0% found this document useful (0 votes)
291 views92 pagesIntroduction To Analog & Digital Communications
Uploaded by
Pritam Deb RoyCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
291 views92 pagesIntroduction To Analog & Digital Communications
Uploaded by
Pritam Deb RoyCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 92
CONTENTS
CHAPTER 1 |NTRODUCTION
1.1 Communicati
1.2. Signals and Their Classifications
13 Fourier Analysis of Signals a
1.4 Elements of a Communication Sy
1.5. Transmission of Message Signals
1.6 Limitations and Resources of Communication Systems
CHAPTER 2 FOURIER ANALYSIS
21 Fourier Series.
22 Fourer Transform i
23 Properties of the Fourier Transform ”
ZA rp Batvwen Time-Domain and Frequency/—
Domain Descriptions
25 Dirac Delta Function
26 _ Fourer Transforms of Perode Signals 7
27 Sampling Theorem
28 Numerical Computation of the Fourier Transform
29 Relationship Between the Fourier and Laplace
Transforms
Problems
CHAPTER 3 FILTERING AND SIGNAL
DISTORTION
Serene
Saas y
33 Linear Distorion and Equalization 2/
34 eal Low-Pass Fiters
3.5 Band-Pass Transmission ~
38. Phase Delay and Group Delay
37 Nonlinear Distorton
Problems
CHAPTER 4 SPECTRAL DENSITY AND
4.1 Energy Spectral Density ‘pafa
42 Correlation of Energy Signals
pisses Seer G eveane
”
102
108
4
8
120
127
128
07 xiv CONTENTS
4.3 Power Spectral Density y
4.4 Correlation of Power Signals ~/
4.5 Flowchart Summaries
4.6 Spectral Characteristics of Periodic Signals
4.7 Spectral Characteristics of Random Signals and Noise
48 — Noise-Equivalent Bandwidth
Problems
CHAPTER § DIGITAL CODING OF ANALOG
WAVEFORMS,
5.1 Digital Pulse Modiation
52 Pulse-Code Mocilation
53. Sampling
54 — Quantizing
55 Coding
5.6 Regeneration
5.7 _ Differential Pulse-Code Modulation
5.8 Delta Modulation
59 Discussion
Time-Division Multiplexing
5.11 Application: Digital Multiplexers for Telephony
Problems
CHAPTER 6 INTERSYMBOL INTERFERENCE
AND ITS CURES
6.1 Baseband Transmission of Binary Data
62 The Intersymbol Interference Problem
63 Ideal Solution
64 — Raised Cosine Spectrum
65 — Correlative Coding
66 Baseband Transmission of M-ary Data
67 Eye Pattern
68 Adaptive Equalization
Problems
CHAPTER 7 MODULATION TECHNIQUES
71 Amplitude Modulation
72 ble-Sideband Suppressed-Carr
73 QuadratureCarier Mutipledng ” Moauaten
7.4 Single-Sideband Modulation
7.5 Vestigial Sideband Modulation
v
141
106
168
177
78
210
213
215
221
8
231
EEESSE
geage 8
7.8 Comparison of Ampli n a
7.1 Frequency Translation
7.8 Frequency Division Muttip
7.9 Application |: Radio Broad
7.10 Angle Modulation: Basic C
7.11 Frequency Modulation
712 4 Loop
7.13. Limiting of FM Wa
7.14 Application Il: FM Radio
7.18 Digital Modulation Techniques
746 Application Il: Digital Communications by Satellite
Problems
TER 8 PROBABILITY THEORY AND
DA RANDOM PROCESSES
8.1 Probablity Theory
82 Random Variables
&3 Gaussian Distribution
84 Transformation of Ranggm Variables
85 Random Processes
86 Stationarty YY
87 Mean, Correlation, and Covariance Functions
88 Random Process Transmission Through Linear Firs
89 Power Spectral Density
Cross-Correlation Functions
Cross-Spectral Densities
Gaussian Process ee
Narrow-band Random Process 7
Envelope and Phase of Narrow-band Random Process ~/
Problems
x
NOISE IN ANALOG
CHAPTERS MODULATION
9.1 Signal-to-Noise Ratios
92 AM Receiver Model
9.3 Signal-to-Noise Ratios for Coherent Reception
9.4 Noise in AM Receivers Using Envelope Detection
95 FM Receiver Model
9.6 Noise in FM Reception
97 FM Threshold Effect
98 — Pre-emphasis and Deemphasis in FM.
99 Discussion
Problems
cod
310
322
327
208
—_—° > . e|€©=6hlUr?d xvi CONTENTS
CHAPTER ONE
CHAPTER 10 OptimUM RECEIVERS FOR
DATA COMMUNICATION 539
10.1 Formulation of the Optimum Receiver Problem 00
10.2 Maximization of Output Signalto-Noise Ratio a1
10.3. Properties of Matched Filters 545,
10.4 Approximations in Matched Filter Design 553
10.5 Probabilistic Approach 558
10.6 Probability of Error for Binary PCM 567
10.7 Noise in Digital Modulation Schemes 874
108 Coherent Detection of Binary Modulated Waves 575
109 Noncoherent Detection of Binary Modulated Waves 581
10.10 Coherent Detection of Quaternary Modulated Waves 583
10.11 Discussion 505
10.12 Tradeoffs in M-ary Data Transmission 508,
Problems 591
APPENDIX A POWER RATIOS AND DECIBELS 997
APPENDIX B BESSEL FUNCTION 598
APPENDIX C SYSTEM NOISE AND.
CALCULATIONS 603
APPENDIX D TABLES 623
GLOSSARY 631
REFERENCES AND BIBLIOGRAPHY SN ee eu GON
ANSWERS TO EXERCISES 640
INDEX 645
1 COMMUNICATIONS
js easy to overlook the multitude of its facets. The telephones in «
homes and offices make it possible for us to communicate with others
no matter how far away. The radio and television sets in our living
rooms bring us entertainment from near as well as far-away places
Communication by radio or satellite provides the mean
high seas, aircraft in flight, and rockets and exploratory
to maintain contact with their home bases. Communic
weather forecaster informed of atmospheric conditions that are
ships
bes in space eee
2 IntRopuction
‘measured by a multitude of sensors. Communication makes it possible for
computers to interact. The list of applications involving the use of com
munications in one way or another goes on.”
In the most fundamental sense, communication involves implicitly the
transmission of information from one point to another through a succession
‘of processes, as described here
‘The generation of a thought pattern or image in the mind of an origi-
2. The description ofthat thought pattern orimage, wih certain measure
of precision, by a set of aural or visual symbols.
3, The encoding ofthese symbolsina form that suitable for ansmison
over a physical medium (channel) of interest.
4,The transmision of the eneoded symbols othe desired destination
5. The decoding and reproduction of the inital symbols
6 The reeeaion ofthe orignal thought pattern or mage
tbl dopredation in gality—in the mind of recipient wth the dep
Tedaton being conse by inperesions inthe system.)
‘The form of communication just described involves a thought pattern or
image originating in a human mind. Of course, there are many other forms
‘of communication that do not directly involve the human mind in real time,
In space exploration, for example, human decisions may enter only the
‘commands sent to the space probe or to the computer responsible for pro-
‘cessing images of far-away planets (e.g., Mars, Jupiter, Saturn) that are
sent back by the probe. In computer communications, human decisions
‘enter only in setting up the computer programs or in monitoring the results
of gomputer processing.
Ihatever form of communication is used, some basic signal-processing.
‘operations are involved in the transmission of information) The next section,
deteribes the different types of signals encountered in the study of com-
‘munication systems. The signal-processing operations of interest are high-
lighted later in the chapter.
AND THEIR CLASSIFICATIONS
Gor our purposes, a signal is defined asa single-valued function of time
that conveys information. Consequently, for every instant of time there is
4 unique value of the function. This value may be a real number, in which
cease we have a real-valued signal, or it may be a complex number, in which
case we have a complex-valued signal, In either case, the independen!
variable (namely, time) is real-valued, )
‘For an essay on communications, see Berkner (1962),
RODUCTION 3
For a given situation, the most useful method of presentation
hinges on the particular type of signal being eon: Depending on the
feature of interest, we may identify four different f dividing
Signals into two classes:
1. PERIODIC SIGNALS, NONPERIODIC SIGNALS
A periodic signal g(t) is function that satisfies the condition
a(t) = glt + Ts)
for all , where denotes time and T, is a constant. The smallest
Ta that satisfies this condition is called the period of g(t). Accordiny
period Ty defines the duration of one complete cycle of g(*)
‘Any signal for which there is no value of T, to satisfy the condition of
Eq. 1.1 is called a nonperiodic or aperiodic signal.
2. DETERMINISTIC SIGNALS, RANDOM SIGNALS
A deterministic signal is a signal about which there is no uncertainty with
Tespect to its value at any time. Accordingly, we find that deterministic
signals may be modeled as completely specified functions of time
‘On the other hand, » random signal is a signal about which there is
uncertainty before its actual occurrence. Such a signal may be viewed as
belonging to an ensemble of signals, with each signal in the collection
hhaving a different waveform. Moreover, each signal within the ensemble
hhas a certain probability of occurrence.
‘8. ENERGY SIGNALS, POWER SIGNALS
In communication systems, a signal may represent a voltage or a current
Consider a voltage v(t) developed across a resistor R, producing a current
(2). The instantaneous power dissipated in this resistor is defined by
a),
Re
on, equivalently,
p= Re) as
In both cases, the instantaneous power pi proportional to the squared
amplitude ofthe signal. Furthermore. fora resstence K of | oho he see
that Eqs. 1.2 and 1.3 take on the same mathematical form. Accordingly,
in signal analysis it is customary to work with a I-ohm resistor, s0 that
* 4 wrropuction
regardless of whether a given signal g(*) represents a voltage or a current
‘We may express the instantancous power associated with the signal as
p= 8 aay
Based on this convention, we define the ‘oral energy of a signal g(") as
B= tim [ode
= i e(Ode as)
and its average power as
P = lim
lim > ‘ g(a (6)
the total
‘We say that the signal g(0) is an energy signal if and onl
‘energy of the signal satisfies the condition
0
\ Let B denote the channel bandwidth, and SNR denote the received signal-
fonoise ratio. The channel capacity theorem states that ideally these ‘wo
parameters are related by
C = Blog,(1 + SNR), bits/s a7)
where C is the channel capacity, and a bit refers to a binary digit. The
‘channel capacity is defined as the maximum rate at which information may
bbe transmied without error through the channel; it is measured in bits
‘per second. Equation 1.7 clearly shows that for a prescribed channel c2-
pacity, ve fay reduce the required SNR by increasing the channel band:
te 144, Shannon pues ppm ta ad foundations of communication
tear (Shannon, 1948). The channel capacity theorem is one penicatin
Presented in that classic paper of three theor
raooucion 19
width B} Moreover, it provides an idealized framework f aning the
[einally, mention should be made of the issue of system ) we
usually find that the efficient exploitat a “ =
mitted power or both is achieved at the 1 increased system com
plexity. We therefore have t he i oun
alongside that of channel bandwidth and transmit ™ $
ctng the various wade ivcved in the design of communication
dl CHAPTER TWO
eer QUEER ANALYSIS
application than the Fourier series.' The primary motivation for
the Fourier series or the Fourier transform is to obtain the spectrum of @
‘The origin of the theory of Fourier series and Fourier transform is fou
J.B. J. Fourier, The Analytical Theory of Heat (trans. A. Freeman),
University Press, London, 1878
13
gaa 14 FOURIER ANALYsis
Biven signal, which describes the frequency content ofthe signal. In effes
this transformation provides an alternative method of viewing the s
that is often more revealing than the original description of the signa
8 function of time.
FOURIER SERIES
Let g(t) denote a periodic signal with period Tp. By using @ Fourier sens
expansion ofthis signal, we are able to resolve the signal into an infnte
sum of sine and cosine terms. Ths expansion may be expressed inthe form
\ wi =a 025 [nerQit) +aae(22)] en
‘where the coefficients a, and b, represent the unknown amplitudes of the
cosine and sine terms, fespectively. The quantity 1/Tp represents the mh
harmonic ofthe fundamental frequency fy = 1/T. Each of the cosine and
sine functions in Eq. 2.1 is called a basis funcion. These basis functions
} form an orthogonal et over the interval T,in that they satis the following
set of relations:
ve Dan) yg (Toman ay
eee aac 2
| J cos) sn(222) = 0 toratmand'n (23)
a (tami) (Ian!) y (Tv2, man
fe snl?) ot) a= [0% moe ew
To determine the coefficient ay, we integrate both sides of Eq. 2.1 over
| ‘2 complete period. We thus find that ay is the mean value of the periodic
ignal g,(0° over one period, as shown by the time average
Lyn
a= Ff aa es)
To determine the coefficient a,, we multiply both sides of Eq, 2.1 by the
cosine function cos(2ant/T,) and integrate over the interval ~T,2 to
T,/2. Then, using Eqs. 2.2 and 23, we find that
ae k [7 nto cos(*2) aa 26
wus 18
TJ
To apply the Fourier r nof E
inside the interval T
following conditions
The function g,(t) is single-valued
The function g,(¢) has a finite nur
1
2
3. The function gy(t) has a finit
4
The function g,(t) is absolut
where g,(t) is assumed to be complex valued
‘These conditions are known as Dirichler’s conditions. They ate
the periodic signals usually encountered in communication
v
COMPLEX EXPONENTIAL FOURIER SERIES
The Fourier series of Eq. 2.1 can be put into a much simpler
elegant form with the use of complex exponentials. W
tuting in Eq. 2.1 the exponential form for the cosine and
1p (item
$ [=0(222) + en
(2) - aloe
We thus obtain
fo this by
exp(
(0 = a + | (0, ~ j6,) exo
+ (a, + 1b.) exp
‘The two product terms inside the square brackets in Eq. 2 S are the complex ‘conjugate of each other, We may also note the following Felson
So. + ing exo( 24) = 3 to. - tooo
Let c, deniote « complex coefficient related to a, and by by
a jb, > 0
c= fay 0 29)
Accordingly, we may simplify Eq, 2.8 as follows:
5 onl") s
Belt)
where
ay
are Olen
TO
“The series expansion of Eq, 2.10 is referred to as the complex exponential
Fourier series. The G ate called the complex Fourier coefficients. Equation
2.11 states that, given a periodie signal g,(#), we may determine the com:
plete set of complex Fourier coefficients. On the other hand, Eq. 2.10
States that, given this set of values, we may reconstruct the original period
signal exactly
“According to this representation, a periodic signal contains all frequen
cies (both positive and negative) that are harmonically elated to the fur»
damental. The presence of negative frequencies is simply a result of th
fact that the mathematical model of the signal as described by Eq. 2.1)
requires the use of negative {cequencies. Indeed, this representation als?
requires the use of a complex-valued basis function exp(j2nmt/T,). which
hhas no physical meaning either. The reason for using complex-valued bass
functions and negative frequency components is merely to provide a com
pact mathematical description of a periodic signal, which is well-suited fo"
both theoretical and practical work,
ouscnere srecraum
TThe representation of a periodic signal by a Fourier serie is equivalent
the revolution of the signal into its various harmonic components. 1
tring the complex exponential Fourier series, we find that a periodic $8
ret
nal g,(0) with period T, has components of f fn *2f
£3fo. «and so forth, where fy = 1/Ty is the quency
That is, while the signal g,(t) exists in the time domain, me may say that
0, * for *2fo, called fe spec . 4
64 (0), we can determine its spectrum, conversely we speci
we can determine the cortesponding signal. Thin mene
Signal g() can be specified in tw
frequency-domain representation where the signal defined in term
itsspectrum, Although the two descriptions are separate aspects of»
phenomenon, they ae not independent of cach ther, bu
Fourier theory shows
In general, the Fourier cocficient cis a
express it inthe form
m. If we specify the periodic signal
plex number; so we may
in| expl argte,)] 2
‘The term |c,| defines the amplitude of the nth harmonic
periodic signal g,(t), so that a plot of |c,| versus frequency yie
‘amplitude spectrum of the signal. A plot of arg(c,) versus trequeney yields
the discrete phase spectrum of the signal. We refer to the spectrum as ®
discrete spectrum because both the amplitude and phase af ¢, have nonze
values only for discrete frequencies that are integer (both positive and
negative) multiples of the fundamental frequency
For a real-valued periodic function g,(), we find from the definition of
the Fourier coefficient c, given by Eq. 2.11 that
ponent
the discrese
where
isthe complex conjugate of c,. We therefore have
le
and
arg(e..
arg(c,)
Thats, the amplitude spectrum ofa real-valued periodic signals sym
(an even function of m) and the phase spectrum is asymmetric (aa odd
function of n) about the vertical axis passing through the origin
"The term “spectrum” comes from the Latin word for “image.” it was originally
introduced by Si lec Newton. For » historical account of spectrum anahyaa. #8
Gardner (1987) 18 FOURIER ANALYSIS
PERIODIC PULSE TRAIN
Consider a periodic train of rectangular pulses of duration 7” and pericd
Tp, as shown in Fig. 2.1, For convenience, the origin has Peen choren
coincide with the eenter of the pulse. This signal may be described ans
Iytically over one period, — (Ty/2) < #'< (Tol2), as follows
Toyet
GO m4 Aig sa Bes 2.15)
0, forthe remainder ofthe period
Using Eq. 2.11 to evaluate the complex Fourier coefficient cx» we Bet
aed” aes *
Tl i
= Aso(2), = 0.1.22 en
‘To simplify notation in the foregoing and subsequent results, we vill
‘use the sinc function defined by
in(nd)
ma
sinc(2) (2.18)
where 2 is the independent variable. The sinc function plays an important
role in communication theory. As shown in Fig. 2.2, it has its maximum
Value of unity at 1 = 0, and approaches zero as 4 approaches infinity,
‘oscillating through positive and negative values. It goes through zero 3!
J = £1, £2... , and so on. Thus, in terms of the sinc function we
may rewrite Eq. 2.17 as follows
w= Aan) Fe
tH.
Figure 21
Faget 2 rain of rectangular pulses of amplitude A, duration T, and period Te
wien senies 19
Figure 22
The sine function.
7 mm re
jae
a eee sO aa.
‘coi ieee
}f 1 T
| HALL |
intone peeps.
| Figure 23
Discrete spectrum of periodic train of rectangular pulses
{ol Armplitade spectrum. (2) Phoce spectrums 20 FOURIER ANALYSIS
where m has discrete values only. In Fig. 2.3 we have plotted the amplitude
spectrum [e,| and phase spectrum arg(c,) versus the discrete trequene,
1n/Ty for a duty cycle T/Ty equal to 0.2. We see that
1. The line spacing in the amplitude spectrum in Fig. 2.3a is determine
by the period 7.
2, The envelope of the amplitude spectrum is determined by the puise
amplitude A and duration 7.
3, Zero-crossings occur in the envelope of the amplitude spectrum at fe
quencies that are multiples of 1/T.
4. The phase spectrum takes on the values 0° and + 180°, depending on
the polarity of sinc(n7/T,); in Fig. 2.3b we have used both 180? and
~ 180° to preserve asymmetry.
EXERCISE 1 Plot the amplitude spectra of rectangular pulses of unit am-
plitude and the following two values of duty cycle:
2 Z-o1
fe
» aoa
swwenu 2:2 FOURIER TRANSFORM
In the previous sections we used the Fourier series to represent a periodic
signal. We now wish to develop a similar representation for a signal g(!)
that is nonperiodic, the representation being in terms of exponential time
functions. In order to do this, we first construct a periodic function &,(0)
of period 7; in such a way that g(t) defines one cycle of this periodic
function, as illustrated in Fig. 2.4. In the limit we let the period 7, become
infinitely large, so that we may write
8(0) = tim g,(0) (2.20)
Representing the periodic function g,(t) in terms of the complex ¢*
ponential form of the Fourier series, we have
a= So eap( Et)
FouniER TRANSFORM «29
Figure 24
The construction of» periodic function trom an arbitrary defined function of time
were
wi
a4)” soa
Detne
1
a
and
Gf) =
Thus, making this change of notation in the Fourier series representation
of g,(), given by Eqs. 2.21 and 2.22, we get the following relations for
the interval —(Ty/2) < ¢ < (T)/2),
8) = LD GUS.) expli2xfat) af 2.23)
where 22 FOURIER ANALYSIS
approach infinity of, equivalently, iy
Suppose we now let the period T>
a ee at, in the limit, the discrete
reciprocal 4f approach zero. Then we find th
frequency f, approaches the continuous frequency variable f, and th
discrete sum in Eq. 2.23 becomes an integral defining the arca under ,
continuous function of frequency f, namely, G(f) exp(j2nfr). A
T, approaches infinity, the function g,(t) approaches g(*). Therefore, in
the limit, Eqs. 2.23 and 2.24 become, respectively.
a) - GOS) expli2aft) df 2.25)
where
Gi) = a ‘g(0) exp(—j2nft) de (2.26)
We have thus achieved our aim of representing an arbitrarily defined signal
‘g(t) in terms of exponential time functions over the entire time interval
from == to =. Note that in Eqs. 2.25 and 2.26 we have used a lowercase
letter to denote the time function and an uppercase letter to denote the
corresponding frequency function.
Equation 2.26 states that, given a time function g(t). we can determine
a new function G(f) of the frequency variable f. Equation 2.25 states
that, given this new or transformed function G(f), we can recover the
original time function g(t). Thus, since from g(t) we can define the function
G(f) and from G(f) we can reconstruct g(t), the time function is also
specified by G(f). The function G(f) can be thought of as a transformed
version of g(¢) and is referred to as the Fourier transform ot g(t). The time
function g(t) is similarly referred to as the inverse Fourier transform of
G(f). The functions g(i) and G(f) are said to constitute a Fourier trans-
‘form pair.
/|uaitibeiaies
For asignal g(t) to be Fourier transformable
Dirichlet's conditions
issuficient that g() satisfies
1, The function g(¢) is single-valued, with a finite number of maxima and
minima and a finite number of discontinuities in any finite time interval.
2. The function g(¢) is absolutely integrable, that i,
[lela <=
The Dirichlet conditions are not strictly necessary but sufficient for the
Fourier transformability of a signal, These conditions include all enerBY
FOURIER TRANSFORM 23.
signals, for which we have
assumed to be
In the two conditions described herein, the signal g(t) is
complex,
NOTATIONS
‘The formulas for the Fourier transform and the inverse Fou
presented in Eqs. 2.25 and 2.26 are written in terms of time ¢
f, with ¢ measured in seconds (s) and f measured in her
frequency f is related to the angular frequency w aso
measured in radians per second (rad/s). We may simplify the expressions
for the exponents in the integrands of Eqs. 2.25 and 2.26 by using « inste
of f. However, the use of fis preferred over « for two reasons. First, we
have the mathematical symmetry of Eqs. 2.25 and 2.26 with respect to each
other. Second, the frequency contents of communication signals (i.¢
speech and video signals) are usually expressed in hertz
‘A convenient shorthand notation for the transforar relations of Eqs
2.26 and 2.25 is
er transform
id frequency
(Hz). The
2nf, which is
Fle] (2.278)
1G) (2.276)
Gf)
8
‘Another convenient shorthand notation for the Fourier transform pair,
represented by g(*) and G(f), is
si = Gi). (228)
“The shorthand notations described herein are used in the text where ap-
propriate.
‘SPECTRUM “
By using Fourier transformation, an energy signal g(¢) is represented by
the Fourier transform G(f), which is @ function of the frequency variable
sit the function gt is such that the value of f=. |i
the Fourier transform Gif) ofthe function g(t) exis
tm [tot [10 eta a] 0
‘This result is known as Plancheral’s theorem,
ccc EE —————
24 FOURIER ANALYSIS
the spectrum of the signal g(0). The SI oe
© hat tis defined for al frequencies. In general, the Fourie! Wansiovm Gif)
; jaa compler function ofthe frequency f. We may therefore expres itn
| the form
i GUS) = |G(D| expN]
2.29
where |G({)] is called the amplitude spectrum of g(t), and Of) is called
the phase spectrum of (0)
Hor the special case of a real-valued function g(#), we have
G(f) = G(-f)
; “Therefore, it follows that if g(t) is a real-valued function of time ¢, then
Ia(- Nl = 1G (2.30)
and
@-f) = - HF) (231)
‘Accordingly, we may make the following statements on the spectrum of
; real-valued signal:
_/ 1. The amplitude spectrum of the signal is an even function of the fre
quency; that is, the amplitude spectrum is symmetric about the vertical
axis
/ 2. The phase spectrum of the signal is an odd function of the frequency’
that is, the phase spectrum is antisymmetric about the vertical axis.
These two statements are often summed up by saying that the spectrum
of a real-valued signal exhibits conjugate symmetry.
==
| EXAMPLE 2 RECTANGULAR PULSE |
Consider a rectangular pulse of duration T and amplitude A, as shown i? |
Fig. 2.5. To define this pulse mathematically in a convenient form, we Y*
the following notation :
“ia
3?)
reet(() = fe
rourrer Taansronm | 20
Figure 25 i
Rectangular pulse j
which stands for a rectangular function of unit amplitude and unit durati
centered at = 0. Then, in terms of this function, we may express the
rectangular pulse of Fig. 2.5 simply as follows
)
cpanel
‘The Fourier transform of this rectangular pulse is given by
Gp) = [7 A exp(—i28F0 at
= AT sinc( fT)
We thus have the Fourier transform pair
area(!) = arses
“The amplitude spectrum |G(J)| of the rectangular pulse g(t) is shown
plotted in Fig. 2.6a. From this spectrum, we may make the following
observations
1. The amplitude spectrum has a main lobe of total width 2/T, centered
on the origin
| _ 26 FOURIER ANALYSIS.
\onl
at
t
Te ee
Fe st
o
Figure 2.6
‘Spectrum of rectangular pulse. (a) Amplitude spectrum. (b) Phase spectrum.
2. The side lobes, on either side of the main lobe, decrease in amplitude
with increasing |f|. Indeed, the amplitudes ofthe side lobes are bounded
by the curve 1)
3. The zero crossings of the spectrum occur at f = #1/T, +2/T,
‘The phase spectrum 0() of the rectangular pulse g(t) is shown plotted
in Fig, 2.65. Depending on the sign of the sinc function sinc( fT), the phase
spectrum takes on the values 0° and +180° in an asymmetric fashion.
A truncated form of decaying exponential pulse is shown in Fig. 2.7a. We
may define this pulse mathematically in a convenient form by using the
unit step function:
1, 1>0
uO = 4h
0,
1=0
234)
<0
BANSFORM 27°
Figure 27
(al Decaying exponential pulse (b) Rising exponential pulse
‘We may then express the exponential pulse of Fig. 2.7a as
(0) = exp(—u(e) (2.38)
‘The Fourier transform of this pulse is
GH) = fF exe(—0) exe(—/2 ft) at
= f exp[—H(l + j2xf)] dt
gi sab
T+ paxf
(2.36)
‘Thus, combining Eqs. 2.35 and 2.36, we obtain the Fourier transform pair
1
exe(=)u() == > ea
Figure 2.8 shows the spectrum of the decaying exponential pulse
‘A truncated rising exponential pulse is shown in Fig. 2.70, which is
defined by
8(0) = exp(u(-1) 2.38)
Note that u(—#) is equal to unity for ¢ < 0, one-half at ¢ = 0, and zero 28 FOURIER ANALYSIS
for ¢ > 0. The Fourier transform of this pulse is given by
G(s) = [expt exp(—j2eft) at
= fi expt = jarpy) ae
1
Dxf
We thus have the Fourier transform pair:
i
exp(t)u(~s) iar
Figure 2.9 shows the spectrum of the rising exponential pulse.
3)
em)
rrore 15 FOURIER TRANSFORM. “2D
Figure 29
Spectrum of ising exponential pulse (a) Amplitude apectrum (b) Phase spectrum
Comparing the spectra of Figs. 2.8 and 2.9, we may make the following
two observations:
1, The decaying and rising exponentials of Fig. 2.7 have the same ampli
tude spectrum,
2. The phase spectrum of the rising exponential is the negative of that of
the decaying exponential
v
123 PROPERTIES OF THE FOURIER TRANSFORM
It is useful to have a feeling for the relationship between a function ¢(:)
and its Fourier transform G(f), and for the effect that various operations
‘on the function g(t) have on the transform G(f). This may be achieved
by examining certain properties of the Fourier transform. This section
describes 10 of these properties, which will be proved, one by one. These
properties are summarized in Table 1 of Appendix D. tat
re
30 FOURIER ANALYSIS
PROPERTY 1 LINEARITY (SUPERPOSITION)
Let git) = GF) and gilt) = Gf). Then for all constants a and b, wo hayy
agit) + bolt) = aGit) + BGT (2a)
“The proof of this property follows simply from the linearity of the jg
tegrals defining G(f) and g(\).
EXAMPLE 4 DOUBLE EXPONENTIAL PULSE
Consider a double exponential pulse defined by (see Fig. 2.10)
exp), <0
= exp(=lel) (2.42)
exp(-, 1 > 0
si) = fi Zo
This pulse may be viewed as the sum of a truncated decaying exponenti
pulse and a truncated rising exponential pulse. Therefore, using the lin
arity property and the Fourier-transform pairs of Eqs. 2.37 and 2.40, we
find that the Fourier transform of the double exponential pulse of Fig. 2.10
is as follows
1 1
GOVSTS fant * 18g
2
+ (Qnfy
‘We thus have the Fourier transform pair
on-— @ay
Figure 210
Double exponential pulse.
PROPERTIES OF THE FOURIER TRANSFORM 34
PROPERTY 2° TIME SCALING
Let git) = Git), Then,
lat) 2.44)
Where ais a time-scaling factor that may be positive or negative
To prove this property, we note that
Fle(at)] = |” g(at) exp(—j2nfr) de
Set r = at. There are two cases that can arise, depending on whether the
scaling factor a is positive or negative. If a > 0, we get
3/2 ex sax (2) «] ae
Flg(an))
On the other hand, if @ < 0, the limits of integration are interchanged so
that we have the multiplying factor ~(1/a) or, equivalently, 1/\a.. This
completes the proof of Eq. 2.44
Note that the function g(at) represents (1) compressed in time by a
factor a, whereas the function G(f/a) represents G(f) expanded in fre
quency by the same, factor a. Thus the scaling property states that the
compression of a function g(t) in the time domain is equivalent to the
expansion of its Fourier transform G(f) in the frequency domain by the
same factor, and vice versa,
EXAMPLE 5 RECTANGULAR PULSE (CONTINUED)
Example 2 evaluated the Fourier transform of a rectangular pulse
the result of the evaluation is given by the Fourier transform pair of
Eq. 2.33. For convenience of presentation, let the rectangular pulse
be normalized to have unit amplitude and unit duration. Then, putting
A= Land T = 1 in Eq. 2.33, we have
rect(t) ==
ef) 32 FOURIER ANALYSIS »
33
Hiense, applying the time-saling property to this Fourier transform pa,
ba) observations are confi
fon the lett side of F
Hass off : if
ct(at) = = sine changes on the amplit
\ me () the right Fig
; clearly show that
Figure 2.11 shows the rectangular pulse and its amplitude spectrum domain de
; three different values of the time-scaling factor a, namely a = 1/2, 1,7 a
t With a = 1 regarded as the frame of reference, we may view the seo
; =12 fon in time, and a = 2 as compression in timo, These
| aE wxencise 2 Example 3 showed that the decaying exp
| ‘ thing expooental ple of ig: have the sume sop
opposa pare meer, Use hs tanec papery of ha agar a
GI pe
m= (em Sr co spite a bt
PROPERTY 3 DUALITY
irate
GIN), then
Gin = gt-r 248
j This property follows from the felation defining the inverse Fourier
transform by writing it in the form
1-0 Gthen(=stxin a
and then interchanging ¢ and f. Note that G(#) is obtaine
Using ran place of f- and g(~f) is obtained from g(*) by using ~ fin place
oft
EXAMPLE 6 SINC PULSE
at Consider a signal g() in the form of a sine function, as shown by
10 ton tthe ty
os :
To evaluate the Fourier transform ofthis function, we apply the duality
and time-scaling properties tothe Fourier transform pair of q 2.33. Then
aoe | Tevognizing that the rectangular function isan even function, we obtaa
Pong z - the following result:
Parameter «2
Frese relation between ime and AsincQwe) = A reer{ 4) e
Fie inverse relation frequency domain descriptions of aw *(3y
rectangular pulse git) = recta,
Gene nll 34° FOURIER ANALYSIS
ow
3
Figure 2:12
{ay Sine pulse gi. (b) Fourie transform Gif)
which is illustrated in Fig. 2.12. We thus see that the Fourier transform o
| {sine pulse is zero for fj > W. Note also that the sine pulse itself ison!
asymptotically limited in time.
spcancis#s Show that the total area under the curve of the sinc function
‘equals one; that is
[i sinc de = 1
EXERCISE 4 Consider a one-sided frequency function G(f), defined by
em-f. 1>0
on = 15, fo
0, f<0
Applying the duality property to the Fourier transform pair of Eq. 2.40,
write the inverse Fourier transform of G(f).
PROPERTY 4 TIME SHIFTING
Gif), then for # constant time shift t,
it
Ut ~ 1) == Git expt - anf 248
om 38
th
positive dic a
its Fourier tra
the amplitu ‘
by the amou
EXAMPLE 7 RECTANGULAR PULSE (CONTINUED)
Consider the rectangular pulse g. 2.130, wh
f= O and terminates at ¢ = T. This pulse is define
This pulse is obtained by shifting the rectangular pulse of Fig. 2.5 t0 the
right by 7/2 seconds. Therefore, applying the time shifting prope
Fourier transform pair of Eq, 2.35, we find that the Fourier
GC) of the rectangular pulse ¢,(r) defined in Eq. 2.49 is give
Gf) = AT sinc( fT) exp( if) 29
Figure 2.13,
Trime-shited versions of # rectangular pulse '
BO FOURIER ANALYSIS.
Consider next the rectangular pulse gy(t) of Fig. 2.136, which starts
T and terminates at ¢ = 0. This second pulse is defined
atin (22)
ular pulse of Fig. 2.5
“The pulse g,(1) is obtained by shifting the rectangular p 2.2
the left by 7/2 seconds. Therefore, applying the time-shifting property
the Fourier transform pair of Eq, 2.33, we find that the Fourier transform
‘G\(f) of the rectangular pulse g4(*) defined in Eq. 2.51 is given by
Gif) = AT sine(fT) exp(jxfT) ¢
PROPERTY 8 FREQUENCY SHIFTING
H git) = Gif), then for a constant frequency shift f.,
explj2nf.t) git) == GIF ~ f) (253)
‘This property follows from the fact that
Flexp(/2xf.t)e(0] = [”_s(0) expl-i2ai(f ~ f.] dt
= GF - fd
That is, multiplication of a function g(t) by the factor exp(j2xf.1) is equiv
alent to shifting its Fourier transform Gi) in the positive direction by the
amount f,..Note the duality between the time-shifting and frequency-shift-
ing operations.
‘Consider the radio frequency (RF) pulse signal g(t) shown in Fig. 2.142,
hich consists ofa sinusoidal wave of amplitude’ and frequency {,. The
pulse occupies the interval from ¢ = ~T/2 to = T/2. This signal is |
referred to as an RF pulse when the frequency f,fallsin the radio-frequency
band. Such a pulse is commonly used in radar forthe detection of toree’>
searing aircraft) and for the estimation of useful target parameters
The signal (0) of Fig. 2.14a may be expressed mathematically as follows
2 ‘
so" wea( 2) cos(2nf.t) as)
ee
en
i
rou 37
a Dan
Figure 214
(a RF puis (0) Amplitude spectrum
To find the Fourier transform of this signal, we note tha
= Herpt22F.)
Therefore, applying the trequency-shifting proper
form pair of Eq. 2.33, we get the desire
ar
Gif) = (sinc TY ~ f)) + wal TU so SD ste
38 FOURIER ANALYSIS
pulse is large, that is,
When the number of cycles within the pul rc
‘may use the approximate result
AT sin Tf = fo > °
oy)
AT sn Uf + fh <9
rum of the RF pulse is shown in Fig. 2.14b, This
Fee rie aon to Fig. 2.64, clearly illustrates the frequency shitting
diagram, in relation 1
property of the Fourier transform
[EXERCISES Consider an exponentially damped sinusoidal wave defined
by
10
aye {reco ene (2° as)
Using the expansion
Sif.) = 5 fexpL/2nft) ~ exp(~/2rf.)]
‘and applying the frequency-shifting property to the Fourier transform pair
of Eq. 2.37, write the Fourier transform of g(:).
PROPERTY 6 DIFFERENTIATION IN THE TIME DOMAIN
Let git) = Gif}, and assume that the first derivative of glt) is Fourier trans
formable. Then
a
Ge 9l!) — j2ntGin) (2.58)
That's, aliferertition of ime function gt) has the efecto ing it
Fourier transform Gif) by the factor anf peaeaecene vino
This result is obtained simply by taking the first der
first derivative of both sides
of the relation defining the inverse Fourier tr of Gi x
Eq. 2.25, and then interchanging the operation of reine an
entiation; we are justified to make this interchange te 0
and diferentiation ate both near operations ne? DESHI nterstion
Multiplication of the Fourier transform G(j) by
the
tor j2nf on the
PROPERTIES OF THE FOURIER TRANSFORM 39
Fight side of Eq. 2.58 implies that differentiation of g(t) with respect to
time enhances the high frequency components of the signal g(t)
may generalize Eq, 2.58 as follow
© et) = (2fO) (259)
a
EXAMPLE 9. GAUSSIAN PULSE
In this example we wish
the ps
> use the differentiation property of the Fourier
transform to det .¢ signal g(t) whose Fourier transform G(f)
has the same form
Let g(0) denote the pulse as a function of time, and G(f) its Fourier
transform. We note that by differentiating the formula for the Fourier
transform Gif) with respect to f we have
Paige) =
4
6 (2)
ap oO)
which expresses the effect of differentiation in the frequency domain. Equa
tion 2.60 is the dual of Eq. 2.58 that describes the time-differentiation
property. Dividing both sides of Eq. 2.60 by j. we may also write
reign =! Lay 61
rin = 14 cy)
Suppose that the pulse-signal g(r) satisfies the First-order differentia equa
tion
2nrgte)
4 ew
dna:
{The imposition of this condition on the pulse signal g(t) is equivalent
‘equating the leftchand members of Eqs. 2.58 and 2.61. Correspondingly
We may equate the right-hand members of Eqs. 2.S8 and 2.61. and thus
write
ld 2 f
jap OW) = RAG 2.63)
Since j? = —1, we may rewrite Eq. 2.63 as
4 Gi) = -25 61 2
FON = —2FG) 264)
ts 4
t
40 FOURIER ANALYSIS
Figure 218
Gaussian pulse.
at if a pulse signal g(*) satisfies the first-order difer-
Meet enon (2.6), then is Four transform G(J) mus satis the
first-order differential equation (2.68). However, these two differential
equations have exactly the same mathematical form. Hence, the pulse
signal and its transform are the same function. In other words, provided
that the pulse signal g(°) satisfies the differential equation (2.62), then
G(f) = a({). Solving Eq, 2.62 for g(¢), we obtain
8(0) = exp(-nr) 265)
This result is shown plotted in Fig. 2.15.
‘The pulse defined by Eq, 2.65 is called a Gaussian pulse, the name being
derived from the similarity of the function to the Gaussian probability
density function. We conclude therefore that the Gaussian pulse is its own
Fourier transform as shown by
exp(= nt) = exp(—nf?) (2.66)
EXERCISE ® Show that
[i ep(-2) a= 1 Qs
Hint: Consider the formula for the Fourier transform of exp( 20) eva
uated at f = 0.
pansrowm 49
PROPERTY 7 INTEGRATION IN THE TIME DOMAIN
transform GU) by the fact
To prove this pr
signal as follows
Al ear] (" exp(-iaas [' lepdee am
On the right side of this relation, we have a
to the variable ¢. Clearly, we may view the corresponding integtan
Product of two time functions: the exponential exp(—/2ft) and th
tegrated signal f'.g(r) dr. Hence, using the formula for integration by
parts and assuming that
GO) a a(t) de = 0
‘and then simplifying the result, we get the relation of Eq. 2.68
dition G(0) = 0 ensures that g() integrates out to zero as + approwches
infinity. The more general case, for which G(0) # 0. is treated
Section 2.5,
Division of the Fourier transform G(f) by the factor /22f on the right
side of Eq. 2.68 implies that integration of g(¢) with respect 10 time »
Dresses the high-frequency components of g(). As expected, this effect
the opposite of that produced by differentiation of g(¢)
EXAMPLE 10 TRIANGULAR PULSE
Consider the doublet pulse g\(t) shown in Fig. 2.16a. By integrating this
pulse with respect to time, we obtain the triangular pulse g.(1) shown in
Fig. 2.160. The duration of this triangular pulse at the half-amplitude pounts
is the same as the duration of the rectangular pulse of Fig 25. We note
that the doublet pulse g\(‘) consists of two rectangular pulses: one of
amplitude A, defined for the interval ~T < ¢ < 0, and the other of am-
plitude ~A, defined for the interval 0 <¢-< T. Therefore, using the results 42 FOURIER ANALYSIS
Figure 2.16
Figs 216 pulse git (0) Trenguiar pulse a9 obtained by integrating o
cof Example 7, we find that the Fourier transform G,(f) of the doublet
pulse g;(t) of Fig. 2.16a is given by
AT sinc(7) [exp (jn fT) ~ exp(—jnfT)]
2jAT sinc fT) sin(afT)
GAA)
(2.70)
We further note that G,(0) is zero. Hence, using Eqs. 2,68 and 2.70, ve
find that the Fourier transform ©,1) of the triangular pulse g() of Fit
2.166 is given by
Te
pap
= are sinc({T)
= AT sinc({7)
GAP)
an
PROPERTIES OF THE FOURIER rou 43
Figure 2.17
Spectrum of triangular pulse
‘The Fourier transform G,(f) is a positive real function of f, which means
that the amplitude spectrum of g(t) is the same as G,(f), and its phase
spectrum is zero for all f. The Fourier transform G,(f) is plotted in Fig.
2.17. Note that the spectrum of the triangular pulse is more tightly centered
‘around the origin.than the spectrum of the rectangular pulse. Also, the
Spectrum of the triangular pulse decreases as 1/f", whereas the spectrum
of the rectangular pulse is discontinuous and decreases as 1/\f
‘a. Show that the Fourier transform of a triangular pulse of unit amplitude
‘and unit duration (measured at the half-amplitude points) is equal to
sinc),
bb. Using the result in part a, show that
ff sae ay =
Hint. For part b, consider the formala for the inverse Fourier transform
‘of sinc*(f) evaluated at ¢ = 0.
PROPERTY 8 CONJUGATE FUNCTIONS
I git) = Gif), then for a complex-valued tin
or = GIN)
where the asterisk denotes the complex conjugate operation
1e function g(t) we have
(27) 44° FOURIER ANALYSIS
“To prove this property, we know from the inverse Fourier transform
that
£00 = [1 oO exo
‘Taking the complex conjugates of both sides:
sey = [or exons a
Neat, replacing f with ~
sr - ‘ Gt(=f) expli2aft) af
[eC emtiasso af
“That is, (0) isthe inverse Fourier transform of G*(—), whichis th
desired result.
EXAMPLE 11. REAL AND IMAGINARY PARTS OF A TIME FUNCTION
Expressing a complex-valued function g(#) in terms of its real and imaginar
parts, we may write
a(t) = Relg(s)] + j Imlg(s)] en
where Re denotes the “real part of” and Im denotes the ‘“imaginary par
of.” The complex conjugate of g(t) is
8°(8) = Relg(s)] ~ j Im[g(s)] cc)
‘Adding Eqs. 2.73 and 2.74:
Re[g(1)] = Sta + 8"(0) 75
Infg(O) = 5 la ~ g*001 on
PROPERTIES OF TH comm 48
Therefore, applying Property 8, we obtain rane
form pair
Relate Gi) + G"
Imig(e GU) - Gt 5
From Eq. 2.78, it is apparent that in the case of a real-val a
function g(1), we have Gif) = G*(—f); that is, G(f) exhibits conjugate
symmetry. This result is a restatement 2'30 and 2.31
EXERCISE 8 Show that for a real-valued signal g(t), Eq. 2.72 may be
rewritten in the equivalent form:
a(-) = Gf)
PROPERTY 9 MULTIPLICATION IN THE TIME DOMAIN
Let gt) = GIF) and gt) = Gif). Then
out) gt) GUAGE ~ a) 8 (2.73)
To prove this property, we first denote the Fourier transform of the
Product gi(t)g.(t) by Gu(f), so that we may write
gi(0) 80) = Gulf)
where
Gf) =
f
For g,(t), we next substitute the inverse Fourier transform
gt)8.(0) exp =F
a0 = f Gif’) expli2xf't) df’
in the integral defining Gx(f) to obtain
* gu(OGF) expl—i2n(f - f')e) df de 46 FOURIER ANALYSIS
Define A = f-~ f'. Then, interchanging the order of integration, we obtain
Guin = [a out 4) [20 en(-Pein
‘The inner integral is recognized simply as Gi(2), $0 we May write
ou = [aaa - ae
which is the desired result. This integral is known as the convolution integral
Expressed in the frequency domain, and the function Gi,(f) is referred to
STihe convolution of G,(f) and G,(f). We conclude that the multiplication
of two signals in the time domain is transformed into the convolution of
their individual Fourier transforms in the frequency domain. This propery
is known as the multiplication theorem.
Tn 2 discussion of convolution, the following shorthand notation is fe
} quently used:
if) He Gf)
where the star 4 denotes convolution, Note that convolution is com:
Gf)
Gulf)
Gf) GAN) = GAN) H GUN)
EXAMPLE 12 TRUNCATED SINC PULSE
Consider the truncation of the sine pulse sine(2W1), so that the resulting
sgn a() i 20 ‘outside the interval =(T/2) < 1 (7/2), as shown in
ig. 2.18a. This signal may be expressed as the ‘i
4 rectangular pulse, as shown by ‘Product of a sinc pulse and
(0) = sinc(2Win reat:
(2.80)
PROPERTIES OF THK F
awa
Gif). -
From Eqs. 2.33 and 2.47, we have
#[rea(7) | = Tsine(fT)
ries
F[sine(2W1)] = wes) )
‘Therefore, using Eq. 2.79, we find that the Fourier transform of the trun:
cated sine pulse g(i) is given by
Ges) = sy Jay) self ~ OT a
) awl sm - aT) aa
rs sin[x(f ~ 4)
~ gar O es 48 FOURIER ANALYSIS
Figure 2.19
‘The sine integral.
“The integral of the function sinx/x from zero up to some upper limit
called the sine integral, which is defined as follows
if ax 28)
Si(w)
‘The sine integral Si(u) cannot be integrated in closed form in terms of
‘elementary functions, but it can be integrated as a power series." It
plotted in Fig. 2.19. We see that: (1) the sine integral Si(u) is odd symmetr:
about u = 0; (2) it has its maxima and minima at multiples of x; and (3)
it approaches the limiting value 1/2 for large values of u.
Substituting x = x(f - A)T in Eq, 2.81, we find that the Fourie:
transform G({) ofthe truncated sine pulse may be expressed convenie%!)
in terms of the sine integral as follows:
GU) = AplSilawT - xfT) + SiaWT + AFT]! C8)
‘This relation is plotted in Fig. 2.18b for the case when T’ = 8/W. West
that G(f) approximates the Fourier transform of a sine pulse sinc(2¥!
of infinite duration in an oscillatory fashion, with a maximum deviatio®
about 9%. Furthermore, for a given value of W, as the pulse duratie? !
“See Goldman (1948, pp. 76-79),
— i 8
Proper 49
is increased, the ripples in th
angular function show
the frequency, f, wher
the discontinuity remain
phenomenon in Fou
fexanciee 8 Using Eq, 2°9, show tha
forma =f cnone
How is the left side of this relation affected bs
5 is relation affected by replacing Gf) with
GA) in the integral on the right side of the relation?
PROPERTY 10 CONVOLUTION IN THE TIME DOMAIN
Let g(t) = Gif) and galt) = Gilt). Then
[/ olngdt ~ 2) = Gi r16, 1) 2.04
‘This result follows directly by combining Property 3 (dualty) and Prop.
ety 9 (time-domain multiplication). We may thus state that she convolution
of two signals in the time domain is transformed into 0
their individual Fourier transforms in the frequency domain. This property
is known as the convolution theorem. Its use permits us to exchange
convolution operation for a transform multiplication, an operation that is
ordinarily easier to manipulate
Using the shorthand notation for convolution, we may rewrite Eq. 2.84
in the form
1¢ multiplication
GAGES) ess)
a) 4 (0)
where the star $- denotes convolution
EXAMPLE 13. DERIVATIVE OF A CONVOLUTION INTEGRAL
F Let gu(t) denote the result of convolving two signals g(t) and g(t). Then
the del ative of gx sequal to the convolution of g(t) with the derivative
of g()_ or vice veosa. That if
gold) = 8:0) $F a) 50 FOURIER ANALYSIS
then
4 a0 = [geo] ed
roperty (i.€-» Ea: 2-58)
To prove this result, we use the diferentition PPE se aning,
in conpancton th the eonveation Proper (E>
tg dr ao) = RAMGMNGAD!
“Assoriating the factor j2nf with Gf), we may write
[Zeca] ae exo = asounieh
which yes he dei rest
Zino denon [feo] 4 ao
Equation 2.86 shows thatthe derivative ofthe conoton of two ine
raetee egal to te convolution of one function wih the deiative
of the other.
“pxEncise 10 Using Ea. 2.84, show that
‘ Seon oa- f CMNCM aF
How is the right side of this relation affected by replacing g,(—t) wit
_gi(0) in the integral on the left side of the relation? How does this result
with that of Exercise 9?
col, 244 INTERPLAY BETWEEN TIME-DOMAIN AND
(DESCRIPTIONS
FREQUENCY-DOMAIN
“The properties of the Fourier transform and the various examples used to
iilustrate them clearly show that the time-domain and frequency-domain
descriptions of a signal are inversely related. In particular, we may make
the following statements
1. If the time-domain description ofa signal is changed, the frequeney-
on of the igha is chnged in an inverse maroc, nod
TERPLA 5t
a signal in both domains. 1 w
2. If a signal is strict
of the signal will tral
assume a progressi
in frequency of stricily ba F
zero outside a finite ban
of a strictly band-limited signal
also shows that the sine pulse ™ m
which confirms the opening statement we made for a
limited signal. In an inverse manner, if signal is strictly limite
(ie, the signal is exactly zero outside a finite time interval).
specirum of the signal is infinite in extent
even though the amplitu
spectrum may assume a progressively smaller value. This behavior
exemplified by both the rectangular pulse (described im Figs. 2.5 and
2.6) and the triangular pulse (described in Figs. 2.165 and 2.17). Ac
cordingly, we may state that a signal cannot be strictly limited in both
time and frequency
sanoworn /
‘The bandwidth of a signal provides a measure of the extent of significant
spectral content of the signal for positive frequencies. When the signal is
strictly band-limited, the bandwidth is well defined. For example, the sine
pulse described in Fig. 2.12 has a bandwidth equal to W. When, however
the signal is not strictly band-limited, which is generally the case, we en
counter difficulty in defining the bandwidth of the signal. The difficulty
arises because the meaning of “significant” attached to the spectral content
(of the signal is mathematically imprevise. Consequently, there is 90 uni
versally accepted definition of bandwidth
Nevertheless, there are some commonly used definitions for bandwidth,
In this section, we consider two such definitions;* the formulation of each
definition depends on whether the signal is low-pass or band-pass. A signal
is said to be low-pass if its significant spectral content is centered around
the origin. A signal is said to be band-pass if its significant spectral content
is centered around f,, where f. 1s a nonzero frequency
‘When the spectrum of a signal is symmetric with a main lobe bounded
by well-defined nulls (ie., frequencies at which the spectrum is zero), we
may use the main lobe as the basis for defining the bandwidth ofthe signal
Specifically, ifthe signal is low-pass, the bandwidth is defined as one half
‘another definition for the bandwidth of a signal s presented in Section 48 ne half of this lobe
a rectangular pulse
jal width 2/T hertz
tral lobe, since only 0”
on. For example
tral lobe of {
tia Accordingly. we may define
the total width of the main spec
Ties inside the positive frequency resi
has a main spec
of duration T seconds
picted in Fig
centered at the origin, as dep
the bandwidth of this rectangular pulse 3s
+e pandvidth is called the ul0-nu
aendwiin, Lorexample, as RF pulse of uration Tsecondseaultregyency
fr hes main spectral Tebes of wth 2/7 hertz cewtered] aroun & J. os
Gepicted in Fig. 2.140, Hence, we may define the null4e-mall bandwith
of this RF pulse as 2/T hertz :
‘Another popular definition of bandwidth is the 3-dB bandwidth’ Spe
cifeally. ifthe signal is low-pass the 3-dB bandwidth is defined as the
Separation between zero frequency, where the amplitude spectrum attains
ita peak vale, and the positive frequency at which the amplitude spectrur
drops to 1/V2 of its peak value. For example, the decaying exponential
nd rising exponential pulses defined in Fig. 2.7 have a 3B bandwith
Of1/2e here Ion the other hand, the signal s band pass centered at
Sf. the 3-dBs bandwidth is defined 2s the separation (along the postive
frequency axis) between the two frequencies a which the amplitude spec-
trum ofthe signal drops to 1/V2 ofthe peak value at f, The 3-dB band-
swith has the advantage in that it can be read directly from a plot of the
Amplitude spectrum, However, it has the disadvantage in that it may be
iisleading ifthe amplitde spectcum has slowly decreasing tails
1D rhertz If,on the other hand,
f. is large, the bandwidth is de
EXERCISE 11 Using the idea of « main spectral lobe, what is the band-
‘width of a triangular pulse defined in Figs. 2.166 and 2.17?
EXERCISE 12 What is the 3-4B bandwidth of the decayis i
Jute cep) that rt regu aying exponential
rme-eanowoTH prooucr —/
For any family of pulse signals that differ by
- y a time-scaling factor, the
product of the signal's duration and its bandwi pecan
pred he and its bandwidth is always a constant, as
(duration) - (bandwidth) = constant
for 9 aiscussion ofthe decibel 28),
Appendix A
. @
sue 2.8 DIRAC DELTA FUNCTION
8 AF 53
The product is called the time-bandwidth product ot bandwidth-duration
product. The constancy of the time-bandw si m
festation of the inverse relationship that ex wee
ationship thal a
and frequency-domain descriptions ofa signal. In pa dura
of a pulse signal is decreased by reducing the time seal a, the
frequency scale of the signal's spectrum, and therefore the bandwidth
the signal, is increased by the same factor a, by virtue
the time-bandwidth product of the signal is thereby maintain
For example, a rectangular pulse of duration T seconds has a bandw
(defined on the basis of the positive-frequency part of the main lobe) equal
to 1/T hertz, making the time-bandwidth product of the pulse equal unity
‘Whatever definition we use for the bandwidth of a signal, the time-band:
width product remains constant over certain classes of pulse signals. The
choice of a particular definition for bandwidth simply changes the value of
the conan
ve
Strictly speaking, the theory of the Fourier transform, as described in
Sections 2.2 and 2.3, is applicable only to time functions that satisfy the
Dirichlet conditions, Such functions include energy signals. However, it
‘would be highly desirable to extend this theory in two ways:
1. To combine the Fourier series and Fourier transform into a unified
theory, s0 that the Fourier series may be treated as a special case of
the Fourier transform.
2. To include power signals in the list of signals to which we may apply
the Fourier transform.
It turns out that both these objectives can be met through the “proper
use” of the Dirac delta function or unit impulse.
"The Dirac delta function belongs to a special class of functions known
‘as generalized distributions that are defined by the use of assignment rules
given in Eqs. 2.87 and 2.88. In particelar, the Dirac delta function,” denoted
by 4(0), is defined as having zero amplitude everywhere except at = 0,
where itis infinitely large in such a way that it contains unit area under its
‘curve, as shown by the pair of rules:
3)
10 sn
>For a detailed treatment ofthe delta function, see Bracewell (1978) or Lighthih
(1950), The notation 47), which was first introduced into quantum mechanics by
Direc is now in general use; see Dirac (1947), 84 FOURIER ANALYSIS
. and
ai) ar = (2.88)
n in the ordinary sense can
ortant to realize that no function
tay te wo Hod 288, However, we can imagine a
ue eer) point, eXcEPL at f= 0 °
ine of fn ew he dea faction he
ete ens ep tee ration approche 2X
timing form of reshape sed Fr example, WE may use c
Setar pute of nite, and ths Wie
449 = tea’) ew) |
“The rectangular pulse is plowed in Fig. 2.20e for « = 5,1,0.2. For another
‘example, we may use a Gaussian pulse of unit area and thus write
(9 = tim on(- x) 2.90)
“The Gaussian pulse is plotted in Fig. 2.206 for r = 5, 1, 0.2. From Fig
2.20, we clearly see that both pulses take on an impulse-like appearance
‘5 the parameter + becomes progressively smaller. Some other examples
‘are considered in Problem 18,
EXERCISE 13 Plot the spectra for the rectan; ssian pulses
forte tert uedlpanaere gee, a
PROPERTIES OF THE DELTA FUNCTION
Tre det function) aera wll ropes that are consequence
re discussed heres Nel Eas. 2.87 and 2.88. These properties Ets
we tt hc ;
40) = 1) =) d= 1a exp et for varying
EE a BBE rounten avatvsts
ime function (1) that is
2. The integral ofthe product of d(t) and any time f a
feontinuous at ¢ = 0 is equal to g{0); thus
{7g s@ dt = 8) (2.92)
ry of the delta function
We refer to this statement as the sifting prope" hala.
Since the operation on g() indicated on the let side of Ex, 2.92 ss ou
fa single value of g(’), namely, g(0). Equation 2.92 may also a
the defining rule for a delta function.
3. The sifting property ofthe delta function may be generalized by writing
fe g(t) 6(¢ ~ &) dt = gt) 12.93)
Since is i ay rewrite Ee
the delta function 6(¢) is an even function of t, we may rewrite Eq
2.93 in a way emphasizing resemblance to the convolution integral, as
follows:
ffs) 8 94e = 800 (2.94)
8 tr 5) = 8) 2.95)
‘That is, the convolution of any function with the delta function leaves that
function unchanged. We refer to this statement asthe replication property
‘of the deta function
4. The Fourier transform of the delta function is given by
FIo(0} = [400 expt j2xft a
‘Using the sifting property of the delta function and noting that the expo-
‘ential function exp( —j2zft) is equal to unity att = 0, we obtain
Fl] = 1
‘We thus have the Fourier transform pair:
ay =1 2.96)
‘This relation states that the spectrum of the delta function 6(1) extends
uniformly over the entire frequency interval from —c to =, as shown if
Fig. 2.21. Y
APPLICATIONS OF THE DELTA FUNCTION
4c Signal By applying the duality property to the Fourier tra ar
Sart aca en porn te Fer ano it
87
Figure 221
(2h Dirac deta function (b) Spectrum
obtain
1s an
Equation 2.97 states that a de signals transformed in the frequency domain
into a delta function 4(f) occurring at zeto frequency, as shown in Fig.
2.22. Of course, this result is intuitively satisfying. From Eq. 2.97 we also
deduce the useful relation
J exo(-i2aso ae = 3) 2.98)
where the integral on the left side is simply the Fourier transform of a
funetion equal to one for all time t.
Complex Exponential Function Next, by applviny the frequency-shifting
property to Eq. 2.97, we obtain the Fourier transform pair
expli2afe) == of — fo) em)
10
Figure 2.22
(al de signal. (0) Spectrum
-_ = —————————— ll
yy fo. Equati
function of frequen
for a complex exponential ustion
that the complex exponential function ¢xpO/24/2) 7
frequency domain into a delta function 3(f — Jo) 8
« problem ofevsluating the Fouric
7), We first note that
joidal Functions Consider next th
ne function cos(2
Sin
transform of the
oO)
fit) + exp(-i2
1
cos(2n fit) = 5 lex?
sTerefore, using Eq, 2.99, we find thatthe cosine funtion c0s(22J.1 is
represented by the Fourier transform pait Figure 2.24
(a} Sine function. (9) Spectrum
cos(2n ft) = 3 [5U — f.) + Uf + fo) (2.100) ‘i
In other words, the spectrum of the cosine function eas(2nJut) consisis of |
I pair of delta functions centered at f = 2f. each of which is weighted 4 pe
by the factor ¥2, as shown in Fig. 2.23. | sent) =} 0, r= 0 ae
Similarly, we may show thatthe sine function sin@2ft) is represented ais a8
by the Fourier transform pair
“The waveform of the signum function is shown in Fig. 2.25a. We may view
the signum function as the limiting form of a time function that consists
Sin(2n ft) = 5 (6 — f) - OF + fo) (2.101) ‘of a positive decaying exponential for positive time and a negative rising
exponential for negative time. That is, we write :
‘which is illustrated in Fig. 2.24.
| sen() = lim g(2,#) 2.103)
ou | ee
Pa en
: et) oe f iP . - 1
lgile ve
i oo
Figure 223 ” a
Pe Ease uncon (e Specrum.
Figure 225,
(a1 Signum function. (B) Spectrum.
_ ae, IO FOURIER Avanysis | ”
{exn(-#0 RY (2.104)
‘ [even <9 \
) = exp(—atyuti) ~ exp(ata(=) Gas) funi
aigs)= ° 03.7
is plotted in Fig
1 s the unit step function. The function g(a, ) is plostes t0 6 transform pa
Sener a' 1 05, oceaileaaaRE nee
parameter a is progressively reduced, the function g(@, 1) Doe O
Pere signa faction in appearance. plying the time-saling ProPes) sen(r) =
to the Fourier transform pairs of Eqs. 2.37 and 2.40, we 62
{0 Eq, 2.107. We this obtain the following You
e i = j seni f (2.108
tere the signum funtion sgn) is defined by
Poo
if =" 0. 9
fcencse 14. Plot the spectrum of the function g(a ) for parameter
= 1,05,001, and compare your results with the spect ofthe sigur
function shown in Fig, 2258
Un stp Raton The unt sp fiction, alti: deined ia Ba, 234
‘eprodiced bie Yor coeaiaa
Leo
wnt, 120 2.109)
The waveform ofthe nit tp function is shun in Fig, 2-71. From
228
The function gl, 0 for varying & 2,102 and 2.109, oF from the corresponding waveforms shown in Figs. 2.288
———— Bix. asscysis
Figure 227
{ai Unit step function, (b} Amplitude spectrum.
and 2,270, we see thatthe unit step funtion and Siem function are
related by
as 3 an a (uno)
> using linearity proy ‘of the Fourier transform and the Fourier
Hence gan 107. we ind tat he ui sep fonction
is represented by the Fourier transform pair
ig
ets an)
WO) = agg * 7°
“This means that the spectrum of the unit step function contains a delta
function weighted by a factor of 1/2 and occurring at zero frequency, as
shown in Fig. 2.27b-
Integration in the Time Domain (Revisited) The relation of Eq, 2.68 de
wirtpes the effect of integration on the Fourier transform of a signal g(t)
suming that G(0) is zero. We now consider the more general case, with
ro such assumption made.
Let
v= ff eter @ny
( The integrated signal
where the time-shifted unit step fun
Recognizing that convolution in the time domain is transformed into ru
tiplication in the frequency domain, and using the Fourier transform paic
of Eq. 2.111 for the unit step function u(t), we find that the Fourier
transform of y(1) is
—
ro-a0[z5-}00] enn
where G(f) is the Fourier transform of g(). Since
GU) 5) = GOO) 5H)
‘we may rewrite Eq. 2.115 in the equivalent form:
1
Ys
1
7 OU) + 5 GO) 30) 2.116)
‘That is, the effect of integrating the signal g() is described by the Fourier
transform pair:
Which is the desired result.
This proof is indirect in that it relies on knowledge of the Fourier trans-
{form of the unit step function. For a direct proof from first principles, refer
to Problem 20.
1 ta
8(2) de = BOW) + 3 GO 5D @u7)
mun 24 FOURIER TRANEFORME OF PERIOOIC HONALS ~
From Section 2.1 we recall that by using the Fourier series, a periodic
signal g, (2) can be represented as a sum of complex exponentials. Also we 64 FOURIER ANALYS!
know that, in a limiting sense, we can define Fourier transforms o
exponentials. Therefore, it seems reasonable that a periodic signal ca
represented in terms of a Fourier transform, provided that this transf
is permitted to include delta functions. «
Consider a periodic signal g,(¢) of period Tye We can represent ¢
terms of the complex exponential Fourier series as in Eq. 2.10, which
reproduced here for convenience
(jannt
BOE st i exo( 2 (2.118)
where c, is the complex Fourier coefficient defined by
es fe 4,(0) e0( = a) de (2.119)
Ty J-tya Ty
Let g(t) be a pulse-like function, which equals g,(¢) over one period
and is zero elsewhere; that is,
ree
go — oO" es (2.120)
0, elsewhere
‘The periodic signal g,(t) may be expressed in terms of the function g(¢)
as an infinite summation, as shown by
(0) = > g(t - mT) (2.121)
Based on this representation, we may view g(t) as a generating function,
which generates the periodic signal g,(t).
The function g(1) is Fourier transformable. Accordingly, we may rewrite
Eq. 2.119 as follows:
1 jaan
% ie 10) exo( - Te ) dt
-h Lo 2 4) (2.122)
where G(n/T,) is the Fourier transform of g(1), evaluated at the frequency
nlTy. We may thus rewrite Eq. 2.118 as
Cy =
1
84(0) 2 3 offen o(222%) 2.123) i
FOURIER TRANSFORMS OF PERIODIC SIGNALS 68
©r, equivalently,
e : 1 fn
e yo 2) 4)
Equation 2.124 is one form of Poisson's sum formula
Finally, using Eq. 2.99, which defines the Fourier transform of a complex
exponential function, and Eq. 2.124, we deduce the following Fourier
transform pair for a periodic signal g,(1) with a generating function ¢(/)
and period 7):
B(t) = Ss g(t — mh) = po o(2) 4(
‘This relation simply states that the Fourier transform of a periodic signal
consists of delta functions occui
ing at integer multiples of the fundamental
frequency 1/7), including the origin, and that each delta function is weighted
by a factor G(n/T,). Indeed, this relation merely provides an alternate
way of displaying the frequency content of a periodic signal g,(:).
Itis of interest to observe that the function g(1), constituting one period
of the periodic signal g,(¢), has a continuous spectrum defined by G(f)
On the other hand, the periodic signal g,(:) itself has a discrete spectrum
We conclude, therefore, that periodicity in the time domain has the effect
of making the spectrum of the signal take on a discrete form, where the
separation between adjacent spectral lines equals the reciprocal of the period.
EXAMPLE 14 IDEAL SAMPLING FUNCTION
An ideal sampling function, ot Dirac comb, consists of an infinite sequence
of uniformly spaced delta functions, as shown in Fig. 2.28a. We will denote
this waveform by
x(t) = Ss} d(t — mT) (2.126)
We observe that the generating function g(¢) for the ideal sampling function
6,,(1) consists simply of the delta function 3(t). Therefore, G(f) = 1, so
that
o(2) 1, foralln (2.127) a F
66 FOURIER ANALYSIS
Se - mty = 2.128)
Equation 2.128 states that the Fourier transform ofa periodic train of delta
functions in the time domain consists of another periodic train of delta
functions in the frequency domain as in Fig, 2.28b. In the special case of
the period T, equal to 1 second, a periodic tran of delta functions is, like
‘a Gaussian pulse, its own Fourier transfor
‘We also deduce from Poisson's sum formula, Eq. 2.124, the following
useful relation
Figure 228
{GY Ovrac come. (0) Spectrum
‘Thus the use of Eq. 2.125 yields the result
i
i
i
i
we cho
lin
all im:
T, seconds
T, seconds apart a
integer values, We refer to T. as
4, AT, as the sampling rate. This
Stantaneous sampling,
Let gy(t) denote the signal obtained by indiv
‘ments of a periodic sequence of delta funct
Figure 229
The sampling process. (a) Analog signa (0) Intantaneculy samgind verion ofthe
signal 2.431)
wsivere Gi) is the Fourier transform ofthe original signa g(t), and fois
Weanling rate, Equation 2.131 states thatthe process of uniformly
ruinous signal of finite energy results ina peridic spec
Taam with a period equal 0 the sampling ate
° 0
“Another useful expression for the Fourier | transform of the ideal sampled
‘signal g,(1) may be ‘obtained by taking the Fourier’ ‘transform of both sides
‘of Eq. 2.130 and noting that ‘the Fourier transform of the delta function
See ait) te equal to exp(j2anfT,). Let Gif) denote the Fourier
ftanaform of g,(0). We may therefore write
GAs) = E alot exo(-j2afT) eam)
This relation is calle the diseretesime Fourier transform. It may be viewed
se Somplex Fourier series representation of the period frequency ane.
2a Sic), with the sequence of samples (g(0,)} defning te coficets
of the expansion. :
ne SApations, as derived here, apply 10 any continuous-time signal g(?)
‘of arte energy and infinite duration, Suppose, however, that the senal
ot iy band mited, with no frequency components hghet than W Rene
‘That is, the Fourier transform G(f) of the signal g(t) has the roperty
That Sj) i ero for fl = Was istrated in Fig, 20s; the shape of ie
that am, show in i iar intended for he ene re
only ee als tat we choose the sampling period T, = 1/2W. Ten
Ghe corresponding {G.{J) of the sampled signal g,() is as shown
te cor sb. Puting T, = 120 in Ea 2.132 yields
oun = 3 (a) e( 4) 18)
Bowe 230
Ota ors samping pro
M2ore 90 (1 Sacrum of saci vernon
From Eq, 2.131, we have
GAS) = 1G) +4, L GY - mip My
Hence, under the following two sonions
1. 6) = Ofor|f >
2 f= 2W
we ind from Eq 2.18 tat
on -+
(= 46M -Wesew ans
Sabottuting Eq. 2.133 n Eq 2.138, we may abo write
innf
) exe( -424) () (QnWi — nn)
(sz) sinc(2W¢ - 1),
‘an interpolation formula for reconstructing the
the sequence of sample values {g(/2W)}, with
jay function.
‘unction sine(2W®) playing the role of an interpolation
aoe erty version of he eon fe
forms are added to obtain (0).
the sampling theorem® for band-limited signals of
nect
