+

=
=
(5.20)
As ( ) ( )
pe
s t s t Re =

, we obtain
( ) ( ) ( )
c n c m
n
s t A J f nf t cos 2
=
= +

(5.21)
Taking the Fourier transform of ( ) s t , we have
( ) ( ) ( ) ( )
c
n c m c m
n
A
S f J f f nf f f nf
2

=
= + + +

(5.22)

Properties of ( )
n
J : The following properties of ( )
n
J can be established.
1) ( )
n
J is always real (For all n and )
2) ( ) ( ) ( )
n
n n
J J 1

=
3) For small values of , ( )
( )
n
n
J
n
2
!

Exercise 5.3
Eq. 5.19(c) gives the Fourier transform of
( )
m
j t
e
sin
. Show that
a)
( )
( ) ( ) ( )
m
n j t
n m
n
e j J f nf
cos

=
+

b) ( ) ( ) ( )
c c m c n c m
n
A t t A J n t n cos cos cos
2
=

+ = + +

Hint:
( )
m
m m
j t
j t
e e
sin
2 cos

=
If ( )
( )
m
j t
x t e
sin
= , then
( )
m
j t
m
e x t
cos
2

.
Now use the Fourier transform properties for time reversal and time shift.

5.23
Hence, for small values of ,
( ) J
0
1
( ) J
1
2

( )
n
J n 0, 1 >
4) ( )
n
n
J
2
1
=
=

Fig 5.9 depicts the behavior of ( )
n
J . We now make the following observations
regarding the spectrum ( ) f 0 > of a tone modulated FM signal.
1) The spectrum of a tone modulated FM wave contains carrier component
and an infinite set of side-frequencies located symmetrically on either side
of the carrier at frequency separations of
m
nf n , 1, 2, =
2) For small , only ( ) J
0
and ( ) J
1
are significant. Then the FM spectrum
has only three components: at
c c m
f f f , . This situation corresponds to the
special case of NBFM.
3) The amplitude of the carrier component varies with according to ( ) J
0
.
Thus, in contrast to AM, this amplitude contains part of the message
information. Moreover, ( ) J
0
0 = for some values ( ) 2.4, 5.5, etc. = .
4) From Fig. 5.9(b), we find that ( )
n
J decays monotonically for
n
1 >
and
that ( )
n
J 1 << for
n
1 >>
.

Fig, 5.10 gives the spectral plots obtained from a spectrum analyzer with
2.4 . As can be seen from the plot, the carrier component is very nearly zero
(approximately 50 dB below the maximum value).


5.24
Fig 5.11(a) gives the theoretical plot of various spectral components for
4 . Fig. 5.11(b) is the spectral plot as observed on a spectrum analyzer. We
see from this plot that the third spectral component has the largest magnitude
which is an agreement with the theory. Let us normalize the dB plot with respect
to the largest magnitude; that is, we treat the largest magnitude as 0 dB and
compare the rest respect to with this value. For n 4 = , Fig. 5.11(a), we have
( ) J
4
0.281 = . Hence
10
0.281
20log 3.7
0.43
= and value indicated by the
spectrum analyzer agrees with this. Similarly, for n 5 = , we expect
10
0.132
20log 10.25
0.430
= ; the value as observed on the plot of Fig. 5.11(b) is in
close agreement with this. For n 1 = , theoretical value is
0.066
20log 17.1
0.43
=
dB. Spectrum analyzer display is in close agreement with this. The values of the
remaining spectral components can be similarly be verified.

From Eq. 5.21, the average power
av
P of the FM signal with tone
modulation is ( )
c
av n
n
A
P J
2
2
2

=
=
.


5.25

Fig. 5.9: Bessel Functions:
(a) ( )
n
J vs. for fixed n
(b) ( )
n
J vs.
n
for fixed


5.26

Fig. 5.10: Spectrum analyzer output with 2.4

But from property 4 of ( )
n
J , we have ( )
n
n
J
2
1
=
=
, which implies that the

av
P of the tone modulated FM signal is
c
A
2
2
. This result is true in general
(whether ( ) m t is a tone or not) because an FM signal is essentially a constant
amplitude cosine signal with time varying frequency. As the RMS value of a
sinusoid is independent of the frequency, we state that
( )
c
av
FM
A
P
2
2
= (5.23)
For large , if we assume that we can neglect ( )
n
J for n 2 > + , then
( ) ( )
T m m
FM
B nf f 2 2 2 = = + (5.24a)
This in agreement with Eq. 5.15, with k 2 = . If we neglect the values of ( )
n
J
for n 1 > + , then,
( ) ( )
T m
FM
B f 2 1 = + (5.24b)
which corresponds to Eq. 5.15 with k 1 = .


5.27

Fig. 5.11(a): ( )
n
J for 4 =
(b) Spectrum analyzer output for 4

Consider 5 = . From Appendix A5.1, we have ( ) J
7
5 0.053 = and
( ) J
8
5 0.018 = . P
1
, the average power of these two is,

5.28
( ) ( )
c c
A A
P
2 2
1
0.0028 0.0003 0.0031
2 2
= + = .
Hence,
( )
av
FM
P
P
1
0.0031 0.31% = = . As ( )
n
J 5 for n 9 are much smaller than
( ) J
8
5 , we see that practically the entire power of the FM signal is confined to the
frequency range
[ ]
c m
f f 6 . As such, we find that even Eq. 5.24(b) is a fairly
good measure of ( )
T
FM
B for tone modulation.

With tone modulation, it is possible to estimate the value of
T
B to the
accuracy desired. In this case, the magnitude of a spectral component depends
only on ( )
n
J and these Bessel coefficients have been tabulated extensively.
For a given , let us define transmission bandwidth such that it includes all those
spectral components whose magnitude is greater than one percent of the
unmodulated carrier amplitude. If we take
c
A 1 = , then

T sig m
B n f 2 = (5.24c)
where
sig
n is such that ( )
n
J 0.01 for
sig
n n . For example, if 20 = ,then
sig
n would be 25. (For 10, 5, = etc.,
sig
n can be found from the table in
Appendix A5.1.) Let
m
f f
0
= when 20 = . Let us tabulate
T
B for a few other
values of . We will keep f f
0
20 = , but reduce the value of by increasing
m
f . These are listed in Table 5.1. Also listed in the table are
T
B
,1
and
T
B
,2
where
( )
T m
B f
,1
2 1 = + (Carsons rule), and
( )
T m
B f
,2
2 2 = +


5.29
m
f 2
sig
n
T
B
,1 T
B
, 2 T
B
f
0

20 50
50 f
0
42 f
0
44 f
0

2 f
0

10 28
56 f
0
44 f
0
48 f
0

4 f
0

5 16
64 f
0
48 f
0
56 f
0

10 f
0

2 8
80 f
0
60 f
0
80 f
0

20 f
0

1 6
120 f
0
80 f
0
120 f
0

40 f
0

0.5 4
160 f
0
120 f
0
200 f
0

60 f
0

0.33 4
240 f
0
160 f
0
280 f
0

200 f
0

0.1 2
400 f
0
440 f
0
840 f
0

Table 5.1:
T
B of Eq. 5.24c for various values of with = f f
0
20

From the above table, we see that
i) For small values of (less than or equal to 0.5), f becomes less and less
significant in the calculation of
T
B . For 0.1 = , we find that

T
f f
B f
0
0
40 2 1
400 10
= =
ii) For small values of , bandwidth is essentially decided by the highest
frequency component in the input spectrum. As such, as 0 ,
T
B does
not go zero. In fact, in absolute terms, it increases. (From the table, with
20 = , we require a
T
B of f
0
50 where as with 0.1 = ,
T
B required is
f
0
400 .)
iii)
T
B
,1
, which is based on Carsons rule is in close agreement with
T
B only
for very small values of . Otherwise,
T
B
,2
is better.

It is interesting to note that for = 5, 10 and 20,
T
B
,2
(which is generally
considered to overestimate the bandwidth requirement) is less than
T
B as given

5.30
by Eq. 5.24(c). This is because
T
B of Eq. 5.24(c) neglects only those frequency
components whose magnitude is less than or equal to
c
A 0.01 . In other words,
the ratio

T
av
B
P
Power of any spectral component not included in

is less than
( )
c
c
A
A
2
4
2
0.01 2
10
2

=
With 5 = ,
T
B
,2
takes into account only upto ( ) J
7
5 and
( ) ( )
( )

=
c
c
J A
A
2
8
4
2
5 2
3.38 10
2
. That is,
T
B as given by Eq. 5.24(c) includes few
more spectral components in the bandwidth than that taken into account by
T
B
,2

resulting in a larger value for the bandwidth.

It is instructive to analyze the FM signal when ( ) m t is a sum of sinusoids.
This has been done in the Appendix A5.3. Another measure of bandwidth that is
useful in the study of frequency modulation is the rms bandwidth. This has been
dealt with in Appendix A5.4.

5.4 Phase Modulation
All the results derived for FM can be directly applied to PM. We know that,
for phase modulation,
( )
( )
( )
p i
i c
k d t
f t f m t
d t
1
'
2 2
= = +

(5.25a)
where ( ) m t
'
is the derivative of ( ) m t . Therefore,

p
p
k
f m
'
2
=
, where ( )
p
m m t
max
' '
= (5.25b)

5.31
( ) [ ]
T
PM
B f kW k 2 , 1 2 = + < <

p p
k m
kW
'
2
2

= +

(5.26)

With respect to f , there is one important difference between PM and
FM. For FM, as
f p
f k m = , it depends only on the peak value of the modulating
signal. Hence the bandwidth of WBFM is essentially independent of the spectrum
of ( ) m t (It has a weak dependence on W ). On the other hand, for PM, f is a
function of
p
m
'
and this value is strongly dependent on ( ) M f . Strong high
frequency components in ( ) M f would result in large values of
p
m
'
. Similarly,
predominance of lower-frequency components will result in a lower value of
p
m
'
.
As a result, for signals with strong high frequency components, ( )
T
PM
B will be
much larger than for the case of the signals with strong spectral components at
lower frequencies.

Example 5.5
To determine the frequency sensitivity of an FM source, the following
method has been used. The FM generator is fed a tone modulating signal
( ) m
A t
3
cos 4 10

. Starting near about zero,
m
A is gradually increased and
when
m
A 2 = V, it has been found that the carrier component goes to zero for
the first time. What is the frequency sensitivity of the source? Keeping
m
A at 2 V,
frequency
m
f is decreased until the carrier component goes zero for the second
time. What is the value of
m
f for this to happen?


5.32
With tone modulation, we know that carrier goes to zero for the first time
when 2.4 = . (That is, the smallest value of for which ( ) J
0
0 = is
2.4 = .)
But
m
f
f
2.4
= = . That is,

m f p
f f k m 2.4 4.8kHz = = =
As
p
m is 2 V, we have
f
k
4.8
2.4
2
= = kHz/V

The carrier component is zero for the second time when 5.5 = . Hence,

m
f
3
4.8 10
5.5

=
or
m
f
3
4.8 10
872
5.5
= = Hz

Example 5.6
A 1.0 kHz tone is used to generate both an AM and an FM signal.
Unmodulated carrier amplitude is the same for both AM and FM. The modulation
index of FM is 8. If the frequency components at ( )
c
f 1000 Hz have the
same magnitude in AM as well as FM, find the modulation index of AM.

For AM, the magnitude of the spectral components at ( )
c
f 1000 Hz is
c
A
2
. For FM, the magnitude of the spectral components at ( )

c
f 1000 Hz is
( )
c
A J
1
8 .
( ) ( )
c
c
A
A J J
1 1
8 2 8
2
= =
0.47 =


5.33
Example 5.7
In this example we will show that the bandwidth of the PM signal is
strongly dependent on ( ) M f whereas this dependence is weak in the case of an
FM signal.

Let ( ) ( ) ( ) m t f t f t
1 2
cos 2 0.85 cos 2 = + be used to generate an FM as
well as a PM signal. Modulator constants are
f
k 5 = kHz/V and
p
k 10 = rad/V.
Let us compute (a)
( )
T
FM
B and (b)
( )
T
PM
B for the following combinations
of f
1
and f
2
.
i) f
1
500 = Hz and f
2
700 = Hz
ii) f
1
1000 = Hz and f
2
1400 = Hz
We will assume that ( )
n
J can be neglected if n 2 > + .

a)
( )
T
FM
B
Case i) f
1
500 = Hz and f
2
700 = Hz
From Eq. A5.3.1 (Appendix 5.3), we have
( ) ( ) ( ) ( )
c m n c
m n
s t A J J m n t
1 2 1 2
cos = + +

3
1
5 10
10
500
= =

3
2
5 10 0.85
6
700

=
We shall take into account ( )
m
J
1
upto ( ) m
1
12 2 = = + and
( )
n
J
2
upto ( ) n
2
8 2 = = + . Then,

( )
( )
T
FM
B 2 12 500 8 700 = +

3
23.2 10 = Hz

5.34
In the equation for ( ) s t above, we see that the magnitude of
any spectral component depends on the product. ( ) ( )
m n
J J
1 2
. Let
us calculate the ratio
( ) ( )
( ) ( )
m n
J J
J J
13 8
max
10 6
10 6

so as to get an idea of the
magnitude of the spectral components that have been neglected in
calculating
( )
T
FM
B .
Let ( ) ( ) A J J
13 9
10 6 0.03 0.02 = =
0.0006 =
From the tables (Appendix A5.1), we find that the maximum value of
the product ( ) ( )
m n
J J 10 6 occurs for m 8 = and n 5 = . Let,
( ) ( ) B J J
8 5
10 6 0.318 0.362 = =
0.115 = ,
Then,
A
B
0.0052 = .
Case ii) f
1
1000 = Hz and f
2
1400 = Hz
Now, we have
1
5 = and
2
3 .
If we account for ( )
m
J
1
and ( )
n
J
2
upto m 7 = and n 5 = ,
then

( ) ( ) T
FM
B
3
2 7 10 5 1400 = +
=
3
28 10 Hz

Of course the maximum value of ( )
m
J
1
and ( )
n
J
2
occurs for
m 4 = and n 2 = , and this product is
( ) ( ) C J J
4 2
5 3 0.391 0.486 0.19 = = =
However, ( ) ( ) D J J
8 6
5 3 0.018 0.0114 = =
0.0002 =

5.35
Hence
D
C
0.0002
0.001
0.19
=
This is much less than ( ) A B which implies that we have taken into
account more number of spectral components than case (i) above. If
we restrict ourselves to the number of spectral components upto
m 1 = + , then,

( ) ( )
( ) ( )
J J
J J
7 6
4 2
5 3
0.003
5 3
= which is closer to ( ) A B than 0.001.
Then,

( )
( )
T
FM
B
3
2 6000 5 1400 26 10 = + = Hz
which is fairly close to the previous value of
3
23.2 10 Hz.
b)
( )
T
PM
B
( ) ( ) ( ) ( ) m t f f t f f t
1 1 2 2
'
2 sin 2 0.85 2 sin 2 = +

Now ( ) m t
'
frequency modulates the carrier
p
p
k
f
k
f
1
1
1
2
10
2

= = =

Similarly,
p
k
2
0.85 8.5 = =
Case i) f
1
500 = Hz and f
2
700 = Hz

( )
( )
T
PM
B
3
2 12 500 10 700 26 10 = + = Hz
Case ii) f
1
1000 = Hz and f
2
1400 = Hz
As
1
and
2
remains unchanged, we have

( ) ( ) T
PM
B
3
2 12 10 10 1400 = +

3
52 10 = Hz

5.36
This is twice
( )
T
PM
B of case (i). This example shows that the
bandwidth of the PM signals is strongly dependent on ( ) M f .

5.5 Generation of FM
We had earlier identified two different categories of FM, namely, NBFM
and WBFM. We shall now present the schemes for their generation.

Exercise 5.4

Fig. 5.12: Scheme for the Exercise 5.4

a) Let
c
f be the frequency of the modulator when ( ) m t 0 = . If ( ) s t can be
expressed in the form
( ) ( )

=

cos ' sin 2
c c m
s t A t t ,
what are the values of
c
f
'
and ?
b) What are the frequency components in the spectrum of ( ) s t ?

Ans:
f m
c c
k A
f f
2
'
2

= +

f m
m
k A
f
2
2
=

5.37
5.5.1 Narrowband FM
One of the principal applications of NBFM is in the (indirect) generation of
WBFM as explained later on in this section. From the approximation (5.10) we
have,
( ) ( ) ( ) ( )
c c f I c
s t A t c m t t cos sin

The system shown in Fig. 5.13 can be used to generate the NBFM signal.

Applying ( ) m t directly to the balanced modulator, results in NBPM with a
suitable value for
p
k .

Fig. 5.13: Generation of NBFM signal

5.5.2 WBFM: Indirect and direct methods
There are two distinct methods of generating WBFM signals: a) Direct FM
b) Indirect FM. Details on their generation are as follows.

a) Indirect FM (Armstrongs method)
In this method - attributed to Armstrong - first a narrowband FM signal is
generated. This is then converted to WBFM by using frequency multiplication.
This is shown schematically in Fig. 5.14.


5.38

Fig. 5.14: Generation of WBFM (Armstrong method)

The generation of NBFM has already been described. A frequency
multiplier is a nonlinear device followed by a BPF. A nonlinearity of order n can
give rise to frequency multiplication by a factor of n . For simplicity, consider a
square law device with output ( ) ( ) y t x t
2
= where ( ) x t is the input. Let ( ) x t be
the FM signal given by,
( ) ( ) x t t cos =

, where ( ) ( )
t
c f
t t k m d 2

= +

Note that we have dropped the subscript i from ( )
i
t . Then,
( ) ( ) y t t
2
cos =

( )
{ }
t
1
1 cos 2
2
= +

( )
t
c f
t k m d
1 1
cos 2 4
2 2

= + +

(5.27)

The DC term in Eq. 5.27 can be filtered out to give an FM output with the
carrier frequency
c
f 2 and frequency deviation twice that of the input FM signal.
An input-output relation of the type ( ) ( ) ( ) ( )
n
n
y t a x t a x t a x t
2
1 2
.... = + + will
give rise to FM output components at the frequencies
c c c
f f nf , 2 , ...., with the
corresponding frequency deviations f f n f , 2 , ...., , where f is the
frequency deviation of the input NBFM signal. The required WBFM signal can be
obtained by a suitable BPF. If necessary, frequency multiplication can be
resorted to in more than one stage. The multiplier scheme used in a commercial
FM transmitter is indicated in Fig. 5.15.


5.39

Fig. 5.15: Multiplier chain used in typical commercial FM transmitter

The carrier frequency of the NBFM signal
c
f
1
, is 200 kHz with the corresponding
f
1
25 = Hz. Desired FM output is to have the frequency deviation f
4
75
kHz and a carrier
( )
c
f
4
of 91.2 MHz.

To obtain f
4
75 = kHz starting from f
1
25 = Hz, we require a total
frequency multiplication of
( )
3
75 10
3000
25
= . In the scheme of Fig. 5.11, this

has been accomplished in two stages, namely, multiplication by 64 followed by
multiplication by 48, giving a total multiplication by the factor 64 48 3072 = .
(Actually each stage of multiplication is implemented by a cascade of frequency
doublers or triplers. Thus multiplication by 64 is obtained by 6 doublers in
cascade where as multiplication by 48 is implemented by a cascade of a
frequency tripler and 4 doublers.) Multiplication of
c
f
1
200 = kHz by 3072 gives a
carrier frequency
c
f
4
614.4 = MHz. As the final required carrier frequency is
91.2 MHz, a frequency conversion stage is used to down convert
c
f
2
(12.8 MHz)
to
c
f
3
(1.9 MHz). In this process of down conversion, frequency deviation is
unaffected ( ) f f
2 3
1.6kHz = = . The possible drawbacks of this scheme are
the introduction of noise in the process of multiplication and distortion in the

5.40
generation of NBFM signal especially for low modulating frequencies as
m
f
f

could become excessive.

Example 5.8
Armstrongs method is to be used to generate a WBFM signal. The NBFM
signal has the carrier frequency
c
f
1
20 = kHz. The WBFM signal that is required
must have the parameters
c
f 6 = MHz and f 10 = kHz. Only frequency
triplers are available. However, they have a limitation: they cannot produce
frequency components beyond 8 MHz at their output. Is frequency conversion
stage required? If so, when does it become essential? Draw the schematic block
diagram of this example.

Total frequency multiplication required
6
3
6 10
300
20 10
= =
. Using only
frequency triplers, we have
5
3 243 = and
6
3 729 = . Hence a set of six
multipliers is required. But these six cannot be used as a single cascade
because, that would result in a carrier frequency equal to
3 6
20 10 3 14.58 =
MHz and the last multiplier cannot produce this output. However, cascade of 5
triplers can be used. After this, a frequency conversion stage is needed.

Fig 5.16: Generation of WBFM from NBFM of example 5.8

Block diagram of this generation scheme is shown in Fig. 5.16. As the final
frequency deviation required is 10 kHz, the NBFM must have a

5.41
( ) f
3
1
10 10
13.71
729
= = Hz. After the frequency conversion stage, we have

one more stage multiplication by 3. If
c
f
3
is the carrier frequency at the mixer
output, then
c
f
3
3 6 = MHz
c
f
3
2 = MHz. Assuming that
Lo
f is greater than
incoming carrier frequency of 4.86 MHz, we require
Lo
f 6.86 = MHz so that the
difference frequency component is 2 MHz.

Exercise 5.5
In the indirect FM scheme shown in Fig. 5.17, find the values of
c i
f
,
and
i
f for i 1, 2and3 = . What should be the centre frequency, f
0
, of the BPF.
Assume that
LO c
f f
,2
> .


Exercise 5.6
In the indirect FM scheme shown in Fig. 5.18, find the values of the
quantities with a question mark. Assume that only frequency doublers are
available. It is required that
LO c
f f
,2
< .


5.42
b) Direct FM
Conceptually, the direct generation of FM is quite simple and any such system
can be called as a Voltage Controlled Oscillator (VCO). In a VCO, the oscillation
frequency varies linearly with the control voltage. We can generate the required
FM signal from a VCO by using the message signal ( ) m t in the control voltage.
We shall explain below two methods of constructing a VCO.

One can construct a VCO by using an integrator and hysteric comparator
(such as Schmitt trigger circuit). Method 1 below gives the details of this scheme.
Another way of realizing a VCO is to vary the reactive component of a tuned
circuit oscillator such as Colpitts oscillator or Hartley oscillator. Method 2 below
gives the details of this scheme.
Method 1: Consider the scheme shown in Fig. 5.19(a). The Schmitt trigger
output ( ) y t is V
0
+ when the integrator output ( ) x t is increasing, and V
0

when ( ) x t is decreasing. Further, ( ) y t changes from V
0
+ to V
0
when ( ) x t
reaches E
0
+ and from V
0
to V
0
+ when ( ) x t reaches E
0
. The relationship
between ( ) x t and ( ) y t is shown in Fig. 5.19(b). The electronic switch is
designed such that it is in position 1 when ( ) y t V
0
= and goes to position 2
when ( ) y t V
0
= . Let us first examine the case when the input ( )
m
v t is a
positive constant v
0
. Consider the situation at t 0 = when ( ) y t has just
switched to V
0
and the switch going from position 2 to position 1. At this point,
( ) x t has attained the value E
0
. Then, for t t
1
0 < , we have
( )
t
x t E v d
RC
0 0
0
1
= +

When t t
1
= , let ( ) x t become E
0
. Then ( ) y t goes to V
0
and the electronic
switch assumes position 2. The value of t
1
can be obtained from

5.43

t
E E v d
RC
1
0 0 0
0
1
= +

or
RCE
t
v
0
1
0
2
=

Fig. 5.19: Direct FM generation (method 1)

The output ( ) x t now keeps decreasing until t t
2
= when ( ) x t E
0
= . It is easy
to see that ( )
RCE
t t
v
0
2 1
0
2
= . That is, ( ) x t and ( ) y t are periodic with period
RCE
v
0
0
4
or the fundamental frequency of these waveforms is
RCE
f
v
1
0
0
0
4

=

.
Note that f
0
depends on the input signal v
0
. If v E
0 0
= then f
RC
0
1
4
= . By
properly choosing the values of R and C, we can have
c
f f
RC
0
1
4
= = , where

5.44
c
f is the required carrier frequency. If the triangular wave ( ) x t is the input to a
narrowband BPF tuned to
c
f , the output ( ) s t is ( )
c c
A f t cos 2 . Then, by making
( ) ( )
m
v t v k m t
0
= + where k is a constant such that ( ) k m t v
0
< , we can have
the instantaneous frequency of ( ) s t as ( )
c f
f k m t + , which is the desired result.
Method 2: The oscillation frequency f
0
of a parallel tuned circuit with inductance
L and capacitance C is given by
f
LC
0
1
2
=
or
LC
0
1
=
Let C be varied by the modulating signal ( ) m t , as given by
( ) ( ) C t C k m t
0
=
where k is an appropriate constant.
Then ( )
( )
i
t
k m t
LC
C
0
0
1
1
=

( ) k m t
C LC
0 0
1
1
2

+

when
( ) k m t
C
0
1 <<
Let
c
LC
0
1
= . Then,
( ) ( )
i c f
t c m t = + , where
c
f
k
c
C
0
2
= .
One of the more recent devices for obtaining electronically variable capacitance
is the varactor (also called varicap, or voltacap). In very simple terms, the
varactor is a junction diode. Though all junction diodes have inherent junction
capacitance, varactor diodes are designed and fabricated such that the value of
the junction capacitance is significant (varactors are available with nominal
ratings from 0.1 to 2000 pF). Varactor diodes, when used as voltage-variable
capacitors are reverse biased and the capacitance of the junction varies
inversely with the applied (reverse) voltage.


5.45

Fig. 5.20: Direct FM generation (method 2)

Consider the scheme shown in Fig. 5.20. The frequency of the oscillator
output depends on L
1
and
eq
C where
eq d
C C C
1
|| = ,
d
C being the capacitance
of the varactor D.
B
V reverse biases D such that when ( ) m t 0 = ,
eq
C is of the
correct value to result in output frequency
c
f . When the message signal ( ) m t is
on,
d
C can be taken as, ( )
d
C C k m t
0
'
= where C
0
'
is the value of
d
C , with
( ) m t 0 = . Hence ( ) C t of the oscillator circuit is,
( )
( )
( ) ( ) C t C C k m t C k m t
1 0 0
'
= + = , where C C C
0 1 0
'
= +

The other components in Fig. 5.14 have the following functions. C
2
is a
blocking capacitor to isolate the DC level of the varactor from the rest of the
modulator circuit. Note that
d
C C
2
>> or C
1
and is essentially a short at the
required FM frequencies. RFC is an RF choke that prevents the RF energy of the
oscillator circuit from feeding into the audio transformer. R
1
limits the current in
the circuit in the event that the peaks of the audio signal voltage exceed the
voltage of the DC source and momentarily forward bias the diode D.

Direct FM generation can produce sufficient frequency deviation and
requires little frequency multiplication. However, it suffers from the carrier

5.46
frequency instability as the carrier frequency is not obtained from a highly stable
oscillator. Some additional circuitry is required to achieve frequency stability.

Voltage controlled oscillators are available in the form of IC chips. For
example XR-2206 of the EXAR Corporation, USA has an operating frequency
range of 0.01 Hz to 1 MHz. Another chip with an operating frequency range of
about 1 MHz is LM 566. Texas Instruments CD 4046 is an inexpensive VCO chip
with a normal operating frequency range upto 1.4 MHz. (Actually CD 4046 is a
PLL and VCO is a part of it. PLLs are discussed in sec. 5.5.4.) To make the VCO
chip functional, what is required is an external capacitor and one or two external
resistors. CD 74HC7046 is another PLL chip with VCO. Free running frequency
of the VCO is 18 MHz. Koster, Waldow and Ingo Wolf describe a VCO operating
in the frequency range 100 MHz to 4 GHz [1]. Donald Tillman describes a new
VCO design (called a quadrature trapezoid VCO) especially suited for electronic
music applications. Details are available at:
http://www.till.com/articles/QuadTrapVCO/discussion.html.


5.47

Exercise 5.7
Consider the circuit of Fig. 5.20 for the direct generation of FM. The
diode capacitance
d
C , is related to the reverse bias as,

d
d
C
v
100
1 2
=
+
pF
where
d
v is the voltage across the varactor. Let
B
V 4 = V and
( )
( )
m t t
3
0.054 sin 10 10

=

. It is given that C
1
250 = pF and the circuit
resonates at 2 MHz when ( ) m t 0 = .
a) Show that, by using the binomial approximation,
d
C can be put in the
form
( ) d
C t
10
12 3
10
0.2 10 sin 10 10
3

=

b) Show that
( )
( ) ( ) i
f t t
6 3
2 10 705.86 sin 10 10

= +

(Note that C
2
is a short circuit at the frequencies being generated.)
c) Let
c
A be the amplitude of oscillations of the VCO. Write the expression
for the generated FM signal.

5.48

Exercise 5.8
A simple variation of the circuit of Fig. 5.20 is shown in Fig. 5.21.

Fig. 5.21: A variation of the VCO circuit of Fig. 5.14

The varactor diodes D
1
and D
2
are connected back-to-back. This
arrangement helps to mitigate the effect of the RF signal of the tuned circuit
(also called tank circuit) driving a single diode into conduction on its peaks
which will change the bias voltage (and thereby the frequency that is
generated). Of course, as far as the tuned circuit is concerned, the two diodes
are in series; this means the capacitance of the combination is one-half that of
a single diode.

Let
B
V 5 = V, ( )
( )
m t t
3
0.5cos 2 10 = . Calculate the carrier
frequency
c
f , ( )
i
f
max
and ( )
i
f
min
from the above scheme, given that C
1
50 =
pF, L
1
50 = H and the capacitance of the varactor diode
d
C , follows the
relation

d
d
C
v
50
= pF
where
d
v is the voltage across the diode. Do not use any approximations.
What are the values of ( )
i
f
max
and ( )
i
f
min
, if you use binomial approximation.
Ans:
c
f 91 kHz, ( )
i
f
max
91.4 kHz and ( )
i
f
min
90.6 kHz, without any
approximation.

5.49
5.6 Demodulation of FM
A variety of techniques and circuits have been developed for
demodulating FM signals. We shall consider a few of these techniques falling
under the following categories:
1) FM-to-AM conversion
2) Phase shift discrimination
3) Zero crossing detection
4) Phase Locked Loop (PLL)

5.6.1 FM-to-AM conversion
The instantaneous frequency of an FM signal is given by ( )
i c f
f f k m t = + .
Hence a frequency selective network with a transfer function of the from
( ) H f f = + , (f 0 > , and and are constants) over the FM band would
yield an output proportional to the instantaneous frequency. That is, the circuit
converts the frequency deviation into a corresponding amplitude change, which
in this case is proportional to ( ) m t , the message signal. It is assumed that the
time constant of the network is small enough in comparison with the variations in
the instantaneous frequency of the FM signal. We shall now indicate three ways
of implementing this demodulation scheme.

Consider the scheme shown in Fig. 5.22 where
d
d t
represents a band-
pass differentiator with the magnitude characteristic ( ) H f f = + , (for f 0 > ),
over the required bandwidth. BPL
1
is a Band-Pass Limiter which eliminates
amplitude fluctuations from the received FM signal.

1
Band-pass limiters have been analyzed in section 5.7.

5.50

Fig. 5.22: Schematic of an FM demodulator based on FM - to - AM conversion

( ) s t is the constant amplitude FM signal to be demodulated. ( ) s t
'
, the
output of the differentiator, is given by
( ) ( ) ( )
t
c c f c f
s t A k m t t k m d
'
2 sin 2

= + +

(5.28)

Eq. 5.28 represents a signal that is both amplitude and frequency modulated.
The envelope of ( ) ( )
c c f
s t A k m t
'
2 = +

(we assume that
f p c
f k m f = ;
hence, ( )
c f
k m t 2 0 +

). As
c c
A represents a DC term, signal ( ) m t can
be obtained from ( ) s t
'
, after the DC-block.

The need for a BPL is as follows. Assume that the received FM signal
(with amplitude fluctuations) is applied directly as the input to the differentiator.
Let ( )
c
A t denote the envelope of the FM signal. Then, there would be an
additional term,
( )
c
d A t
d t
on the RHS of Eq. 5.28. Even if this term were to be
neglected, the envelope of ( ) s t
'
would be ( ) ( )
c c f
A t k m t 2 +

, which implies
that the envelope of ( ) s t
'
does not contain a term proportional to ( ) m t .
Therefore, it is essential to maintain the FM envelope at a constant level.
(Several factors such as channel noise, fading etc. cause variations in
c
A ).
Band-pass limiter eliminates the amplitude fluctuations, giving rise to an FM
signal with constant envelope.
We shall now indicate some schemes to implement this method of demodulation.


5.51
Scheme 1: Circuit implementation of the scheme of Fig. 5.22, can be carried out
by constructing an op-amp differentiator circuit followed by an envelope detector.
Scheme 2 (Slope detection): Another way of implementing the FM-to-AM
conversion scheme is through a simple tuned circuit followed by an envelope
detector. The transfer characteristic of a tuned circuit in a small region off
resonance is approximately linear.

Fig. 5.23: Magnitude characteristics of a tuned circuit

In Fig. 5.23, we have shown the resonance characteristic of a tuned
circuit. The parts of the characteristic shown in red have been drawn as straight
lines. (This is a good approximation to the actual characteristic.) Assuming that
( ) m t 0 > produces an increase in the instantaneous frequency, we can use the
straight line segment between A and B (for f 0 > ; for f 0 < , we have the
corresponding segment, A
'
to B
'
) for demodulation purposes. Let us assume
that the FM signal to be demodulated has the carrier frequency
c
f f
f
1 2
2
+
= and
T
c
B
f f
1
2
where as
T
c
B
f f
2
2
+ . This is shown in Fig. 5.24.


5.52

Fig. 5.24: Segment A to B of Fig. 5.23

As can be seen from the figure, changes in the instantaneous frequency
will give rise to corresponding changes in the output amplitude. Envelope
detection of this output will produce the required message signal.

Fig. 5.25: Tuned circuit demodulator

Consider the demodulator circuit shown in Fig. 5.25. The primary of the
coupled circuit is tuned to
c
f
1
whereas the secondary is tuned to f
0
, where
c
f f
0
> . If, over the frequency range
T
c
B
f f
2
, the output of the primary is
fairly constant, then, we can expect the ( ) m t
to resemble ( ) m t fairly closely.

This method of demodulating an FM signal is also known as slope detection.

Though the method is fairly simple, it suffers from the following
disadvantages: the resonant circuits on the primary and secondary side are

1
In a superheterodyne receiver, the detector stage follows the IF stage. As such,
c
f gets
converted to
IF
f .

5.53
tuned to two different frequencies; the frequency range of linear amplitude
response of a tuned circuit is somewhat limited, making the demodulation of a
WBFM unsatisfactory.
Scheme 3 (Balanced slope detection): This latter problem is partially overcome
by using a balanced configuration (balanced slope detection). The scheme,
shown in Fig. 5.26(a) has three tuned circuits: two on the secondary side of the
input transformer and one on the primary. The resonant circuit on the primary is
tuned to
c
f whereas the two resonant circuits on the secondary side are tuned to
two different frequencies, one above
c
f and the other, below
c
f . The outputs of
the tuned circuits on the secondary are envelope detected separately; the
difference of the two envelope detected outputs would be proportional to ( ) m t .

Though the balanced configuration has linearity over a wider range (as
can be seen from Fig. 5.26(b), the width of linear frequency response is about
B 3 , where B 2 is the width of the 3-dB bandwidth of the individual tuned circuits)
and does not require any DC bock (The two resonant frequencies of the
secondary are appropriately selected so that output of the discriminator is zero
for
c
f f = ), it suffers from the disadvantage that the three tuned circuits are to be
maintained at three different frequencies.


5.54

Fig. 5.26: Balanced slope detection (a) circuit schematic (b) response curve

Example 5.9
Let the FM waveform of Fig. 5.4(b) be the input to a differentiator followed
by an envelope detector. Let us find an expression for the output of the
differentiator and sketch the output of the envelope detector. We shall assume
that the differentiator will produce the appropriate step change in the output for
sudden changes in the input frequency.


5.55
Let the FM waveform between the time instants ( ) t t
1 2
, be taken as
( ) f t
1
cos 2 and that between time instants ( ) t t
2 3
, be taken as ( ) f t
2
cos 2 ,
where f f
2 1
> .
Then the output of the differentiator is,
t t t
1 2
< < : ( ) ( ) f f t A f t
1 1 1 1
2 sin 2 sin 2 =
t t t
2 3
< < : ( ) ( ) f f t A f t
2 2 2 2
2 sin 2 sin 2 =
with A A
2 1
> and A A
1 2
, 0 > .
(Imagine Figure 5.4(a), with two different frequencies.) The output of the ED is
proportional to A
1
during t t t
1 2
< < and proportional to A
2
during t t t
2 3
< < .
Taking the constant of proportionality as unity, we have the output of ED as
shown in Fig 5.27

Fig. 5.27: Output of the ED

After DC block, we obtain the modulating square wave signal. Hence,
d
d t

followed by ED with a DC block, would act as a demodulator for the FM.

Example 5.10
a) Consider the RC network shown in Fig. 5.28. For the values of R and C
given, we will show that for frequencies around 1.0 MHz, this can act as a
differentiator.


5.56

Fig. 5.28: The RC network of Example 5.10

b) Let us find the condition on R and C such that this network can act as a
differentiator for frequencies around some frequency
c
f .

c) For the above network, let ( ) ( ) ( )
in c f I
v t s t A t c m t
6
cos 2 10

= = +

where ( )
f
k m t
6
10 << Hz. If ( )
out
v t is envelope detected, we will show
that, we can recover ( ) m t from the ED output.

a) ( )

=
+
j f RC
H f
j f RC
2
1 2

RC
12 9
100 50 10 5 10

= = and for frequencies around 1.00 MHz,
f RC
6 9 3
2 2 10 5 10 10 10
100

= = =
As 1
100
<< , we can take ( ) H f as

( ) H f j f RC 2 , which is a differentiator.

( )
j f
9
5 10 2
=
b) for frequencies around some
c
f , we require, << f RC 2 1.
c) With ( ) s t as the input, ( )
out
v t is
( ) ( )
out
d
v t s t
d t
9
5 10
=


5.57
( ) ( )
{ } c f I f
A t c m t k m t
9 6 6
5 10 sin 2 10 2 10 2

= + +

Output of the ED ( )
c f
A k m t
9 6
5 10 2 10 2

= +

( )
f
c
k
A m t
6 9
6
2 10 5 10 1
10

= +

.
( )
c f
A k
m t
6
1
100
10

= +

5.6.2 Phase shift discriminator
This method of FM demodulation involves converting frequency variations
into phase variations and detecting the phase changes. In other words, this
method makes use of linear phase networks instead of the linear amplitude
characteristic of the circuits used in the previous method. Under this category, we
have the Foster-Seely discriminator (and its variant the ratio detector) and the
quadrature detector. Foster-Seely discriminator and the ratio detector have
been discussed in appendix A5.2. We shall now explain the operation of the
quadrature detector.

Quadrature Detector (QD): Consider the FM signal
( ) ( )
c c
s t A t t cos = +

where
( ) ( )
t
f
t k m d 2

=

Then ( )
( )
( )
f
d t
t k m t
d t
'
2
= =
( ) ( ) t t t
t
1

, provided t is small.

( ) t t can be obtained from ( ) t with a delay line or a network with
linear phase.


5.58
Consider the scheme shown in Fig. 5.29. ( ) H f has the transfer
characteristic given by
( )
( ) j A f
H f e =
where ( ) A f , the phase function, can be well approximated by a linear phase
function, namely,
( )
( )
( )
T
c c
T
c c
B
f f t f f f
A f
B
f f t f f f
2 , 0,
2 2
2 , 0,
2 2

> <
+ < + <
(5.29)

Fig. 5.29: Block diagram of a quadrature detector

With ( ) H f specified as above, let us calculate the output of the network with the
input ( )
( ) ( )
c c
j t t j t t
c c c
e e
A t t A cos
2
+ +

+
+ =

. We can take
( )
c
j t t
c
A e
1
2
+

to represent the positive part of the spectrum. As ( ) H f has the
linear phase term
j f t
e
2
, (which will contribute a delay of t ), the term at the
output of the filter is

( ) ( ) ( )
c c c
j t t t t t j t t t
c c
A e A e
2 2
1 1
2 2

+ + +

= .

5.59
Similarly, the output corresponding to the negative part of the input spectrum
would be
( )
c
j t t t
c
A e
2
1
2

+

. Combining these two terms, we have at the
output of the filter, the quantity
( ) ( )
c c c c
A t t t A t t t cos sin
2

+ = +

.
The
2
phase shift provided by ( ) H f at

c
f f = gives rise to the term
quadrature detector. Multiplication of this output by ( )
c c
A t t cos +

followed
by low pass filtering yields the output ( ) y t proportional to ( ) ( ) t t t sin

.
Assuming t to be very small, ( ) y t can be approximated as,
( ) ( ) ( ) { }
y t k t t t
1

( ) k t t
1
'
=
( ) k m t
2
=
where k
1
and k
2
are constants with =
f
k c k t
2 1
.
Several tuned circuits, when properly designed, can provide the band-
pass response with the phase characteristic given by Eq. 5.29. Consider the
series RLC circuit shown in Fig. 5.30.

Fig. 5.30: A network to provide the phase of Eq. 5.29


5.60
( )
( )
( )
i
V f
j f C
H f
V f
R j f L
j f C
0
1
2
1
2
2
= =
+ +

( ) j f RC f LC
2
1
2 2 1
=
+
(5.30)
Let f
LC
0
1
2
=
and
b
R
f
L 2
=
. Then, Eq. 5.30 reduces to

( )
( )
( )
b
f
H f
f f j f f
2
0
2 2
0
=
+

Consider f 0 >
( )
b
f f
H f
f f
1
2 2
0
arg tan

b
f f
f f
2 2
1
0
tan
2

=

( ) ( )
b
f f f f
f f
1 0 0
tan
2

+
=

Let the circuit be operated in a small frequency interval, around f
0
so that f f
0
.
Then, f f f
0 0
2 + ,
( )
b b
f f
f f f
0
2
+
and
( )
( )
b
f
H f
f
1
2
arg tan
2

( ) f Q
f
1
0
2
tan
2

where
b
f
Q
f
0
= and ( ) f f f
0
=
If
( ) f Q
f
0
2
1
<< , then,
( )
( ) Q f
H f
f
0
2
arg
2

(5.31)

5.61
As ( ) H f is the frequency response of a network with real impulse response, we
will have
( ) ( )
=

H f H f arg arg
By choosing
c
f f
0
= , we have ( )
c
f f f = and
c
Q
t
f
1
=

There are other circuit configurations (other than the one given in Fig.
5.30) that can provide the required phase shift for the quadrature detector. The
circuit given in Fig. 5.31 is another possibility.

Fig. 5.31: Another phase shift circuit for the QD

Here C provides a very high reactance at the carrier frequency and the parallel
tuned circuit resonates at
c
f f = .

Quadrature detector is well suited to IC construction. Companies such as
Signetics, have built high quality FM receivers using QD as the FM demodulator.
Some details on these FM receivers can be found in Roddy and Coolen [2].

5.6.3 Zero-crossing detection
Consider the scheme shown in Fig. 5.32(a). When the input to the hard
limiter is a sine wave of period T
0
, it produces at its output a square wave of the
same period, with the transitions in the square wave occurring at the zero

5.62
crossings of the input sine wave. When the input to the hard limiter is an FM
signal, the hard limiter output appears as a square wave of varying frequency.
The hard limiter output ( )
H
v t , triggers a monostable pulse generator, which
produces a short pulse of amplitude A and duration at each upward (or
downward) transition of ( )
H
v t .
Consider a time interval T such that
c
T
W f
1 1
>> >> , where W is the
highest frequency present in the input signal. We shall assume that during the T
sec. interval, the message signal ( ) m t is essentially constant which implies that
instantaneous frequency ( )
i
f t is also a near constant (Fig. 5.32(b)). Then the
monostable output, ( )
p
v t , looks like a pulse train of nearly constant period. The
number of such pulses in this interval ( )
T i
n T f t with an average value,
( ) ( )
t
I p
t T
v t v d
T
1

( )
T i
n A A f t
T
1
=
After the DC block, we will have ( ) y t being proportional to ( ) m t , which is the
desired result.


5.63

Fig 5.32: Zero-crossing detector

5.6.4 FM demodulation using PLL
PLL is a versatile building block of the present day communication
systems. Besides FM demodulation, it has a large number of other applications
such as carrier tracking (in schemes with a pilot carrier and even suppressed
carrier; recall the squaring loop, Section 4.2.3), timing recovery, frequency
synthesis etc.

The basic aim of a PLL is to lock (or synchronize) the instantaneous angle
of a VCO output to the instantaneous angle of a signal that is given as input to
the PLL. In the case of demodulation of FM, the input signal to PLL is the
received FM signal.

In its simplest form, PLL consists of a phase detector and a VCO
connected as shown in Fig. 5.33(a).


5.64

Fig. 5.33: Phase lock loop (a) Basic configuration
(b) PD of (a) in functional form
(c) PLL as a negative feedback loop

PD makes the comparison of the instantaneous phase of ( ) x t and ( ) v t ,
and is designed such that ( )
v
t , the instantaneous phase of ( ) v t locks on to
( )
x
t , the instantaneous phase of ( ) x t , if necessary with some fixed phase
difference. (This will become evident later.)

A number of circuits are available which have been used as phase
detectors. In the context of FM demodulation, the most common PD is that of an
analog multiplier followed by a LPF (Fig. 5.33(b)). The scheme resembles closely
that of a negative feedback amplifier configuration, shown in Fig. 5.33(c). In this
figure, s is the variable of the Laplace transform, ( ) G s is system function in the

5.65
forward path whereas ( ) H s is the network in the feedback path. A properly
designed negative feedback system ensures that the error quantity ( ) e t
2
is fairly
close to zero so that ( ) ( ) v t x t
1 1
. This is ensured by providing sufficiently high
loop gain. Similarly, by making the amplifier gain
a
g sufficiently large, it is
possible to make ( )
v
t follow the changes in ( )
x
t .

Let ( ) ( )
c
x t t t cos = +

and let the VCO output be,
( ) ( )
c
v t t t cos = +

. Then, from Fig. 5.33(b),
( ) ( ) ( ) ( ) ( )
c c
w t x t v t t t t t cos cos = = + +

( ) ( ) ( ) ( )
{ }
c
t t t t t
1
cos 2 cos
2
= + + +

Only the term ( ) ( ) t t
1
cos
2

will appear at the output of the LPF; that is
( ) ( ) ( ) e t t t
1
1
cos
2
=

(5.32)
(We are assuming that the LPF has unit gain)
As the phase detector, we want ( ) e t to be zero where ( ) ( ) t t = ; but from Eq.
5.32, ( ) e t
1
is maximum when ( ) ( ) t t = . This is not the characteristic of a
proper phase detector. This anomaly can be corrected, if the loop provides a
2

phase shift so that the output of the VCO is ( )
c
t t sin +

. That is, the loop
locks in phase quadrature. Here after, we shall assume this
2
phase shift in the

VCO output.

Now let us look at the demodulation of FM. Let
( ) ( ) ( )
c c
x t s t A t t cos = = +


5.66
where ( ) ( )
t
f
t k m d 2

=
. The VCO is designed such that when ( ) y t , the

control voltage is zero, its frequency is
c
f . (As mentioned earlier, in a superhet,
the demodulator follows the IF stage and
c
f is actually
IF
f , which is a known
quantity). This is called the free running frequency of the VCO. Hence, the
VCO output can be written as
( ) ( )
v c
v t A t t sin = +

(5.33a)
where ( ) ( )
t
v
t k y d 2 =
(5.33b)
and
v
k is the voltage sensitivity of the VCO, in units of Hz/volt.
Or ( ) ( )
v
t k y t
'
2 = (5.33c)
( ) K y t
1
= , where
v
K k
1
2 =
( ) e t
1
of Fig. 5.33(b) is,
( ) ( ) ( ) ( )
c v
A A
e t t t h t
1
sin
2
=

(5.34a)
where ( ) h t is the impulse response of the LPF,
and ( ) ( )
a
y t g e t
1
= (5.34b)
Let ( ) ( ) ( )
e
t t t = (5.35)
Then ( ) ( ) ( )
c v a
e
A A g
y t t h t sin
2
=

( ) ( )
e
K t h t
2
sin =

, where
c v a
A A g
K
2
2
= (5.36)
Using Eq. 5.35, 5.36 and 5.33(c), we can draw following block diagram (Fig.
5.34) for the PLL.


5.67

Fig. 5.34: Equivalent circuit of a PLL

Fig. 5.34 brings out much more clearly the negative feedback nature of the PLL,
where in the quantities involved are instantaneous phase deviations of the input
and the VCO output.

Let the loop be in lock so that ( ) <<
e
t 1 for all t . Then
( ) ( )
e e
t t sin , ( ) ( ) t t and ( ) ( ) t t
' '
. As ( ) ( )
t
f
t k m d 2

=
,
we have
( ) ( )
f
t k m t
'
2 =
Hence ( )
( )
( )
f
t
k
y t m t
K K
1 1
'
2

= . That is, ( ) ( ) y t m t .

The above has been a very elementary analysis of the operation of PLL.
Literature on PLL is very widespread. For a more detailed and rigorous analysis,
refer Taub and Shilling [3].

Summary of the Detectors:
Slope detection using a single tuned circuit has been presented to show
how simple an FM demodulator could be; it is not used in practice because of its
limited range of linearity. Though balanced slope detection offers linearity over a
wider frequency range, it suffers from the problem of tuning. The Foster-Seely
discriminator and the ratio detector have been the work horses of the FM industry

5.68
until recently; they are now becoming less and less important as better circuit
configurations have been developed especially from the point of view of IC
design. The quadrature detector offers very high linearity and is commonly used
in high quality receivers. Except the phase-shifting network, the rest of the
detector, including the IF amplifier is available in a single chip (e.g. RCA CA
3089E). Commercial zero crossing detectors have better than 0.1 % linearity and
can operate from 1Hz to 10 MHz. A divide by ten counter inserted after the hard
limiter extends the range up to 100 MHz. This type of detector is best suited
when exceptional linearity over a very large frequency deviation is required. It is
less useful when the frequency deviation is a small fraction of the carrier
frequency. PLL performs better than other demodulators when the signal-to-
noise ratio at the input to the detector is somewhat low.

5.7 BandPass Limiter (BPL)
We know that information in an FM wave resides in the instantaneous
frequency (or in the zero-crossings) of the signal. As such, the amplitude
changes of the carrier are irrelevant. In fact, as was pointed out in section 5.6.1,
if the envelope ( ) A t is not a constant, it will give rise to distortion in the
demodulated output. In other words, from the point of view of proper
demodulation, we want ( ) A t A = , a constant.

Though the FM signal that is generated at the transmitter has a constant
envelope, the received signal may not possess this property. This is due to
various impairments during propagation on the channel, namely, channel noise,
distortion introduced by the channel, etc. BPL helps us to recover the constant
envelope FM signal from the one that has envelope fluctuations.

A band-pass limiter consists of a hard limiter followed by a band-pass
filter. The input-output relationship of a hard-limiter is given by

5.69
( )
( )
( )
A x t
y t
A x t
, 0
, 0
>
=

<
(5.37)
where ( ) x t is the input, ( ) y t is the output and A is a constant. Let ( ) x t be as
shown in Fig. 5.35(a).

Fig. 5.35: (a) Input to the hard limiter
(b) Output of the hard limiter
Then the output with A taken as 1, would be as shown in Fig. 5.35(b). If the input
to the hard limiter is the FM signal (with or without envelope fluctuations), then
the output would be a sequence of alternate positive and negative rectangular
pulses with durations that are not uniform. Such a wave is difficult to analyze.
However, cos as a function of is always periodic with period 2. Hence the
hard limiter output, when considered as a function of , will be a periodic square
wave (with period 2) when the input to the limiter is a cosine signal. Hence, if
( ) x t t cos cos = = , then

5.70
( ) y
1, cos 0
1, cos 0
>
=

<

Fig. 5.36: Hard limiter output with cos as input

( ) y being a periodic signal (Fig. 5.36), we can expand it in terms of Fourier
series, namely
( ) y
4 1 1
cos cos3 cos5
3 5

= + +

(5.38)
(Note that ( ) y is real, and has half-wave symmetry. Hence Fourier series
consists of only cosine terms with the even harmonics missing. See Sec. 1.2.2.)

Let be the instantaneous angle of the FM signal; that
( ) ( ) = = + = +
t
i c f c
t t c m d t t ( )
Then, from Eq. 5.38, we have
( ) [ ] ( )
= + + +

c c
y t t t t
4 4
cos ( ) cos 3 ( )
3

At the output of BPL, we have the constant envelope FM waves with carrier
frequencies
c
nf and frequency n f respectively, where n 1, 3, 5 = etc. With an
appropriate BPF, it is possible for us to obtain constant envelope FM signal with
carrier frequency
c
f and deviation f . We will assume that BPF will pass the
required FM signal, suppressing the components with spectra centered at the

5.71
harmonics of
c
f . Incidentally, BPL can be used as a frequency multiplier,
multiplication factors being 3, 5, 7, etc.

A hard limiter, as defined by Eq. 5.37 can be easily be realized in practice.
One such circuit is shown in Fig. 5.37.

Fig. 5.37: Circuit realization of a hard limiter

As shown in the figure, the circuit consists of a high gain amplifier, a current
limiting resistor R and two zener diodes arranged in a back-to-back
configuration. Zeners can be chosen to have the appropriate break down
voltages.

5.8 Broadcast FM
5.8.1 Monophonic FM Reception
FM stations operate in the frequency range 88.1 to 107.9 MHz with
stations being separated by 200 kHz; that is, the transmission bandwidth
allocation for each station is about 200 kHz. The receiver for the broadcast FM is
of the superheterodyne variety with the intermediate frequency of 10.7 MHz. As
the audio bandwidth is 15 kHz, these stations can broadcast high quality music.

Let us now look at the receiver for the single channel (or monophonic)
broadcast FM. Like the superhet for AM, the FM receiver (Fig. 5.38) also has the

5.72
front end tuning, RF stage, mixer stage and IF stage, operating at the allocated
frequencies. This will be followed by the limiter-discriminator combination, which
is different from the AM radio.

Fig. 5.38: FM broadcast superhet receiver

The need for a band-pass limiter has already been explained. As far as the
frequency discriminator is concerned, we have a lot of choices: Foster-Seely,
ratio detector, quadrature detector, PLL etc. This is unlike AM, where the
envelope detector is the invariable choice. Also, the output of the discriminator is
used in a feedback mode to control the frequency stability of the local oscillator.
This is very important as a change in the frequency of the LO can result in
improper demodulation. (Recall that, in the case AM, the envelope detector
output is used for AVC.) However, most of the present day receivers might be
using frequency synthesizers in place of a LO. These being fairly stable, AFC is
not a requirement.

Basic operation of the AFC block is as follows. Assume that the receiver is
properly tuned so that
LO c IF
f f f = . Then the discriminator input will have equal
frequency variations with respect to
IF
f ; hence the discriminator output will vary
symmetrically with respect to zero output. As such, the net DC voltage is zero
and LO frequency will not be changed. Assume, however, that LO frequency is
not correct. Let
LO c IF
f f f
'
= , where
IF IF
f f
'
< . Then the input to the S-curve of

5.73
the discriminator will be operating mostly in the negative output region. This
means that the discriminator output will have a net negative voltage. If this
voltage is applied to the varicap in the LO circuit,
LO
f will increase which implies
IF
f
'
will be increased. In other words,
IF
f
'
tends towards
IF
f . Similarly, if
LO c IF
f f f
'
= where
IF IF
f f
'
> , then the discriminator output is positive most of
the time which implies a net positive DC value. This voltage will increase the
capacitance of the varicap which implies
LO
f decreases and this makes
IF
f
'
tend
to
IF
f . Discriminator output goes through a base-band amplifier
1
(with a
bandwidth of 15 kHz), whose output drives the speaker.

5.8.2 Two-channel (stereo) FM
Two-channel (stereo) FM is fairly common these days and stereo
transmission has been made compatible with mono-aural reception. As the name
indicates, in the two-channel case, audio signal is derived as the output of two
separate microphones. These are generally called as left microphone and the
right microphone. Let the corresponding output signals be denoted by ( )
L
m t
and ( )
R
m t . The two- channel transmitter is shown in Fig. 5.39.

Fig. 5.39: FM stereo transmission scheme

1
Output of the discriminator goes through a de-emphasis network before being applied to a base-
band amplifier. Pre-emphasis at the transmitter and de-emphasis at the receiver are used to
improve the signal-to-noise ratio performance. Pre-emphasis and de-emphasis will be discussed
in Chapter 7.

5.74
As shown in the figure, ( ) ( ) ( )
L R
v t m t m t
1
= + and ( ) ( ) ( )
L R
v t m t m t
2
= . The
signals ( ) v t
1
and ( ) v t
2
go through a pre-emphasis stage. The signal ( ) y t
2
is
applied to a balanced modulator, the other input to modulator being the carrier
with a frequency of 38 kHz. Hence ( ) y t
3
is a DSB-SC signal. The carrier 38 kHz
is derived from the primary source at 19 kHz and a frequency doubler. The final
base-band signal ( )
B
m t consists of the sum of ( ) y t
1
, ( ) y t
3
and 19 kHz primary
carrier. Typical spectrum (for f 0 > ) of the base-band signal as shown in Fig.
5.40.

( ) ( ) ( )
L R
V f m f m f
1
= +
( ) ( ) ( ) Y f K Y f f Y f f
3 2 0 2 0
= + +

f
0
38 = kHz and K is a constant
Fig. 5.40: Spectrum of the final base-band signal

The signal ( )
B
m t is used to frequency modulate the carrier allotted to the station.
The resulting signal ( ) s t is transmitted on to the channel.

The block schematic of the stereo receiver upto the discriminator is the
same as shown in Fig. 5.38 (that is, monophonic case). Hence, let us look at the
operations performed by the stereo receiver after recovering ( )
B
m t from ( ) s t .

5.75
These operations are indicated in Fig. 5.41. The DSB-SC signal is coherently
demodulated by generating the 38 kHz carrier from the pilot carrier of 19 kHz.
(Note that if a pilot carrier of 38 kHz had been sent, it would have been difficult to
extract it at the receiver.) ( ) r t
1
and ( ) r t
2
, after de-emphasis will yield ( ) v t
1
and
( ) v t
2
. (Note that constants of proportionality are ignored and are taken as 1.)

Fig. 5.41: Scheme to recover ( )
L
m t and ( )
R
m t

From ( ) v t
1
and ( ) v t
2
, the individual channels signals, namely ( )
L
m t and ( )
R
m t
are obtained. These signals, after suitable power amplification, will drive the two
speakers, arranged such that one is on the left and the other is on the right. If the
receiver is not stereophonic, it would respond only to ( ) v t
1
thereby making it
stereo transmission and monophonic reception compatible.


5.76
Appendix A5.1
Table of Bessel Functions

( )
n
J

n
0.5 1 2 3 4 5 6 8 10
0 0.9385 0.7652 0.2239 - 0.2601 - 0.3971 - 0.1776 0.1506 0.1717 - 0.2459
1 0.2423 0.4401 0.5767 0.3391 - 0.0660 - 0.3276 - 0.2767 0.2346 0.0435
2 0.0306 0.1149 0.3528 0.4861 0.3641 0.0465 -0.2429 - 0.1130 0.2546
3 0.0026 0.0196 0.1289 0.3091 0.4302 0.3648 0.1148 - 0.2911 0.0584
4 0.0002 0.0025 0.0340 0.1320 0.2811 0.3912 0.3576 - 0.1054 - 0.2196
5 - 0.0002 0.0070 0.0430 0.1321 0.2611 0.3621 0.1858 - 0.2341
6 - 0.0012 0.0114 0.0491 0.1310 0.2458 0.3376 - 0.0145
7 0.0002 0.0025 0.0152 0.0533 0.1296 0.3206 0.2167
8 - 0.0005 0.0040 0.0184 0.0565 0.2235 0.3179
9 0.0001 0.0009 0.0055 0.0212 0.1263 0.2919
10 - 0.0002 0.0014 0.0070 0.0608 0.2075
11 - - 0.0020 0.0256 0.1231
12 0.0005 0.0096 0.0634
13 0.0001 0.0033 0.0290
14 - 0.0010 0.0120


5.77
Appendix A5.2
Phase Shift Discriminator
(i) Foster-Seely Discriminator
(ii) The Ratio Detector

A5.2.1 Foster-Seely discriminator
Fig. A5.2.1 illustrates the circuit diagram of this discriminator where all the
resonant circuits involved are tuned to the same frequency. Note the similarity
between this circuit and the circuit of Fig. 5.26. Major differences are a by-pass
capacitor C between the primary and secondary, an additional inductance L and
only a single tuned circuit on the secondary ( ) L C
2 2
|| .

Fig. A5.2.1: Circuit schematic of Foster-Seely discriminator

In the frequency band of operation, C, C
3
and C
4
are essentially short circuits,
which implies that the entire input voltage
in
V would appear across L. The
mutually coupled double tuned circuit has high primary and secondary Q and
low mutual inductance. When evaluating the primary current, we may neglect the
primary resistance and any impedance coupled from the secondary. Hence, the
primary current

5.78

in
p
V
I
j L
1
=
(A5.2.1)
(Note that all the voltages and currents are phasor quantities)

The voltage induced in series with the secondary as a result of the primary
current is given by

s p
V j MI = (A5.2.2)
where the sign depends on the direction of the winding. Taking the negative sign
and using Eq. A5.2.1 in Eq. A5.2.2, we have

s in
M
V V
L
1

=

Assuming the diode circuit will draw very little current, we calculate the
current in the secondary because of
s
V as,

( )
s
s
L C
V
I
R j X X
2 2
2
=
+

where R
2
is the resistance associated with the inductance L
2
,
L
X L
2
2
= and
C
X
C
2
2
1
=
. Hence, the voltage across the terminals 2, 3 is given by

( )
s C
V I j X
2
23
=

( )
( )
s C
L C
V j X
R j X X
2
2 2
2
=
+

in C
V X
j M
L R j X
2
1 2 2
=
+
where
( )
L C
X X X
2 2
2
=
The voltage applied to diode D
1
, V
62
, is

L
V V V
62 23
1
2
= +

in
V V
23
1
2
= + (A5.2.3)

5.79
Similarly, the voltage applied to diode D
2
, V
63
, is
=
in
V V V
63 23
1
2
(A5.2.4)
The final output voltage V
54
is, V V V
54 64 65
= which is proportional to
{ }
V V
62 63
. We will consider three different cases:
c
f f =
1
,
c
f f > and
c
f f < .
i) When the input frequency
c
f f = we have

in C c
in
V X M X
j M
V j V
L R L R
2 2
23
1 2 1 2

= =

(A5.2.5a)
That is, the secondary voltage V
23
leads the primary voltage by 90
.
Thus V
23
1
2
will lead
in
V by 90
and V
23
1
2
will lag
in
V by 90
. Let us
construct a phasor diagram by taking
in
V as reference. This is shown in Fig.
A5.2.2(a). As the magnitude of the voltage vectors applied to the diodes D
1

and D
2
, V
62
and V
63
respectively are equal, the voltages V
64
and V
65
are
equal and hence the final output V
54
is zero.

1
Note that in a superheterodyne receiver, the demodulator follows the IF stage. Hence,
c
f is
actually
IF
f , and the discriminator circuit is always tuned to
IF
f , irrespective of the incoming
carrier frequency.

5.80

Fig. A5.2.2: Phasor diagram illustrative of the operation of Foster-Seely
discriminator

ii) when the input frequency exceeds
c
f ,
L C
X X X
2 2
2
= is positive. Let
j
R j X Z e
2 2 2

+ = . Then,
( )
in C in C j
V X V X M
j M
V e
L R j X L Z
2 2
90
23
1 2 2 1 2

= =
+
(A5.2.5b)
That is, V
23
leads in
in
V by less than 90
and V
23
lags in
in
V by
more than 90
. This is shown in Fig. A5.2.2(b). As the magnitude of the

vector V
62
is greater than that of V
63
, V V
64 65
> which implies the final
output V V V
54 64 65
= is positive.


5.81

Fig. A5.2.3: Response curve of the Foster-Seely discriminator

iii) Similarly, based on the phasor diagram of Fig. A5.2.2(c), we can easily
argue that the final output would be negative when
c
f f < . The actual value
of the final output depends on how far away the input frequency is from
c
f .
Fig. A5.2.3 gives the plot of the frequency response of the Foster-Seely
discriminator, which is usually termed as the S-curve of the discriminator.
Useful range of the discriminator (frequency range of linear response,
shown in red in Fig. A5.2.3) normally lies between the 3 dB points of the
tuned circuit which forms part of the discriminator circuit.

Foster-Seely discriminator responds also to input amplitude variations. Let
the input to the discriminator be ( )
i c
f t f = . Then, the voltages across R
3
and R
4

are equal and let this value be 3 V. Now, let ( )
i
f t be such that voltage across R
3

increases while that across R
4
decreases. Let the voltage increase on R
3
be 2
Volts. We can take the voltage decrease on R
4
also as 2 V. In other words, for
the given frequency deviation, say f
1
, we have the voltage at point 4 equal to 5
volts where as the voltage at point 5 equal to 1 V. This implies

5.82
( )
out
V 5 1 4 = = . Let V
64
denote voltage across R
3
and V
65
, the voltage
across R
4
. Then
V
V
64
65
5 = .

Let the input signal strength be increased such that, when ( )
i c
f t f = ,
V V
64 65
6 = = . Now let ( )
i
f t change such that we have the deviation f
1
as in
the previous case. Then V
64
will become 10 Volts whereas V
65
becomes 2 V,
with their difference being equal to 8 V. Though the ratio
V
V
64
65
remains at 5,
out
V
changes from the previous value. That is, the circuit responds not only to
frequency changes but also to changes in the incoming carrier strength. Hence,
Foster-Seely discriminator has to be the preceded by a BPL.

A5.2.2 Ratio Detector
By making a few changes in the Foster-Seely discriminator, it is possible
to have a demodulator circuit which has built in capability to handle the amplitude
changes of the input FM signal, thereby obviating the need for an amplitude
limiter. The resulting circuit is called the ratio detector which has been shown in
Fig. A5.2.4.

Comparing the ratio detector circuit with that of the Foster-Seely
discriminator, we find the following differences: direction of D
2
is reversed, a
parallel RC combination consisting of ( ) R R
5 6
+ and C
5
has been added and
the output
out
V is taken across a different pair of points. We shall now briefly
explain the operation of the circuit.


5.83

Fig. A5.2.4: Circuit schematic of a ratio detector

Reexamining Fig.A5.2.1 and the corresponding phasor diagrams, we find that by
and large, the sum V V
62 63
+ remains constant. Hence, any variation in the
magnitude of this sum voltage can be considered to be spurious. Suppression of
these spurious variations would result in a detector that is unaffected by input
voltage fluctuations which implies that circuit does not require a separate limiter
stage. How the sum voltage is kept constant would be explained a little later.

With the diode D
2
being reversed, we find that the voltages V
64
and V
65

are series aiding rather than series opposing and as such, the voltage V
54

represents the sum voltage. Taking R R
5 6
= , we find

out
V V V V V
64 47 64 74
= + =
V V
64 54
1
2
=

V V
V
56 64
64
2
+
=

[ ]
V V
64 56
1
2
=
k V V
62 63
=


5.84
Usually, C C
3 4
= and R R
3 4
= . Hence at resonance, V V
64 56
= which implies
that
out
V is zero. Above resonance, as V V
64 56
> , the output is positive whereas
below resonance V V
56 64
> , and the output is negative.

In the circuit Fig. A5.2.4, C
5
is a capacitor of a rather large value. For
example, C
5
is of the order of 5 F whereas C
3
and C
4
are of the order 300
pF. If
in
V is constant, C
5
charges to the full potential existing between the points
5 and 4, which, as indicated earlier is essentially a constant. If
in
V tries to
increase, C
5
will tend to oppose any rise in
out
V . This is because as the input
voltage tries to rise, extra diode current flows trying to charge C
5
. But V
54

remains constant at first because C
5
is a fairly large capacitance and it is not
possible for the voltage across it to change instantaneously. The situation now is
that the current in the diodes' load has risen but the voltage across the load has
not changed. This being so, the secondary of the ratio detector transformer is
more heavily damped, Q falls and so does the gain of the amplifier driving the
detector. This nearly counteracts the rise in the input voltage. Similarly, when
in
V
increases, the damping is reduced. The gain of the driving amplifier increases
thereby counteracting the fall in the input voltage. Thus the ratio detector
provides variable damping.

For a large number of years, the Foster-Seely discriminator and the ratio
detector have been the work horses of the FM industry. As these circuit
configurations are not very convenient from the point of view of IC fabrication, of
late, their utility has come down. Companies such as Motorola have built high
quality FM receivers using the Foster-Seely discriminator and the ratio detector.
Some details can be found in Roddy and Coolen [2].


5.85
Appendix A5.3
Multi-tone FM
Let ( ) m t A t A t
1 1 2 2
cos cos = + where f
1
and f
2
are arbitrary. Then,
( )
( ) ( )
c
j t j t j t
pe c
s t A e e e
1 1 2 2
sin sin

=

where
f
A k
f
1
1
1
= and
f
A k
f
2
2
2
=
( ) ( ) ( )
c
j t j m t j n t
pe c m n
m n
s t A J e J e e
1 2
1 2

=

Hence, ( ) ( ) ( ) ( )
c m n c
m n
s t A J J m n t
1 2 1 2
cos = + +

(A5.3.1)
The (discrete) spectrum of ( ) s t can be divided into 4 categories:
1) Carrier component: amplitude ( ) ( ) J J
0 1 0 2
= when
c
f f =
2) A set of side frequency components due to f
1
: These components have
amplitudes ( ) ( )
m
J J
1 0 2
at frequencies ( )
c
f mf m
1
, 1, 2, 3, =
3) A set of side frequency components due to f
2
: These components have
amplitudes ( ) ( )
n
J J
0 1 2
at frequencies ( )
c
f nf n
2
, 1, 2, 3, =
4) A set of cross modulation (or beat frequency) terms with amplitudes
( ) ( )
m n
J J
1 2
at frequencies ( )
c
f mf nf
1 2
where m 1, 2, 3, = ;
n 1, 2, 3, =

Cross spectral terms of the type given at (4) above clearly indicate the
non-linear nature of FM. These terms are not present in linear modulation. Even
with respect to the terms of the type (2) and (3), linear modulation generates only
those components with m 1 = and n 1 = .


5.86
Appendix A5.4
RMS Bandwidth of WBFM
We have already used a number of measures for the bandwidths of
signals and systems, such as 3-dB bandwidth, null-to-null bandwidth, noise
equivalent bandwidth, etc. In the context of WBFM, another meaningful and
useful bandwidth quantity is the r ms bandwidth. The basic idea behind the rms
bandwidth is as follows.

Let ( )
X
S f denote the power spectral density of a random process ( ) X t .
Then, the normalized PSD, ( )
( )
( )
X
X N
X
S f
S f
S f d f
,

=
, has the properties of a

PDF. Let
f
denote the standard deviation of ( )
X N
S f
,
. Then,
f
2 can be used
as a measure of the spectral width of the process.

The WBFM process, with
c
f f >> , can be treated as a band-pass
random process. We shall define the r ms bandwidth of a band-pass process as
( )
rms
B f f
1
2 2
0
2

=

(A5.4.1a)
or ( )
rms
B f f
2
2
0
4 = (A5.4.1b)
where
( )
( ) ( )
( )
X
X
f f S f d f
f f
S f d f
2
0
2
0
0
0

( ) ( )
X
T
f f S f d f
P
2
0
0
2
=

(A5.4.2a)
where
T
P is the total power of the process and

5.87
( )
X
T
f f S f d f
P
0
0
2

=

(A5.4.2b)

Let ( ) M t be a strict sense stationary message process with ( ) m t as a
sample function. Let ( )
M
p m denote the PDF of the process. (Note that we are
using the symbol ( ) p to denote the PDF, instead of the earlier symbol ( ) f ; this
has been done to avoid confusion. In this derivation, f denotes only the
frequency variable.) For the FM case, we have

i c f
f f k m = +
where m is a specific value of ( ) m t for some t . Then

( )
i c
f
f f
m
k
= (A5.4.3)
Using quasi-static approximation, it has been shown by Peebles [4] that,
( )
c c T
X M M
WBFM
f f f
f f f f P
S f p p
k k k 2
+
= +

(A5.4.4)
(Note that in quasi-static approximation, it is assumed that ( )
i
f t remains constant
for a sufficiently long period; as such, FM wave appears to be a regular sinusoid
and
i
f can be replaced by f .) For the WBFM process,
c
T T
A
P S
2
2
= = .

In Eq. A5.4.4,
c
M
f
f f
p
k

is the positive part of the spectrum. We will now
show that
( )
( )
rms f M
WBFM
B k R
1
2 2 0 =

where ( ) ( )
M
R M t
2
0 = , the mean
square value of the process. (Note that ( )
M
R is the ACF of the process.)


5.88
For f 0 > , ( )
X
S f is symmetrical with respect to
c
f f = (we assume that
the input PDF is symmetric about zero) and hence
c
f f
0
= . Using Eq. A5.4.4 in
A5.4.2(a), we have

( )
( )
c T
rms c M
WBFM
T f f
f f P
B f f p d f
P k k
2
2
0
8
2
.
Let
c
f
f f
k
= . Then,

( )
( ) ( )
c
f
rms f M
WBFM
f
k
B k p d
2
2
4

( ) ( )
f M f
k p d k M t
2 2 2 2
4 4

( )
f M
k R
2
4 0 =
That is,

( )
( )
rms f M
WBFM
B k R 2 0 = (A5.4.5)

Example A5.4.1
Let ( ) m t be a sample function of a strict sense stationary process ( ) M t .
A WBFM signal is generated using ( ) m t as the message signal. It is given that
( )
M
m
p m
otherwise
1
, 1
2
0 ,
<

a) Find ( )
X
WBFM
S f

.
b) What is the value of
rms
B ?

As ( )
M
p m is uniform, we expect the PSD of the resulting WBFM also to
be uniform over the appropriate frequency range. When m 0 = , we have
c
f f = ,

5.89
whereas m 1 = results in the instantaneous frequency to be
c f
f k
respectively. In other words, the frequency range over which ( )
M
S f exists is
c f c f
f k f f k < < + . Let
c c
M M
f f
f f f f
p p
k k
+
+

be as shown in Fig. A5.4.1.

Fig. A5.4.1: Shape of the PSD of Example A5.4.1

To find the value of , we require that the RHS of Eq. A5.4.4, when integrated
over the entire range should be equal to
c
A
2
2
. That is,

c f
c f
f k
c c
f
f k
A A
d f
k
2 2
2
4 2
+

or ( )
c c
f
f
A A
k
k
2 2
2
2 2
=
That is,
1
2
=
Hence, ( )
c
c f c f
X
f
WBFM
A
f k f f k
S f
k
otherwise
2
,
8
0 ,
< < +

In a few situations
( )
rms
B might be quite meaningful. For example, if ( ) m t
is the sample function of a Guassian process, there is a small but finite

5.90
probability that ( ) m t will assume very large values. This implies that f
becomes excessive and as such
T
B given by Carsons rule or its variants would
be extremely high. As
( )
rms
B weights the large frequency derivations with small
probabilities,
( )
rms
B will not be unduly excessive. For example if ( ) m t belongs
to a Guassian process and say ( )
M
p m is ( ) N 0, 9 , then

( )
rms f f
WBFM
B k k 2 3 6 = =
(Note that ( )
M
R 0 9 3 = = )

Exercise A5.4.1
We define the RMS bandwidth of any lowpass process ( ) M t as,

( )
( )
( )
M
rms
M
M
f S f d f
B
R
2
2
0
(A5.4.6)
Let
( )
rms
PM
B denote the RMS bandwidth of the PM signal. Show that

( )
( )
( )

=

rms p M rms
PM M
B k R B 2 0 (A5.4.7)

5.91
Appendix A5.5
Modulation Techniques in TV
Color television is an engineering marvel. At the flick of a button on a
remote, we have access to so much information and entertainment in such a fine
detail (audio as well as video with all its hues and colors) that the angels could
envy the humans on this count. Distances are no longer a barrier; we have real
time reception almost at any point on this globe. Maybe, not too far into the
future, we may be able to watch any program of our choice in the language we
prefer to listen either in real time or near real time. We shall now take a closer
look at the modulation techniques used in (commercial) TV, based on the NTSC
standard used in North America and J apan. Our purpose here is to illustrate how
various modulation techniques have been used in a practical scheme. Hence
PAL and SECAM systems have not been discussed.

As the TV transmission and reception started with monochrome (black
and white) signals we shall begin our discussion with this scheme. (Note that
color transmission can be viewed on a monochrome receiver. Similarly, black-
and-white transmission can be viewed on a color receiver.)

The bandwidth of a monochrome video signal is about 4.2 MHz. Let
( )
v
m t denote this signal. The bandwidth allocated to each TV station by the
regulatory body is about 6 MHz. Hence the use of DSB is ruled out. It is very
difficult to generate the SSB of the video signal, ( )
v
m t . (Filtering method is ruled
out because of appreciable low frequency content in ( )
v
M f and because of the
fairly wide bandwidth, designing the HT with required specifications in extremely
difficult.) As such, VSB becomes the automatic choice. With a suitable carrier
component, we have seen that VSB can be envelope detected. Actually at the
transmitter, it is only partially VSB; because of the high power levels at the
transmitter, it is difficult to design a filter with an exact vestigial sideband. It is at

5.92
the IF stage of the receiver that perfect VSB shaping is achieved; subsequently it
is demodulated.

Fig. A5.5.1: Modulation stages in a monochrome TV transmitter

Fig. A5.5.1 shows the block diagram of the modulation scheme of a
monochrome TV transmitter. Details of the magnitude spectrum of the
transmitted signal is given in Fig. A5.5.2.

Fig. A5.5.2: Audio and video spectrum of a monochrome TV signal


5.93
In this figure,
cv
f is the video carrier and
ca
f is the carrier for audio, which is
higher than
cv
f by 4.5 MHz. (The video carrier
cv
f is 1.25 MHz above the lower
frequency limit allotted to the station)

The USB of the video occupies the full bandwidth of 4.2 MHz. That part of
LSB spectrum between ( )
cv
f 0.75 to
cv
f is transmitted without any attenuation.
Below the frequency ( )
cv
f 0.75 MHz, LSB is gradually attenuated reaching
almost zero level at ( )
cv
f 1.25 MHz . As this is not exactly a VSB characteristic,
it is termed as partial VSB.

The audio signal, band-limited to 10 kHz frequency modulates the audio
carrier. The maximum frequency deviation is 25 kHz. Hence, the audio
bandwidth can be taken as ( ) ( ) f W 2 2 25 10 70 + = + = kHz.

TV receiver is of the superheterodyne variety. A part of the reciever structure is
shown in Fig. A5.5.3.

Fig. A5.5.3: (Partial) Block diagram of monochrome TV receiver

The mixer output, ( ) v t
1
, is applied to an IF amplifier and VSB shaping
network. The IF amplifier has a pass-band of 41 to 47 MHz. The characteristic of
the VSB shaping filter is shown in Fig. A5.5.4.


5.94

Fig. A5.5.4: VSB shaping in TV receiver

The frequency modulated audio signal is also passed by the IF stage but with
much less gain than the video part. That is, ( ) v t
2
consists of the video signal of
the (VSB+C) type and the FM audio signal, with a carrier at 4.5 MHz and with
ca cv
A A << where
ca
A is the audio carrier amplitude and
cv
A is the amplitude of
the video carrier. Under these conditions, it can be shown that ( ) v t
3
, the output
of the envelope detector, has the video as well as required audio. The video
amplifier removes the audio from ( ) v t
3
. The output of the video amplifier is
processed further and is displayed on the picture tube. ( ) v t
3
is also applied as
input to an IF stage, with the IF of 4.5 MHz. The audio part of the ( ) v t
3
is passed
by this IF stage; the FM demodulator that follows produces the audio signal.

Color TV: The three primary colors, whose linear combination can give rise to
other colors are: Red, Blue and Green. These color signals, pertaining to the
scene that is being shot, are available at the outputs of three color cameras. Let
us denote these signals as ( )
R
m t , ( )
G
m t and ( )
B
m t respectively. These basic
color components are linearly combined to produce (i) the video signal of the
monochrome variety (this is called the luminance signal and is denoted by
( )
L
m t )) and (ii) two other independent color signals (called the in-phase
component of the color signal and the quadrature component of the color signal).

5.95
Let us denote these components as ( )
I
m t and ( )
Q
m t respectively. The
equations of the linear transformation are given below in the form of a matrix
equation.

( )
( )
( )
( )
( )
( )
L
R
I G
B
Q
m t
m t
m t m t
m t
m t
0.3 0.59 0.11
0.6 0.28 0.32
0.21 0.52 0.31

=

M
(A5.5.1)
From Eq. A5.5.1, we obtain
( ) ( ) ( ) ( )
L R G B
m t m t m t m t 0.3 0.59 0.11 = + + (A5.5.2)
It has been found that ( )
L
m t as given by Eq. A5.5.2 closely resembles ( )
v
m t of
the monochrome system. ( )
L
m t has the same bandwidth as ( )
v
m t namely, 4.2
MHz. This is required to preserve the sharp transitions in the intensity of light at
the edges in a scene. However, the eye is not as sensitive to color transitions in
a scene and is possible to reduce the bandwidth occupancy of ( )
I
m t and ( )
Q
m t .
In the NTSC system, bandwidth allocation for ( )
I
m t is 1.5 MHz and that of
( )
Q
m t is 0.5 MHz. These chrominance signals are quadrature multiplexed
(QAM) on the color subcarrier as shown in Fig. A5.5.5.

Fig. A5.5.5: Generation of composite baseband video


5.96
The composite video signal ( )
bv
m t is modulated on to the carrier
cv
f .
This implies that color subcarrier,
cc
f , is 3.58 MHz above
cv
f and
ca
f is 4.5 MHz
above
cv
f . As such, the color information gets interleaved in between the spectral
lines of the luminance signal and no additional bandwidth is required for the color
TV system. In order to facilitate coherent demodulation of the QAM signal, a few
cycles of the color subcarrier (called color burst) is sent along with the
transmitted signal. This reference carrier is tracked by a PLL in the receiver. The
VCO output of the PLL is used in the demodulation of the chrominance signal.
For more details on TV transmission and reception, refer to Carlson, Crilly and
Rutledge [5] or Leon Couch [6].

TV modulator circuit is available in an IC chip. For example, Motorola MC
1374 includes an FM audio modulator, sound carrier oscillator, RF dual input
modulator. It is designed to generate a TV signal from audio and video inputs. It
is also suited for applications such as video tape recorders, video disc players,
TV games etc. The FM system can also be used in the base station of a cordless
telephone. Circuit details and other parameters can be obtained from the manual.

Exercise A5.5.1
Indirect method can used to generate the FM signal for the audio in TV.
Required carrier is 4.5 MHz and f 25 = kHz. Using
c
f
1
200 = kHz and
f
1
20 < Hz, design the modulator such that frequency at any point in the
modulator does not exceed 100 MHz. Use the shortest possible chain of
frequency doublers and triplers.

5.97
References
1) Bettina Koster, Peter Waldow and Ingo Wolff, A unique low voltage,
source-coupled J -FET VCO, RF signal processing, April 2001, PP58-66
(website: http://www.rfdesign.com)
2) Dennis Roddy and J ohn Coolen, Electronic communications, (4
th
ed.), PHI,
1995
3) Herbert Taub and Donald L. Schilling, Principles of communication
systems, (2
nd
ed.), McGraw Hill International ed., 1986
4) Peebles P. Z., Communication system principles, Addison-Wesley, 1976
5) A. Bruce Carlson, Paul B. Crilly and J anet C. Rutledge, Communication
systems (4
th
ed.), McGraw Hill International ed., 2002
6) Leon W. Couch II, Digital and Analog Communication systems (6
th
ed.),
Pearson Education-Asia, 2001

6.1
CHAPTER 6 CHAPTER 6

Digital Transmission of Analog Signals:
PCM, DPCM and DM

6.1 Introduction
Quite a few of the information bearing signals, such as speech, music,
video, etc., are analog in nature; that is, they are functions of the continuous
variable t and for any t t
1
= , their value can lie anywhere in the interval, say
A to A. Also, these signals are of the baseband variety. If there is a channel
that can support baseband transmission, we can easily set up a baseband
communication system. In such a system, the transmitter could be as simple as
just a power amplifier so that the signal that is transmitted could be received at
the destination with some minimum power level, even after being subject to
attenuation during propagation on the channel. In such a situation, even the
receiver could have a very simple structure; an appropriate filter (to eliminate the
out of band spectral components) followed by an amplifier.

If a baseband channel is not available but have access to a passband
channel, (such as ionospheric channel, satellite channel etc.) an appropriate CW
modulation scheme discussed earlier could be used to shift the baseband
spectrum to the passband of the given channel.

Interesting enough, it is possible to transmit the analog information in a
digital format. Though there are many ways of doing it, in this chapter, we shall
explore three such techniques, which have found widespread acceptance. These
are: Pulse Code Modulation (PCM), Differential Pulse Code Modulation (DPCM)

6.2
and Delta Modulation (DM). Before we get into the details of these techniques, let
us summarize the benefits of digital transmission. For simplicity, we shall assume
that information is being transmitted by a sequence of binary pulses.
i) During the course of propagation on the channel, a transmitted pulse
becomes gradually distorted due to the non-ideal transmission
characteristic of the channel. Also, various unwanted signals (usually
termed interference and noise) will cause further deterioration of the
information bearing pulse. However, as there are only two types of signals
that are being transmitted, it is possible for us to identify (with a very high
probability) a given transmitted pulse at some appropriate intermediate
point on the channel and regenerate a clean pulse. In this way, we will be
completely eliminating the effect of distortion and noise till the point of
regeneration. (In long-haul PCM telephony, regeneration is done every few
kilometers, with the help of regenerative repeaters.) Clearly, such an
operation is not possible if the transmitted signal was analog because there
is nothing like a reference waveform that can be regenerated.
ii) Storing the messages in digital form and forwarding or redirecting them at a
later point in time is quite simple.
iii) Coding the message sequence to take care of the channel noise,
encrypting for secure communication can easily be accomplished in the
digital domain.
iv) Mixing the signals is easy. All signals look alike after conversion to digital
form independent of the source (or language!). Hence they can easily be
multiplexed (and demultiplexed)

6.2 The PCM system
Two basic operations in the conversion of analog signal into the digital is
time discretization and amplitude discretization. In the context of PCM, the former
is accomplished with the sampling operation and the latter by means of
quantization. In addition, PCM involves another step, namely, conversion of

6.3
quantized amplitudes into a sequence of simpler pulse patterns (usually binary),
generally called as code words. (The word code in pulse code modulation refers
to the fact that every quantized sample is converted to an R -bit code word.)

Fig. 6.1: A PCM system

Fig. 6.1 illustrates a PCM system. Here, ( ) m t is the information bearing
message signal that is to be transmitted digitally. ( ) m t is first sampled and then
quantized. The output of the sampler is ( ) ( )
s
s
t nT
m nT m t
=
= .
s
T is the
sampling period and n is the appropriate integer.
s
s
f
T
1
= is called the sampling
rate or sampling frequency. The quantizer converts each sample to one of the
values that is closest to it from among a pre-selected set of discrete amplitudes.
The encoder represents each one of these quantized samples by an R -bit code
word. This bit stream travels on the channel and reaches the receiving end. With
s
f as the sampling rate and R -bits per code word, the bit rate of the PCM
system is
s
s
R
Rf
T
= bits/sec. The decoder converts the R -bit code words into
the corresponding (discrete) amplitudes. Finally, the reconstruction filter, acting
on these discrete amplitudes, produces the analog signal, denoted by
( ) m t . If
there are no channel errors, then
( ) ( ) m t m t .


6.4
6.3 Sampling
We shall now develop the sampling theorem for lowpass signals.
Theoretical basis of sampling is the Nyquist sampling theorem which is stated
below.

Let a signal ( ) x t be band limited to W Hz; that is, ( ) X f 0 = for f W > .
Let ( ) ( )
s
s
t nT
x nT x t n ,
=
= < < represent the samples of ( ) x t at
uniform intervals of
s
T seconds. If
s
T
W
1
2
, then it is possible to reconstruct
( ) x t exactly from the set of samples, ( ) { }
s
x nT .

In other words, the sequence of samples ( ) { }
s
x nT can provide the
complete time behavior of ( ) x t . Let
s
s
f
T
1
= . Then
s
f W 2 = is the minimum
sampling rate for ( ) x t . This minimum sampling rate is called the Nyquist rate.
Note: If ( ) x t is a sinusoidal signal with frequency f
0
, then
s
f f
0
2 > .
s
f f
0
2 = is
not adequate because if the two samples per cycle are at the zero crossings of
the tone, then all the samples will be zero!

We shall consider three cases of sampling, namely, i) ideal impulse
sampling, ii) sampling with rectangular pulses and iii) flat-topped sampling.

6.3.1 Ideal impulse sampling
Consider an arbitrary lowpass signal ( ) x t shown in Fig. 6.2(a). Let
( ) ( ) ( )
s s
n
x t x t t nT
=

=

(6.1a)

6.5
( ) ( )
s
n
x t t nT
=
=
(6.1b)
( ) ( )
s s
n
x nT t nT
=
=
(6.1c)
where ( ) t is the unit impulse function of section 1.5.1. ( )
s
x t , shown in red in
Fig. 6.2(b) consists of a sequence of impulses; the weight of the impulse at
s
t nT = is equal to ( )
s
x nT . ( )
s
x t is zero between two adjacent impulses.

Fig. 6.2: (a) A lowpass signal ( ) x t
(b) ( )
s
x t , sampled version of ( ) x t

It is very easy to show in the frequency domain that ( )
s
x t preserves the
complete information of ( ) x t . As defined in Eq. 6.1(a), ( )
s
x t is the product of
( ) x t and
( )
s
n
t nT
. Hence, the corresponding Fourier relation is

convolution. That is,

6.6
( ) ( )
s
s s n
n
X f X f f
T T
1

=

=

(6.2a)

s s n
n
X f
T T
1

=

=

(6.2b)
( )
s
s n
X f nf
T
1

=
=
(6.2c)
From Eq. 6.2(c), we see that ( )
s
X f is a superposition of ( ) X f and its shifted
versions (shifted by multiples of
s
f , the sampling frequency) scaled by
s
T
1
. This
is shown in Fig. 6.3. Let ( ) X f be a triangular spectrum as shown in Fig. 6.3(a).

Fig. 6.3: Spectra of ( ) x t and ( )
s
x t
(a) ( ) X f (b) ( )
s
X f ,
s
f W 2 >
(c) ( )
s
X f ,
s
f W 2 = (d) ( )
s
X f ,
s
f W 2 <

6.7
From Fig. 6.3(b) and 6.3(c), it is obvious that we can recover ( ) x t from ( )
s
x t by
passing ( )
s
x t through an ideal lowpass filter with gain
s
T and bandwidth W , as
shown in Fig. 6.4.

Fig. 6.4: Reconstruction of ( ) x t from ( )
s
x t

Of course, with respect to Fig. 6.3(b), which represents the over-sampled case,
reconstruction filter can have some transition band which can fit into the gap
between f W = and ( )
s
f f W = . However, when
s
f W 2 < , (under-sampled
case) we see that spectral lobes overlap resulting in signal distortion, called
aliasing distortion. In this case, exact signal recovery is not possible and one
must be willing to tolerate the distortion in the reconstructed signal. (To avoid
aliasing, the signal ( ) x t is first filtered by an anti-aliasing filter band-limited to
s
f
W
2
and the filtered signal is sampled at the rate of
s
f samples per second.
In this way, even if a part of the signal spectrum is lost, the remaining spectral
components can be recovered without error. This would be a better option than
permitting aliasing. See Example 6.1.)

It is easy to derive an interpolation formula for ( ) x t in terms of its samples
( )
s
x nT when the reconstruction filter is an ideal filter and
s
f W 2 . Let ( ) H f
represent an ideal lowpass filter with gain
s
T and bandwidth
s
f
W
'
2
= where

6.8
s
s
f
W f W
2
. Then, ( ) h t the impulse response of the ideal lowpass filter
is, ( )
( ) s
h t T W c W t
' '
2 sin 2 = . As ( ) ( ) ( )
s
x t x t h t = and
s
W T
'
2 1 = , we have
( ) ( ) ( )
s s
n
x t x nT c W t nT
'
sin 2
=

=

(6.3a)
If the sampling is done at the Nyquist rate, then W W
'
= and Eq. 6.3(a) reduces
to
( ) ( )
n
n
x t x c W t n
W
sin 2
2
=

=

(6.3b)
That is, the impulse response of the ideal lowpass filter, which is a ( ) c sin
function, acts as the interpolating function and given the input,
( ) ( ) { }
s s
x nT t nT , it interpolates the samples and produces ( ) x t for all t .

Note that ( )
s
x t represents a sequence of impulses. The weight of the
impulse at
s
t nT = is equal to ( )
s
x nT . In order that the sampler output be equal
to ( )
s
x nT , we require conceptually, the impulse modulator to be followed by a
unit that converts impulses into a sequence of sample values which are basically
a sequence of numbers. In [1], such a scheme has been termed as an ideal C-
to-D converter. For simplicity, we assume that the output of the sampler
represents the sample sequence ( ) { }
s
x nT .

To reconstruct ( ) x t from ( ) { }
s
x nT , we have to perform the inverse
operation, namely, convert the sample sequence to an impulse train. This has
been termed as an ideal D-to-C converter in [1]. We will assume that the
reconstruction filter in Fig. 6.1 will take care of this aspect, if necessary.


6.9
6.3.2 Sampling with a rectangular pulse train
As it is not possible in practice to generate impulses, let look at a more
practical method of sampling, namely, sampling with a rectangular pulse train.
(Note that an impulse is a limiting case of a rectangle pulse as explained in Sec
1.5.1.)

Let ( )
p
y t represent the periodic rectangular pulse train as shown in Fig.
6.5(b). Let
( ) ( ) ( )
s p
x t x t y t =

(6.4a)

Fig. 6.5: Sampling with a rectangular pulse train
(a) ( ) x t , (b) the pulse train, (c) the sampled signal

6.10
Then, ( ) ( ) ( )
s p
X f X f Y f = (6.4b)
But, from exercise 1.1, we have
( ) ( ) ( )
p s s
s n
Y f c nf f nf
T
sin

Hence,
( ) ( ) ( )
s s s
s n
X f c nf X f nf
T
sin
(6.5)
( ) ( ) ( )
s s
s s s
c X f f X f c X f f
T T T
sin sin

= + + + + +

As
s
n
c
T
sin

is only a scale factor that depends on n , we find that ( ) X f and its
shifted replicas are weighted by different factors, unlike the previous case where
all the weights were equal. A sketch of ( )
s
X f is shown in Fig. 6.6 for
s
T
1
10
=
and,
s
f W 2 > and ( ) X f of Fig. 6.3(a).

Fig. 6.6: Plot of Eq. 6.5

In the Fig. 6.6,
( ) c
1
sin 0.1
0.0983
10
= = ,
( ) c
2
sin 0.2
0.0935
10
= = etc. From
Eq. 6.5 and Fig. 6.6, it is obvious that each lobe of ( )
s
X f is multiplied by a
different number. However as the scale factor is the same for a given spectral

6.11
lobe of ( )
s
X f , there is no distortion. Hence ( ) x t can be recovered from ( )
s
x t of
Eq. 6.4(a).
From Fig. 6.5(c), we see that during the time interval , when ( )
p
y t 1 = ,
( )
s
x t follows the shape of ( ) x t . Hence, this method of sampling is also called
exact scanning.

6.3.3 Flat topped sampling

Fig. 6.7: Flat topped sampling

ZOH is the zero order hold (Example 1.15) with the impulse response
( )
t
h t ga
2

=

( )
s
x t is the same as the one given by Eq. 6.1.
As ( ) ( ) ( )
s
y t x t h t = , (6.6a)
( ) y t will be as shown (in red color) in Fig. 6.8.


6.12

Fig. 6.8: Sampled waveform with flat top pulses

Holding the sample value constant provides time for subsequent coding
operations. However, a large value of would introduce noticeable distortion and
exact recovery of ( ) x t is not possible unless ( ) y t is passed through an
equalizer, as will be seen shortly.

Taking the Fourier transform of Eq. 6.6(a), we have
( ) ( ) ( )
s
Y f X f H f = (6.6b)
where ( ) ( )
j f
H f e c f sin

=
As ( ) ( )
s s
s n
X f X f nf
T
1
=
, we have
( ) ( ) ( )
j f
s
s n
Y f X f nf e c f
T
sin

(6.7)
Let ( ) Y f
0
denote the term for n 0 = in Eq. 6.7. That is,
( ) ( ) ( )
j f
s
Y f e X f c f
T
0
sin

= (6.8)
By holding the samples constant for a fixed time interval, we have introduced a
delay of
2
and more importantly ( ) X f is multiplied with ( ) c f sin , which is a

function of the variable f ; that each spectral component of ( ) X f is multiplied by
a different number, which implies spectral distortion. Assuming that the delay of

6.13
2
is not serious, we have to equalize the amplitude distortion caused and this
can be achieved by multiplying ( ) Y f
0
by
( ) H f
1
. Of course, if
s
T
0.1
< , then the

amplitude distortion may not be significant. The distortion caused by holding the
pulses constant is referred to as the aperture effect.

6.3.4 Undersampling and the problem of aliasing
When
s
f W 2 < , we have seen that there is spectral overlap (Fig. 6.3d)
and the resulting distortion is called the aliasing distortion. Let us try to
understand this in some detail.

Fig. 6.9: A part of the aliased spectrum

Let ( )
s
X f represent the aliased spectrum and let us consider the three
lobes from it, namely, ( )
s
X f
T
1
, ( )
s
s
X f f
T
1
and ( )
s
s
X f f
T
1
+ , indicated by 1,
2, and 3 respectively in Fig. 6.9. The left sides of these triangular spectra are
shown in red. As can be seen from the figure, the complex exponential with the
(negative) frequency f
1
'
is interfering with the frequency f
1
. That is,
j f t
e
1
'
2
is
giving rise to the same set of samples as that
j f t
e
1
2
, for the
s
T chosen. We

6.14
shall make this clear by considering the sampling of a cosine signal. Let the
signal to be sampled be the cosine signal,
( ) ( ) x t t 2cos 2 4 =

Its spectrum consists of two impulses, one at 4 Hz and the other at - 4 Hz as
shown in Fig. 6.10(a), in which the negative frequency component has been
shown with a broken line. Let ( ) x t be sampled at the rate of 6 samples/sec.

Fig. 6.10: Sampling of a cosine signal with aliasing

(Note that the Nyquist rate is greater than 8 samples per second.) The resulting
spectrum is shown in Fig. 6.10(b). Notice that in ( )
s
X f , there are spectral
components at 2 Hz. The impulse with a broken line at 2 Hz is originally - 4 Hz
component. In other words,
( ) j t
e
2 4
is giving rise to same set of samples as
( ) j t
e
2 2
when sampled with
s
T
1
6
= sec. This is easy to verify.

( )
j n j n
j t j n
n
t
e e e e
4 4
2 4 2
3 3
6

=
= =

6.15

( )
j n
j t
n
t
e e
2
2 2
3
6
=
= =
Similarly, the set of values obtained from
j t
e
8
for
s
t nT = is the same as those
from
j t
e
4
(Notice that in Fig. 6.10(b), there is a solid line at f 2 = ). In other
words, a 4 Hz cosine signal becomes an alias of 2 Hz cosine signal. If there were
to be a 2 Hz component in ( ) X f , the 4 Hz will interfere with the 2 Hz component
and thereby leading to distortion due to under sampling. In fact, a frequency
component at f f
2
= is the alias of another lower frequency component at f f
1
=
if
s
f k f f
2 1
= , where k is an integer. Hence, with
s
f 6 = , we find that the 4 Hz
component is an alias the of 2 Hz component, because 4 =6 - 2. Similarly,
( )t cos 2 8

as 8 =6 +2. A few of the aliases of the 2 Hz component with
s
f 6 = have been shown on the aliasing diagram of Fig. 6.11.

Fig. 6.11: Aliasing diagram with
s
f 6 =

As seen from the figure, upto 3 Hz there is no aliasing problem. Beyond that, the
spectrum folds back, making a 4 Hz component align with a 2 Hz component.
With repeated back and forth folding, we find that 8 Hz, 10 Hz, 14 Hz, etc. are
aliases of 2 Hz and all of them will contribute to the output at 2 Hz, when ( )
s
X f
is filtered by an ILPF with cutoff at
s
f
3
2
= Hz.

6.16
Note: Reconstruction of the analog signal is done by placing an ideal LPF with a
cutoff at
s
f
2
. If ( ) t cos 8 were to be sampled at 6 samples/sec, the output of the
lowpass filter (with a cutoff at 3 Hz) is cosine signal with the frequency of
2 Hz!

Fig. 6.12: Aliasing picture in the time domain

The concept of aliasing can also be illustrated in the time domain. This is shown
in Fig. 6.12. The red dots represent the samples from the two cosine signals (one
at 4 Hz and the other at 2 Hz) at t
1 1 2 3
, 0, , ,
6 6 6 6
= . By extrapolation, we
find that ( ) t cos 8 will give rise to same set of samples as ( ) t cos 4 at
n
t n , 0 , 1, 2 etc.
6
= =

We will now demonstrate the effect of aliasing on voice and music signals.
There are two voice recordings: one is a spoken sentence and the other is part of
a Sanskrit sloka rendered by a person with a very rich voice. The music signal
pertains to sa re ga ma pa da ne ... sounds generated by playing a flute. (The
recording of the flute signal was done on a cell phone.)
1) a. The signal pertaining to the spoken sentence God is subtle but not
malicious, is sampled at 44.1 ksps and reconstructed from these

6.17
samples.
b. The reconstructed signal after down sampling by a factor of 8.
2) a. Last line of a Sanskrit sloka. Sampling rate 44.1 ksps.
b. Reconstructed signal after down sampling by a factor of 8.
c. Reconstructed signal after down sampling by a factor of 12.
3) a. Flute music sampled at 8 ksps.
b. Reconstructed signal after down sampling by a factor of 2.

0 1000 2000 3000 4000 5000 6000
0
100
200
300
400
500
600
frequency (Hz)
m
a
g
n
i
t
u
d
e
Magnitude spectrum of original version

Fig. 6.13(a): Spectrum of the signal at (2a)

6.18
0 500 1000 1500 2000 2500
0
20
40
60
80
100
120
140
160
180
200
frequency (Hz)
m
a
g
n
i
t
u
d
e
Magnitude spectrum of aliased version (downsampling factor = 8)

Fig. 6.13(b): Spectrum of the signal at (2b)

0 200 400 600 800 1000 1200 1400 1600 1800
0
20
40
60
80
100
120
140
frequency (Hz)
m
a
g
n
i
t
u
d
e
Magnitude spectrum of aliased version (downsampling factor = 12)

Fig. 6.13(c): Spectrum of the signal at (2c)

6.19
With respect to 1b, we find that there is noticeable distortion when the word
malicious is being uttered. As this word gives rise to spectral components at the
higher end of the voice spectrum, fold back of the spectrum results in severe
aliasing. Some distortion is to be found elsewhere in the sentence.

Aliasing effect is quite evident in the voice outputs 2(b) and (c). (It is much
more visible in 2(c).) We have also generated the spectral plots corresponding to
the signals at 2(a), (b) and (c). These are shown in Fig. 6.13(a), (b) and (c). From
Fig. 6.13(a), we see that the voice spectrum extends all the way upto 6 kHz with
strong spectral components in the ranges 200 Hz to 1500 Hz and 2700 Hz to
3300 Hz. (Sampling frequency used is 44.1 kHz.)

Fig. 6.13(b) corresponds to the down sampled (by a factor of 8) version of
(a); that is, sampling frequency is about 5500 samples/sec. Spectral components
above 2750 Hz will get folded over. Because of this, we find strong spectral
components in the range 2500-2750 Hz.

Fig. 6.13(c) corresponds to a sampling rate of 3600 samples/sec. Now
spectrum fold over takes place with respect to 1800 Hz and this can be easily
seen in Fig. 6.13(c).

As it is difficult to get rid of aliasing once it sets in, it is better to pass the
signal through an anti-aliasing filter with cutoff at
s
f
2
and then sample the output
of the filter at the rate of
s
f samples per second. This way, we are sure that all
the spectral components upto
s
f
f
2
will be preserved in the sampling process.
Example 6.1 illustrates that the reconstruction mean square error is less than or
equal to the error resulting from sampling without the anti-aliasing filter.


6.20
Example 6.1
An analog signal ( ) x t is passed through an anti-aliasing filter with cutoff
at
c
f and then sampled with
s c
f f 2 = . (Note that ( ) X f is not band limited to
c
f .)
Let ( ) y t
1
be the signal reconstructed from these samples and
( ) ( ) ( )
e x t y t d t
2
1 1

Now the anti-aliasing filter is withdrawn and ( ) x t is sampled directly with
s c
f f 2 = as before. Let ( ) y t
2
be the signal reconstructed from these samples
and let ( ) ( ) ( )
e x t y t d t
2
2 2
. Show that e e
2 1
.

Let ( ) X f be as shown in the figure 6.14(a), with
N
f being the highest
frequency in it. ( ) Y f
1
is as shown in Fig. 6.14(b).

As can be shown from the figure, ( ) Y f
1
does not have any aliasing but
that part of ( ) X f for
c N
f f f is missing. This introduces some distortion and
the energy of the error signal,
( ) ( ) ( )
e y t x t d t
2
1 1

( ) ( ) Y f X f d f
2
1

But ( )
( )
c
c
X f f f
Y f
f f
1
,
0 ,

=

>
.
Hence, ( ) ( )
c
c
f
f
e X f d f X f d f
2 2
1

= +

(6.9a)


6.21

Fig. 6.14: (a) Original spectrum, ( ) X f
(b) Spectrum of the reconstructed signal after
( ) x t is passed through an anti-aliasing filter
before sampling
(c) Spectrum of the reconstructed signal with
aliasing

It is evident from Fig. 6.14(c), that ( ) y t
2
suffers from aliasing and
( ) ( ) ( ) ( )
c c
c c
f f
f f
e X f d f X f d f Y f X f d f
2 2 2
2 2

= + +

(6.9b)

6.22
But ( ) ( ) Y f X f
2
for
c
f f . Comparing Eqs. 6.9(a) and (b), we find that Eq.
6.9(b) has an extra term which is greater or equal to zero. Hence e e
2 1
.

Example 6.2
Find the Nyquist sampling rate for the signal
( ) ( ) ( ) x t c t c t
2
sin 200 sin 1000 = .

( ) c t sin 200 has a rectangular spectrum in the interval f 100 Hz and
( ) ( ) ( ) c t c t c t
2
sin 1000 sin 1000 sin 1000 =

has a triangular spectrum in the
frequency range f 1000 Hz. Hence ( ) X f has spectrum confined to the range
f 1100 Hz. This implies the Nyquist rate is 2200 samples/sec.

Example 6.3
a) Consider the bandpass signal ( ) x t with ( ) X f as shown in Fig. 6.15. Let
( ) x t be sampled at the rate of 20 samples per second. Sketch ( )
s
X f (to
within a scale factor), for the frequency range f 20 30 Hz.
b) Let ( ) x t be now sampled as the rate of 24 samples per second. Sketch
( )
s
X f (to within a scale factor) for f 20 30 Hz.

Fig. 6.15: ( ) X f for example 6.3

6.23
a) Let P denote ( ) X f for f 0 and let N denote ( ) X f for f 0 < . (N is
shown with a broken line.) Consider the right shifts of ( ) X f in multiples of
20 Hz, which is the sampling frequency. Then P will be shifted away from
the frequency interval of interest. We have to only check whether shifted N
can contribute to spectrum in the interval f 20 30 . It is easy to see
that this will not happen. This implies, in ( )
s
X f , spectrum has the same
shape as ( ) X f for f 20 30 . This is also true for the left shifts of ( ) X f .
That is, for f 20 30 , ( )
s
X f , to within a scale factor, is the same as
( ) X f .
b) Let
s
f 24 = . Consider again right shifts. When N is shifted to the right by
s
f 2 48 = , it will occupy the interval ( ) 30 48 18 + = to ( ) 20 48 28 + = .
That is, we have the situation shown in Fig. 6.16.

Fig. 6.16: The two lobes contributing to the spectrum for f 20 30

6.24
Let the sum of (a) and (b) be denoted by ( ) Y f . Then, ( ) Y f is as shown in Fig.
6.17.

Fig. 6.17: Spectrum for f 20 30 when
s
f 24 =

For f 30 20 , ( ) Y f is the mirror image of the spectrum shown in Fig.
6.17. Notice that for the bandpass signal, increasing
s
f has resulted in the
distortion of the original spectrum.

Example 6.4
Let ( ) ( ) ( ) x t t t 2cos 800 cos 1400 = + . ( ) x t is sampled with the
rectangular pulse train ( )
p
x t shown in Fig. 6.18. Find the spectral components in
the sampled signal in the range 2.5 kHz to 3.5 kHz.

Fig. 6.18: ( )
p
x t of the Example 6.4

6.25
( ) X f has spectral components at 400 Hz and 700 Hz. The impulses
in ( )
p
X f occur at 1 kHz, 2 kHz, 4 kHz, 5 kHz, etc. (Note the absence
of spectral lines in ( )
p
X f at f 3 = kHz, 6 kHz, etc.) Convolution of ( ) X f
with the impulses in ( )
p
X f at 2 kHz will give rise to spectral components at 2.4
kHz and 2.7 kHz. Similarly, we will have spectral components at ( ) 4 0.4 3.6 =
kHz and ( ) 4 0.7 3.3 = kHz. Hence, in the range 2.5 kHz to 3.5 kHz, sampled
spectrum will have two components, namely, at 2.7 kHz and 3.3 kHz.

Exercise 6.1
A sinusoidal signal ( )
( )
x t A t
3
cos 2 10

=

is sampled with a
uniform impulse train at the rate of 1500 samples per second. The sampled
signal is passed through an ideal lowpass filter with a cutoff at 750 Hz. By
sketching the appropriate spectra, show that the output is a sinusoid at 500
Hz. What would be the output if the cutoff frequency of the LPF is 950 Hz?

Exercise 6.2
Let ( ) x t be the sum of two cosine signals. ( ) X f is shown in Fig. 6.19.
Let ( ) x t be sampled at the rate of 300 samples per second. Let ( )
s
x t be
passed through an ideal LPF with a cutoff at 150 Hz.
a) Sketch the spectrum at the output of the LPF.
b) Write the expression for the output of the LPF.

Fig. 6.19: ( ) X f for the Exercise 6.2

6.26
6.4 Quantization
6.4.1 Uniform quantization
An analog signal, even it is limited in its peak-to-peak excursion, can in
general assume any value within this permitted range. If such a signal is
sampled, say at uniform time intervals, the number of different values the
samples can assume is unlimited. Any human sensor (such as ear or eye) as the
ultimate receiver can only detect finite intensity differences. If the receiver is not
able to distinguish between two sample amplitudes, say v
1
and v
2
such that
v v
1 2
2
< , then we can have a set of discrete amplitude levels separated by

and the original signal with continuous amplitudes, may be approximated by a
signal constructed of discrete amplitudes selected on a minimum error basis from
an available set. This ensures that the magnitude of the error between the actual
sample and its approximation is within
2
and this difference is irrelevant from the

receiver point of view. The realistic assumption that a signal ( ) m t is (essentially)
limited in its peak-to-peak variations and any two adjacent (discrete) amplitude
levels are separated by will result in a finite number (say L) of discrete
amplitudes for the signal.

The process of conversion of analog samples of a signal into a set of
discrete (digital) values is called quantization. Note however, that quantization is
inherently information lossy. For a given peak-to-peak range of the analog signal,
smaller the value of , larger is the number of discrete amplitudes and hence,
finer is the quantization. Sometimes one may resort to somewhat coarse
quantization, which can result in some noticeable distortion; this may, however
be acceptable to the end receiver.

The quantization process can be illustrated graphically. This is shown in
Fig. 6.20(a). The variable x denotes the input of the quantizer, and the variable

6.27
y represents the output. As can be seen from the figure, the quantization
process implies that a straight line relation between the input and output (broken
line through the origin) of a linear continuous system is replaced by a staircase
characteristic.

The difference between two adjacent discrete values, , is called the step
size of the quantizer. The error signal, that is, difference between the input and
the quantizer output has been shown in Fig. 6.20(b). We see from the figure that
the magnitude of the error is always less than or equal to
2
. (We are assuming

that the input to the quantizer is confined to the range
7 7
to
2 2

.

Fig. 6.20: (a) Quantizer characteristic of a uniform quantizer (b) Error signal

6.28
Let the signal ( ) m t , shown in Fig. 6.21(a), be the input to a quantizer with
the quantization levels at 0, 2 and 4 . Then, ( )
q
m t , the quantizer output is
the waveform shown in red. Fig. 6.21(b) shows the error signal
( ) ( ) ( )
q
e t m t m t = as a function of time. Because of the noise like appearance
of ( ) e t , it is a common practice to refer to it as the quantization noise. If the input
to the quantizer is the set of (equispaced) samples of ( ) m t , shown in Fig.
6.21(c), then the quantizer output is the sample sequence shown at (d).

Fig. 6.21: (a) An analog signal and its quantized version (b) The error signal


6.29

Fig. 6.21: (c) Equispaced samples of ( ) m t (d) Quantized sample sequence

For the quantizer characteristic shown in Fig. 6.20(a), the maximum value
of the quantizer output has been shown as 3 and the minimum as ( ) 3 .
Hence, if the input x is restricted to the range 3.5 (that is,

6.30

=

x x
max min
7
2
, the quantization error magnitude is less than or equal to
2
. Hence, (for the quantizer shown), 3.5 is treated as the overload level.

The quantizer output of Fig. 6.20(a) can assume values of the form
i
H
where
i
H 0, 1, 2, = . A quantizer having this input-output relation is said to
be of the mid-tread type, because the origin lies in the middle of a tread of a
staircase like graph.

A quantizer whose output levels are given by
i
H
2
, where
i
H 1, 3, 5, = is referred to as the mid-riser type. In this case, the origin lies
in the middle of the rising part of the staircase characteristic as shown in Fig.
6.22. Note that overload level has not been indicated in the figure. The difference
in performance of these two quantizer types will be brought out in the next sub-
section in the context of idle-channel noise. The two quantizer types described
above, fall under the category of uniform quantizers because the step size of
the quantizer is constant. A non-uniform quantizer is characterized by a variable
step size. This topic is taken up in the sub-section 6.4.3.


6.31

Fig. 6.22: Mid-riser type quantizer

6.4.2 Quantization Noise
Consider a uniform quantizer with a total of L quantization levels,
symmetrically located with respect to zero. Let x represent the quantizer input,
and ( ) y QZ x = denote its output. Consider a part of the quantizer operating
range shown in Fig. 6.23.

Let
k
I be the interval defined by

{ }
k k k
I x x x k L
1
, 1, 2, ,
+
= < = .
Then
k
y y = , if
k
x I ; we can also express
k
y as

k
y x q = + , if
k
x I
where q denotes the quantization error, with q
2
.


6.32

Fig. 6.23: Input intervals and the corresponding output levels of a quantizer

Let us assume that the input x is the sample value of a random variable
X with zero mean and variance
X
2
. When the quantization is fine enough, the
distortion produced by the quantization affects the performance of a PCM system
as though it were an additive, independent source of noise with zero mean and
mean square value determined by the quantizer step size .

Let the random variable Q denote the quantization error, and let q denote
its sample value. Assuming Q is uniformly distributed in the range ,
2 2

, we
have
( )
Q
q
f q
otherwise
1
,
2 2
0 ,

where ( )
Q
f q is the PDF of the quantization error. Hence, the variance of the
quantization error,
Q
2
is

6.33

Q
2
2
12
= (6.10)
The reconstructed signal at the receiver output, can be treated as the sum of the
original signal and quantization noise. We may therefore define an output signal-
to-quantization noise ratio ( )
q
SNR
0,
as
( )
X X
q
Q
SNR
2 2
2 2 0,
12
= =

(6.11a)
As can be expected, Eq. 6.11(a) states the result for a given
X
, smaller the
value of the step size , larger is the ( )
q
SNR
0,
. Note that
Q
2
is independent of
the input PDF provided overload does not occur. ( )
q
SNR
0,
is usually specified in
dB.
( ) ( ) ( )
q q
SNR SNR
10
0, 0,
in dB 10log

=

(6.11b)

Idle Channel Noise
Idle channel noise is the coding noise measured at the receiver output with zero
input to the transmitter. (In telephony, zero input condition arises, for example,
during a silence in speech or when the microphone of the handset is covered
either for some consultation or deliberation). The average power of this form of
noise depends on the type of quantizer used. In a quantizer of the mid-riser type,
zero input amplitude is coded into one of two innermost representation levels,
2
. Assuming that these two levels are equiprobable, the idle channel noise for
mid-riser quantizer has a zero mean with the mean squared value

2 2
2
1 1
2 2 2 2 4

+ =

On the other hand, in a quantizer of the mid-tread type, the output is zero for zero
input and the ideal channel noise is correspondingly zero. Another difference
between the mid-tread and mid-riser quantizer is that, the number of quantization
levels is odd in the case of the former where as it is even for the latter. But for

6.34
these minor differences, the performance of the two types of quantizers is
essentially the same.

Example 6.5
Let a baseband signal ( ) x t be modeled as the sample function of a zero
mean Gaussian random process ( ) X t . ( ) x t is the input to a quantizer with the
overload level set at
X
4 , where
X
is the variance of the input process (In
other words, we assume that the ( )
X
P x t 4 0

). If the quantizer output
is coded with R -bit sequence, find an expression for the ( )
q
SNR
0,
.

With R -bits per code word, the number of quantization levels,
R
L 2 = .
For calculating the step size , we shall take ( ) x x t
max
max
= as
X
4 .
Step size
X X
R
x
L L
max
2 8 8
2

= = =
( )
X X
q
Q
SNR
2 2
2 2 0,
12

= =

( )
R
X
X
2 2
2
12 2
64

R 2
3
2
16
=
( ) ( )
q
SNR R
0,
in dB 6.02 7.2 =
(6.12)

This formula states that each bit in the code word of a PCM system contributes 6
dB to the signal-to-noise ratio. Remember that this formula is valid, provided,
i) The quantization error is uniformly distributed.
ii) The quantization is fine enough (say n 6 ) to prevent signal correlated
patterns in the error waveform.

6.35
iii) The quantizer is aligned with the input amplitude range (from
X
4 to
X
4 ) and the probability of overload is negligible.

Example 6.6
Let ( ) x t be modeled as the sample function of a zero mean stationary
process ( ) X t with a uniform PDF, in the range ( ) a a , . Let us find the
( )
q
SNR
0,
assuming an R -bit code word per sample.

Now the signal ( ) x t has only a finite support; that is, x a
max
= . Its
variance,
X
a
2
2
3
= .
Step size
R
a 2
2
=

( ) R
a
1
2

=
Hence, ( )
( )
= =

M
q R
a
SNR
a
2
2
0, 2 1 2 2
3
2
12 12

=
R 2
2
( ) ( )
q
SNR R
0,
in dB 6.02 = (6.13a)
Even in this case, we find the 6 dB per bit behavior of the ( )
q
SNR
0,
.

Example 6.7
Let ( ) m t , a sinusoidal signal with the peak value of A be the input to a
uniform quantizer. Let us calculate the ( )
q
SNR
0,
assuming R -bit code word per
sample.


6.36
Step size
R
A 2
2
=
Signal power
A
2
2
=
( )
R
q
A
SNR
A
2 2
2 0,
2 12
2
4
=

( )
R 2
3
2
2
=
( ) ( )
q
SNR R
0,
in dB 6.02 1.8 = +
(6.13b)

We now make a general statement that ( )
q
SNR
0,
of a PCM system using a
uniform quantizer can be taken as R 6 + , where R is the number of bits/code
word and is a constant that depends on the ratio
rms X
x
x x
max max

=

, as shown
below.

R
x
max
2
2
=

Q
R
x
2 2
2 max
2
12
2 3
= =

( )
R
X X
q
Q
SNR
x
2
2
2
2 0,
max
3 2

= =

( )
X
q
SNR R
x
10 10
0,
max
10log 4.77 6.02 20log

= + +

R 6.02 = + (6.14a)
where
X
x
10
max
4.77 20log

= +

(6.14b)


6.37
Example 6.8
Let ( ) m t be a real bandpass signal with a non-zero spectrum (for f 0 > )
only in the range
L
f to
H
f , with
H L
f f . Then, it can be shown that, the minimum
sampling frequency
1
to avoid aliasing is,
( )
L
H
s
L
f
f B
f B
f k
B
min
1
2
2
1

+

= =

+

(6.15)
Where
H L
B f f = ,
H
f
k
B
= and x

denotes the integer part of x . Let ( ) m t be
a bandpass signal with
L
f 3 = MHz and
H
f 5 = MHz. This signal is to be sent
using PCM on a channel whose capacity is limited to 7 Mbps. Assume that the
samples of ( ) m t are uniformly distributed in the range - 2 V to 2 V.
a) Show that the minimum sampling rate as given by Eq. 6.15 is adequate to
avoid aliasing.
b) How many (uniform) quantization levels are possible and what are these
levels so that the error is uniform with in
2
.
c) Determine the ( )
q
SNR
0,
(in dB) that can be achieved.

a) Eq. 6.15 gives ( )
s
f
min
as
( )
s
f
6
min
3
1
2
2 2 10
3
1
2
+
=

+

6
5 10 = samples/sec
Let ( ) M f , the spectrum of ( ) m t be as shown in Fig. 6.24.

1
Note that a sampling frequency ( )
s s
f f
min
> could result in aliasing (see Example 6.3) unless
s H
f f 2 , in which case the sampling of a bandpass signal reduces to that of sampling a
lowpass signal.

6.38

Fig. 6.24: Bandpass spectrum of Example 6.8

As
s
f
6
5 10 = samples/sec, ( ) M f is to be repeated at intervals of
6
5 10
Hz. Shifting N to the right by 5 MHz will create a spectral lobe between 0 to
2 MHz whereas shifting by 10 MHz will result in a spectral lobe occupying
the range 5 to 7 MHz. That is, right shifts of N will not interfere with P .
Similarly, left shifts of P (in multiples of 5 MHz) will not fall in the range - 5
MHz to - 3 MHz. Hence, there is no aliasing in the sampling process.
b) As the sampling rate is
6
5 10 samples/sec and the capacity of the
channel is limited to 7Mbps; it is possible to send only one bit per sample;
that is, we have only a two-level quantization. The quantizer levels are at
1 so that step size is 2 V and the error is uniform in the range 1 .
c) As the signal is uniformly distributed in the range ( ) 2, 2 , we have
( )
M
2
2
4
4
12 3
= = .
Variance of the quantization noise,
( )
Q
2
2
2
1
12 3
= =
Hence ( )
q
SNR
0,
4
3
4
1
3
= =
( ) ( ) = =
q
SNR
10
0,
dB 10 log 4 6.02 dB


6.39
When the assumption that quantization noise is uniformly distributed
between
2
, is not valid, then we have to take a more basic approach in

calculating the noise variance. This is illustrated with the help of Example 6.9.

Example 6.9
Consider the quantizer characteristic shown in Fig. 6.25(a). Let X be the
input to the quantizer with ( )
X
f x as shown at Fig. 6.25(b). Find
a) the value of A
b) the total quantization noise variance,
Q
2

c) Is it is the same as
2
12
?

Fig. 6.25(a): Quantizer characteristic (b) the input PDF (Example 6.9)


6.40
a) As ( )
X
f x d x
4
4
1
, A
1
4
=
b) ( )
X
x x
f x
otherwise
1 1
, 4
4 16
0 ,

Let us calculate the variance of the quantization noise for x 0 . Total
variance is twice this value. For x 0 > , let
( ) ( ) ( ) ( )
Q X X
x f x d x x f x d x
2 4
2 2
2
0 2
'
1 3 = +

Carrying out the calculations, we have
Q
2
1
'
6
= and hence,
Q
2
1
3
=
c) As
2
4 1
12 12 3
= = , in this case we have

Q
2
the same as
2
12
.

Exercise 6.3
The input to the quantizer of Example 6.9 (Fig. 6.25(a)) is a random
variable with the PDF,
( )
x
X
Ae x
f x
otherwise
, 4
0 ,

Find the answers to (a), (b) and (c) of Example 6.9.
Answers:
Q
2
2
0.037
12
= .

Exercise 6.4
A random variable X , which is uniformly distributed in the range 0 to
1 is quantized as follows:
( ) QZ x x 0, 0 0.3 =
( ) QZ x x 0.7, 0.3 1 = <
Show that the root-mean square value of the quantization is 0.198.

6.41

6.4.3 Non-uniform quantization and companding
In certain applications, such as PCM telephony, it is preferable to use a
non-uniform quantizer, where in the step size is not a constant. The reason is
as follows. The range of voltages covered by voice signals (say between the
peaks of a loud talk and between the peaks of a fairly weak talk), is of the order
of 1000 to 1. For a uniform quantizer we have seen that ( )
X
q
SNR
2
2 0,
12
=
,
where is a constant. Hence ( )
q
SNR
0,
is decided by signal power. For a
person with a loud tone, if the system can provide an ( )
q
SNR
0,
of the order of
40 dB, it may not be able to provide even 10 dB ( )
q
SNR
0,
for a person with a
soft voice. As a given quantizer would have to cater to various speakers, (in a
commercial setup, there is no dedicated CODEC
1
for a given speaker), uniform
quantization is not the proper method. A non-uniform quantizer with the feature
that step size increases as the separation from the origin of the input -output
characteristic is increased, would provide a much better alternative. This is
because larger signal amplitudes are subjected to coarser quantization and the
weaker passages are subjected to fine quantization. By choosing the
quantization characteristic properly, it would be possible to obtain an acceptable
( )
q
SNR
0,
over a wide range of input signal variation. In other words, such a

1
CODEC stands for COder-DECoder combination
Exercise 6.5
A signal ( ) m t with amplitude variations in the range
p
m to
p
m is to
be sent using PCM with a uniform quantizer. The quantization error should be
less than or equal to 0.1 percent of the peak value
p
m . If the signal is band
limited to 5 kHz, find the minimum bit rate required by this scheme.
Ans:
5
10 bps

6.42
quantizer favors weak passages (which need more protection) at the expense of
loud ones.

The use of a non-uniform quantizer is equivalent to passing the baseband
signal through a compressor and then applying the compressed signal to a
uniform quantizer. In order to restore the signal samples to their correct relative
level, we must of course use a device in the receiver with a characteristic
complimentary to the compressor. Such a device is called an expander. A non-
uniform quantization scheme based on this principle is shown in Fig. 6.26. In the
scheme shown, ( ) C x denotes the output of the compressor for the input x . The
characteristic ( ) C x is a monotonically increasing function that has odd
symmetry, ( ) ( ) C x C x = . Ideally, the compression and expansion laws are
exactly inverses so that except for the effect of quantization, the expander output
is equal to the compressor input. In the scheme of Fig. 6.26, ( ) C
1
.
denotes the
expander characteristic. The combination of a COMpressor and an exPANDER
is called a compander. We shall now derive the compressor characteristic ( ) C . ,
which is capable of giving rise to a fairly constant ( )
q
SNR
0,
over a wide range of
the input signal variation.

Fig. 6.26: Non-uniform quantization through companding

Let the input signal x be bounded in the range ( ) x x
max max
to . The
characteristic ( ) C x analytically defines non-uniform intervals
k
via uniform
intervals of
x
L
max
2
(Fig. 6.27). L is the number of quantization levels and is
assumed to be large. Note that the uniform quantizer in Fig. 6.27 is the mid-riser

6.43
variety. When x I
6
, ( ) C x I
6
'
and the input to the UQ will be in the range
( ) , 2 . Then the quantizer output is
3
2
. When
k
x I , the compressor
characteristic ( ) C x , may be approximated by a straight line segment with a
slope equal to
k
x
L
max
2
, where
k
is the width of the interval
k
I . That is,
( )
( )
k
k
dC x x
C x x I
d x L
max
2
'
, =

As ( ) C x
'
is maximum at the origin, the equivalent step size is smallest at
x 0 = and
k
is the largest at x x
max
= . If L is large, the input PDF ( )
X
f x can
be treated to be approximately constant in any interval
k
I k L , 1, , = . Note
that we are assuming that the input is bounded in practice to a value x
max
, even
if the PDF should have a long tail. We also assume that ( )
X
f x is symmetric; that
is ( ) ( )
X X
f x f x = . Let ( ) ( )
X X k
f x f y constant = ,
k
x I , where

k k
k
x x
y
1
2
+
+
=
and
k k k k
length of I x x
1 +
= =
Then,
[ ] ( )
k
k
x
k k X
x
P P x I f x d x
1 +
= =

, and,
L
k
k
P
1
1
=
=

Let
k
q denote the quantization error when
k
x I .
That is, ( )
k k
q QZ x x x I , =

k k
y x x I , =


6.44

Fig. 6.27: A typical compression characteristic and non-uniform quantization
levels

Let the corresponding random variable be
k
Q . We can assume
k
Q to be
uniformly distributed in the range
k k
,
2 2

.

k
k
Q
2
2
12
=

Q
2
= variance of the total quantization noise

L L
k
k k k
k k
P Q P
2
2
1 1
12
= =
= =

Note that if
k
= for all k , then
Q
2
2
12
= , as expected.
But
( )
k k
x
x I
LC x
max
2
,
'

Hence
( )
L
Q k
k
x
P
L C x
2
2
max
2
2
1
4 1
12
'
=


6.45
( )
L
k
k
x
P C x
L
2 2
max
2
1
'
3

( ) ( )
k
k
x
L
X
k
x
x
f x C x d x
L
1
2 2
max
2
1
'
3
+

=

( ) ( )
x
X
x
x
f x C x d x
L
max
max
2 2
max
2
'
3
(6.16a)
This approximate formula for
Q
2
is referred to as Bennets formula.
( )
( )
( ) ( )
X
X
x q
Q
X
x
x f x d x
L
SNR
x
f x C x d x
max
max
2
2 2
2 2 0,
2
max
3
'

( )
( ) ( )
x
X
x
x
X
x
x f x d x
L
x
f x C x d x
max
max
max
max
2
2
2
2
max
3
'
(6.16b)
From the above approximation, we see that it is possible for us to have a
constant ( )
q
SNR
0,
i.e. independent of
X
2
, provided
( ) C x
K x
1
'
= where K is a constant, or ( ) C x K x
2
2 2 '

=

Then, ( )
q
L
SNR
K x
2
2 2 0,
max
3
=
As ( ) C x
K x
1
'
= , we have
( )
x
C x
K
ln
= + , with x 0 > and being a constant. (Note that
e
x x ln log = .)

6.46
The constant is arrived at by using the condition, when x x
max
= ,
( ) C x x
max
= . Therefore,
x
x
K
max
max
ln
= .
Hence, ( )
x x
C x x
K K
max
max
ln ln
= +

x
x x
K x
max
max
1
ln , 0

= + >

As ( ) ( ) C x C x = we have
( ) ( )
x
C x x x
K x
max
max
1
ln sgn

= +

, (6.17)
where, ( )
x
x
x
1, 0
sgn
1, 0
>
=

<

As the non-uniform quantizer with the above ( ) C x results in a constant
( )
q
SNR
0,
, it is also called robust quantization.
The above ( ) C x is not realizable because ( )
x
C x
0
lim
. We want
( )
x
C x
0
lim 0
. That is, we have to modify the theoretical companding law. We

discuss two schemes, both being linear at the origin.

(i) A-law companding:
( )
( )
( )
A x x
x
A x A
A x
C x
x x
x x
A A x
max
max
max
max
1
sgn , 0
1 ln
1 ln
1
sgn , 1
1 ln

= +

+

(6.18)
The constant A decides the compression characteristic, with A 1 = providing a
linear I-O relationship. The behavior of ( ) C x for A 2 = and 87.56 are shown in
Fig. 6.28.

6.47

Fig. 6.28: A-law compression characteristic
A 87.56 = is the European commercial PCM standard which is being followed
in India.

(ii) -law companding:
( )
( )
( )
x
x
C x x x
max
max
ln 1
sgn
ln 1

+

=

+

(6.19)
( ) C x is again quasi-logarithmic. For small values of ( ) x x 0 > , such that
x x
max
<< , ( )
( )
C x x
ln 1
+
, which indicates a linear relationship between
the input x and output ( ) C x . Note that ( ) ( ) C x C x = . If x is such that
x x
max
>> , then ( ) C x is logarithmic, given by
( )
( )
( )
x x
C x x
x
max
max
ln sgn
ln 1

+


6.48
is a constant, whose value decides the curvature of ( ) C x ; 0 = corresponds
to no compression and as the value of increases, signal compression
increases, as shown in Fig. 6.29.

Fig. 6.29: -law compression characteristic

255 = is the north American standard for PCM voice telephony. Fig. 6.30
compares the performance of the 8-bit -law companded PCM with that of an 8
bit uniform quantizer; input is assumed to have the bounded Laplacian PDF
which is typical of speech signals. The ( )
q
SNR
0,
for the uniform quantizer is
based on Eq. 6.14. As can be seen from this figure, companding results in a
nearly constant signal-to-quantization noise ratio (within 3 dB of the maximum
value of 38 dB) even when the mean square value of the input changes by about
30 dB (a factor of
3
10 ).
c
G is the companding gain which is indicative of the
improvement in SNR for small signals as compared to the uniform quantizer. In
Fig. 6.30,
c
G has been shown to be about 33 dB. This implies that the smallest
step size,
min
is about 32 times smaller than the step size of a corresponding

6.49
uniform quantizer with identical x
max
and L, the number of quantization levels.
min
would be smaller by a factor more than 32, as a
c
G of 30 dB would give
rise to
min
1
32
= . (
uniform
) The A - law characteristic, with A 87.56 = has a
similar performance as that of 255 = law, with a companding gain of about 24
dB; that is, for low amplitude signals, a uniform quantizer would require four
additional bits to have the same ( )
q
SNR
0,
performance of the A 87.56 =
companded quantizer. (See Exercise 6.9)

Fig. 6.30: ( )
q
SNR
0,
performance of -law companded and uniform quantization

A number of companies manufacture CODEC chips with various
specifications and in various configurations. Some of these companies are:
Motorola (USA), National semiconductors (USA), OKI (J apan). These ICs include
the anti-aliasing bandpass filter (300 Hz to 3.4 kHz), receive lowpass filter, pin
selectable A-law / -law option. Listed below are some of the chip numbers
manufactured by these companies.
Motorola: MC145500 to MC145505

6.50
National: TP3070, TP3071 and TP3070-X
OKI: MSM7578H / 7578V / 7579
Details about these ICs can be downloaded from their respective websites.

Example 6.10
A random variable X with the density function ( )
X
f x shown in Fig. 6.31 is
given as input to a non-uniform quantizer with the levels q
1
, q
2
and q
3
.

Fig. 6.31: Input PDF of Example 6.10

q
1
, q
2
and q
3
are such that
( ) ( ) ( )
q q q
X X X
q q q
f x d x f x d x f x d x
3 2 1
3 2 2
1
6

= = = =

(6.20)
a) Find q
1
, q
2
and q
3
.
b) Suggest a compressor characteristic that should precede a uniform
quantizer such that Eq. 6.20 is satisfied. Is this unique?

a) ( ) ( )
q q
X X
q
f x d x f x d x
1 1
1
0
1
2
6
= =

Hence, ( )
q
x d x
1
0
1
1
12
=


6.51
Solving, we obtain q
1
0.087 =
Similarly we have equations
( )
q
X
f x d x
2
0.087
1
6
=
(6.21a)
( )
q
X
q
f x d x
3
2
1
6
=
(6.21b)
From Eq. 6.21(a), we get q
2
0.292 = and solving Eq. 6.21(b) for q
3
yields
q
3
0.591 = .
b) A uniform quantizer will have levels at
1
6
,
1
2
and
5
6
.
Let ( ) C x be the compressor characteristic which is a non-decreasing
function of its argument and is anti-symmetrical about x 0 = . Any
compression characteristic that does the following mapping will take care of
the requirements of the problem.

x ( ) C x
0.0087
1
6

0.292
1
2

0.591
5
6

1 1

As such, ( ) C x is not unique.

Example 6.11
Consider a companded PCM scheme using the -law characteristic as
given by

6.52
( )
( )
( )
( )
x
C x x x
ln 1
sgn , 0, 1
ln 1
+
= >
+

Find the ( )
q
SNR
0,
expected of this system when the input X is uniformly
distributed in the range ( ) 1, 1 .

Using Bennets formula, Eq. 6.16(a), let us compute
Q
2
. From the ( ) C x
given, we have

( )
( )
( )
dC x
C x
d x x
1
'
ln 1 1
= =
+ +

( )
( )
( )
C x x
2
2
2 ln 1
'
1
+

= +

( ) ( )
Q X
f x C x d x
L
1
2
2
2
0
2
'
3

But ( )
X
x
f x
otherwise
1
, 1
2
0 ,

Carrying out the integration,

( )
Q
L
2
2
2
2 2
ln 1
1
3
3
+

= + +

X
2
1
3
=
( )
( )
q
L
SNR
1 2
2
0,
1
ln 1 3

= + +

+

If 100 , then
2
3
>> and 1 >> , and ( )

( )
q
L
SNR
2
2 0,
3
ln 1 +


6.53
Example 6.12
Let ( )
q
SNR
0,
of a -law companded PCM be approximated as
( )
( )
q
L
SNR
2
2 0,
3
ln 1 +

We will show that ( )
q
SNR
0,
(in dB) follows the ( ) R 6 + rule.

Let
R
L 2 = . Then
( )
( )
( )
R
q
SNR
2
2 0,
3
2
ln 1

+

R
C 4 , where
( )
C
2
3
ln 1
=
+

Then ( ) ( )
q
SNR C R
10
0,
in dB 10log 6 = +
R 6 = +

Example 6.13
A music signal band-limited to 15 kHz is sampled at the rate of
3
45 10
samples/sec and is sent using 8 bit -law ( = 255) companded PCM.
( )
q
SNR
0,
of this system was found be inadequate by atleast 10 dB. If the
sampling rate is now reduced to
3
35 10 samples/sec, let us find the expected
improvement in ( )
q
SNR
0,
, when the bit rate is not to exceed the previous case.

With
s
f
3
45 10 = samples/sec, and assuming 8 bits/samples, we have
the transmitted bit rate as
3 4
45 8 10 36 10 = bits/sec.

With ( )
s
f
3
35 10 = , number of bits/sample that can be used

6.54
R
4
3
36 10
10.28
35 10
= =

As R has to be an integer, it can be taken as 10. As R has increased by two
bits, we can expect 12 dB improvement in SNR.

Exercise 6.6
Consider the three level quantizer (mid-tread type) shown in Fig. 6.32.
The input to the quantizer is a random variable X with the PDF ( )
X
f x give by
( )
X
x
f x
x
1
, 1
4
1
, 1 3
8
<

Fig, 6.32: Quantizer characteristic for the Exercise 6.5

a) Find the value of A such that all the quantizer output levels are
equiprobable. (Note that A has to be less than 1, because with A 1 = ,
we have ( ) P QZ x
1
0
2
= =

)
b) Show that the variance of the quantization noise for the range x A ,
is
4
81
.

6.55

Exercise 6.7
Let the input to -law compander be the sample function ( )
j
x t of a random
process ( ) ( )
m
X t f t cos 2 = + where is uniformly distributed in the
range ( ) 0, 2 . Find the ( )
q
SNR
0,
expected of the scheme.
Note that samples of ( )
j
x t will have the PDF,
( )
X
x
f x
x
otherwise
2
1
, 1
1
0 ,

Answer: ( )
( )
q
L
SNR
1 2
2
0,
3
1
2 ln 1 2

4
= + +

+

Note that if
2
1
2
>> >> , we have ( )

( )
q
L
SNR
2
2 0,
3
ln 1 +

Exercise 6.8
Show that for values of x such that A x x
max
>> , ( )
q
SNR
0,
of the
A-law PCM is given by
( )
q
SNR R
0,
6 = +
where
[ ]
A
10
4.77 20log 1 ln = +

Exercise 6.9
Let ( ) C x denote the compression characteristic. Then
( )
x
dC x
d x
0

is called the companding gain,
c
G . Show that
a)
c
G (A-law) with A 87.56 = is 15.71 (and hence
c
G
10
20log 24 dB)
b)
c
G (-law) with 255 = is 46.02 (and hence
c
G
10
20log 33 dB.)


6.56

6.5 Encoding
The encoding operation converts the quantized samples into a form that is
more convenient for the purpose of transmission. It is a one-to-one
representation of the quantized samples by using code elements or symbols of
the required length per sample.

In the binary code, each symbol may be either of two distinct values or
kinds, which are customarily denoted as 0 and 1. In a ternary code, each symbol
may be one of three distinct values; similarly for the other codes. By far, the most
popular from the point of view of implementation are the binary codes. With R -
binary digits (bits) per sample, we can have
R
2 distinct code words and we
require
R
2 (number of quantization levels), so that it is possible for us to
maintain a one-to-one relationship between the code words and the quantization
levels.

Let us identify the R -bit sequence as
R R
b b b b b
1 3 2 1
. In the natural
binary code, this sequence represents a number (or level) N , where

( ) ( )
R R
R R
N b b b b
1 2 1 0
1 2 1
2 2 2 2

= + + + + (6.22)
Natural binary code results when the codeword symbols or digits are
assigned to N , with N listed in an increasing or decreasing (decimal) order; that
is, though the quantized samples could be either positive or negative, we simply
Exercise 6.10
Show that for the -law,

( )
( )
( )
x
x x
C x
C x
max
0
max
min
'
lim
maximum step size
1
'
minimum step size
lim
= = = +
.

6.57
label the quantized levels decimally without regard to the polarity of the samples.
The first column in Table 6.1 illustrates a decimal assignment for a 16-level
quantization in which half the levels are for positive samples (say 8 to 15) and the
remaining half for negative samples. Column 2 shows the corresponding natural
binary code.

Table 6.1: Various binary code words for a 4-bit Encoder
1 2 3 4 5
Natural
Binary code
Gray code Decimal
Level
No. b
4
b
3
b
2
b
1

Folded
Binary
code
Inverted
folded
binary g
4
g
3
g
2
g
1

Comments
15 1 1 1 1 1 1 1 1 1 0 0 0 1 0 0 0
14 1 1 1 0 1 1 1 0 1 0 0 1 1 0 0 1
13 1 1 0 1 1 1 0 1 1 0 1 0 1 0 1 1
12 1 1 0 0 1 1 0 0 1 0 1 1 1 0 1 0
11 1 0 1 1 1 0 1 1 1 1 0 0 1 1 1 0
10 1 0 1 0 1 0 1 0 1 1 0 1 1 1 1 1
9 1 0 0 1 1 0 0 1 1 1 1 0 1 1 0 1
8 1 0 0 0 1 0 0 0 1 1 1 1 1 1 0 0
Levels
assigned
to positive
message
samples
7 0 1 1 1 0 0 0 0 0 1 1 1 0 1 0 0
6 0 1 1 0 0 0 0 1 0 1 1 0 0 1 0 1
5 0 1 0 1 0 0 1 0 0 1 0 1 0 1 1 1
4 0 1 0 0 0 0 1 1 0 1 0 0 0 1 1 0
3 0 0 1 1 0 1 0 0 0 0 1 1 0 0 1 0
2 0 0 1 0 0 1 0 1 0 0 1 0 0 0 1 1
1 0 0 0 1 0 1 1 0 0 0 0 1 0 0 0 1
0 0 0 0 0 0 1 1 1 0 0 0 0 0 0 0 0
Levels
assigned
to
negative
message
samples


6.58
The other codes shown in the table are derived from the natural binary
code. The folded binary code (also called the sign-magnitude representation)
assigns the first (left most) digit to the sign and the remaining digits are used to
code the magnitude as shown in the third column of the table. This code is
superior to the natural code in masking transmission errors when encoding
speech. If only the amplitude digits of a folded binary code are complemented
(1's changed to 0's and 0's to 1's), an inverted folded binary code results; this
code has the advantage of higher density of 1's for small amplitude signals,
which are most probable for voice messages. (The higher density of 1's relieves
some system timing errors but does lead to some increase in cross talk in
multiplexed systems).

With natural binary encoding, a number of codeword digits can change
even when a change of only one quantization level occurs. For example, with
reference to Table 6.1, a change from level 7 to 8 entails every bit changing in
the 4-bit code illustrated. In some applications, this behavior is undesirable and a
code is desired for which only one digit changes when any transition occurs
between adjacent levels. The Gray Code has this property, if we consider the
extreme levels as adjacent. The digits of the Gray code, denoted by
k
g , can be
derived from those of the natural binary code by

R
k
k k
b k R
g
b b k R
1
,
,
+
=
=

<

where denotes modulo-2 addition of binary digits.
(0 0 0; 0 1 1 0 1 and 1 1 0 = = = = ; note that, if we exclude the sign
bit, the remaining three bits are mirror images with respect to red line in the
table.)

The reverse behavior of the Gray code does not hold. That is, a change in
anyone of code digits does not necessarily result in a change of only one code
level. For example, a change in digit g
4
from 0 when the code is 0001 (level 1) to

6.59
1 will result in 1001 (the code word for level 14), a change spanning almost the
full quantizer range.

6.6 Electrical Waveform Representation of Binary
Sequences
For the purposes of transmission, the symbols 0 and 1 should be
converted to electrical waveforms. A number of waveform representations have
been developed and are being currently used, each representation having its
own specific applications. We shall now briefly describe a few of these
representations that are considered to be the most basic and are being widely
used. This representation is also called as line coding or transmission coding.
The resulting waveforms are called line codes or transmission codes for the
reason that they are used for transmission on a telephone line.

In the Unipolar format (also known as On-Off signaling), symbol 1
represented by transmitting a pulse, where as symbol 0 is represented by
switching off the pulse. When the pulse occupies the full duration of a symbol,
the unipolar format is said to be of the nonreturn-to-zero (NRZ) type. When it
occupies only a fraction (usually one-half) of the symbol duration, it is said to be
of the return-to-zero (RZ) type. The unipolar format contains a DC component
that is often undesirable.

In the polar format, a positive pulse is transmitted for symbol 1 and a
negative pulse for symbol 0. It can be of the NRZ or RZ type. Unlike the unipolar
waveform, a polar waveform has no dc component, provided that 0's and 1's in
the input data occur in equal proportion.

In the bipolar format (also known as pseudo ternary signaling or Alternate
Mark Inversion, AMI), positive and negative pulses are used alternatively for the

6.60
transmission of 1's (with the alternation taking place at every occurrence of a 1)
and no pulses for the transmission of 0s. Again it can be of the NRZ or RZ type.
Note that in this representation there are three levels: +1, 0, -1. An attractive
feature of the bipolar format is the absence of a DC component, even if the input
binary data contains large strings of 1's or 0's. The absence of DC permits
transformer coupling during the course of long distance transmission. Also, the
bipolar format eliminates ambiguity that may arise because of polarity inversion
during the course of transmission. Because of these features, bipolar format is
used in the commercial PCM telephony. (Note that, some authors use the word
bipolar to mean polar)

In the Manchester format (also known as biphase or split phase signaling),
symbol 1 is represented by transmitting a positive pulse for one-half of the
symbol duration, followed by a negative pulse for the remaining half of the
symbol duration; for symbol 0, these two pulses are transmitted in the reverse
order. Clearly, this format has no DC component; moreover, it has a built in
synchronization capability because there is a predictable transition during each
bit interval. The disadvantage of the Manchester format is that it requires twice
the bandwidth when compared to the NRZ unipolar, polar and bipolar formats.
Fig. 6.33 illustrates some of the waveform formats described above. Duration of
each bit has been taken as
b
T sec and the levels as 0 or a .


6.61

Fig. 6.33: Binary data waveform formats
NRZ: (a) on-off (b) polar (c) bipolar
(d) Manchester format

6.7 Bandwidth requirements of PCM
In Appendix A6.1, it has been shown that if
( ) ( )
k d
k
X t A p t kT T
=
=
,
then, the power spectral density of the process is,

6.62
( ) ( ) ( )
j nf T
X A
n
S f P f R n e
T
2
2
1

=
=

By using this result, let us estimate the bandwidth required to transmit the
waveforms using any one of the signaling formats discussed in Sec. 6.6

6.7.1 Unipolar Format
In this case,
k
A 's represent an on-off sequence. Let us assume that 0's
and 1's of the binary sequence to be transmitted are equally likely and '0' is
represented by level 0 and '1' is represented by level 'a'. Then,
( ) ( )
k k
P A P A a
1
0
2
= = = =
Let us compute ( )
A k k n
R n E A A
+

=

. For n 0 = , we have ( )
A k
R E A
2
0

=

.
That is,
( ) ( ) ( ) ( ) ( )
A k k
R P A a P A a
2 2
0 0 0 = = + =

a
2
2
=
Next consider the product
k k n
A A n , 0
+
. This product has four possible
values, namely, 0, 0, 0 and a
2
. Assuming that successive symbols in the binary
sequence are statistically independent, these four values occur with equal
probability resulting in,

k k n
E A A a
2
3 1
0
4 4
+

= +

a
n
2
, 0
4
=
Hence,
( )
A
a
n
R n
a
n
2
2
, 0
2
, 0
4


6.63
Let ( ) p t be rectangular pulse of unit amplitude and duration
b
T sec. Then, ( ) P f
is ( ) ( )
b b
P f T c f T sin =
and ( )
X
S f can be written as
( ) ( ) ( ) [ ]
b b
X b b b
n
a T a T
S f c f T c f T j nf T
2 2
2 2
sin sin exp 2
4 4

=

= +

But from Poisson's formula,

[ ]
b
b b n m
m
j nf T f
T T
1
exp 2

= =

=

Noting that ( )
b
c f T sin has nulls at ( )
X
b
n
f n S f
T
, 1, 2, , = = can be
simplified to
( ) ( ) ( )
b
X b
a T a
S f c f T f
2 2
2
sin
4 4
= + (6.23a)
If the duration of ( ) p t is less than
b
T , then we have unipolar RZ sequence. If
( ) p t is of duration
b
T
2
seconds, then ( )
X
S f reduces to
( )
b b
X
b b m
a T f T m
S f c f
T T
2
2
1
sin 1
16 2

=

= +

(6.23b)
From equation 6.23(b) it follows that unipolar RZ signaling has discrete spectral
components at
b b
f
T T
1 3
0, , = etc. A plot of Eq. 6.23(b) is shown in Fig. 6.34(a).

6.7.2 Polar Format
Assuming,
[ ] [ ]
k k
P A a P A a
1
2
= + = = = and 0's and 1's of the binary
data sequence are statistically independent; it is easy to show,
( )
A
a n
R n
n
2
, 0
0 , 0
=
=


6.64
With ( ) p t being a rectangular pulse of duration
b
T sec. (NRZ case), we have
( ) ( )
X b b
S f a T c f T
2 2
sin = (6.24a)
The PSD for the case when ( ) p t is duration of
b
T
2
(RZ case), is given by
( )
b b
X
T f T
S f a c
2 2
sin
4 2

=

(6.24b)

6.7.3 Bipolar Format
Bipolar format has three levels: a, 0 and ( ) a . Assuming that 1's and 0's
are equally likely, we have
[ ] [ ]
k k
P A a P A a
1
4
= = = = and
[ ]
k
P A
1
0
2
= = . We shall now compute the autocorrelation function of a bipolar
sequence.
( ) ( ) ( ) ( ) ( )
A
R a a a a
1 1 1
0 0 0
4 4 2
= + +

a
2
2
=
To compute ( )
A
R 1 , we have to consider the four two bit sequences, namely, 00,
01, 10, 11. As the binary 0 is represented by zero volts, we have only one non-
zero product, corresponding to the binary sequence (11). As each one of these
two bit sequences occur with a probability
1
4
, we have
( )
A
a
R
2
1
4
=
To compute ( )
A
R n , for n 2 , again we have to take into account only those
binary n-tuples which have 1 in the first and last position, which will result in the
product a
2
. It is not difficult to see that, in these product terms, there are as
many terms with a
2
as there are with a
2
which implies that the sum of the
product terms would be zero. (For example, if we take the binary 3-tuples, only

6.65
101 and 111 will result in a non-zero product quantity. 101 will yield a
2

whereas 111 will result in a
2
). Therefore ( )
A
R n 0 = for n 2 . That is, for the
bipolar format,
( )
A
a
n
a
R n n
otherwise
2
2
, 0
2
, 1
4
0 ,
= =

For the NRZ case, with a rectangular ( ) p t , we obtain
( ) ( ) ( )
X b b b
S f a T c f T f T
2 2 2
sin sin = (6.25a)
Similarly for the return-to-zero case, we have
( ) ( )
b b
X b
a T f T
S f c f T
2
2 2
sin sin
4 2

=

(6.25b)

6.7.4 Manchester Format
If the input binary data consists of independent equally likely symbols,
( )
A
R n for the Manchester format is the same as that of the polar format. The
pulse ( ) p t for the Manchester format is a doublet of unit amplitude and duration
b
T . Hence,
( )
b b
b
f T f T
P f j T c sin sin
2 2

=

and ( )
b b
X b
f T f T
S f a T c
2 2 2
sin sin
2 2

=

(6.26)
Plots of Eq. 6.24(b), 6.25(b) and 6.26 are shown in Fig. 6.34(b).

From Fig. 6.34(a), we see that most of the power of the unipolar return- to-
zero signal is in a bandwidth of
b
T
2
(of course, for the NRZ case, it would be
b
T
1
).

6.66
Similarly, the bandwidth requirements of the Manchester format and return-to-
zero polar format can be taken as
b
T
2
; the spectral width of the return-to-zero
bipolar format is essentially
b
T
1
.

Fig. 6.34: Power spectral density of some signaling schemes
(a) RZ unipolar (b) polar (RZ), bipolar (RZ) and split phase (Manchester) formats


6.67
Bipolar signaling has several advantages:
1) The spectrum has a D.C. null; hence it is possible to transformer-couple
such a signal
2) Its transmission bandwidth is not excessive
3) It has single error detection capability because an isolated error, whether it
causes the deletion or erection of a pulse will violate the pulse alternation
property of the bipolar format
4) It is insensitive to polarity inversion during the course of transmission
5) If a return-to-zero bipolar pulse is rectified, we get an on-off signal that has
a discrete component at the clock frequency. This feature can be used to
regenerate the clock at the receiving end. (See Example 6.15.)
For these reasons, bipolar signaling is used in the PCM voice telephony.

Now let us look at the bandwidth requirement of a PCM system. Consider
the case of a speech signal. In the commercial PCM telephony, this signal is
sampled at 8 kHz and each sample is converted to an 8 bit sequence which
implies a bit rate of 64 kbps. Assuming bipolar signaling, we require a bandwidth
of 64 kHz which is an order of magnitude larger than analog baseband
transmission whose bandwidth requirement is just about 4 kHz!

Example 6.14
A PCM voice communication system uses bipolar return-to-zero pulses for
transmission. The signal to be transmitted has a bandwidth of 3.5 kHz with the
peak-to-peak and RMS values of 4 V and 0.2 V respectively. If the channel
bandwidth is limited to 50 kHz, let us find the maximum ( ) ( )
q
SNR
0,
in dB that
can be expected out of this system.

Assuming that the bandwidth requirements of return-to-zero bipolar pulse
as
b
T
1
, it would be possible for us to send upto 50,000 pulses per second on this
channel. As the bit rate is the product of sampling rate and the number of

6.68
bits/sample, we choose the minimum sampling rate that is permitted so that the
number of bits/sample can be as high as possible. (This would result in the
maximization of ( )
q
SNR
0,
). Minimum sampling rate =7000 samples/sec. Hence,
we could choose a 7-bit quantizer resulting in
7
2 quantization levels.)
Step size
5
7
4
2
2

= =
Noise variance,
Q
2 10
2
2
12 12
= =

( )
x
q
Q
SNR
2 10
8
2 0,
48 2
48 2
100

= = =

( ) ( )
q
SNR
0,
in dB 26.8 =

Example 6.15
In a digital communication system - such as a PCM system - bit clock
recovery at the receiver is very important. Bipolar pulses enable us to recover the
clock without too much difficulty.


Fig. 6.35: Clock recovery scheme for bipolar signals


6.69
The bit rate of the bipolar waveform is
5
10 bps. Let us find the output of
the system when
(a) f
0
100 = kHz (b) f
0
200 = kHz

a) ( ) v t
1
, output of the full wave rectifier, is a RZ, on-off waveform. From Eq.
6.23(b), we know that this waveform process has a discrete spectral
component at
b
f
T
1
= which is 100 kHz in this case. This component will be
selected by the tuned filter; hence ( ) v t
2
is a sinusoidal signal with
frequency 100 kHz. The output of the hard limiter, ( ) v t
3
, is a square wave
at 100 kHz. Differentiation of this result in a sequence of positive and
negative going impulses with two positive (or negative) impulses being
separated by
5
10
sec. This can be used as the clock for the receiver.

b) When the centre frequency of the tuned filter is 200 kHz, the output ( ) v t
2
is
zero as there is no discrete component at 200 kHz in ( ) v t
1
.


6.70

6.8 Differential Pulse Code Modulation (DPCM)
DPCM can be treated as a variation of PCM; it also involves the three
basic steps of PCM, namely, sampling, quantization and coding. But, in the case
of DPCM, what is quantized is the difference between the actual sample and its
predicted value, as explained below.

Exercise 6.11
Let a random process ( ) X t be given by
( ) ( )
k b d
k
X t B p t kT T
=
=

It is given that
k k k
B A A
1
= + where
k
A s are discrete random variables
with ( )
k A
k n
A A R n
= .
k
A a = if the input binary symbol is 1 and
k
A a = if the input symbol is 0. 1s and 0s are equally likely and are
statistically independent.
b
T is the bit duration and
d
T is a random delay with
a uniform distribution in the range
b b
T T
,
2 2

.
k
A s are independent of
d
T .
Show that
a) ( )
k B
k n
a n
B B R n a n
otherwise
2
2
2 , 0
, 1
0 ,
= = =

b) ( ) ( )
b
X b b
f T
S f a T c f T
2 2 2
sin cos
2

=

when ( ) p t is a unit amplitude
rectangular pulse of duration
b
T
2

seconds.

6.71
Let ( ) x t represent the analog signal that is to be DPCM coded, and let it
be sampled with a period
s
T . The sampling frequency
s
s
f
T
1
= is such that there
is no aliasing in the sampling process. Let ( ) ( )
s
t nT
x n x t
=
= . As
s
T is a fixed
quantity, which is known apriori, it is the integer n that carries the time
information of the signal ( ) x t . Quite a few real world signals such as speech
signals, biomedical signals (ECG, EEG, etc.), telemetry signals (temperature
inside a space craft, atmospheric pressure, etc.) do exhibit sample-to-sample
correlation. This implies that ( ) x n and ( ) x n 1 + (or ( ) x n and ( ) x n 1 ) do not
differ significantly. In fact, given a set of previous M samples, say ( ) x n 1 ,
( ) ( ) x n x n M 2 , , , it may be possible for us to predict (or estimate) ( ) x n to
within a small percentage error. Let
( ) x n denote the predicted value of ( ) x n and

let ( ) ( )

( ) e n x n x n = (6.27)
In DPCM, it is the error sequence ( ) e n that is quantized, coded and sent on the
channel. At the receiver, samples are reconstructed using the quantized version
of the error sequence. If there are no transmission errors, the reconstructed
sample sequence ( ) y n will follow the sequence ( ) x n to within the limits of
quantization error.

From the point of view of calculating the ( )
q
SNR
0,
, we can treat ( ) e n to
be a specific value of the random variable ( ) n E . Similarly, we will assume that
( ) x n and
( ) x n represent specific values of the random variable ( ) X n and
( ) X n . Hence
( ) ( )

( ) n X n X n = E (6.28)


6.72
The rationale behind quantizing the sequence ( ) n E is the following: if the
assumption that ( ) X n sequence is correlated is true in practice, then the
variance of ( ) n E would be less that ( ) X n . As such, for a given number of
quantization levels, the quantization step size
to discretize ( ) n E would be
smaller than
x
, the step size required to quantize ( ) X n . This implies that
x
2 2
12 12

< and there would be net improvement in the overall signal-to-
quantization noise ratio. In other words, for a given signal-to-quantization ratio,
DPCM requires fewer bits/input sample than what is required by PCM.

We shall now look at a few prediction schemes and show that the variance
of the error sequence,
2
, is less than
x
2
, the variance of the sequence ( ) X n .

Case i)
( ) X n is a scaled version of ( ) X n 1 .
Let
( ) ( ) X n X n
1
1 = , (6.29)
where
1
is a constant. Then,
( ) ( )

( ) n X n X n = E
( ) ( ) X n X n
1
1 =
( ) ( ) ( )
{ }
E n E X n X n
2
2 2
1
1

= =

E
( ) ( ) ( ) ( ) E X n X n X n X n
2 2 2 2
1 1
1 2 1

= +

But, ( ) ( )
t nT
X n X t
=
= and if ( ) X t is zero mean, WSS process, then
( ) ( )
x
E X n E X n
2 2 2
1

= =

. Therefore,

( ) ( )
x
x
E X n X n
2 2 2
1 1
2
1
1 2

= +


6.73
Let
1
denote the adjacent sample correlation coefficient; that is,

( ) ( )
x
E X n X n
1
2
1

=
.
Hence,
x
2 2 2
1 1 1
1 2

= +

(6.30)
Let us find the value of
1
that minimizes
2
. Differentiating Eq. 6.30 with

respect to
1
and setting the result to zero, we will end up with the result,

1 1
= (6.31a)
Using this value in Eq. 6.30, gives us the result

( ) x
2 2 2
1
1
= (6.31b)
If
1
is 0.8, then
x
2 2
0.36
= , which is about a third of the input signal

variance.

Case ii) Let us generalize the predictor of Eq. 6.29. Let

( ) ( ) ( ) ( )
M
X n X n X n X n M
1 2
1 2 = + + +
( )
M
i
i
X n i
1 =
=
(6.32)
where
1
to
M
are constants.
Eq. 6.32 expresses
( ) X n in terms weighted linear combination of past M input

samples and represents the I-O relation of a linear predictor of order M .
1
to
M
are called Linear Predictor Coefficients (LPCs).

We would like to choose
M 1 2
, , , such that

6.74
( ) ( ) ( )
M
i
i
E n E X n X n i
2
2
1 =

=

E is minimized. To obtain the

optimum ( )
M 1 2
, , , = , let us introduce some notation. Let
n i n j
E X X

be denoted by ( )
x
r j i and let

( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
x x x
x x x
X
x x x
r r r M
r r r M
R
r M r M r
0 1 1
1 0 2
1 2 0

( )
( )
( )
x
x
X
x
r
r
r M
1
2

t
Then, it has been shown that (see Haykin [2], J ayant and Noll [3]), the optimum
predictor vector
( ) ( )
T T
M
opt
1 2
, , , = , where the superscript T denotes the
transpose, is given by
( )
T
X X
opt
R
1
= t (6.33)
and
( )
( ) ( )
M
X i X
i
r r i
2
min
1
0
=
=
(6.34a)

M
X i i
i
2
1
1
=

=

(6.34b)
where
( )
( )
X
j
X
r j
r 0
= (6.34c)


6.75
Let us look at some special cases of Eq. 6.33 and 6.34.
i) First order predictor ( ) M 1 =
Then, is the scalar
1
and

( )
( )
x
x
r
r
1 1
1
0
= =
and
( )
( ) ( )
x x
r r
2
1
min
0 1
=
( )
( )
( )
x
x
x
r
r
r
1
1
0 1
0

=

( ) x
2 2
1
1 =
which are the same as given by Eq. 6.31.
ii) Second order predictor ( ) M 2 =

T
opt
1
2

=

is obtained from
( ) ( )
( ) ( )
( )
( )
x x x
x x x
r r r
r r r
1
2
0 1 1
1 0 2

=

Solving for
1
and
2
, we obtain

1 1 2
1
2
1
1

=

(6.35a)

2
2 1
2
2
1
1

=

(6.35b)
where
( )
( )
x
x
r
r
2
2
0
=
( )
2
min
is obtained by using 6.35(a) and 6.35(b) in Eq. 6.34.

The block diagram of a DPCM transmitter is shown in Fig. 6.36(a) and the
corresponding receiver is shown in Fig. 6.36(b).

6.76

Fig. 6.36: (a) DPCM transmitter (b) DPCM receiver

In the scheme of Fig. 6.36(a), ( ) y n , the input to the predictor, is a quantized
version of ( ) x n . That is,

( ) ( )
M
i q
i
x n x n i
1 =
=
(6.36)
This is essential to enable the receiver to track the predictor in the transmitter, as
will be made clear later on in this section. Let us denote the quantizer operation
by
( ) ( )
q
e n QZ e n =

( ) ( ) e n q n = + (6.37)
where ( ) q n is the quantization error. Then,
( )

( ) ( )
q
y n x n e n = +

( ) ( ) ( ) x n e n q n = + +
( ) ( ) ( )
q
x n q n x n = + = (6.38)

6.77
That is, the predictor input differs from the input sequence ( ) x n by the
quantization error, which can be assumed to be fairly small (much smaller than
( ) x n most of the time). In such a situation, predictor output
( ) x n can be taken
as being predicted from the input sequence ( ) x n itself. If the predictor is good,
then
( ) ( ) x n x n . As such,
2
, the variance of the error quantity ( ) n E should

be much less than
x
2
. Therefore, the quantizer with a given number of levels will
produce much less quantization error, when compared to quantization of ( ) x n
directly.

Let us assume that there are no transmission errors on the channel. Then
the decoder output in the receiver ( )
q
e n is the same as one in the transmitter.
Hence the sequence ( ) y n at the receiver will follow the sequence ( ) x n within
the limits of the quantizer error.

Let
Q
2
denote the variance of the quantization error. Then,
( )
x
q
DPCM
Q
SNR
2
2 0,

=

x
Q
2 2
2 2

=

, where ( ) n
2 2
= E
( )
p
p
G SNR = (6.39a)

x
p
G
2
2
, is called the prediction gain and ( )

p
Q
SNR
2
2

From Eq. 6.39(a), we find that the signal-to-quantization noise ratio of the
DPCM scheme depends on two things: the performance of the predictor
(indicated by
p
G ) and how well the quantizer performs in discretizing the error

6.78
sequence ( ) e n , indicated by ( )
p
SNR . A value of
p
G greater than unity
represents the gain in SNR that is due to the prediction operation of the DPCM
scheme.

x X
p
M
X i i
i
G
2 2
2
2
1
1
=

= =

M
i i
i 1
1
1
=
=

(6.39b)
where
i
s are the optimum predictor coefficients. J ayant and Noll [3], after a
great deal of empirical studies, have reported that for voice signals
p
G can be of
the order of 5-10 dB. In the case of TV video signals, which have much higher
correlation than voice, it is possible to achieve a prediction gain of about 12 dB.
With a 12 dB prediction gain, it is possible to reduce the number of bits/ sample
by 2.

Example 6.16
The designer of a DPCM system finds by experimental means that a third
order predictor gives the required prediction gain. Let
pi
denote the
th
i
coefficient of an optimum
th
p order predictor. Given that
11
2
3
= ,
22
1
5
=
and
33
1
4
= , find the prediction gain given by the third order predictor. you can
assume ( )
x
r 0 1 = .

We have
( )
( )
x
x
r
r
11 1
1
2
0 3
= = =

2
2 1
22
2
1
1

=

(See Eq. 6.35(b))

6.79

2
2
2
2
1 3
5
2
1
3

=

Solving for
2
, we have ( )
X
r
2
5
2
9
= = . We can obtain
31
and
32
from the
following equation.

( ) ( ) ( )
( ) ( ) ( )
( )
( )
x x x
X
X
x x x
r r r
r
r
r r r
31
32
33
0 1 2
1
2
1 0 1

=

That is,

31
32
33
2 5 2
1
3 9 3
2 2 5
1
3 3 9

=

Solving for
31
and
32
, we obtain
31
7
12
= and
32
1
3
=
From Eq. 6.39(b),

p
i i
i
G
3
3
1
1
1
=
=

To calculate
p
G , we require ( )
X
r
3
3 = . This can be obtained from the equation

2 31 1 32 33 3
+ + = . That is,

3
5 7 2 1 1
9 12 3 3 4

= + +

8
27
=
Hence,
p
G 2 = .

DPCM scheme can be made adaptive (ADPCM); that is, the quantizer and
the predictor are made to adapt to the short term statistics of their respective

6.80
inputs. Evidently, we can expect better performance from a ADPCM than a
scheme that is fixed. 32 kbps (8 kHz sampling, 4 bits/sample coding) ADPCM
coders for speech have been developed whose performance is quite comparable
with that of 64 kbps companded PCM. Also 64 kbps DPCM has been developed
for encoding audio signals with a bandwidth of 7 kHz. This coder uses a
sampling rate of 16 kHz and a quantizer with 16 levels. Details can be found in
Benevuto et al [4] and Decina and Modena [5].

At the start of the section, we had mentioned that DPCM is a special case
of PCM. If we want, we could treat PCM to be a special case of DPCM where the
predictor output is always zero!

6.9 Delta Modulation
Delta modulation, like DPCM is a predictive waveform coding technique
and can be considered as a special case of DPCM. It uses the simplest possible
quantizer, namely a two level (one bit) quantizer. The price paid for achieving the
simplicity of the quantizer is the increased sampling rate (much higher than the
Nyquist rate) and the possibility of slope-overload distortion in the waveform
reconstruction, as explained in greater detail later on in this section.

In DM, the analog signal is highly over-sampled in order to increase the
adjacent sample correlation. The implication of this is that there is very little
change in two adjacent samples, thereby enabling us to use a simple one bit
quantizer, which like in DPCM, acts on the difference (prediction error) signals.

In its original form, the DM coder approximates an input time function by a
series of linear segments of constant slope. Such a coder is therefore referred to
as a Linear (or non-adaptive) Delta Modulator (LDM). Subsequent developments
have resulted in delta modulators where the slope of the approximating function

6.81
is a variable. Such coders are generally classified under Adaptive Delta
Modulation (ADM) schemes. We use DM to indicate either of the linear or
adaptive variety.

Fig. 6.37: Waveforms illustrative of LDM operation

6.9.1 Linear Delta Modulation (LDM)
The principle of operation of an LDM system can be explained with the
help of Fig. 6.37. The signal ( ) x t , band-limited to W Hz is sampled at the rate
s
f W 2 >> . ( ) x n denotes the sample of ( ) x t at
s
t nT = . The staircase
approximation to ( ) x t , denoted by
( ) x t , is arrived as follows. One notes, at

s
t nT = , the polarity of the difference between ( ) x n and the latest staircase
approximation to it; that is,
( ) x t at
s
t nT = which is indicated by ( ) y n 1 in the
figure. ( ) y n 1 is incremented by a step of size if ( ) ( ) x n y n 1

is
positive or decremented by , if this is of negative polarity. Mathematically, let
( ) ( )

( ) e n x n x n = (6.40a)
( ) ( ) x n y n 1 = (6.40b)

6.82
and ( ) ( ) b n e n sgn =

(6.40c)
Then, ( ) ( ) ( ) y n y n b n 1 = + (6.41)
For each signal sample, the transmitted channel symbol is the single bit ( ) b n
and the bit rate of the coder is
s
f .

Figure 6.38 shows the (discrete-time) implementation of an LDM system
as implied by Eq. 6.41.

Fig. 6.38: Discrete-time LDM system (a) Transmitter (b) Receiver

In the transmitter we have the delay element acting as the first order predictor so
that
( ) ( ) x n y n 1 = . The boxed portion of the transmitter is generally referred

to as the accumulator. Assuming that the initial contents of the accumulator are
zero, it can easily be shown that ( ) ( )
n
i
y n b i
1 =
=

, justifying the use of the term
accumulator for the boxed portion. At each sampling instant, the accumulator

6.83
increments the approximation to the input signal by , depending on the output
of the quantizer. (Note that the quantizer step size is 2 = ). In the receiver,
the sequence ( ) y n is reconstructed by passing the incoming sequence of
positive and negative pulses through an accumulator in a manner similar to that
used in the transmitter. The sequence ( ) y n is passed through a discrete time
LPF resulting in
( ) y n . This lowpass filtering operation is quite essential in the

case of DM because it improves the ( )
q
SNR
0,
of a DM scheme as explained
later.

Figure 6.39 shows a single integration DM coder-decoder system in an
analog implementation notation that is very common in the DM literature. Note
that the coder input is ( ) x t , a function of the continuous parameter t , and that
the sampling clock appears after the two level quantizer. The outputs
corresponding to
( ) x n and ( ) y n of Fig. 6.38 have not been indicated in Fig.

6.39(a). These quantities
( ) x n and ( ) y n , are realized at the output of the

integrator just before and just after each clock operation. The time-continuous
notation in Fig. 6.39 is very relevant for implementation, though for analytical
reasons, the use of discrete time notation may be preferred.

Figure 6.39 also depicts the waveforms involved at various stages of the
LDM coder. These illustrations are self explanatory.


6.84

Fig. 6.39: LDM system (analog) with representative waveforms

6.85
Figures 6.37 and Fig. 6.39 indicate two types of quantization errors in DM:
slope overload distortion (noise) and granular noise. Slope-overload is said to
occur when the step size is too small to follow a steep segment of the input
waveform ( ) x t . Granularity, on the other hand, refers to a situation where the
staircase function
( ) x t hunts around a relatively flat segment of the input

function, with a step size that is too large relative to the local slope characteristic
of the input. It is therefore clear that for a given statistics of the input signal slope
relatively small values of accentuate slope-overload while relatively large
values of increase granularity.

Slope overload distortion is a basic drawback of the LDM system. This
distortion can be avoided provided ( )
s
f x t
max
'
where ( ) x t
'
refers to the
slope of the input waveform. This is because
( ) x t (or ( ) y t ) changes by
every
s
s
T
f
1
= seconds; hence the maximum LDM slope is
s
f and if this
quantity is greater than or equal to ( ) x t
max
'
for all t , then LDM will be able to
track the slope changes of ( ) x t . Consider the case of tone signal,
( ) ( )
m m
x t A f t cos 2 = . Then
( ) ( )
m m m
x t f A f t
'
2 sin 2 = ,
and ( )
m m m m
x t f A A
max
'
2 = =
Hence to avoid slope overload, we require
( )
s m m
f x t f A
max
'
2 = (6.42a)
or
s m
m
f A
f
1
2

(6.42b)

6.86
The RHS of Eq. 6.42(b) may be viewed as a kind of spectral characteristic for a
basic delta modulator, since it is a function of
m
f . Taken in this light, we expect
that (single integration) LDM is quite suitable to waveforms that have PSDs that
decrease as the reciprocal of the square of the frequency. Examples of such
signals are speech and monochrome TV video.

Example 6.17
Find the minimum sampling frequency ( )
s
f
min
, to avoid slope overload
when ( ) ( ) x t t cos 2 800 = and 0.1 = .

To avoid slope-overload, we require (Eq. 10.42(a)),

m
s m
A
f f 2

Hence, for
m
f 800 = Hz and
m
A
10 =
,
( )
s
f
min
2 800 10 =
50 kHz
Note that this sampling rate is 10 times the Nyquist rate of 1.6 kHz.

Example 6.18
Let a message signal ( ) m t be the input to a delta modulator where
( )
( ) ( )
m t t t
3 3
6 sin 2 10 4 sin 4 10

= +

V, with t in seconds. Determine
the minimum pulse rate that will prevent slope overload, if the step size is
0.314 V.

( )
( ) ( )
m t t t
3 3
6 sin 2 10 4 sin 4 10

= +

( )
( ) ( )
d m t
t t
d t
3 3 3 3
12 10 cos 2 10 16 10 cos 4 10

= +


6.87
( )
( )
d m t
m t
d t
'
= is maximum at t 0 = .
( )
( )
m t
3
max
'
28 10 =
To avoid slope overload, we require

s
f
3
28 10

s
f
4
3
28 10 0.314
280 10
0.314

=
Pulse rate
3
280 10 = /sec.

The performance of the DM system depends on the granular noise, slope-
overload noise (together they are termed as quantization noise in the context of
DM) and regeneration errors (that is, errors due to channel noise). In the analysis
that follows, we shall derive an expression for the output SNR of a DM system,
neglecting the slope-overload-noise and regeneration errors.

Exercise 6.12
The input to a linear delta modulator is a sinusoidal signal whose
frequency can vary from 200 Hz to 4000 Hz. The input is sampled at eight
times the Nyquist rate. The peak amplitude of the sinusoidal signal is 1 Volt.
a) Determine the value of the step size in order to avoid slope overload
when the input signal frequency is 800 Hz.
b) What is the peak amplitude of the input signal to just overload the
modulator, when the input signal frequency is 200 Hz.
c) Is the modulator overloaded when the input signal frequency is 4 kHz.
Ans: (a)
40
(b) 4 V (c) Yes


6.88
For negligible slope-overload, we require ( ) ( )

( ) e t x t x t = . Let
( ) t E denote the random process of which ( ) e t is a sample function. We will
assume that ( ) t E is zero mean WSS process and the samples of ( ) E t are
uniformly distributed in the range ( ) , . Then
2
, the variance of ( ) t E is,

2
2
3

= . Experimental results confirm that the PSD of ( ) t E is essentially flat for
s
s
f f
T
1
= . That is,
( )
s s
S f
f f
2 2
2 6

= = , for
s
f f
Lowpass filtering the output of the accumulator in the receiver rejects the out-of-
band noise components (remember that in DM,
s
f W 2 >> ), giving the output
granular noise component,
g
N , as
( )
W
g
W
N S f d f
=

s
W
f
2
3
= (6.43a)
and ( )
s
X
g
f
SNR
W
2
2 0,
3
=
(6.43b)
where ( )
X
X t
2 2
=
Recall that in order to avoid slope-overload, and
s
f must satisfy Eq. 6.42.


6.89

Now, if the PSD of ( ) X t is ( )
X
S f , then the PSD of
( ) d X t
d t
is ( ) ( )
X
f S f
2
2 .
Hence the mean square signal slope can be put in the form

( )
( ) ( )
X
d X t
f S f d f
d t
2
2
2

=

( )
X rms
B
2
2 =
where
rms
B , the RMS bandwidth of the lowpass process ( ) X t , is defined as
( )
rms X
x
B f S f d f
1
2
2
1

=

(6.44)
We now introduce the so called slope loading factor

s
X rms
f
B 2

=
(6.45)
which is the ratio of maximum DM slope to the RMS signal slope. For any given
signal ( ) x t , the parameters
x
and
rms
B are fixed. Hence, given the sampling
frequency, is a function of only . A reasonably large value of ensures
negligible slope-overload. Expressing in terms of and using this value in Eq.
6.43(b), we obtain
Exercise 6.13
The input to an LDM scheme is the sinusoidal signal ( )
m m
A f t cos 2

.
The scheme uses a sampling frequency of
s
f and the step size is such that it
is the minimum value required to avoid slope overload. Show that the ( )
g
SNR
0,

expected of such a system, at the output of the final lowpass filter with cutoff at
m
f , is
( )
s
g
m
f
SNR
f
3
2 0,
3
8

=


6.90
( )
s
g
rms
f
SNR
B W
3
2 2 2 0,
3
4
=

rms
W F
B
2
3
2 2
6

=

(6.46)
where
s
f
F
W 2
= is the over-sampling ratio.
Equation 6.46 indicates that for a given value of , the SNR in LDM with single
integration increases as the cube of the bit rate (9 dB/octave). The bit rate, in
turn, decides the bandwidth requirements of the system.

Computer simulations indicate that Eq. 6.46 is also valid for ( )
q
SNR
0,

provided F ln 2 8 < . If F ln 2 < , then slope-overload noise dominates (
too small) and SNR drops off quite rapidly. Figure 6.40 illustrates the typical
variation of ( )
q
SNR
0,
with .

Fig. 6.40: LDM performance versus slope loading factor


6.91
For a specified value of F , DM performance is maximized by taking the
empirically determined optimum slope loading factor,

opt
F ln 2
The maximum value of ( ) ( )
g q
SNR SNR
0, 0,
is then given by Eq. 6.46 with
opt
= . Note that as the over sampling ratio is fixed for each curve, is
dependent only on . Hence, when
opt
< , is not large enough to prevent
slope overload. Similarly, as the value of increases away from
opt
, the size of
increases which implies an increase in the granular noise level.

Example 10.20
A signal ( ) x t with
X
2
1 = has the spectral density ( )
X
S f given by
( )
X
f
S f
f
1
, 125 Hz
500
1
, 125 375 Hz
1000
< <

( ) x t is over sampled by a factor of 10 and the samples are applied to DM coder.
Find
opt
, the optimum value of .
We will take
opt
ln20 3 = = . From Eq. 6.45,

rms opt
opt
s
B
f
2
= .
We need to compute
rms
B .

rms
B f d f f d f
125 375
2 2 2
0 125
1 1
2
500 1000

= +

f f
125 275
3 3
0 125
1 1
250 3 500 3

= +


6.92

[ ] [ ]
1 1
1953125 50781250
750 1500
= +
7812.5 33854.16 41666.66 = + =

rms
B 204.12 =

opt
2 3 204.12
7500

=
0.512 =

6.9.2 Adaptive delta modulation
Consider Fig. 6.40. We see that for a given value of F , there is an
optimum value of which gives rise to maximum ( )
q
SNR
0,
from a DM coder.
Let us now change from
opt
by varying
X
with fixed. We then find that
the range of
X
for which near about this maximum ( )
q
SNR
0,
is possible is very
limited. (This situation is analogous to the uncompanded PCM). Adaptive Delta
Modulation, a modification of LDM, is a scheme that circumvents this deficiency
of LDM.

The principle of ADM is illustrated in Fig. 6.41. Here the step size of the
quantizer is not a constant but varies with time. We shall denote by
( ) ( ) n n 2 = , the step size at ( )
s
t nT n = increases during a steep
segment of the input and decreases when the modulator is quantizing a slowly
varying segment of x(t). A specific step-size adaptation scheme is discussed in
detail later on in this section.

The adaptive step size control which forms the basis of an ADM scheme
can be classified in various ways such as: discrete or continuous; instantaneous
or syllabic (fairly gradual change); forward or backward. Here, we shall describe
an adaptation scheme that is backward, instantaneous and discrete. For
extensive coverage on ADM, the reader is referred to J ayant and Noll [3].

6.93

Fig. 6.41: Waveforms illustrative of ADM operation

Figure 6.42 gives the block diagram of an ADM scheme. In practical
implementation, the step size ( ) n is constrained to be between some
predetermined minimum and maximum values. In particular, we write
( ) n
min max
(6.47)
The upper limit,
max
, controls the amount of slopeoverload distortion. The lower
limit,
min
, controls the granular noise. Within these limits, the adaptive rule for
( ) n can be expressed in the general form
( ) ( ) ( ) n g n n 1 = (6.48)
where the time varying gain ( ) g n depends on the present binary output ( ) b n
and M previous values, ( ) ( ) b n b n M 1 , , . The algorithm is usually initiated
with
min
.


6.94

Fig. 6.42: ADM (a) Transmitter (b) Receiver

In a conceptually least complex ( ) M 1 = realization of the logic implied by
Eq. 6.48, successive bits ( ) b n and ( ) b n 1 are compared to detect probable
slope-overload ( ) ( ) b n b n 1 =

, or probable granularity ( ) ( ) b n b n 1

.
Then, ( ) g n is arrived at
( )
( ) ( )
( ) ( )
P if b n b n
g n
if b n b n
P
, 1
1
, 1
=
(6.49)
where P 1 . Note that P 1 = represents LDM. Typically, a value of
opt
P 1.5 =
minimizes the quantization noise for speech signals. In general, for a broad class
of signals, P 1 2 < < , does seem to yield good results. In an illustrative
computer simulation with real speech, with W 3.3 = kHz and
s
f 60 = kHz, ADM
( ) P 1.5 = showed an SNR advantage of more than 10 dB over LDM. The ADM

6.95
scheme with ( ) g n given by Eq. 6.49 is referred to as constant factor ADM with
one bit memory.

Results have been reported in the literature which compare the ( )
q
SNR
0,

performance of -law PCM and the ADM scheme discussed above. One such
result is shown in Fig. 6.43 for the case of bandpass filtered (200-3200 Hz)
speech. For PCM telephony, the sampling frequency used is 8 kHz. As can be
seen from the figure, the SNR comparison between ADM and PCM is dependent
on the bit rate. An interesting consequence of this is, below 50 kbps, ADM which
was originally conceived for its simplicity, out-performs the logarithmic PCM,
which is now well established commercially all over the world. A 60 channel ADM
(continuous adaptation) requiring a bandwidth of 2.048 MHz (the same as used
by the 30 channel PCM system) was in commercial use in France for sometime.
French authorities have also used DM equipment for airborne radio
communication and air traffic control over Atlantic via satellite. However, DM has
not found wide-spread commercial usage simply because PCM was already
there first!

Fig. 6.43: Performance of PCM and ADM versus bit rate

6.96
Instantaneously adapting delta modulators (such as the scheme described
above) are more vulnerable to channel noise than the slowly adapting or linear
coders. Therefore, although instantaneously adapting delta modulators are very
simple and efficient (SNR-wise) in a relatively noise-free environments (say, bit
error probability
4
10
< ), a number of syllabically adaptive delta modulators

have been designed for use over noisy channels and these coders are
particularly attractive for low-bit-rate ( 20 < k bits/sec) applications. For details on
syllabic adaptation, the reader is referred to [3].


6.97
Appendix A6.1
PSD of a waveform process with discrete amplitudes
A sequence of symbols are transmitted using some basic pulse ( ) p t . Let
the resulting waveform process be denoted by ( ) X t , given by
( ) ( )
k d
k
X t A p t kT T
=
=
(A6.1)
k
A s are discrete random variables, and
d
T
1
is a random variable uniformly
distributed in the interval
T T
,
2 2

.
k
d
T . T is the
symbol duration. (For the case of binary transmission, we use
b
T instead of T ;
bit rate
b
T
1
= .) We shall assume that the autocorrelation of the discrete
amplitude sequence is a function of only the time difference. That is,
( )
k k n A k k n
E A A R n E A A
+

= =

(A6.2)

With these assumptions, we shall derive an expression for the
autocorrelation of the process ( ) X t
( ) ( ) ( )
X
R t t E X t + X t , + =

( ) ( )
k j d d
k j
E A A p t kT T p t j T T

= =

= +

Let j k n = + for some n . Then,
( ) ( ) ( )
+

+ =

T
X k k d
n
k n T
R t t A A p p d t
T
2
2
1
,
where ( )
d
t kT t = + and ( )
d
t j T t = .

1
For an alternative derivative for the PSD, without the random delay
d
T in ( ) X t , see [6].

6.98
(Note that
k
d
T ; hence the expectation with respect to
d
T
can be separated from
k j
A A )
( ) ( ) ( ) ( )
T
X A d
n k T
R t t R n p p d t
T
2
2
1
,

+ =

Let
d
t kT t u = ; then,
d
d u d t =
( ) ( ) ( ) ( )
t k T
X A
n k
t k T
R t t R n p u p u nT d u
T
1
2
1
2
1
,

+

+ = +

( ) ( ) ( )
t k T
A
n k
t k T
R n p u p u nT d u
T
1
2
1
2
1

+

= +

( ) ( ) ( )
A
n
R n p u p u nT d u
T
1

= +

Let ( )
p
r be the autocorrelation of the pulse ( ) p t ; that is,
( ) ( ) ( )
p
r p u p u d u

= +
.
Then, ( ) ( ) ( )
X A p
n
R t t R n r nT
T
1
, + = +
(A6.3)
As the RHS of Eq. A6.3 is only a function of , we have,
( ) ( ) ( )
X A p
n
R R n r nT
T
1

=
= +
(A6.4)
As ( ) ( )
p
r P f
2
, Fourier transform of A6.4 results in the equation
( ) ( ) ( )
j nf T
X A
n
S f P f R n e
T
2
2
1

=
=

where ( )
X
S f is the power spectral density of the process.

6.99
References
1) A. V. Oppenheim, R. W. Schafer and J . R. Buck, Discrete-Time signal
processing, (2
nd
ed.), Pearson Education, 2000
2) Simon Haykin, Communication Systems, (4
th
ed.) J ohn Wiley, 2001
3) N. S. J ayant and P. Noll, Digital Coding of Waveforms: Principles and
applications to speech and video, Prentice Hall, 1984
4) Benevuto et al, The 32 kbps coding standard, AT & T technical journal, vol.
65, no.5, pp12-22, Sept-Oct, 1986
5) M. Decina and G. Modena, CCITT standards on digital speech processing,
IEEE journal on selected areas of communication, Vol. 6, Feb. 1988, pp
227-234
6) Leon W. Couch, II, Digital and analog communication systems, Pearson
Education Asia, 2001

7.1
CHAPTER 7

Noise Performance of Various Modulation
Schemes

7.1 Introduction
The process of (electronic) communication becomes quite challenging
because of the unwanted electrical signals in a communications system. These
undesirable signals, usually termed as noise, are random in nature and interfere
with the message signals. The receiver input, in general, consists of (message)
signal plus noise, possibly with comparable power levels. The purpose of the
receiver is to produce the desired signal with a signal-to-noise ratio that is above
a specified value.

In this chapter, we will analyze the noise performance of the modulation
schemes discussed in chapters 4 to 6. The results of our analysis will show that,
under certain conditions, FM is superior to the linear modulation schemes in
combating noise and PCM can provide better signal-to-noise ratio at the receiver
output than FM. The trade-offs involved in achieving the superior performance
from FM and PCM will be discussed.

We shall begin our study with the noise performance of various CW
modulations schemes. In this context, it is the performance of the detector
(demodulator) that would be emphasized. We shall first develop a suitable
receiver model in which the role of the demodulator is the most important one.


7.2
7.2 Receiver Model and Figure of Merit: Linear
Modulation

7.2.1 Receiver model
Consider the superheterodyne receiver shown in Fig. 4.75. To study the
noise performance we shall make use of simplified model shown in Fig. 7.1.
Here, ( )
eq
H f is the equivalent IF filter which actually represents the cascade
filtering characteristic of the RF, mixer and IF sections of Fig. 4.75. ( ) s t is the
desired modulated carrier and ( ) w t represents a sample function of the white
Gaussian noise process with the two sided spectral density of
N
0
2
. We treat
( )
eq
H f to be an ideal narrowband, bandpass filter, with a passband between
c
f W to
c
f W + for the double sideband modulation schemes. For the case of
SSB, we take the filter passband either between
c
f W and
c
f (LSB) or
c
f and
c
f W + (USB). (The transmission bandwidth
T
B is W 2 for the double sideband
modulation schemes whereas it is W for the case of SSB). Also, in the present
context,
c
f represents the carrier frequency measured at the mixer output; that is
c IF
f f = .

Fig. 7.1: Receiver model (linear modulation)


7.3
The input to the detector is ( ) ( ) ( ) x t s t n t = + , where ( ) n t is the sample
function of a bandlimited (narrowband) white noise process ( ) N t with the power
spectral density
( )
N
N
S f
0
2
= over the passband of ( )
eq
H f . (As ( )
eq
H f is
treated as a narrowband filter, ( ) n t represents the sample function of a
narrowband noise process.)

7.2.2 Figure-of-merit
The performance of analog communication systems are measured in
terms of Signal-to-Noise Ratio ( ) SNR . The SNR measure is meaningful and
unambiguous provided the signal and noise are additive at the point of
measurement. We shall define two ( ) SNR quantities, namely, (i) ( ) SNR
0
and
(ii) ( )
r
SNR .

The output signal-to-noise ratio is defined as,
( ) SNR
0
Average power of the message at receiver output
Average noise power at the receiver output
= (7.1)
The reference signal-to-noise ratio is defined as,
( )
r
SNR
Average power of the modulated
message signal at receiver input
Average noise power in the message
bandwidth at receiver input

=

(7.2)
The quantity, ( )
r
SNR can be viewed as the output signal-to-noise ratio which
results from baseband or direct transmission of the message without any
modulation as shown in Fig. 7.2. Here, ( ) m t is the baseband message signal
with the same power as the modulated wave. For the purpose of comparing
different modulation systems, we use the Figure-of-Merit ( ) FOM defined as,

7.4

( )
( )
r
SNR
FOM
SNR
0
= (7.3)

Fig. 7.2: Ideal Baseband Receiver

FOM as defined above provides a normalized ( ) SNR
0
performance of
the various modulation-demodulation schemes; larger the value of FOM, better is
the noise performance of the given communication system.

Before analyzing the SNR performance of various detectors, let us
quantify the outputs expected of the (idealized) detectors when the input is a
narrowband signal. Let ( ) x t be a real narrowband bandpass signal. From Eq.
1.55, ( ) x t can be expressed as
( )
( ) ( ) ( ) ( ) ( )
( ) ( ) ( )
c c s c
c
x t t x t t
x t
A t t t
cos sin 7.4a
cos 7.4b

=

+

( )
c
x t and ( )
s
x t are the in-phase and quadrature components of ( ) x t . The
envelope ( ) A t and the phase ( ) t are given by Eq. 1.56. In this chapter, we will
analyze the performance of a coherent detector, envelope detector, phase
detector and a frequency detector when signals such as ( ) x t are given as input.
The outputs of the (idealized) detectors can be expressed mathematically in
terms of the quantities involved in Eq. 7.4. These are listed below. (Table 7.1)


7.5
Table 7.1: Outputs from various detectors

( ) x t is input to an ideal
Detector output proportional to
i) Coherent detector ( )
c
x t
ii) Envelope detector ( ) A t
iii) Phase detector ( ) t
iv) Frequency detector
( ) d t
d t
1
2

( ) x t could be used to represent any of the four types of linear modulated signals
or any one of the two types of angle modulated signals. In fact, ( ) x t could even
represent (signal +noise) quantity, as will be seen in the sequel.

Table 7.2 gives the quantities ( )
c
x t , ( )
s
x t , ( ) A t and ( ) t for the linear
and angle modulated signals of Chapter 4 and 5.


7.6
Table 7.2: Components of linear and angle modulated signals

Signal ( )
c
x t ( )
s
x t ( ) A t ( ) t
1
DSB-SC
( ) ( )
c c
A m t t cos
( )
c
A m t zero ( )
c
A m t
( ) m t 0, 0 >
( ) m t , 0 <
2 DSB-LC (AM)
( ) [ ] ( )
c m c
A g m t t 1 cos + ,
( ) [ ]
c m
A g m t 1 0 +
( ) [ ]
c m
A g m t 1+ zero ( ) [ ]
c m
A g m t 1+ zero
3 SSB
( ) ( )
( ) ( )
c
c
c
c
A
m t t
A
m t t
cos
2
sin
2

( )
c
A m t
2

( )
c
A m t
2
( )

( )
c
A
m t t m
2
2
2
+
( )
( )
m t
m t
1
tan

4 Phase modulation
( ) [ ]
c c
A t t cos + ,
( ) ( )
p
t k m t =
( )
c
A t cos ( )
c
A t sin
c
A ( )
p
k m t
5 Frequency modulation
( ) [ ]
c c
A t t cos + ,
( ) ( )
t
f
t k m d 2

=

( )
c
A t cos ( )
c
A t sin
c
A ( )
t
f
k m d 2

Example 7.1
Let ( ) ( ) ( )
c m c
s t A t t cos cos = where
m
f
3
10 = Hz and
c
f
6
10 = Hz.
Let us compute and sketch the output ( ) v t of an ideal frequency detector when
( ) s t is its input.

From Table 7.1, we find that an ideal frequency detector output will be
proportional to
( ) d t
d t
1
2

. For the DSB-SC signal,


7.7
( )
( )
( )
m t
t
m t
0, 0
, 0
>
=

<

For the example, ( )
( )
m t t
3
cos 2 10

=

. Hence ( ) t is shown in Fig. 7.3(b).

Fig. 7.3: (Ideal) frequency detector output of example 7.1

Differentiating the waveform in (b), we obtain ( ) v t , which consists of a
sequence of impulses which alternate in polarity, as shown in (c).


7.8
Example 7.2
Let ( ) m t
t
2
1
1
=
+
. Let ( ) s t be an SSB signal with ( ) m t as the message
signal. Assuming that ( ) s t is the input to an ideal ED, let us find the expression
for its output ( ) v t .

From Example 1.24, we have

( ) m t
t
2
1
1
=
+

As the envelope of ( ) s t is ( )

( ) m t m t
1
2
2 2

+

, we have
( ) v t
t
2
1
1
=
+
.

7.3 Coherent Demodulation
7.3.1 DSB-SC
The receiver model for coherent detection of DSB-SC signals is shown in
Fig. 7.4. The DSB-SC signal is, ( ) ( ) ( )
c c
s t A m t t cos = . We assume ( ) m t to
be sample function of a WSS process ( ) M t with the power spectral density,
( )
M
S f , limited to W Hz.

Fig. 7.4: Coherent Detection of DSB-SC.


7.9
The carrier,
( )
c c
A t cos , which is independent of the message ( ) m t is actually
a sample function of the process
( )
c c
A t cos + where is a random
variable, uniformly distributed in the interval 0 to 2 . With the random phase
added to the carrier term, ( )
s
R , the autocorrelation function of the process ( ) S t
(of which ( ) s t is a sample function), is given by,
( ) ( ) ( )
c
s M c
A
R R
2
cos
2
= (7.5a)
where ( )
M
R is the autocorrelation function of the message process. Fourier
transform of ( )
s
R yields ( )
s
S f given by,
( ) ( ) ( )
c
s M c M c
A
S f S f f S f f
2
4
= + +

(7.5b)
Let
M
P denote the message power, where
( ) ( )
W
M M M
W
P S f d f S f d f

= =

Then, ( ) ( )
c
c
f W
c c M
s M c
f W
A A P
S f d f S f f d f
2 2
2
4 2
+

= =

.
That is, the average power of the modulated signal ( ) s t is
c M
A P
2
2
. With the (two
sided) noise power spectral density of
N
0
2
, the average noise power in the
message bandwidth W 2 is
N
W W N
0
0
2
2
= . Hence,
( )
c M
r
DSB SC
A P
SNR
W N
2
0
2

=

(7.6)
To arrive at the FOM , we require ( ) SNR
0
. The input to the detector is
( ) ( ) ( ) x t s t n t = + , where ( ) n t is a narrowband noise quantity. Expressing ( ) n t
in terms of its in-phase and quadrature components, we have

7.10
( ) ( ) ( ) ( ) ( ) ( )
c c c c s c
x t A m t t n t t n t cos cos sin = +
Assuming that the local oscillator output is
( )
c
t cos , the output ( ) v t of the
multiplier in the detector (Fig. 7.4) is given by

( ) ( ) ( ) ( ) ( ) ( )
( ) ( )
c c c c c
c s c
v t A m t n t A m t n t t
A n t t
1 1 1
cos 2
2 2 2
1
sin 2
2
= + + +

As the LPF rejects the spectral components centered around
c
f 2 , we have
( ) ( ) ( )
c c
y t A m t n t
1 1
2 2
= + (7.7)

From Eq. 7.7, we observe that,
i) Signal and noise which are additive at the input to the detector are additive
even at the output of the detector
ii) Coherent detector completely rejects the quadrature component ( )
s
n t .
iii) If the noise spectral density is flat at the detector input over the passband
( )
c c
f W f W , + , then it is flat over the baseband ( ) W W , , at the
detector output. (Note that ( )
c
n t has a flat spectrum in the range W to
W .)
As the message component at the output is
( )
c
A m t
1
2

, the average
message power at the output is
c
M
A
P
2
4

. As the spectral density of the in-phase
noise component is N
0
for f W , the average noise power at the receiver
output is
( )
W N
W N
0
0
1
2
4 2
= . Therefore,
( )
( )
( )
c M
DSB SC
A P
SNR
W N
2
0
0
4
2

=


7.11

c M
A P
W N
2
0
2
= (7.8)
From Eq. 7.6 and 7.8, we obtain
[ ]
( )
( )
DSB SC
r
SNR
FOM
SNR
0
1
= = (7.9)

7.3.2 SSB
Assuming that LSB has been transmitted, we can write ( ) s t as follows:

( ) ( ) ( )

( ) ( )
c c
c c
A A
s t m t t m t t cos sin
2 2
= +
where
( ) m t is the Hilbert transform of ( ) m t . Generalizing,

( ) ( ) ( )

( ) ( )
c c
c c
A A
S t M t t M t t cos sin
2 2
= + .
We can show that the autocorrelation function of ( ) S t , ( )
s
R is given by
( ) ( ) ( )

( ) ( )
c
M
s M c c
A
R R R
2
cos sin
4

= +

where
( ) M
R t is the Hilbert transform of ( )
M
R t . Hence the average signal
power, ( )
c
s M
A
R P
2
0
4
=
and ( )
c M
r
A P
SNR
W N
2
0
4
= (7.10)
Let ( ) ( ) ( ) ( ) ( )
c c s c
n t n t t n t t cos sin =
(Note that with respect to
c
f , ( ) n t does not have a locally symmetric spectrum).
( ) ( ) ( )
c c
y t A m t n t
1 1
4 2
= +
Hence, the output signal power is
c M
A P
2
16
and the output noise power as
W N
0
1
4

. Thus, we obtain,

7.12
( )
c M
SSB
A P
SNR
W N
2
0,
0
4
16
=

c M
A P
W N
2
0
4
= (7.11)
From Eq. 7.10 and 7.11,
( )
SSB
FOM 1 = (7.12)
From Eq. 7.9 and 7.12, we find that under synchronous detection, SNR
performance of DSB-SC and SSB are identical, when both the systems operate
with the same signal-to-noise ratio at the input of their detectors.

In arriving at the RHS of Eq. 7.11, we have used the narrowband noise
description with respect to
c
f . We can arrive at the same result, if the noise
quantity is written with respect to the centre frequency
c
W
f
2

.

7.4 Envelope Detection
DSB-LC or AM signals are normally envelope detected, though coherent
detection can also be used for message recovery. This is mainly because
envelope detection is simpler to implement as compared to coherent detection.
We shall now compute the ( )
AM
FOM .

The transmitted signal ( ) s t is given by
( ) ( ) ( )
c m c
s t A g m t t 1 cos = +

Then the average signal power in ( )
c m M
A g P
s t
2 2
1
2

+

= . Hence
( )
( ) c m M
r DSB LC
A g P
SNR
W N
2 2
,
0
1
2
+
= (7.13)

7.13
Using the in-phase and quadrature component description of the narrowband
noise, the quantity at the envelope detector input, ( ) x t , can be written as
( ) ( ) ( ) ( ) ( ) ( )
c c s c
x t s t n t t n t t cos sin = +
( ) ( ) ( ) ( ) ( )
c c m c c s c
A A g m t n t t n t t cos sin = + +

(7.14)
The various components of Eq. 7.14 are shown as phasors in Fig. 7.5. The
receiver output ( ) y t is the envelope of the input quantity ( ) x t . That is,
( ) ( ) ( ) ( )
{ } c c m c s
y t A A g m t n t n t
1
2
2 2
= + + +

Fig. 7.5: Phasor diagram to analyze the envelope detector

We shall analyze the noise performance of envelope detector for two different
cases, namely, (i) large SNR at the detector input and (ii) weak SNR at the
detector input.

7.4.1 Large predetection SNR
Case (i): If the signal-to-noise ratio at the input to the detector is sufficiently
large, we can approximate ( ) y t as (see Fig. 7.5)
( ) ( ) ( )
c c m c
y t A A g m t n t + + (7.15)
On the RHS of Eq. 7.15, there are three quantities: A DC term due to the
transmitted carrier, a term proportional to the message and the in-phase noise
component. In the final output, the DC is blocked. Hence the average signal

7.14
power at the output is given by
c m M
A g P
2 2
. The output noise power being equal
to W N
0
2 we have,
( )
c m M
AM
A g P
SNR
W N
2 2
0
0
2

(7.16)
It is to be noted that the signal and noise are additive at the detector output and
power spectral density of the output noise is flat over the message bandwidth.
From Eq. 7.13 and 7.16 we obtain,
( )
m M
AM
m m
g P
FOM
g P
2
2
1
=
+
(7.17)
As can be seen from Eq. 7.17, the FOM with envelope detection is less than
unity. That is, the noise performance of DSB-LC with envelope detection is
inferior to that of DSB-SC with coherent detection. Assuming ( ) m t to be a tone
signal, ( )
m m
A t cos and
m m
g A = , simple calculation shows that ( )
AM
FOM is
( )
2
2
2
+
. With the maximum permitted value of 1 = , we find that the
( )
AM
FOM is
1
3
. That is, other factors being equal, DSB-LC has to transmit three
times as much power as DSB-SC, to achieve the same quality of noise
performance. Of course, this is the price one has to pay for trying to achieve
simplicity in demodulation.

7.4.2 Weak predetection SNR
In this case, noise term dominates. Let ( ) ( ) ( )
n c
n t r t t t cos = +

. We
now construct the phasor diagram using ( )
n
r t as the reference phasor (Fig. 7.6).
Envelope detector output can be approximated as
( ) ( ) ( ) ( ) ( )
n c c m
y t r t A t A g m t t cos cos + +

(7.18)
From Eq. 7.18, we find that detector output has no term strictly proportional to
( ) m t . The last term on the RHS of Eq. 7.18 contains the message signal ( ) m t

7.15
multiplied by the noise quantity, ( ) t cos , which is random; that is, the message
signal is hopelessly mutilated beyond any hope of signal recovery. Also, it is to
be noted that signal and noise are no longer additive at the detector output. As
such, ( ) SNR
0
is not meaningful.

Fig. 7.6: Phase diagram to analyze the envelope detector for case (ii)

The mutilation or loss of message at low input SNR is called the
threshold effect. That is, there is some value of input SNR, above which the
envelope detector operates satisfactorily whereas if the input SNR falls below
this value, performance of the detector deteriorates quite rapidly. Actually,
threshold is not a unique point and we may have to use some reasonable
criterion in arriving it. Let R denote the random variable obtained by observing
the process ( ) R t (of which ( ) r t is a sample function) at some fixed point in time.
It is quite reasonable to assume that the detector is operating well into the
threshold region if
[ ]
c
P R A 0.5 ; where as, if the above probability is 0.01 or
less, the detector performance is quite satisfactory. Let us define the quantity,
carrier-to-noise ratio, as

average carrier power
average noise power in the transmission bandwidth
=

c c
A A
W N W N
2 2
0 0
2
2 4
= =
We shall now compute the threshold SNR in terms of defined above. As R is
Rayleigh variable, we have

7.16
( )
N
r
R
N
r
f r e
2
2
2
2

where
N
W N
2
0
2 =
[ ] ( )
c
c R
A
P R A f r d r
=

c
A
W N
e
2
0
4
=
e

=
Solving for from e 0.5

= , we get ln 2 0.69 = = or - 1.6 dB. Similarly,
from the condition
[ ]
c
P R A 0.01 = , we obtain ln100 4.6 = = or 6.6 dB.

Based on the above calculations, we state that if 1.6 dB, the
receiver performance is controlled by the noise and hence its output is not
acceptable whereas for 6.6 dB, the effect of noise is not deleterious.
However, reasonable intelligibility and naturalness in voice reception requires a
post detection SNR of about 25 dB. That is, for satisfactory reception, we require
a value of much greater than what is indicated by the threshold considerations.
In other words, additive noise makes the signal quality unacceptable long before
multiplicative noise mutilates it. Hence threshold effect is usually not a serious
limitation in AM transmission.

We now present two oscilloscope displays of the ED output of an AM
signal with tone modulation. They are in flash animation.
TUED - Display 1UT: SNR at the input to the detector is about 0 dB. ( ( ) m t is a tone
signal at 3 kHz.) Output resembles the sample function of the
noise process. Threshold effect is about to be set in.
TUED - Display 2UT: SNR at the detector input is about 10 dB. Output of the detector,
though resembling fairly closely a tone at 3 kHz, is still not a

7.17
pure tone signal. Some amount of noise is seen riding on the
output sine wave and the peaks of the sinewave are not
perfectly aligned.

Example 7.3
In a receiver meant for the demodulation of SSB signals, ( )
eq
H f has the
characteristic shown in Fig. 7.7. Assuming that USB has been transmitted, let us
find the FOM of the system.

Fig. 7.7: ( )
eq
H f for the Example 7.3

Because of the non-ideal ( )
eq
H f , ( )
c
N
S f will be as shown in Fig. 7.8.

Fig. 7.8: ( )
c
N
S f of Example 7.3

For SSB with coherent demodulation, we have

7.18
Signal quantity at the output
( )
c
A
m t
4
=
Noise quantity at the output
( )
c
n t
2
=
Output noise power
c
W
N
W
S d f
1
4

=

N W
0
5
16
=

( )
c M
A P
SNR
N W
2
0
0
16
5
16

=

c M
A P
N W
2
0
5
=
( )
c M
r
A P
SNR
W N
2
0
4
=
Hence
( )
( )
r
SNR
FOM
SNR
0
4
0.8
5
= = = .


7.19

Example 7.4
In a laboratory experiment involving envelope detection, AM signal at the
input to ED, has the modulation index 0.5 with the carrier amplitude of 2 V. ( ) m t
is a tone signal of frequency 5 kHz and
c
f 5 >> kHz. If the (two-sided) noise
PSD at the detector input is
8
10
Watts/Hz, what is the expected ( ) SNR

0
of this
scheme? By how many dB, this scheme is inferior to DSB-SC?

Spectrum of the AM signal is as shown in Fig. 7.10.

Exercise 7.1
In a receiver using coherent demodulation, the equivalent IF filter has
the characteristics shown in Fig. 7.9. Compute the output noise power in the
range f 100 Hz assuming N
3
0
10
= Watts/Hz.

Fig. 7.9: ( )
eq
H f for the Exercise 7.1

Ans: 0.225 Watts

7.20

Fig. 7.10: Spectrum of the AM signal of Example 7.4

( )
c m M
AM
A g P
SNR
W N
2 2
0,
0
2
=
As
m
M
A
P
2
2
= ,
( )
( )
c m m
AM
A g A
SNR
W N
2
2
0,
0
4
=
But
m m
g A 0.5 = = . Hence,
( )
AM
SNR
3 8 0,
1
4
4
4 5 10 2 10
=

5
1
40 10

4
10
4
=
36 = dB
( )
AM
FOM
2
2
1
1
4
1
9
2
2
4
= = =
+
+

( )
DSB SC
FOM 1
=
DSB-SC results in an increase in the ( ) SNR
0
by a factor of 9; that is by 9.54 dB.


7.21
7.5 Receiver Model: Angle Modulation
The receiver model to be used in the noise analysis of angle modulated
signals is shown in Fig. 7.11. (The block de-emphasis filter is shown with broken
lines; the effect of pre-emphasis, de-emphasis will be accounted for
subsequently).

Fig. 7.11: Receiver model for the evaluation of noise performance

The role of ( )
eq
H f is similar to what has been mentioned in the context of
Fig. 7.1, with suitable changes in the centre frequency and transmission
bandwidth. The centre frequency of the filter is
c IF
f f = , which for the commercial
FM is 10.7 MHz. The bandwidth of the filter is the transmission bandwidth of the
angle modulated signals, which is about 200 kHz for the commercial FM.
Nevertheless, we treat ( )
eq
H f to be a narrowband bandpass filter which passes
the signal component ( ) s t without any distortion whereas ( ) n t , the noise
component at its output is the sample function of a narrowband noise process
with a flat spectrum within the passband. The limiter removes any amplitude
variations in the output of the equivalent IF filter. We assume the discriminator to
be ideal, responding to either phase variations (phase discriminator) or derivative
of the phase variations (frequency discriminator) of the quantity present at its
input. The figure of merit ( ) FOM for judging the noise performance is the same
as defined in section 7.2.2, namely,
( )
( )
r
SNR
SNR
0
.


7.22
7.6 Calculation of FOM
Let,
( ) ( )
c c
s t A t t cos = +

(7.19)
where

( )
( ) ( )
( ) ( )
p
t
f
k m t
t
k m d
, forPM 7.20a
2 , for FM 7.20b

The output of ( )
eq
H f is,
( ) ( ) ( ) x t s t n t = + (7.21a)
( ) ( ) ( )
c c n c
A t t r t t t cos cos = + + +

(7.21b)
where, on the RHS of Eq. 7.21(b) we have used the envelope ( ) ( )
n
r t and phase
( ) ( )
t representation of the narrowband noise. As in the case of envelope
detection of AM, we shall consider two cases:
i) Strong predetection SNR,
( ) ( )
>>
c n
A r t most of the time and
ii) Weak predetection SNR, ( ) ( )
<<
c n
A r t most of the time .

7.6.1 Strong Predetection SNR
Consider the phasor diagram shown in Fig. 7.12, where we have used the
unmodulated carrier as the reference. ( ) r t represents the envelope of the
resultant (signal +noise) phasor and ( ) t , the phase angle of the resultant. As
far as this analysis is concerned, ( ) r t is of no consequence (any variations in
( ) r t are taken care of by the limiter). We express ( ) t as
( ) ( )
( ) ( ) ( )
( ) ( ) ( )
n
c n
r t t t
t t
A r t t t
1
sin
tan
cos

= +

+

(7.22a)


7.23

Fig. 7.12: Phasor diagram for the case of strong predetection SNR.

If we make the assumption that ( )
c n
A r t >> most of the time, we can write,
( ) ( )
( )
( ) ( )
n
c
r t
t t t t
A
sin +

(7.22b)
Notice that the second term on the RHS of Eq. 7.22(b) has the factor
( )
n
c
r t
A
. Thus
when the FM signal is much stronger than the noise, it will suppress the small
random phase variations caused by noise; then the FM signal is said to capture
the detector. ( ) v t , the output of the discriminator is given by,
( ) ( )
d
v t k t = (phase detector)

( )
d
d t k
d t 2

=
(frequency detector)
where
d
k is the gain constant of the detector under consideration.

a) Phase Modulation
For PM, ( ) ( )
p
t k m t = . For convenience, let
p d
k k 1 = . Then,
( ) ( )
( )
( ) ( )
d n
c
k r t
v t m t t t
A
sin +

(7.23)
Again, we treat ( ) m t to be a sample function of a WSS process ( ) M t . Then,

7.24
output signal power = ( ) ( )
M M
P M t R
2
0 = = (7.24)
Let ( )
( )
( ) ( )
d n
P
c
k r t
n t t t
A
sin =

(7.25)

To calculate the output noise power, we require the power spectral density
of ( )
P
n t . This is made somewhat difficult because of ( ) t in ( )
P
n t . The analysis
becomes fairly easy if we assume ( ) t 0 = . Of course, it is possible to derive
the PSD of ( )
P
n t without making the assumption that ( ) t 0 = . This has been
done in Appendix A7.1. In this appendix, it has been shown that the effect of
( ) t is to produce spectral components beyond W , which are anyway removed
by the final, LPF. Hence, we proceed with our analysis by setting ( ) t 0 = on
the RHS of Eq. 7.25. Then ( )
P
n t reduces to,
( )
( )
( )
d n
P
c
k r t
n t t
A
sin =

( )
d
s
c
k
n t
A
=
Hence,
( ) ( )
P s
d
N N
c
k
S f S f
A
2

=

But, ( )
s
T
N
B
N f
S f
otherwise
0
,
2
0 ,

T
B is the transmission bandwidth, which for the PM case can be taken as the
value given by Eq. 5.26.

Post detection LPF passes only those spectral components that are within
( ) W W , . Hence the output noise power
d
c
k
W N
A
2
0
2

=

, resulting in,

7.25
( )
M
PM
d
c
P
SNR
k
W N
A
2 0,
0
2
=

c
p M
A
k P
W N
2
2
0
2
= (7.26)
As, ( )
( ) c
r PM
A
SNR
N W
2
,
0
2
=
we have,
( )
( )
( )
p M
PM
r
SNR
FOM k P
SNR
2
0
= = (7.27a)
We can express ( )
PM
FOM in terms of the RMS bandwidth. From Eq. A5.4.7,
(Appendix A5.4), we have

( )
( )
( )
rms p M rms
PM M
B k R B 2 0 =

( )
p M rms
M
k P B 2 =
Hence
( )
( )
rms
PM
p M
rms
M
B
k P
B
2
2
2
4
=
Using this value in Eq. 7.27(a), we obtain
( )
( )
( )
rms
PM
PM
rms
M
B
FOM
B
2
2
4

=

(7.27b)

b) Frequency Modulation
( )
( )
d
d t k
v t
d t 2

=
(7.28a)
( )
( )
s d
f d
c
d n t k
k k m t
A d t 2
= +
(7.28b)
Again, letting
f d
k k 1 = , we have

7.26
( ) ( )
( )
s d
c
d n t k
v t m t
A d t 2
= +

output signal power
M
P =
Let ( )
( )
s d
F
c
d n t k
n t
A d t 2
=

Then, ( ) ( )
F S
d
N N
c
k
S f j f S f
A
2
2
2
2

=

The above step follows from the fact that
( )
s
d n t
d t
can be obtained by
passing ( )
s
n t through a differentiator with the transfer function j f 2 . Thus,
( ) ( )
F S
d
N N
c
k f
S f S f
A
2 2
2
=

Fig. 7.13: Noise spectra at the FM discriminator output

As ( )
S
N
S f is flat for
T
B
f
2
, we find that ( )
F
N
S f is parabolic as shown
in Fig. 7.13.

The post detection filter eliminates the spectral components beyond
f W > . Hence,

7.27
The output noise power
W
d
c
W
k f N
d f
A
2 2
0
2
=

d
c
k N
W
A
2
3
0
2
2
3

=

(7.29)
This is equal to the hatched area in Fig 7.13.

Again, as in the case of PM, we find that increasing the carrier power has
a noise quietening effect. But, of course, there is one major difference between
( )
P
N
S f and ( )
F
N
S f ; namely, the latter is parabolic whereas the former is a fiat
spectrum.

The parabolic nature of the output FM noise spectrum implies, that high
frequency end of the message spectrum is subject to stronger degradation
because of noise. Completing our analysis, we find that
( )
c M
FM
d
A P
SNR
k N W
2
2 3 0,
0
3
2
= (7.30a)

c M
f
A P
k
N W
2
2
3
0
3
2
= (7.30b)

c f M
A k P
N W
W
2 2
2
0
3
2

=

(7.30c)
Let us express ( )
FM
FOM in terms of
( )
rms
FM
B . From Appendix A5.4, Eq. A5.4.5,

( )
( )
rms f M
FM
B k R 2 0 =

f M
k P 2 =
That is,
( )
rms
FM
f M
B
k P
2
2
4

=

7.28
Hence, ( )
( )
rms
c FM
FM
B
A
SNR
N W W
2
2
0,
0
3
2 4

=

As ( )
r
SNR is the same as in the case of PM, we have
( )
( )
rms
f M FM
FM
B
k P
FOM
W
W
2
2
2
3
3
4

= =

(7.31)
For a given peak value of the input signal, we find that the deviation ratio D is
proportional to
f
k
W
; hence ( )
FM
FOM is a quadratic function of D. The price paid
to achieve a significant value for the FOM is the need for increased transmission
bandwidth, ( )
T
B D k W 2 = + . Of course, we should not forget the fact that the
result of Eq. 7.31 is based on the assumption that SNR at the detector input is
sufficiently large.

How do we justify that increasing D, (that is, the transmission bandwidth),
will result in the improvement of the output SNR? Let us look at Eq. 7.28(b). On
the RHS, we have two quantities, namely ( )
f d
k k m t and
( )
s d
c
d n t k
A d t 2
. The
latter quantity is dependent only on noise and is independent of the message
signal.
d
c
k
A 2
being a constant,
( )
s
d n t
d t
is the quantity that causes the
perturbation of the instantaneous frequency due to the noise. Let us that assume
that it is less than or equal to ( )
n
f , most of the time. For a given detector,
d
k is
fixed. Hence, as
f
k increases, frequency deviation increases, thereby increasing
the value of D. Let
f f
k k
,2 ,1
> . Then, to transmit the same ( ) m t , we require
more bandwidth if we use a modulator with the frequency sensitivity
f
k
,2
instead
of
f
k
,1
. In other words, ( ) ( )
f p f p
f k m f k m
,2 ,1
2 1
= > = . Hence,

7.29

( )
( )
( )
( )
n n
f f
f f
2 1

<

In other words, as the frequency derivation due to the modulating signal keeps
increasing, the effect of noise becomes less and less significant , thereby
increasing the output SNR.

Example 7.5
A tone of unit amplitude and frequency 600 Hz is sent via FM. The FM
receiver has been designed for message signals with a bandwidth upto 1 kHz.
The maximum phase deviation produced by the tone is 5 rad. We will show that
the ( ) SNR
0
31.3 = dB, given that
c
A
N
2
5
0
10
2
= .

From Eq. 7.29, output noise power for a message of bandwidth W is
d
c
k N
W
A
2
3
0
2
2
3

. For the problem on hand, W 1 = kHz. Hence output noise
power
( )
d
k
3
2
5
1000
1
3
10
= . We shall assume
f d
k k 1 = so that
d
f
k
k
1
= .
( ) ( )
c c m
FM
s t A t t cos sin = +

This maximum phase deviation produced is .
But
f m
m m
k A f
f f
= = .
As
m
A 1 = , we have

f
k
5
600
= . That is,
f
k 3000 = . Then,

d
k
1
3000
= .
Output noise power
( )
9
2 5
1 1 10
3
10
3000
=

7.30

1
2700
=
Output signal power
1
2
=
Hence ( ) SNR
0
1350 = .
31.3 = dB

Example 7.6
Compare the FOM of PM and FM when
( )
( )
m t t
3
cos 2 5 10 = . The
frequency deviation produced in both cases is 50 kHz.

For the case of PM, we have,
p
p
k
f m
'
2

=

As
p
m
3 '
2 5 10 = and f
3
50 10 = ,

p
k
3
3
2 50 10
10
2 5 10

= =

Therefore,
( )
p M
PM
FOM k P
2
=

1
100 50
2
= =
For the case of FM,

f p
f k m =
As
p
m 1 = , we have
f
k f
3
50 10 = =
Therefore,
( )
f M
FM
k P
FOM
W
2
2
3 =

7.31

( )
( )
2
3
2
3
1
50 10
2
3
5 10

3
100 150
2
= =

The above result shows, that for tone modulation and for a given
frequency deviation, FM is superior to PM by a factor of 3. In fact, FM results in
superior performance as long as
( )
( ) p p
W m m
2
2
'
2 3 < . Evidently, the Example
7.6 falls under this category.

Example 7.7
Let
( )
( ) ( )
m t t t
3 3
3cos 2 10 cos 2 5 10 = + . Assuming that the
frequency deviation produced is 50 kHz, find
( )
( )
PM
FM
FOM
FOM
.

p
m
3 3 3 '
6 10 10 10 16 10 = + =

p
m 3 1 4 = + =
We have
p
p f p
k
m k m
3 '
50 10
2
= =
. That is,

p p
f
p
k m
k
m
3
2
1
'
2 10
= =

Hence,

( )
( )
p
PM
f
FM
FOM k
W
FOM k
2
2 6
6
1 1 1
25 10
3 3
4 10

= =

25
2.1
12
=


7.32

This example indicates that PM is superior to FM. It is the PSD of the input
signal that decides the superiority or otherwise of the FM over PM. We can gain
further insight into this issue by looking at the expressions for the FOM in terms
of the RMS bandwidth.

From Eq. 7.27(b) and 7.31, we have

( )
( )
( )
( ) ( )
rms
PM PM
FM
rms rms
M FM
B W
FOM
FOM
B B
2
2
2 2
1
3

=

Assuming the same RMS bandwidth for both PM and FM, we find that PM is
superior to FM, if

( )
rms
M
W B
2
2
3

>

If
( )
rms
M
W B
2
2
3

=

, then both PM and FM result in the same performance.
This case corresponds to the PSD of the message signal, ( )
M
S f being uniformly
distributed in the range ( ) W W , . If ( )
M
S f decreases with frequency, as it does
in most cases of practical interest, then
( )
rms
M
W B
2
2
3

>

and PM is superior
to FM. This was the situation for the Example 7.7. If, on the other hand, the
spectrum is more heavily weighted at the higher frequencies, then
( )
rms
M
W B
2
2
3

<

, and FM gives rise to better performance. This was the
situation for the Example 7.6, where the entire spectrum was concentrated at the
tail end (at 5 kHz) with nothing in between.

In most of the real world information bearing signals, such as voice, music
etc. have spectral behavior that tapers off with increase in frequency. Then, why
not have PM broadcast than FM transmission? As will be seen in the context of
pre-emphasis and de-emphasis in FM, the so called FM transmission is really a

7.33
combination of PM and FM, resulting in a performance which is better than either
PM or FM alone.

We have developed two different criteria for comparing the SNR
performance of PM and FM, namely, PM is superior to FM, if either
C1)
p
p
m
W
m
2
2 2
'
3
4
>
, or
C2)
( )
rms
M
W B
2
2
3

>

is satisfied.

Then which criterion is to be used in practice? C1 is based on the
transmission bandwidth where as C2 is based on the RMS bandwidth. Though
C1 is generally preferred, in most cases of practical interest, it may be difficult to
arrive at the parameters required for C1. Then the only way to make comparison
is through C2.

7.6.2 Weak predetection SNR: Threshold effect
Consider the phasor diagram shown in Fig. 7.14, where the noise phasor
is of a much larger magnitude, compared to the carrier phasor. Then, ( ) t can
be approximated as
( ) ( )
( )
( ) ( )
c
n
A
t t t t
r t
sin +

(7.32)


7.34

Fig. 7.14: Phasor diagram for the case of weak predetection SNR
As can be seen from Eq. 7.32, there is no term in ( ) t that represents
only the signal quantity; the term that contains the signal quantity in ( ) t is
actually multiplied by
( )
c
n
A
r t
, which is random. This situation is somewhat
analogous to the envelope detection of AM with low predetection SNR. Thus, we
can expect a threshold effect in the case of FM demodulation as well. As
( )
c
n
A
r t
,
is small most of the time, phase of ( ) r t is essentially decided by ( ) t . As ( ) t
is uniformly distributed, it is quite likely that in short time intervals such as ( ) t t
1 2
, ,
( ) t t
3 4
, etc., ( ) t changes by 2 (i.e., ( ) r t rotates around the origin) as shown
in Fig. 7.15(a).


7.35

Fig. 7.15: Occurrence of short pulses at the frequency discriminator output for
low predetection SNR.

When such phase variations go through a circuit responding to
d
d t
, a series of
short pulses appear at the output (Fig. 7.15(b)). The duration and frequency
(average number of pulses per unit time) of such pulses will depend on the
predetection SNR. If SNR is quite low, the frequency of the pulses at the
discriminator output increases. As these short pulses have enough energy at the
low frequencies, they give rise to crackling or sputtering sound at the receiver
(speaker) output. The ( ) SNR
0
formula derived earlier, for the large input SNR
case is no longer valid. As the input SNR keeps decreasing, it is even

7.36
meaningless to talk of ( ) SNR
0
. In such a situation, the receiver is captured by
noise and is said to be working in the threshold region. (To gain some insight into
the occurrence of the threshold phenomenon, let us perform the following
experiment. An unmodulated sinewave +bandlimited white noise is applied as
input to an FM discriminator. The frequency of the sinusoid can be set to the
centre frequency of the discriminator and the PSD of the noise is symmetrical
with respect to the frequency of the sinusoid. To start with, the input SNR is
made very high. If the discriminator output is observed on an oscilloscope, it may
resemble the sample function of a bandlimited white noise. As the noise power is
increased, impulses start appearing in the output. The input SNR value at which
these spikes or impulses start appearing is indicative of the setting in of the
threshold behavior).

We now present a few oscilloscopic displays of the experiment suggested
above. Display-1 and Display-2 are in flash animation.
TUFM: Display - 1UT: (Carrier +noise) at the input to PLL with a Carrier-to-Noise Ratio
(CNR) of about 15 dB.
TUFM: Display - 2UT: Output of the PLL for the above input. Note that the response of
the PLL to a signal at the carrier frequency is zero. Hence,
display-2 is the response of the PLL for the noise input which
again looks like a noise waveform.
FM: Display - 3: Expanded version of a small part of display - 2. This could be
treated as a part of a sample function of the output noise
process.
FM: Display - 4: (Carrier +noise) at the input to the PLL. CNR is 0 dB. The effect
of noise is more prominent in this display when compared to the
15 dB case.
FM: Display - 5: Output of the PLL with the input corresponding to 0 dB CNR.
Appearance of spikes is clearly evident.


7.37

FM: Display - 3

FM: Display - 4

FM: Display - 5


7.38
As in the case of AM noise analysis, if we set the limit that for the FM
detector to operate above the threshold as,
[ ]
n c
P R A 0.01 > , then we find
that the minimum carrier-to-noise ratio
c
T
A
B N
2
0
2
= required is about 5. But,
experimental results indicate that to obtain the predicted SNRimprovement of
the WBFM, is of the order of 20, or 13 dB. That is, if
c
T
A
B N
2
0
20
2
> , then the
FM detector will be free from the threshold effect.

Fig. 7.16 gives the plots of ( ) SNR
0
vs. ( )
c
r
A
SNR
N W
2
0
2
= for the case of
FM with tone modulation. If we take ( )
T
B W 2 1 = + , threshold value of ( )
r
SNR
will approximately be ( ) 13 10log 1 + +

dB. More details on the threshold
effect in FM can be found in [1, 2].

Fig. 7.16: ( ) SNR
0
performance of WBFM


7.39
For the FM demodulator operating above the threshold, we have
(Eq. 7.31),

( )
( )
f m
r
SNR
k P
SNR
W
2
0
2
3
=
For a tone signal, ( )
m m
A t cos ,
m
m
A
P
2
2
= ,
m
W f = and
f m
k A f = . As
m
f
f
= , we have

( )
( )
r
SNR
SNR
2
0
3
2
= .
That is,
( ) ( )
r
FM FM
SNR SNR
2
10 10 10
0
3
10log 10log 10log
2

= +

(7.33)
That is, WBFM operating above threshold provides an improvement of
2
10
3
10log
2

dB, with respect to ( )
r
SNR . For 2 = , this amounts to an
improvement of about 7.7 dB and 5 = , the improvement is about 15.7 dB. This
is evident from the plots in Fig. 7.16.

We make a few observations with respect to the plots in Fig. 7.16.
(i) Above threshold [i.e. ( )
r
SNR above the knee for each curve), WBFM gives
rise to impressive ( ) SNR
0
performance when compared to DSB-SC or SSB
with coherent detection. For the latter, ( ) SNR
0
is best equal to ( )
r
SNR .
(Using pre-emphasis and de-emphasis, the performance of FM can be
improved further).
(ii) Simply increasing the bandwidth without a corresponding increase in the
transmitted power does not improve ( ) SNR
0
, because of the threshold

7.40
effect. For example with ( )
r
SNR about 18dB, 2 = and 5 = give rise to
the same kind of performance. If ( )
r
SNR is reduced a little, say to about
16dB, the ( ) SNR
0
performance with 5 = is much inferior to that of
2 = .

7.7 Pre-Emphasis and De-Emphasis in FM
For many signals of common interest, such as speech, music etc., most of
the energy concentration is in the low frequencies and the frequency components
near about W have very little energy in them. When these low energy, high-
frequency components frequency modulate a carrier, they will not give rise to full
frequency deviation and hence the message will not be utilizing fully the allocated
bandwidth. Unfortunately, as was established in the previous section, the noise
PSD at the discriminator output increases as f
2
. The net result is an
unacceptably low SNR at the high frequency end of the message spectrum.
Nevertheless, proper reproduction of the high frequency (but low energy) spectral
components of the input spectrum becomes essential from the point of view of
final tonal quality or aesthetic appeal. To offset this undesirable occurrence, a
clever but easy-to-implement signal processing scheme has been proposed
which is popularly known as pre-emphasis and de-emphasis technique.

Pre-emphasis consists in artificially boosting the spectral components in
the latter part of the message spectrum. This is accomplished by passing the
message signal ( ) m t , through a filter called the pre-emphasis filter, denoted
( )
PE
H f . The pre-emphasized signal is used to frequency modulate the carrier at
the transmitting end. In the receiver, the inverse operation, de-emphasis, is
performed. This is accomplished by passing the discriminator output through a
filter, called the de-emphasis filter, denoted ( )
DE
H f . (See Fig. 7.11.) The de-

7.41
emphasis operation will restore all the spectral components of ( ) m t to their
original level; this implies the attenuation of the high frequency end of the
demodulated spectrum. In this process, the high frequency noise components
are also attenuated, thereby improving the overall SNR at the receiver output.

Let ( )
F
N
S f denote the PSD of the noise at the discriminator output. Then
the noise power spectral density at the output of the de-emphasis filter is
( ) ( )
F
DE N
H f S f
2
. Hence,
Output noise power with de-emphasis ( ) ( )
F
W
DE N
W
H f S f d f
2
=

(7.34)

As the message power is unaffected because of PE-DE operations.
( )
( )
DE
PE
H f
H f
1
Note that

=

, it follows that the improvement in the output
SNR is due to the reduced noise power after de-emphasis. We quantify the
improvement in output SNR, produced by PE-DE operation by the improvement
factor I , where
I
average output noise power without PE - DE
average output noise power with PE - DE
= (7.35)
The numerator of Eq. 7.35 is
d
c
k N W
A
2 3
0
2
2
3
. As the frequency range of interest is
only f W , let us take
( )
F
d
N
c
k N f
f W
S f
A
otherwise
2 2
0
2
2
,
0 ,

We can now compute the denominator of Eq. 7.35 and thereby the improvement
factor, which is given by

7.42

( )
W
DE
W
W
I
f H f d f
3
2
2
2
3
(7.36)

We shall now describe the commonly used PE-DE networks in the
commercial FM broadcast and calculate the corresponding improvement in
output SNR.

Fig 7.17(a) gives the PE network and 7.17(b), the corresponding DE
network used in commercial FM broadcast. In terms of the Laplace transform,
( )
PE
s
r C
H s K
R r
s
r RC
1
1
+
=
+
+
(7.37)
where K
1
is a constant to be chosen appropriately. Usually R r << . Hence,
( )
PE
s
r C
H s K
s
RC
1
1
1
+
+
(7.38)

Fig. 7.17: Circuit schematic of a PE-DE network

7.43
The time constant
C
T r C
1
= normally is 75 sec. lf
C
f
T
1 1
1
1
2 = = ,
then f
1
2.1 = kHzTP
1
PT. The value of
C
T RC
2
= is not very critical, provided
C
f
T
2
2
1
2
=
is not less than the highest audio frequency for which pre-emphasis
is desired (15 kHz).

Bode plots for the PE and DE networks are given in Fig. 7.18. Eq. 7.38
can be written as
( )
PE
RC s r C
H s K
r C sRC
1
1
1
+
+
(7.39a)
with s j f 2 = , ( )
PE
f
j
f
H f K
f
j
f
1
2
1
1

+

+

(7.39b)
where
R
K K
r
1
= .

For f f
2
, we can take
( )
PE
f
H f K j
f
1
1

= +

Hence ( )
DE
f
H f j
f K
1
1
1
1

= +

TP
1
PT The choice of f
1
was made on an experimental basis. It is found that this choice of f
1

maintained the same peak amplitude
p
m with or without PE-DE. This satisfies the constraint of a
fixed
T
B .


7.44

Fig. 7.18: Bode plots of the response of PE-DE networks

The factor K is chosen such that the average power of the emphasized
message signal is the same as that of the original message signal ( ) m t . That is,
K is such that
( ) ( ) ( )
W W
M PE M M
W W
S f d f H f S f d f P
2

= =

(7.40)
This will ensure the same RMS bandwidth for the FM signal with or without PE.
Note that
( )
rms f M
FM
B k P 2 = .

Example 7.8
Let ( )
M
f W
f
S f
f
outside
2
1
1
,
1
0 ,
= +

and ( )
PE
f
H f K j
f
1
1

= +


7.45
Let us find (i) the value K and (ii) the improvement factor I , assuming f
1
2.1 =
kHz and W 15 = kHz.

From Eq. 7.40, we have

W W
W W
d f K d f
f
f
2
1
1
1

=

+

or
f W
K
W f
1
1
1
tan

=

and
W
W
W
W
K Cf d f
I
f
Cf d f
f
2
1
2
2
1
1
=

+

(7.41a)
where ( )
N
S f is taken as Cf
2
, C being a constant.
Carrying out the integration, we find that

W
f W
I
f
W W
f f
1
2
1
1 1
1 1
tan
3 tan

=

(7.41b)
With W 15 = kHz and f
1
2.1 = kHz, I 4 6 = dB.

Pre-emphasis and de-emphasis also finds application in phonographic
and tape recording systems. Another application is the SSB/FM transmission of
telephone signals. In this, a number of voice channels are frequency division
multiplexed using SSB signals; this composite signal frequency modulates the
final carrier. (See Exercise 7.3.) PE-DE is used to ensure that each voice
channel gives rise to almost the same signal-to-noise ratio at the destination.


7.46
Example 7.9
Pre-emphasis - de-emphasis is used in a DSB-SC system. PSD of the
message process is,
( )
M
S f f W
f
f
2
1
1
,
1
=

+

Let ( )
PE
f
H f K j
f
1
1

= +

where f
1
is a known constant.
Transmitted power with pre-emphasis remains the same as without pre-
emphasis. Let us calculate the improvement factor I .

As the signal power with pre-emphasis remains unchanged, we have
( ) ( )
W W
M M
W W
f
S f d f S f K d f
f
2
1
1

= +

f
K d f
f
f
f
2
2
1
1
1
1
1

= +

+

That is,
( )
W
M
W
f W
K S f d f
W f
1
1
1
tan

= =

Noise power after de-emphasis,
( )
W
out DE
W
N
N H f d f
2
0
4
=

W
W
f
f
N
d f
K
1
2
1
0
1
4

+

=


7.47

N
W
0
2
=
Note that the noise quantity at the output of the coherent demodulator is ( )
c
n t
1
2
.
Noise power without de-emphasis
N W N
W
0 0
2
4 2
= =
Hence,
W N
I
W N
0
0
2
1
2
= =
This example indicates that PE-DE is of no use in the case of DSB-SC.

Exercise 7.2
A signal ( ) ( ) m t t 2cos 1000 =

is used to frequency modulate a
very high frequency carrier. The frequency derivation produced is 2.5 kHz. At
the output of the discriminator, there is bandpass filter with the passband in
the frequency range f 100 900 < < Hz. It is given that
c
A
N
2
5
0
2 10
2
= and
d
k 1 = .
a) Is the system operating above threshold?
b) If so, find the ( ) SNR
0
, dB.
Ans: (b) 34 dB

7.48

Exercise 7.3

Fig. 7.19: Transmission scheme of Exercise 7.3

Each one of the USSB signals occupies a bandwidth of 4 kHz with respect to
its carrier. All the message signals, ( )
i
m t i , 1, 2, , 10 = , have the same
power. ( ) m t frequency modulates a high frequency carrier. Let ( ) s t
represent the FM signal.
a) Sketch the spectrum of ( ) s t . You can assume suitable shapes for
( ) ( ) ( ) M f M f M f
1 2 10
, , ,
b) At the receiver ( ) s t is demodulated to recover ( ) m t . (Note that from
( ) m t we can retrieve ( )
j
m t j , 1, 2, , 10 =

.) If ( ) m t
1
can give rise to
signal-to-noise ratio of 50 dB, what is the expected signal-to-noise ratio
from ( ) m t
10
?

7.49
7.8 Noise Performance of a PCM system
There are two sources of error in a PCM system: errors due to
quantization and the errors caused by channel noise, often referred to as
detection errors. We shall treat these two sources of error as independent noise
sources and derive an expression for the signal-to-noise ratio expected at the
output of a PCM system. As we have already studied the quantization noise, let
us now look into the effects of channel noise on the output of a PCM system.

We will assume that the given PCM system uses polar signaling. Even if
the transmitted pulse is rectangular, the received pulse ( )
r
p t will be distorted
due to a band-limited and imperfect channel; hence ( )
r
p t may look like as
shown in Fig. 7.20 (a).

Fig. 7.20: (a) Typical received pulse (without noise)
(b) Received pulse with noise

The input to the PCM receiver will be ( ) ( ) ( )
r
r t p t n t = + , where ( ) n t is
a sample function of a zero-mean Gaussian process with variance
N
2
. A
possible pulse shape of ( ) r t is shown in Fig. 7.20(b). The detection process of
the PCM scheme consists of sampling ( ) r t once every
b
T seconds and
comparing it to a threshold. For the best performance, it would be necessary to
sample ( )
r
p t at its peak amplitude
( )
s b p
k t kT t = + resulting in a signal

7.50
component of
p
A . Hence ( )
s p
r k t A N = + whose N is a zero mean
Gaussian variable, representing the noise sample of a band-limited white
Gaussian process with a bandwidth greater than or equal to the bandwidth of the
PCM signal. If binary '1' corresponds to
p
A + and '0' to
p
A , then,
( )
R
f r '1' transmitted is
( ) p N
N A
2
, and
( )
R
f r '0' transmitted is
( ) p N
N A
2
,
where R (as a subscript) represents the received random variable. We assume
that 1's and 0's are equally likely to be transmitted. The above conditional
densities are shown in Fig. 7.21. As can be seen from the figure, the optimum
decision threshold is zero. Let
e
P
,0
denote the probability of wrong decision,
given that '0' is transmitted (area hatched in red); similarly
e
P
,1
(area hatched in
blue). Then
e
P , the probability of error is given by
e e e
P P P
,0 ,1
1 1
2 2
= + .
From Fig. 7.21, we have
( )
p
e R
N
A
P f r d r Q
,0
0
'0'

= =

Fig. 7.21: Conditional PDFs at the detector input

Similarly
p
e
N
A
P Q
,1

=

, which implies

7.51

p
e
N
A
P Q

=

(7.42)
For optimum results, the receiver uses a matched filter whose output is sampled
once every
b
T seconds, at the appropriate time instants so as to obtain the best
possible signal-to-noise ratio at the filter output. In such a situation, it can be
shown that we can replace
p
N
A

by
b
E
N
0
2
where
b
E is the energy of the
received binary pulse and
N
0
2
represents the spectral height of the, band-limited
white Gaussian noise process. Using the above value for
p
N
A

yields,

b
e
E
P Q
N
0
2

=

(7.43)
Assuming that there are R binary pulses per sample and W 2 samples/second,
we have
( )
=
b
T
RW
1
2
where
b
T represents the duration of each pulse. Hence
the received signal power
r
S is given by, = =
b
r b
b
E
S RW E
T
2 or =
r
b
S
E
RW 2
.
Therefore Eq. 7.43 can also be written as

=

r
e
S
P Q
RW N
0

=

Q
R
(7.44)
where
r
S
W N
0
= . Eq. 7.42 to 7.44 specify the probability of any received bit
being in error. In a PCM system, with
R
bits per sample, error in the
reconstructed sample will depend on which of these R bits are in error. We would
like to have an expression for the variance of the reconstruction error. Assume
the following:

7.52
i) The quantizer used is a uniform quantizer
ii) The quantizer output is coded according to natural binary code
iii)
e
P is small enough so that the probability of two or more errors in a block
of
R
bits is negligible.

Then, it can be shown that
c
2
, the variance of the reconstruction error
due to channel noise is,

( )
( )

=
R
p e
c
R
m P
2 2
2
2
4 2 1
3 2
(7.45)
Details can be found in [3]. In addition to the reconstruction error due to channel
noise, PCM has the inevitable quantization noise with variance
Q
2
2
12
= , where
=
p
m
L
2
and =
R
L 2 . Treating these two error sources as independent noise
sources, total reconstruction noise variance
e
2
, can be written as

e Q c
2 2 2
= +

( )
= +
p e
p
m P L
m
L L
2 2
2
2 2
4 1
3 3
(7.46)
Let ( )
M
M t P
2
=
Then, ( )
M
e
P
SNR
2 0
=

( )

=

+

M
p e
P L
m P L
2
2 2
3
1 4 1
(7.47)
Using Eq. 7.44 in Eq. 7.47, we have
( )
( )

=

+

M
p
P L
SNR
m
L Q
R
2
0 2
2
3
1 4 1
(7.48)

7.53
Figure 7.22 shows the plot of ( ) SNR
0
as a function of for tone modulation
M
p
P
m
2
1
2

=

. However, with suitable modifications, these curves are applicable
even in a more general case.

Fig. 7.22: ( ) SNR
0
performance of PCM

Referring to the above figure, we find that when is too small, the
channel noise introduces too many detection errors and as such reconstructed
waveform has little resemblance to the transmitted waveform and we encounter
the threshold effect. When is sufficiently large, then
e
P 0 and
e Q
2 2

which is a constant for a given n . Hence ( ) SNR
0
is essentially independent of
resulting in saturation.


7.54
For the saturation region, ( ) SNR
0
can be taken as
( )
( )

=

R M
p
P
SNR
m
2
0 2
3 2 (7.49)
The transmission bandwidth
PCM
B of the PCM system is
( ) k WR 2 where k is a
constant that is dependent on the signal format used. A few values of k are
given below:

S. No. Signal Format k
1 NRZ polar 1
2 RZ polar 2
3 Bipolar (RZ or NRZ) 1
4 Duobinary (NRZ) 1
2

(For a discussion on duobinary signaling, refer Lathi [3]). As
( ) =
PCM
B k W R 2 ,
we have
( )
=
PCM
B
R
k W 2
. Using this in the expression for ( ) SNR
0
in the saturation
region, we obtain
( )
PCM
B
kW M
p
P
SNR
m
2 0
3 2

=

(7.50)
It is clear from Eq. 7.50 that in PCM, ( ) SNR
0
increases exponentially with the
transmission bandwidth. Fig. 7.22 also depicts the ( ) SNR
0
performance of DSB-
SC and FM ( ) 2, 5 = . A comparison of the performance of PCM with that of
FM is appropriate because both the schemes exchange the bandwidth for the
signal -to -noise ratio and they both suffer from threshold phenomenon. In FM,
( ) SNR
0
increases as the square of the transmission bandwidth. Hence,
doubling the transmission bandwidth quadruples the output SNR. In the case of

7.55
PCM, as can be seen from Eq. 7.49, increasing R by 1 quadruples the output
SNR, where as the bandwidth requirement increases only by
R
1
. As an
example, if R is increased from 8 to 9, the additional bandwidth is 12.5% of that
required for R =8. Therefore, in PCM, the exchange of SNR for bandwidth is
much more efficient than that in FM, especially for large values of
R
TP
1
PT. In addition,
as mentioned in the introduction, PCM has other beneficial features such as use
of regenerative repeaters, ease of mixing or multiplexing various types of signals
etc. All these factors put together have made PCM a very important scheme for
modern-day communications.

The PCM performance curves of Fig. 7.22 are based on Eq. 7.48 which is
applicable to polar signaling. By evaluating
e
P for other signaling techniques
(such as bipolar, duobinary etc.) and using it in Eq. 7.47, we obtain the
corresponding expressions for the ( ) SNR
0
. It can be shown that the PCM
performance curves of Fig.7.19 are applicable to bipolar signaling if 3 dB is
added to each value of .

Example 7.10
A PCM encoder produces ON-OFF rectangular pulses to represent 1 and
0 respectively at the rate of 1000 pulses/sec. These pulses amplitude modulate
a carrier,
( )
c c
A t cos , where
c
f 1000 >> Hz. Assume that 1s and 0s are
equally likely. Consider the receiver scheme shown in Fig. 7.23.

TP
1
PT Note that the FM curve for 2 = and PCM curve for = R 6 intersect at point A . The
corresponding is about 30 dB. If we increase beyond this value, there is no further
improvement in ( ) SNR
0
of the PCM system where as no saturation occurs in FM. If we take
( )
T m
B f 2 1 = + , then the transmission bandwidth requirements of PCM with = R 6 and FM
with 2 = are the same. This argument can be extended to other values of R and .

7.56

Fig. 7.23: Receiver for the Example 7.11

( )
( )
c
t
s t
2cos , if the input is '1'
0 , if the input is '0'

( ) w t : sample function of a white Guassian noise process, with a two sided
spectral density of
4
0.25 10
Watts/Hz.
BPF : Bandpass filter, centered at
c
f with a bandwidth of 2 kHz so that the
signal is passed with negligible distortion.
DD : Decision device (comparator)

a) Let Y denote the random variable at the sampler output. Find ( )
Y
f y
0
0
and ( )
Y
f y
1
1 , and sketch them.
b) Assume that the DD implements the rule: if Y 1 , then binary 1 is
transmitted, otherwise it is binary 0.
If ( )
Y
f y
1
1 can be well approximated by ( ) N 2, 0.1 , let us find the overall
probability of error.

( ) ( ) ( ) x t s t n t = + , where ( ) n t is the noise output of the BPF.

( ) ( ) ( ) ( ) ( )
k c c c s c
A t n t t n t t cos cos sin = +
where
k
A
th
th
2, if the k transmission bit is '1'
0, if the k transmission bit is '0'

The random process ( ) Y t at the output of the ED, is
( ) ( ) [ ]
c
Y t R t t cos = +

7.57
( ) ( ) ( )
{ } k c s
R t A n t n t
1
2 2
2
= + +

If 0 is transmitted, ( ) Y t represents the envelope of narrowband noise. Hence,
the random variable Y obtained by sampling ( ) Y t is Rayleigh distributed.
( )
y
N
Y
y
f y e y
N
2
0
2
0
0
0 , 0
=
where
( )
N
4 3
0
0.25 10 4 10 0.1
= =
when 1 is transmitted, the random variable Y is Rician, given by
( )
y
N
Y
y y
f y I e y
N N
2
0
4
2
1 0
0 0
2
1 , 0

+

=

These are sketched in Fig. 7.24.

Fig. 7.24: The conditional PDFs of Example 7.10

b) As the decision threshold is taken as 1, we have

y
e
P y e d y e
2
5 5
, 0
1
10

= =

( ) y
e
P e d y
2
2
1
0.2
, 1
1
2 0.1


7.58

( )
Q 10 =
( )
e
P error P e Q
5
1
10
2

= = +


7.59
Appendix A7.1
PSD of Noise for Angle Modulated Signals
For the case of strong predetection SNR, we have (Eq. 7.22b),
( ) ( )
( )
( ) ( )
n
c
r t
t t t t
A
sin +

Let ( )
( )
( ) ( )
n
c
r t
t t t
A
sin =

( )
( )
( ) j t j t
m n
c
I e r t e
A
1

=

Note that
( ) j t
e

is the complex envelope of the FM signal and ( )
( ) j t
n
r t e

is
( )
ce
n t , the complex envelope of the narrow band noise.

Let ( )
( )
( ) ( ) ( )
j t
ce ce c s
x t e n t x t j x t

= = + .
Then,
( ) ( )
s
c
t x t
A
1
=
We treat ( )
ce
x t to be a sample function of a WSS random process ( )
ce
X t
where,
( )
( )
( ) ( ) ( )
j t
ce ce c s
X t e N t X t j X t

= = + (A7.1.1)
Similarly,

( ) ( )
s
c
t X t
A
1
= (A7.1.2)
Eq. A7.1.2 implies, the ACF of ( ) t , ( ) R
is
( ) ( )
s
X
c
R R
A
2
1
= (A1.7.3a)
and the PSD
( ) ( )
s
X
c
S f S f
A
2
1
= (A7.1.3b)
We will first show that

7.60
( ) ( )
ce ce
E X t X t 0 + =

( ) ( ) ( ) ( )
( ) ( ) j t t
ce ce ce ce
E X t X t E N t N t e
+ +

+ = +

(A7.4.1a)
( ) ( )
( ) ( ) j t t
ce ce
E N t N t E e
+ +

= +

(A7.4.1b)
Eq. A7.4.1(b) is due to the condition that the signal and noise are statistically
independent.

( ) ( ) ( ) ( ) ( ) ( )

+ = + +

c s c s s c
ce ce N N N N N N
E N t N t R R j R R
As ( ) ( )
c s
N N
R R = and ( ) ( )
c s s c
N N N N
R R = ,
we have ( ) ( )
+ =

ce ce
E N t N t 0
Therefore
( ) ( ) ( ) ( ) ( ) ( )
c s c s s c
ce ce X X X X X X
E X t X t R R j R R 0

+ = + + =

This implies ( ) ( )
c s
X X
R R = .

Now consider the autocorrelation of ( )
ce
X t ,

( ) ( )
( )
( )
( )
( )
{ }
j t j t
ce ce ce ce
E X t X t E e N t e N t
+

+ = +

(A7.1.5)
( ) ( ) ( ) ( )
( ) ( ) j t t
ce ce ce ce
a j b
a j b
E X X E N t N t E e
1 1
2 2
+

+
+

= +

(A7.1.6)

( ) ( ) ( ) ( )
c s s c c s
X X X X X X
R R j R R

= + +

( ) ( ) ( )
s s c c s
X X X X X
R j R R 2

= +

(Note that ( ) ( )
c s
X X
R R = as proved earlier.)
( ) ( ) ( )
s
X
R a j b a j b
1 1 2 2
1
Re
2
= + +


7.61
( ) a a b b
1 2 1 2
1
2
=
Now ( )
s c
N N
b R
1
2 0 = = for symmetric noise PSD, and ( )
s
n
a R
1
2 = . Hence,
( ) ( ) ( ) ( )
{ }
( )
s s
X N
g
R R E t t
1
2 cos
2

= +

That is,
( ) ( ) ( )
s s
X N
S f S f G f =
where ( ) ( ) g G f .
As ( ) ( )
s
X
c
S f S f
A
2
1
= , we have,
( ) ( ) ( )
s
N
c
S f S f G f
A
2
1

=

By definition, ( ) g is the real part of the ACF of
( ) j t
e

. We know that
( ) j t
e

is
the complex envelope of the FM process. For a wideband PM or FM, the
bandwidth of ( ) G f W >> , the signal bandwidth. The bandwidth of ( ) G f is
approximately
T
B
2
, as shown in Fig. A7.1.

Fig. A7.1: Components of ( ) S f
: (a) typical ( ) G f (b) ( )

s
N
S f


7.62
Note that ( ) ( ) ( ) g G f d f E 0 cos 0 1

= = =

.

For f W , ( ) ( ) ( )
s
N
G f S f N G f d f N
0 0
. That is, ( ) t is
immaterial as far as PSD of ( )
s
N t in the range f W is concerned. Hence, we
might as well set ( ) t 0 = in the Eq. 7.22 (b), for the purpose of calculating the
noise PSD at the output of the discriminator.


7.63
References
1) Herbert Taub and D. L. Shilling, Principles of Communication systems, (2P
nd
P
ed.) Mc Graw Hill, 1986
2) Simon Haykin, Communication systems, (4P
th
P ed.) J ohn Wiley, 2001
3) B. P. Lathi, Modern digital and analog communication systems, Holt-
Saunders International ed., 1983

Principles of Communication (By Prof. v. V Rao, IIT Madras)

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Principles of Communication (By Prof. v. V Rao, IIT Madras)

Încărcat de

Drepturi de autor:

Formate disponibile

Principles of Communication Prof. V.

denotes the integral over any one period of ( )

= . Hence the spectral plot has only

, then the series converges in the mean. That is,

is zero. There could be

; but they occur at isolated points, resulting in the

(Finite energy signal), then we have the convergence in the

. Then, 90 percent of the energy of

shown in Fig 1.16 below.

is a conjugated and shifted version of ( ) y t ,

. Note that the variable of integration in Eq. 1.35

, we find it to be zero as ( ) x t and ( ) y t do not

, which would be about the maximum of ( )

are quite similar, we can expect large value for ( )

is a right shifted version of ( ) x t

to the right by and take the

fixed and move ( ) x t to the left by . This

. This product is obtained from Eq.

are orthogonal for all

. (The student is advised to

cos 2 sin 8 2 sin 2

sgn X f j f X f = . That is,

will have spectrum only for 0 f ).

are zero for f W > , ( )

= < < where a and b are positive constants.

and to show that R and are

= < < >

< < < <

whereas at the height of

indicates summation with

, the complex conjugate of

= for all f . That is,

is taken as the standard room temperature).

( ) X t is the Hilbert transform of ( ) X t . The process ( ) X t

( ) n t is the Hilbert transform of ( ) n t .

(transformation by a multiplicative constant).

, Eq. 3.53 can

. Then, the Rician density of Eq.

( ) N t is obtained as the output of an LTI system with ( )

( ) N R is the Hilbert transform of ( )

( ) N t have the same PSD, we have ( )

1 , 98.5 kHz 101.5 kHz

can be represented as a rotating vector with a magnitude

, rotating counterclockwise at the rate of

can be shown as a vector with clockwise rotational speed of

( ) m t as the quadrature component. Similarly

( ) m t is the quadrature component.

( ) m t , the HT of ( ) m t . ( ) m t is the product of a lowpass and

( ) y t , from the result of example 1.25, is

phase shifters and an adder. One of the phase shifter

phase shift for all the

phase shifter over a 5

tends to become an impulse.

with respect to the carrier. This makes the time

, which implies that the

. For FM, the magnitude of the spectral components at ( )

= . In the scheme of Fig. 5.11, this

= = Hz. After the frequency conversion stage, we have

to resemble ( ) m t fairly closely.

<< , we can take ( ) H f as

phase shift provided by ( ) H f at

. Then, Eq. 5.30 reduces to

phase shift in the