APPENDIX A

A Review of Fourier Techniques

A.1 SIGNALS, SPECTRA, AND LINEAR SYSTEMS Electrical communication signals consist of time-varying voltage or current waveforms, typically described in the time domain. It is also convenient to describe such signals in the frequency domain. A signal's frequency-domain description is called its spectrum. Spectral concepts are important in communication analysis and design; they can describe a signal by its average power or energy content at various frequencies, and they illustrate how much of the electromagnetic spectrum (bandwidth) the signal occupies. Broadcast stations are required by the Federal Communications Commission (FCC) to operate at their assigned frequency with very tight tolerances on the occupied bandwidth; for example, amplitude-modulated (AM) radio channels are spaced 10 kHz apart, and television channels are spaced 6 MHz apart. Our interest in spectra and Fourier techniques has to do with the real-world constraints of ensuring that our communication signals are confined to specified spectral boundaries. Frequency spectral characteristics can be ascribed to both signal waveforms and to circuits. When we say that a particular spectrum describes a signal, we mean that one way of characterizing the signal waveform is to specify its amplitude and phase as a function of frequency. However, when we talk about the spectral attributes of a circuit we are referring to the output versus input frequency-domain transfer function of the circuit; in other words, we are characterizing the circuit by how much of a specific input signal spectrum is allowed to pass through it.

results. Item 1, predicting system response, is illustrat~d schematically in Figure A.I. Let the input be an arbitrary periodic waveformWIthperiod equal to To seconds. Founer techmques allow us to describe such an input as a Sumof sinusoidal waveforms, as shown in the figure. The lowest-frequency sinusoid, or the fundamental frequency of the input periodic, has frequency liTo hertz; the balance of the sinusoids have frequencies that are integral harmonics (2ITo, 3/To, .. ) of this fundamental frequency. An important attribute of a linear system is that superposition applies, which means that the response to the sum of excitations is the sum of the responses to the individually applied excitations. In fact, this is used as a definition of linearity. Specifically, if

YI(t) = system response tOXI(t) Y2(t) = system response to x,(t)

and

aYI(t) + bY2(t)

system response to aXI(t) + bx,(t)

for all a, b, XI(t), and x,(t), then the system is linear. A ~onsequenceof this definition is that the output response of a linear system Withsinusoidal input waveforms must be made up of sinusoidal waveforms having the same frequencies a .the input waveforms' such a system is typically specified by an output versus input f~e, . . quency transfer function (magmtu de and p hase versus frequency) as shown m Fig. . ure A.2. Figure A.2a illustrates a typical example of SIgnal magnitude versus
Output waveform

t
)

Input
waveform

JUL
---j ~
To seconds nEqUalS

A.f\J
~EqUaIS

f = - hertz
To
2

f= - hertz To f= - hertz
To
2

f= - hertz To f = ]. hertz To
+

f= -3 hertz
To
+

A.2 FOURIER TECHNIQUES FOR LINEAR SYSTEM ANALYSIS Fourier techniques are often used for analyzing linear circuits or systems in the following ways: (1) by predicting the system response, (2) by determining the system dynamic specification (transfer function), and (3) by evaluating or interpreting test
1012

Figure A.1

Predictingsystemresponse.
1013

A.2

. S stem Analysis Fourier Techniques for Linear Y

Itrz...

APPENDIX A

A Review of Fourier Techniques

A.1 SIGNALS, SPECTRA, AND LINEAR SYSTEMS Electrical communication signals consist of time-varying voltage or current waveforms, typically described in the time domain. It is also convenient to describe such signals in the frequency domain. A signal's frequency-domain description is called its spectrum. Spectral concepts are important in communication analysis and design; they can describe a signal by its average power or energy content at various frequencies, and they illustrate how much of the electromagnetic spectrum (bandwidth) the signal occupies. Broadcast stations are required by the Federal Communications Commission (FCC) to operate at their assigned frequency with very tight tolerances on the occupied bandwidth; for example, amplitude-modulated (AM) radio channels are spaced 10 kHz apart, and television channels are spaced 6 MHz apart. Our interest in spectra and Fourier techniques has to do with the real-world constraints of ensuring that our communication signals are confined to specified spectral boundaries. Frequency spectral characteristics can be ascribed to both signal waveforms and to circuits. When we say that a particular spectrum describes a signal, we mean tbat one way of characterizing the signal waveform is to specify its amplitude and phase as a function of frequency. However, when we talk about the spectral attributes of a circuit we are referring to the output versus input frequency-domain transfer function of the circuit; in other words, we are characterizing the circuit by how much of a specific input signal spectrum is allowed to pass through it.

results. Item 1, predicting system response, is illustrated schematically in Figure A.l. Let the input be an arbitrary periodic waveformwith period equal to To seconds. Founer techmques allow us to describe such an input as a Sumof sinusoidal waveforms, as shown in the figure. The lowest-frequency sinusoid, or the [undamental frequency of the input periodic, has frequency liTo hertz; the balance of the sinusoids have frequencies that are integral harmonics (2ITo, 31To, .. ) of this fundamental frequency. An important attribute of a linear system is that superposition applies, which means that the response to the sum of excitations is the sum of the responses to the individually applied excitations. In fact, this is used as a definition of linearity. Specifically, if

Yl(t) = system response tOXI(t) Y2(t) = system response to x,(t)

and

aYI(t)

+ bY2(t)

system response to aXI(t)

+ bx,(t)

for all a, b, XI(t), and xit), then the system is linear. A consequence of this definition is that the output response of a linear system with sinusoidal input waveforms must be made up of sinusoidal waveforms having the same [requencies a .the mput waveforms' such a system is typically specified by an output ver us input !~equency transfer function (magnitude and phase versus frequency) as shown m Figure A.2. Figure A.2a illustrates a typical example of signal magnitude versus Output
waveform

Input
waveform

J1SL
-1
To seconds

/\.f\J
t> -1
f= 2
hertz

nEqUalS

f = - hertz
To
2

To
To

hertz

f= - hertz To f= - hertz To
+
3

t> -3
+

To

hertz

A.2 FOURIER TECHNIQUES FOR LINEAR SYSTEM ANALYSIS Fourier techniques are often used for analyzing linear circuits or systems in the following ways: (1) by predicting the system response, (2) by determining the system dynamic specification (transfer function), and (3) by evaluating or interpreting test
1012

Figure A.1

Predictingsystemresponse.
1013

A.2

. S stem Analysis Fourier Techniques for Linear Y

APPENDIX A

A Review of Fourier Techniques

A.1 SIGNALS, SPECTRA, AND LINEAR SYSTEMS Electrical communication signals consist of time-varying voltage or current waveforms, typically described in the time domain. It is also convenient to describe such signals in the frequency domain. A signal's frequency-domain description is called its spectrum. Spectral concepts are important in communication analysis and design; they can describe a signal by its average power or energy content at various frequencies, and they illustrate how much of the electromagnetic spectrum (bandwidth) the signal occupies. Broadcast stations are required by the Federal Communications Commission (FCC) to operate at their assigned frequency with very tight tolerances on the occupied bandwidth; for example, amplitude-modulated (AM) radio channels are spaced 10 kHz apart, and television channels are spaced 6 MHz apart. Our interest in spectra and Fourier techniques has to do with the real-world constraints of ensuring that our communication signals are confined to specified spectral boundaries. Frequency spectral characteristics can be ascribed to both signal waveforms and to circuits. When we say that a particular spectrum describes a signal, we mean that one way of characterizing the signal waveform is to specify its amplitude and phase as a function of frequency. However, when we talk about the spectral attributes of a circuit we are referring to the output versus input frequency-domain transfer function of the circuit; in other words, we are characterizing the circuit by how much of a specific input signal spectrum is allowed to pass through it.

results. Item .1, predicting system response, is illustrated schematically in Figure A.1. Let the Input be an arbitrary periodic waveform with period equal to To seconds. Founer techniques allow us to describe such an input as a Sumof sinusoidal waveforms, as shown in the figure. The lowest-frequency sinusoid, or the fundamental frequency of the input periodic, has frequency liTo hertz; the balance of the sinusoids have frequencies that are integral harmonics (2fTo, 3fTo, ... ) of this fundamental frequency. An important attribute of a linear system is that superposition applies, which means that the response to the sum of excitations is the sum of the responses to the individually applied excitations. In fact, this is used as a definition of linearity. Specifically, if Yl(t) y,(t) and aYI(t)
= =

system response to xI(r) system response to x,(t)

+ by,(t)

system response to aXI(t)

+ bx,(r)

for all a, b, XI (z), and x,(t), then the system is linear. A ~onsequenceof this definition is that the output response of a linear system with sinusoidal input waveforms must be made up of sinusoidal waveforms having the same frequencies as.the input waveforms' such a system is typically specified by an output versus mp~t f~e, . . quency transfer function (rnagmtu de and p hase versus frequency) as shown mig. . ure A.2. Figure A.2a illustrates a typical example of signal magnitude versus Output
waveform

Input
waveform

J1SL
~ ~
To seconds

fV\./

nEQUalS

Equals

f: - hertz To f:
2 - hertz To

[> - hertz

To

2 f: -

To

hertz

A.2 FOURIER TECHNIQUES FOR LINEAR SYSTEM ANALYSIS Fourier techniques are often used for analyzing linear circuits or systems in the following ways: (1) by predicting the system response, (2) by determining the system dynamic specification (transfer function), and (3) by evaluating or interpreting test
1012

3 f:-hertz To

t> -3
+

To

hertz

FigureA.1
A.2

Predictingsystem response.
1013

. S stem Analysis Fourier Techniques for Linear Y

~t---~
.'='
::l

'" ~
Frequency
(a)

"

The terms cos x and sin X are called the [und I . 'f I amenia terms; the terms cos nl- and sin ns, or n > ,are called harmonic terms where u l . . . ,illS an mteger. The terms a and b; represent the coefficients of the fundamental and harmonics an 1. n stant or dc term. ' d ,aD IS the conThe function x(>..) must have a period of 211, or a submultiple thereof and it must. be single valued. The.. Fourier series can be thought of as a " recipe"f' or syn. . . thesizing any arbitrary periodic waveform using sinusoidalcomponents. To be useful, the senes n:'ust converge; that IS, the sum ofthe series, as more and more of the higher harmorucs are added, must approach a limit. The process of synthesizing an arbitrary periodic waveform,from the coefficient values describing the mix of harmonics, is termed synthesis. The inverse process of calculating the coefficient values is termed analysis. Calculation of the coefficients is facilitated by the fact that the average of the sine and cosine crossproducts is zero, as well as the average of any sinusoid.The followingequations il lustrate the basic averaging properties of the sine, cosine, and their products and cross-products:

rn

Frequency
(b)

A.2 System transfer function. (a) Magnitude response. (b) Phase response. Figure

frequency; similarly, Figure A2b illustrates a typical example of signal phase versus frequency. The system transfer function serves as a performance specification; it describes the system response to each of the component sinusoids. Therefore, with the system transfer function in hand, one can predict each of the resulting output components. Using the principle of superposition, the final step of the analysis is to sum the individual output responses, thus forming the resulting overall response to the input periodic. (See Figure A1.) In a similar manner, one can determine a system's transfer function, or evaluate a system's test results from knowledge of the input and output waveforms. The development of Fourier methods had a major impact on the analysis of linear systems; it provided the translation between transient phenomena and sinusoidal techniques, and it simplified the analysis of linear systems under the excitation of any arbitrary input waveform. Just as logarithms allow the operation of multiplication to be treated as addition, so Fourier techniques allow the replacement of complex waveforms with sinusoidal components and sinusoidal methods. A.2.1 Fourier Series Transform Signals that are periodic with finite energy within each period can be represented by the Fourier series. The following equation describes such an arbitrary periodic waveform x(>..) in terms of an infinite number of increasing harmonic sine and cosine components: x(>..) = tao

l: l: l: l: l:
A.2

l: l:

sin m): dl-

=0
(A.2)

cos m): dl- = 0 where m and n are any integers

=

sin ms: cos n); dl-

sin m>..sin nl-dl-

=0
for m
=0

'* n

(A.3)

cosm>..cosn>..dl-

(sin m>..)' dl- = (cos m>..)' dl- =

11

form
11

(AA)

+ al

cos x

+ a,

cos 2>" + a3 cos 3>"

(A.I)
>..+ b, sin 2>" + b-; sin 3>" +

+ ... + b I sm

...

. d. the value of the coefficient,an or bn, Consider how one could go about fin 109f ample we can multiply both in Equation (AI). To find the coefficien\a3'ra~~:~ follo;s: sides of Equation (AI) by cos 31-dl- and 10eg ,
1015

1014

A Review of Fourier Techniques

App.A

. S stem Analysis Fourier Techniques for Linear Y

f_~~

x()..) cos 3)..d)"

i:~
+

aor)..

ds: +

f_~~j a
3)" d)"

co~s

3)" d)"

I:

a
a2 c00s

a
+

f~

a, (cos 3),,)' ds.

+ ...

.r
I:

a
blS~OS3)"d)"+

a
b'Si~OS3)"d)"

i:
+ ...

b'Si~OS3)"d)"

A periodic function with period To seconds has frequency components of ff> 2/0, 3/0, ... , where 10 = liTo IS called the fundamental frequency. We also refer 10 the frequency components as "'0, 2"'0' 3"'0' ... , where "'0 = 271ITo is called the fundamental radian frequency. The terms f and", are each used to denote frequency. When I is used, frequency in hertz is intended; when", is used, frequency i radio anslsecond is intended. Let us replace the n).. terms of Equations (A.S) to (A.8) with 21Tn/ot = 21TntlTo as the general argument of the sinusoidalcomponents, where n is an integer. For n = 1, nlo represents the fundamental frequency; for n > 1, nfo represents harmonics of the fundamental frequency. Using Equations (A.8) to (A.10), we can express x(t) in exponential form as follows:

x(t) = a 0 + 2
and b; by

~~I

[(a, - jb,)el'''f,'

+ (a, + jb,,)e'j'''f.,]

(A.lt)

i:
aD

x()..) cos 3)" dt. =

a,
=

i:
'IT

Let c, denote the complex coefficients, or spectral componentsof X(I), related t

a,

a3 (cos 3),,)' d)" = a,1T

x()..) cos 3)" d)" c ,=

~(a, - jb,,)

for n for n for n

>0
=a
(A. t2)

1c [ "

ao
2

J-1r

We can generalize the preceding analysis to get 1 a, = 1T b; is found by solving Equation 1

l(a, + jb,)
(AS) (A6)

<a
(A.t3)

i~ =:;;- i~ _~
_~

Then we can simplify Equation (A.ll), writing

x()..) cos ni: d)"

x()..) sin ns: di:

where the coefficients of the exponential harmonics are 1 c =-

(A.S) with n = ~ aD
=

1 2
1T

i~
_~

a. This

results in

TJl

x()..) d)"

(A7)

To -TJl

xle

()

-j'''f,' dl

(A.14)

which represents the zero-frequency term, or the average value of the periodic waveform. The synthesis process of Equation (AI) can be expressed in more compact form as follows: x()..)
= ~aD

I' I both sides of Equation (A.13) by To verify Equation (A14), -;e mu t(l~i '2 T[J2), and use the relationship e-j'~mfo'dt/Ti; integrate over the interval ~, T.fl {I for n = III (A.lS) i...-m1lrrfo'dt = 8,m = 0 for n m

1.

L (a,
n=l

cos ns;

+ b ; sin n)..)

(A8)
where 1\
nm

To - TJl 1 B multiplyingand integrating in this is known as the Kronecker de tao y 1

There are several ways to express the transform pair (analysis and synthesis) of the Fourier series. The most common form makes use of the following identities to express the sine and cosine in exponential form: cos ).. =
.

way we obtain, for all integers m,

-=---'---=--2
ej"he-j"A

e'': + e-j).

(A9) (Ala)

(A,16) dt == ~ c,.8nm:::;: em _ x(l)e ,=,00 To -To/' . b expressedin the form . om lex number; It can e In general, the coefficient c, \S a c P, (A.17) l c" = Ic,le , (A.l8)
00

ToI2

-j2'ffmfol

sm)..

-=--_...::..2j A Review of Fourier Techniques

c_,=lc,e

-j',
1

1017

1016

App. A

A.2

. Linearsystem Analysis Fourier Techniques for

where

ICnl

t Va~+ b~
I

(A.19) (A.20)

A .....

+-.,

en = tanbo

--

b;
an

and

Co =-

ao 2

The value of Icnl defines the magnitude of the nth harmonic component of the periodic waveform, so that a plot of Icnl versus frequency, called the magnitude spectrum, yields the magnitude of each of the n discrete harmonics in the signal. Similarly, a plot of versus frequency, called the phase spectrum, yields the phase of each harmonic component in the signal. The Fourier coefficients of a real-valued periodic time function exhibit the relationship

en

FigureA.3 Pulsetrain.

(A.21) where c~ is the complex conjugate of

Cn'

negative values. It goes through zero at y = 1, 2, ... The pulse train magnitude s ectrum Ic I as a function of nlTo, is plotted in Figure A.5a, and the pha e specp '. n d i F A 5b The positive and negative frequencies of the ~~:~i~;~/::el~:~m ~:pr~~~t a ~seful way of expr:~~n~o~:e~~~~trau~b:::t~er;ati'

We therefore have (A.22)

ICnl=lcnl

and the magnitude spectrum is an even function of frequency. Similarly, the phase spectrum is an odd function of frequency, because from Equation (A.20),

en

cally; of course, only the positive f;~~~~~~~:s e:~ebcoeirieients of Equation (A.24) SynthesIs IS performed b Y ies yi:ldS the original ideal pulse train, X/I), into Equation (A.13). The resu lting sen synthesized from its component parts:

(A.23) The Fourier series is particularly useful in characterizing arbitrary periodic waveforms, with finite energy in each period, as presented above. The Fourier series can also be used to characterize nonperiodic signals having finite energy over a finite interval. However, a more convenient frequency-domain representation for such signals uses the Fourier integral transform. (See Section A.2.3.) A.2.2 Spectrum of a Pulse Train A signal of great interest in digital communications is an ideal periodic sequence of rectangular pulses, called a pulse train, illustrated in Figure A.3. For the pulse train, xp(t), with pulse amplitude A, pulse width T, and period To, the reader can verify, using Equations (A.14) and (A.I0), the following expression for the Fourier series coefficients: c =n

sincy

AT sin(TInT/To) To TInT/To

= -StnC-

AT. To

nT To

(A.24)

In this expression, . sin TIy smcy =-TIy The sine function, as shown in Figure AA, has a maximum value of unity at y = 0 and approaches zero as y approaches infinity, oscillating through positive and 1018 A Review of Fourier Techniques App.A
A.2

-0.5 . A4 FIgure . Sinefunction. 1019

. Linearsystem AnalysiS Fourier Techniques for

, , ,
-31T -21T -11T

, ATITo
\ \ \ \

the Fourier transform, We can describe the no . di , , .. npeno IC SIgnalas a iodi . the limiting sense. For example consider the pit . peno IC one, III , U se rain shown in F A3 To -7 00 and the pulse train approaches a single 1 igure " As lines approaches infinity and the spectral plotPU se, X(l ), the number of spectral h '. " approac es a smooth freque spectrum X(f) . For this limiting case, we can define th F . . ncy . e ouner mtegral transform pan.

(A26) and

_...t..l~LL.u~LL.u:.,J,-JJ-LL.L..!--!-J~:-"-...L.J:':':'::-"-...L.J~;-niTa

X(t)
(al

1:
f-+

X(nei"!' dt

(A27)

en _. 1 1 1

.-----1

lIr

-11 t-

1 1

i
1

1 _____ 1

1 1

:
I

niTa

where f is frequency measured in hertz, This pair can be used to describe the time-frequency relationship for nonperiodic signals. Henceforth, the Fourier integral transform operation wili be designatedby the notation ~l'), and the inverse Fourier integral transformwilibe designatedby ~'I!'J. he relationship between the time and frequencydomainswilibe indicated T by using the double arrow as follows:

I. _

X(I)

X(fJ

o
Figure A.5 Spectrum of a pulse train. (a) Magnitude spectrum. (b) Phase spectrum.

This notation indicates that X(f) is the Fourier transformof X(I) and that X(I) is the inverse Fourier transform of X(f). In the typical communicationscontext,X(I) is a real-valued function and X(f) is a complex function, having reai and imaginary components; in polar form, the spectrum, X(f), can be specifiedby a magnitude characteristic and a phase characteristic:

AT xp(t) = To

cc

sinc -

nT, To

X(n e ) 2.nl. ,
(A.25)

!x(fJI ei~f)

(A.28)

n=-OO

The ideal periodic pulse train contains frequency components at all integer multiples of the fundamental. In communication systems the significant portion of a baseband signal's power or energy is often assumed to be contained within the frequencies from zero to the first null of the magnitude spectrum (see Figure A.5a). Therefore, liT is often used as a measure of signal bandwidth, in hertz, for a pulse train with pulse width T. Note that bandwidth is inversely proportional to pulse width; the narrower are the pulses, the wider is the bandwidth associated with these pulses. Also, notice that the spacing between spectral lines Af = liTo is inversely proportional to the pulse period; as the period increases, the lines move closer together. A.2.3 Fourier Integral Transform In communication systems we often encounter nonperiodic signals having finite energy in a finite interval, and having zero energy outside this interval. Such signals can be conveniently characterized using the Fourier integral transform, or simply
1020

The properties of X(j) the spectrum of a nonperiodic waveform,are similarto , 'E . (A 17) to those of the spectrum for a periodic waveform, presented III quauons ' (A.23); that is, when x(t) is real valued,

X(-n = X'(n

(A,29)
e'i~n

= IX(nl

(AJO)

where X* is the complex conjugate of X. The magnitudespectrum IX(f)1 isaX(fev)el'~ . d f ti off 10 manycases function of f and the phase spectrum ISan od unc 100 ff', t describeit 'h .' . eit er purely real or purely unagmary, andcly=~w=o 0

A,3 FOURIERTRANSFORM

PROPERTIES

A Review of Fourier Techniques

App.A

, ith the detailsof Fourier transforms There are many excellent references deah~g WI. 1 h size the propertiesthat . hi dix we wil ernp a . . and their properties [1-4]. In t IS appen f the key features affectlllgsigare fundamental to communication systems. Somea 1021
A.3 Fourier Transform Properties

lenl

, ATITo
I I I I \ \ \ \

the Fourier transform. We can describe the nonperiodic signalas a periodic one, in the limiting sense. For example, consider the pulse train shown in Figure A.3. As To ~ cc and the pulse train approaches a single pulse, x(l), the number of spectral lines approaches infinity and the spectral plot approaches a smooth frequency spectrum X(f). For this limiting case, we can define the Founer Integral transform pair. (A.26)

-31T

-21T

-liT

o~
(a)

~/T liTo

21T

31T

nlTo

and

x(t) =

XU)ej2rr/t dt

(A.27)

en --- 1
I I

.-----1
I I

nI-

I I

:
I

,
I I

nlTo

. d i hertz This pair can be used to describe the where f IS frequency measure In .... time-frequency relationship for nonpenodlc ;Ignals. tion will be designated by Henceforth, the Fou~ler integral trans tor~lot~:~:form will be designated by the notation ~[.), and the inverse Fo~ner m def:equencydomains will be indicated ~-I('l. he relationship between the tune an T by using the double arrow as follows:

x(t) ..... X(tl

. . ). the Fourier transform of x(t) and that X(I) is the This notation mdlcates that X(j IS . I ommunicalions context, X(I) ISa inverse Fourier transform of X(j). In the ItYPlcfall'on having real and imaginary c . d X(j)' a comp ex unc,. . d real-valued function an IS X(j) can be specifIed by a magmtu e . 1 f m the spectrum, , components; in po ar or , . '. .' d hase charactenstlc. charactenstic an a p . (A.28) X(f) = lXU)l el~fI . d' waveform are similar to h trum of a nonpeno IC '. (A 17) to The properties of X(j), t e spec ic waveform, presented in EquallOnS . those of the spectrum for a penod (A.23); that is, when x(t) is real valued, X( -f) = X*(f)
=

-n

'_____1

I- _

(b)

Figure A.S Spectrum of a pulse train. (a) Magnitude spectrum. (b) Phase spectrum.

xp(t)

=-

AT
To

n~-~

sine - eJ2.nfot To

nT

(A.25)

The ideal periodic pulse train contains frequency components at all integer multiples of the fundamental. In communication systems the significant portion of a baseband signal's power or energy is often assumed to be contained within the frequencies from zero to the first null of the magnitude spectrum (see Figure A.5a). Therefore, liT is often used as a measure of signal bandwidth, in hertz, for a pulse train with pulse width T. Note that bandwidth is inversely proportional to pulse width; the narrower are the pulses, the wider is the bandwidth associated with these pulses. Also, notice that the spacing between spectral lines Ilf = liTo is inversely proportional to the pulse period; as the period increases, the lines move closer together. A,2.3 Fourier Integral Transform

(A.29) . (A.30)

\XU'l1 is an even . e of X. The magnitude spectrum cases X(f) is where X* is the complex conJugat . an odd function of f In rna';! 'be it function of f and the phase spectrumv .and only one plot sufficesto escn . rely imagmar J' either purely rea I or pu
PROPERTIES il of Fourier transform A.3 FOURIER TRANSFOR M . tth the deta s . that ferences deahng WI. basize the properlles . There are many e~cel[le~t]r~n this appendix we Will key features affecting sigand their propertles 1 '. ti on systems. Some 0 are fundamental to commuOlca I 1021

IX(tll e-}~fI

In communication

systems we often encounter nonperiodic signals having finite energy in a finite interval, and having zero energy outside this interval. Such signals can be conveniently characterized using the Fourier integral transform, or simply
1020

t~~

A Review of Fourier Techniques

App. A

A.3

Fourier Transform Properties

AA USEFUL FUNCTIONS nal transmission in communication systems are time delay, phase shift, multiplication by other signals, frequency translation, waveform convolution, and spectral convolution. We shall focus on the Fourier properties (shifting and convolution) needed to describe these key communication features. A.3.1 Time Shifting Property If x(t)
H

A.4.1 Unit Impulse Function A useful function in communication theory is the unit impulse or Dirac delra function, 1'>(t). The impulse function can he developed from any of several fundamental functions (e.g., a rectangular pulse or a triangular pulse). In each development, the impulse function is defined in the limiting sense (the pulse amplitude approaches . finit the pulse width approaches zero, hut the area under the pulse IS conm im y, , f II " t strained to be unity) [5]. The unit impulse function has the 0 owmg importan properties:

X(j), then ~{x(t - to)}

=

i:
=

x(t - t.)e-j2!,

dt

(A.31)

Let f.L = t - to; then

i:
o(t) = 0 g;{o(r)}

S(r)dr=1 for t

(A.34)
(AJ5)

*0
=

o(t) is unbounded at r = 0 Xf..!)e-j2.!,.

=

(A.36)
(AJ7)

g;-1{3(fl}

As a signal is delayed in time, the magnitude of its frequency spectrum remains unchanged, but its phase spectrum experiences a phase shift, A time shift of to in the time domain is equivalent to multiplication by e-j2 (a phase shift of -2TIfto) in the frequency domain.

Loo

(00

x(t)o(t - to) dt

= x(to

(AJ)

.!,.

A.3.2 Frequency Shifting Property If x(t)

H

X(j), then ~{x(t)ej2.!",}

I~

I ro erty; the unit impulse mulEquation (A.38) is known as the sifting or sami ;::at ~tiplier selects a sample of the functl~~ ~ollOWlOg ~q~JValent ntegrals fran i In some problems, It ISuseful . thefrequency domain [3). f d i the time domam or impulse function, de me in (AJ9) 3(t) = eM'd!

:~l ~~:

t;

i:
-~

x(t)ej2.!"

e-j2.!, dt o(f) =
(A.32) A.4.2 Spectrum of a Sinusoid

i: i:

(A.40)
e-M'dt

x(t)e-j2U-!')'dt

= XU - fo)
This is the basic frequency translating property that describes the shifted spectrum resulting from multiplying a signal by eP'!o'. Equation (A.32) can be used in conjunction with Equation (A.9) to yield the Fourier transform of a waveform multiplied by a cosine wave, as follows:

x(t) cos 2TIfot

Hx(t)ej2.!"

+ x(t)e-j2.!,,]

(A.33)

, f rm the f by a Founer tranS 0 , ting a sinusoidal wave orm TrJ2 < t < TrJ2). Under For the purpose of represe~ to exist only in the interval (- as T is finite. In the waveform may be assume. has a Fourier transform as long f~rm x(1) = A cos these conditions the functIOn fi ite The spectrum of the wave I rge but 101 . ( 26)' limit, To is made very a. ' uations (A.9) and A. ' 2TIfot can be found by usmg Eq -vt , oo A " -j2'/o')e I ! dr

x(t) cos 2TIfot - Hxu

- fo) +

XU + fo)]

X(f)

= .00

(eJ2'ITJOI

+e

This property is also called the mixing or modulation theorem. Multiplication of an arbitrary signal by a sinusoid of frequency fo translates the original signal spectrum by fo, and also by - fo.
1022

A (

00
e

_j2n(f-!,)1

+ e .' I 2n(f +!oJ'dr

1023

=2"100
AA
Useful Functions

A Review of Fourier Techniques

App. A

As des~ribed in Equation (AAO), the foregoing integral expression can be equated to umt Impulse functions located at frequencies fo as follows: X(n

A.S CONVOLUTION
Convolution was used by Oliver Heaviside in the late nineteenth century to calculate electncal circuit output current when the input voltage waveform was more complicated than a simple battery source. The use of the methods of Heaviside predates the use of the analytical methods developed by Fourier and Laplace (even though publications by Fourier and Laplace came earlier). The response of a circuit to an impulse voltage V(I) = S(I) is called the impulse response and is denoted by h(I), as shown in Figure A.S; it is simplythe output voltage that would result if the input were a delta function. Heaviside approximated an arbitrary voltage waveform, like the one shown in Figure A.9a, by a set of equally spaced pulses. Such pulses of finite height and nonzero duration are shown in Figure A.9b. In the limit as the pulse width ~T approaches zero, each pulse approaches an impulse function with weight equal to the area under that pulse. In the following discussion we shall refer to these equally spaced pulses as impulses even though they are impulses only in the limit. . Care needs to be taken with the notation of time, since we are interested In the times at which impulses are applied and also the times at which their output responses are observed. We need to identify these two different time sequences; we

= 2" [o(f - fa) + o(f + fa)]

(A.41)

Similarly, the spectrum of a sine waveform yet) = A sin 2rrfot can be shown to be equal to A y(n = 2j

[o(f - fa) - o(f + fa)]

(A.42)

The cosine waveform spectrum is shown in Figure A.6, and the sine waveform spectrum is shown in Figure A. 7. Each of the impulse functions shown on these spectral plots is depicted as a spike with a weight of AI2 or -AI2.

X(f)

shall use the following notation:

AI2

A/2

1. Time of the input application will be termed T, so that the input voltage impulses are designated VeT,), VeT,), ... ,V(TN)'
2. Time of the output response will be termed I, so that the output currents are designated i(l,), i(I,), ... ,i(IN)' d ed by each input impulse inHeaviside found the response or current pro uc Th dependently; then he added the individual res~:s~~l:~ ~e:tt~i~~O~~liSc~~:e;:~dUC~ weight of the Impulse produced by the rectang t the ~rbitrary input voltage as VeT,) AT. The series of impulses can approxh,mae Note again that the instant at wing M to approac the instance at which the system reo close Iy as desi d b Y aIIo. esire and zero. which an impulse is applied IS c. iled Ti, . th . put time variable, I is the output a . . di lied t where 1 IS e to sponse IS determme [Sca " time variable, and i = 1, ... ,N.

---_~fo------lO----fLo--f

FigureA.6

Spectrum for

x(t) = A cos 2rrfa/.

jY(f)

A/2
v(l) = 0(1)

Linear

fo --_...Lf:--o----.l--~~-

~,'O'"'
o
Figure A.7 Spectrum for y( t) = A sin 2rrfot. FigureA.S I App.A
A.5

network

L--'::""- t

mpulseresponse ofa linearsystem.

1025

-A/2 1024

A Review of Fourier Techniques

Convolution

vet)

Input impulse (at time 1,) = V(1,)

:l :l

Output response i(t) =A,h(1-1,) Output response (at time I,) = i(tl)
=A,h(t, -1,)

C.

a
"0

c;
:l

'"

(a)

C. C

'Time
Figure A.10

Output responseto an impulseat time1,.

i(t) =A,h(t-1,) Figure A.9 (a) Input voltage waveform. (b) Approximate input voltage waveform.

+Azh(I-1z) i(t,) =A,h(t,-1,) +Azh(I,-1Z)

(b)

Figure A.IO illustrates the output response i(l) = Alh(1 - 71) to an impulse with height V(7,). Since the input impulse at 71 is not a unit impulse, we weight it with its strength or area, A, ~ V(71) ~7. At some time I" where II > 7" the output response to the impulse V(7,) is expressed as
i(I,)
= A]h(11 -

'Time Output response to twoimpulses.

71)

for I,

> 71

Figure A.11

as shown in Figure A.IO. When there are several input impulses, the total output response for a linear system is simply the sum of the individual responses. Figure A.ll illustrates the response of the network to two input impulses. For N impulses, the output current measured at time II can be expressed as
i(ll) = A,h(I, - 71)

or

i(t) ~

:s V(1
N

(A.43)
1)

(J.1h(t - 'i)

Azh(t,

- 72)

+ ... + AN(I,

- 7N)

where the impulses are applied at 7" 72, ... ,7N and where I, > 7N' Any impulses applied at times greater than I, are disregarded, for they contribute nothing to i(I,). This corresponds to the causality requirement for physically realizable systems, which states that the system response must be zero prior to the application of the excitation. By generalizing, we get the output current at any time t, namely,
1026

i"' ( ) As M approaches zero, the . I t IS equal to v 1} . ( ). can re1 since the height of the Impu sc a ' h ctual applied voltage v 1 , we . I pproaches tea . . al: sum of the input rmpu ses a the convo[Ulwn mlegr . mmallon becomes place ~1 with dt, and the su (A.44a) i(l) ~ v(1)h(t - 1) dn

1:

1027
A.5

A Review of Fourier Techniques

App. A

Convolution

or itt) In shorthand
=

i:
=

v(t - T) h(T) d-r

(A.44b)

notation, this is expressed as i(t) v(t)

* h(t)

(A45)

_'dJ'---.- _kL,
-1 0 1

vet)

hit)

In summary, i(t) is the sum of the individual impulse responses as a function of output time t. Each impulse response is due to an impulse applied at some input time T and is weighted by the strength of that impulse. A.5.1 Graphical Illustration of Convolution Consider that an input square pulse v(t) is applied to a linear network whose impulse response is labeled h(t) as shown in Figure A12a. The output response is characterized by the convolution integral expressed in Equation (A.44). The independent variable in the convolution integral is T. The functions V(T) and h(-T) are shown in Figure A12b. Note that h(-T) is obtained by folding h(T) about T =0. The term h(t-T) represents the function zf-cr) shifted by tseconds along the positive T axis. Figure Al2c shows the function h(tj - T). The value of the convolution integral at t = tj is given by Equation (A.44) evaluated at t = t,. This is simply the area under the product curve of V(T) and h(tj -T), shown shaded in Figure A12d. Similarly, the convolution integral evaluated at t = t, is equal to the shaded area in Figure A12e. Figure Al2f is a plot of the output response as a result of the square pulse input to the circuit with impulse response shown in Figure A12a. Each evaluation of the convolution integral, at some time t., yields one point, i(t;), on the plot of Figure A12f. A.5.2 Time Convolution Property If Xj(t)
H

(a)

-1 0

1
(b)

-3

_Ll_..n.---~
-1 0 1
(c)

-3

0 t,

Xj(f), and x,(t)

X,(f), then

x,(t)*x,(t)

~{Xj(t)*x,(t)}

i: i: i:
=
= =

X,(T)X,(t

- T) ds

r~
dt as follows:
-j2nf'

o
(dl

t,

v(~)h(t, -~) d~

r~

-1

1I,

v(t)h(t, - t) dt
(e)

Xj(T) x,(t - T) dTe-j2nf'

For linear systems, we may exchange the order of integration ~{x,(t)*X,(t)}

i: i:
Xj(T) ds X,(f)

x,(t - T)e

dt

(A46) of the right-

By the Fourier time shifting property, the second integral expression hand side is equal to X,(f)e-j'TIf<:

~{x,(t)*X,(t)}

i:

---

""1/\
t,
(f)

.... -1-0 1 2 3 4

Xj(T)e -j'TIf' dT

(A47)
Figure A.12

Graphical example of convolution. 1028

Xj (f)X,(f)
A Review of Fourier Techniques

1028

App. A

A.5

Convolution

Therefore, the operation of convolution multiplication in the frequency domain.

in the time domain can be replaced by

and

X(f) * &(f) = X(f)

(A.52)

A.5.3 Frequency Convolution Property Because of the symmetry of the Fourier transform pair in Equations (A.26) and (A.2?), it can be shown that multiplication in the time domain transforms to convolution in the frequency domain: (A.48) The properties that transform multiplication in one domain to convolution in the other domain are particularly useful, since one operation is often easier to perform than the other. For example, we discussed earlier that Heaviside used convolution to solve for the output current of a linear system when the input was excited by an arbitrary voltage waveform. Such methods involve the (sometimes tedious) convolution of an input waveform with the impulse response of a system. Since convolution in the time domain is transformed into multiplication in the frequency domain, as shown in Equation (A.47), for a linear system we can simply multiply the input waveform spectrum by the system transfer function. The output waveform is then found by taking the inverse Fourier transform of the product: itt)

We therefore conclude that ~onvolution of a function with a unit impulse function reproduces the original function. A simple extension of Equation (A.52) yield

X(f) * 8(f - fo) = X(f - fo)

(A.53)

Figure A.13 illustrates the ease of convolving the spectrum of ao arbitrary waveform with the spectrum of a cosine wave. Figure A.l3a shows an arbitrary baseband spectrum X(f). Figure A.l3b shows a spectrum, Y(f) = &(f - fo) + 8(f + fo) = ~(2 cos 2'lTfotl.The output, Z(f) = X(f) * Y(f), in Figure A.13c i obtained by convolving the waveform spectrum with the impulse functions of Y(f) according to Equation (A.53), where the impulses act as sampling functions. Hence, for this simple example, convolution can be performed graphically by sweeping the sampling impulses past the waveformspectrum. Multiplication by the impulse functions at each step in the sweep yields replications of the waveform spectrum. The result, shown in Figure A.l3c, is a shifted version of the riginal spectrum X(!) to the locations of the impulse functions in Figure A.13b. A.5.5 Demodulation Applicationof Convolution In Section A.5.4 we examined a waveform multiplied by 2 cos 2-rrfor.We illustrated the frequency-domain view of convolving the waveform spectrum WItha cosinewave spectrum. In this section we look at the reverse process. A wa.veformthat has been multiplied by 2 cos 2'lTfot is to be demodulated (the waveform IS to be restored to its baseband frequency range). Figure A.14a represents the spectrum, Z(f), of the waveform thatdbas been upshifted in frequency. We can demodulate this upshifted waveform a~1il;~~~:~~ the baseband waveform, the detection process m trum of the carrier, Y(f) A simple extension by multiplymg it by 2 cos ~~h the specthe frequency domam by. g 14b = B(f - fo) + 8(f +fo)),sh~w(1~3f~~;~sa. . of EquatlOnS(A.52 an .

= ~-l{V(f)H(f)}

(A.49)

Solutions of the form shown in Equation (A.49) are often much easier to perform than those described by Equation (A.45). However, under certain circumstances the operation of convolution is so simple that it can be performed graphically, by mspection. For example, suppose that we wished to multiply an arbitrary waveform by some fixed frequency cosine wave, such as a carrier wave, in the case of modulation. By applying Equation (A.48), we can convolve the spectrum of the arbitrary waveform with the spectrum of the cosine wave. This is easily accomplished, as is shown in the next section. A.5.4 Convolution of a Function with a Unit Impulse By the property shown in Equation (A.47), it should be clear that if

2:~;I~:~ea~(fi

X(f - fo)

* &(f -

fl) = X(f - fo - fil (A.54) I' X(f) - Z(f) * Y(f) is obtained by ap-

x(t) _XCI)
and since B(t) _1 then

Therefore, the result of demodu at;~ndals e~rum appears at baseband (deplying Equatlon (A.54). The resultIng hg p Figure A 14c As in the prevIouS . + 2t as sown in .' . A 14c tected) and also at frequenCles - 0' ed ra hically.The resulting FIgure . section , the convolutlOn can be perform g p contains the following terms:

Z(f + f) ] * [8(f - fo) + 8(f + fo) 1 [Z(f - fo) + JO = Z(f - fo) 8(f - fo) + Z(f - fo) * 8(f + fo)

(A.55)

x(t)
It should also be evident that

* B(t) _
=

XU)

(A. 50)

+ Z(f + fo) 8(f - fo) + Z(f + fo) * 8(f + fo) = 2Z(f) + Z(f - 2fo) + Z(f + 2fo)
1031

x(t) * B(t)
1030

x(t)

(A.51) App. A A.S Convolution

A Review of Fourier Techniques

Notice that the resulting terms consist of tbe baseband spectrum plus terms associated with higher-frequency components. Tbe result is typical of the dete ti n process; the higher-frequency terms are filtered and discarded,leaving the demodulated baseband spectrum.

0
(a)

A.6 TABLES

OF FOURIER TRANSFORMS AND OPERATIONS

used Fourier transforms are listed in Table A.1 and Fourier operation'

Commonly Y(f) Z(f) ;X(f)

* Y(f)

in Table

A.2.
TABLE A.1 Fourier Transforms X(fj 1 8(f) 1 - [8(f - 10) 2 1 2j [8(f -

-Ie

'l
0
(b) Z(f)

t-t to

/\
-to

't /\
0

x(t) I. B(t) 2. 1 3. cos 2-rrfor

to

Ic)
Convolving a signal spectrum with a cosine-wave spectrum.

+ 8(f + 10) + fo)]

Figure A.13

4. sin 2'ITfot 5. B(t - to)

10) - 8(f

exp(-j2'ITfto)
8(f- fo) 2a

/\
-fo

t..:
0

6. exp (j2'ITlot)
7. exp (-altl), a> 0

8. exp [
9. u(t)

-'IT( ~

y]
Ior r > 0
fort

a' + (2~n' T exp [--'IT(fT)'1

~8(f) 2 1

fo

~ ~

(a)
Y(f)

+ _1_
jhf

<0

10. exp (-at) u(t), 11. t exp (-at) u(t), 12. reet (~)

a> 0 a>O

+ j2~f
1

--.l

-fo

t
0
(b)

(n + j2~n' TsincfT

2" [sine
13. cos hID t [ rect ( ~ ) ]

(f - fo)T

+ sine(f + I,)T]

X(f) ; Z(f) * Ylf)

14. Wsinc WI

re,,(fv)
lor ItI '" T lor ItI > T
Tsinc2
0

115.
_.L...JL.l.. ---..I~....l..__L-':-~f

tI
T 8(1
0

IT
(

-2fu

0
Ie)

2fu

Figure A.14

Demodulation

16.
m

2:

mTo)

.!- ~
To "
0

t::e
"

application. A Review of Fourier Techniques App. A A.6

f W
< ,

0 for Ifl > W, and sine x::: (sin 1fx)I1Tx. 1033

-llor-W< Note: fee t(fl2W) -

1032

Tables of Fourier Transforms

and operations

APPENDIX B
TABLE A.2 Operation 1. Scaling 2. Time shifting 3. Frequency shifting Fourier Operations x(t) x(at) x(t - to) x(t) exp (jhfo') X(f)

2.X
lal

(f);;

X(f) exp (-jhfto) X(f - fo) (j2Trf)"X(f)

dllX

4. TIme differentiation .
5. Frequency differentiation 6. Time integration 7. Time convolution 8. Frequency convolution

d'x
dt (-j27rt)"(x(t))

df'
_1_ j2Trf X(f) XI(f)X,(f) XI(f)

L~
XI(') XI (t)x,(t)

x(T)dT

+ "2 X(0)8(f)

* x,(,)

Fundamentals of Statistical Decision Theory

The basic elements of a statistical decision problem are (1) a set of hypothesesthat charactenze the possible true states of nature, (2) a test in whichdata are obtained from which we wish to infer the truth, (3) a decisionrule that operates on the data to decide in an optimal fashion which hypothesisbest describesthe true slate of nature, and (4) a criterion of optimality. These fundamentalsteps are treated in the material that followS. The optimality criterion we willchoosefor the decisionrule is to minimize the probability of making an erroneous decision,although other erile'

* X,(f)

REFERENCES
1. 2. , Anal . PapoulisP AS' M dYStS, Panter ,.., F., ignalulatio N' McGraw-Hill Book Compan N ew York, 1965.a n, Olse, and Spectral Anal si y, ew York, 1977. 3. N Bracewell R Th Y IS, McGraw-Hill Book C , ., e Fourier T

ompany

pany, New York ,.1978 4 H' . aykin, S., Communication

5.

ransform and Its A pp I'tcations M G . ' C raw-Hill

Book Com-

Schwartz ,.,MIn' [ormation Systems, John Wiley & Sons In N Company New Y k . ' . Transmission M d I' C., OlSe, York 1983 ew McGraw-Hill Book ' ' , or ,1980 ' 0 u ation and N . '

na areThe subject of statistical decision theory and hypothesislesling builds on lhe possible [1]. mathematical discipline of probability theory and random variables. II is assumed that the reader has a familiarity with these subjects;if not, Reference [2] is a suggested resource.

8.1 BAYES' THEOREM

l Th' mamerna tica ,""d. '" bYOO rhes' rest B.,d'b"rem . b'" ti is derived from the definition of the relationship between the condItIOnaland Jom (B.I) probability of the random variables A and B: P(AIB)P(B) ~ P(BIA)P(A) ~ P(A,B)

0'

"".g 0"

A statement of the theorem is P(AIB)~~

p(BIA)P(A) .' I

(B.2) bability P(A\B) (rom the con'

1035

Bayes' theorem allows us to infer the condltlona pro ditional probability P(BIA)
1034 ARe' view of Fourier Techniques App.A