Contents

Chapter I : Signals and Systems /

1.1 Signal Analysis 1

1.2 Complex Frequency 4
1.3 Network Analysis 7
1.4 Network Synthesis 14

Chapter 2 : Signals and Waveforms 20

2.1 General Characteristics of Signals 20

2.2 General Descriptions of Signals 24
2.3 The Step Function and Associated Waveforms 28
2.4 The Unit Impulse 33
\

Chapter 3 : The Frequency Domain : Fourier Analysis 46

,
3.1 Introduction 46
y
3.2 Orthogonal Functions 47
3.3 Approximation Using Orthogonal Functions 48
3.4 Fourier Series 50
3.5 Evaluation of Fourier Coefficients 52
3.6 Evaluation of Fourier Coefficients Using Unit
Impulses 58
3.7 The Fourier Integral 63
3.8 Properties of Fourier Transforms 67
xiii

8 '

Contents xv
xiv Contents

Differential Equations 75 Chapter 8 : Amplitude, Phase, and Delay 212

Chapter 4:

4.1 Introduction 75
76 8.1 Amplitude and Phase Response 212
4.2 Homogeneous Linear Differential Equations 8.2 Bode Plots 221
82
4.3 Nonhomogeneous Equations 8.3 Single-Tuned Circuits 229
85
4.4 Step and Impulse Response 8.4 Double-Tuned Circuits 238
4.5 Integrodifferential Equations 91
93 8.5 On Poles and Zeros and Time Delay 245
4.6 Simultaneous Differential Equations

Network Analysis: I 100

Chapter 5 : Chapter 9 : Network Analysis : II 253
100
5.1 Introduction
103 9.1 Network Functions 253
5.2 Network Elements
106 9.2 Relationships Between Two-Port Parameters 264
5.3 Initial and Final Conditions
111 9.3 Transfer Functions Using Two-Port Parameters 266
5.4 Step and Impulse Response
5.5 Solution of Network Equations
114 9.4 Interconnection of Two-Ports 271
5.6 Analysis of Transformers 122 9.5 Incidental Dissipation 276
9.6 Analysis of Ladder Networks 279

134
Chapter 6 : The Laplace Transform
Chapter 10 : Elements of Realizability Theory 290
6.1 The Philosophy of Transform Methods 134
The Laplace Transform 135
6.2
137 10.1 Causality and Stability 290
6.3 Properties of Laplace Transforms
6.4 Uses of Laplace Transforms 144 10.2 Hurwitz Polynomials 294
6.5 Partial-Fraction Expansions 148 10.3 Positive Real Functions 299
Poles and Zeros 155 10.4 Elementary Synthesis Procedures 308
6.6
6.7 Evaluation of Residues 162
6.8 The Initial and Final Value Theorems 165
Chapter II : Synthesis of One- Port Networks with Two Kinds
of Elements 315
175
Chapter 7 : Transform Methods in Network Analysis

7.1 The Transformed Circuit 175

V
I 11.1 -
Properties of L C Immittance Functions 315
7.2 Thevenin’s and Norton’s Theorems 180
187
11.2
11.3
-
-
-
Synthesis of L C Driving Point Immittances
Properties of R C Driving-Point Impedances
319
325
7.3 The System Function
7.4 The Step and Impulse Responses 194 11.4 -
Synthesis of R C Impedances or R-L Admittances 329
7.5 The Convolution Integral 197 11.5 - -
Properties of R L Impedances and R C Admittances 331
7.6 The Duhamel Superposition Integral 201 11.6 --
Synthesis of Certain R L C Functions 333

S
 ;
xvi Contents Contents xvii

Appendix A: Introduction to Matrix Algebra 461

Chapter 12 : Elements of Transfer Function Synthesis 341
A.l Fundamental Operations 461
12.1 Properties of Transfer Functions 341 A.2 Elementary Concepts
345 462
12.2 Zeros of Transmission A.3 Operations on Matrices 464
12.3 Synthesis of T21 and Z21 with a 1-0 Termination 347 A.4 Solutions of Linear Equations 468
12.4 Synthesis of Constant-Resistance Networks 352 A.5 References on Matrix Algebra 469
'

Chapter 13 : Topics in Filter Design 365 Appendix B : Generalized Functions and the Unit Impulse 470

365 B.l Generalized Functions 470

13.1 The Filter Design Problem
365 B.2 Properties of the Unit Impulse 476
13.2 The Approximation Problem in Network Theory
13.3 The Maximally Flat Low-Pass Filter Approximation 368
13.4 Other Low-Pass Filter Approximations 373 Appendix C : Elements of Complex Variables 481
13.5 Transient Response of Low-Pass Filters 388
13.6 A Method to Reduce Overshoot in Filters 392 C. l Elementary Definitions and Operations 481
13.7 A Maximally Flat Delay and Controllable Magnitude C.2 Analysis 483
Approximation 395 C.3 .Singularities and Residues 486
13.8 Synthesis of Low-Pass Filters 397 C.4 Contour Integration 487
13.9 Magnitude and Frequency Normalization 402
13.10 Frequency Transformations 404
Appendix D : Proofs of Some Theorems on Positive Real
Functions 490
I
Chapter 14 : The Scattering Matrix 413

413
Appendix E : An Aid to the Improvement of Filter Approxi
mation
-
14.1 Incident and Reflected Power Flow 493
14.2 The Scattering Parameters for a One-Port Network 415
E. l Introduction
14.3 The Scattering Matrix for a Two-Port Network 419 493
426 E.2 Constant Logarithmic Gain Contours 494
14.4 Properties of the Scattering Matrix
429 E.3 Constant Phase Contours 495
14.5 Insertion Loss ! E.4 Contour Drawings
14.6 Darlington’s Insertion Loss Filter Synthesis 431 496
E.5 Correction Procedure 498
E.6 Correction Network Design 502
438 E.7 Conclusion 504
Chapter 15 : ' Computer Techniques in Circuit Analysis
:
15.1 The Uses of Digital Computers in Circuit Analysis 438 Bibliography 505
15.2 Amplitude and Phase Subroutine 450
15.3 A Fortran Program for the Analysis of Ladder Name Index 509
Networks 453
15.4 Programs that Aid in Darlington Filter Synthesis 457 Subject Index 511

r
chapter I
Signals and systems

This book is an introduction to electric network theory. The first half

of the book is devoted to network analysis and the remainder to network
synthesis and design. What are network analysis and synthesis ? In a
generally accepted definition of network analysis and synthesis, there are
three key words : the excitation, the network , and the response as depicted
in Fig. 1.1. Network analysis is concerned with determining the response,
given the excitation and the network. In network synthesis, the problem
is to design the network given the excitation and the desired response.
In this chapter we will outline some of the problems to be encountered
in this book without going into the actual details of the problems. We
,
will also discuss some basic definitions.

.
l l SIGNAL ANALYSIS

For electric networks, the excitation and response are given in terms of
.
voltages and currents which are functions of time, t In general, these
functions of time are called signals. In describing signals, we use the two
—
universal languages of electrical engineering time and frequency. Strictly
speaking, a signal is a function of time. However, the signal can be
described equally well in terms of spectral or frequency information. As
between any two languages, such as French and German, translation is
needed to render information given in one language comprehensible in the

Excitation Response
Network

. ..
FIG l l The objects of our concern.
I
 Signals and systems 3
2 Network analysis and synthesis
|A
s( t )
Ao

o
/
t

t —* — — —1 —
CO CO 4 CO 3 0) 2 — 0)1 Ctfo oil 0)2
“
i
3 0> 4 +01

/ . .
FIG 1.4a Discrete amplitude spectrum .
,'H 3 6

FIG 1.2 . . Sinusoidal signal.

—————
0 )2 013 0> 4

, the translation is effected

by the CO 0) 4 0)3 0)2 0)1 |0 T + 0)
other . Between time and frequency . We shall
series , the Fourier integral , and the Laplace transform
Fourier and study these terms later in the
book.
have ample opportunity to define terms of . .
FIG 1.4b Discrete phase spectrum .
how a signal can be described in
At the moment, let us examine
the sinusoidal signal
both frequency and time. Consider let the angular frequency co be the independent variable . In this case, the
(1.1)
s( t ) —
A o sin (co0 t + 0O) signal is described in terms of A0 , co0 , and 0O, as shown in Fig. 1.3a, where
amplitude is plotted against frequency, and in Fig. l .3b , where phase shift
the angular
is the phase shift , and co0 is
where A a is the amplitude , 0O is plotted .
frequency as given by the equation Now suppose that the signal is made up of 2n + 1 sinusoidal components
217 • ( 1.2)
( 1.3)
-2 ^- + °)
s( t ) = 1 sin (“M i
i n
. The signal is plotted
against time
where T is the period of the sinusoid we The spectral description of the signal would then contain 2n + 1 lines at
description of the signal is obtained if
in Fig. 1.2. An equally complete ± co1, ± co2> • • • > ± <y„, as given in Figs. 1.4a and b . These discrete spectra
of amplitude A versus co and phase shift 6 versus co are sometimes called
„A line spectra . Consider the case when the number of these spectral lines
° —
become infinite and the intervals «j+1 ooi between the lines approach
I
Q-
zero . Then there is no longer any discrimination between one frequency
<
and another, so that the discrete line spectra fuse into a continuous spectra,
0 0)0
O)
as shown by the example in Figs. 1.5a and b . In the continuous case, the
Angular frequency sum in Eq . 1.3 becomes an integral
. .
FIG 1.3a Plot of amplitude A versus angular frequency co.
s( t ) = J A{ co ) sin [ cot + 0(a>)] dm (1.4)

' where A( ao ) is known as the amplitude spectrum and 0(co) as the phase
£ spectrum .
CO As we shall see later, periodic signals such as the sine wave in Fig. 1.2
o OX )
can be described in terms of discrete spectra through the use of Fourier
Angular frequency
. versus angular frequency co. series. On the other hand, a nonperiodic signal such as the triangular
FIG . 1.3b Plot of phase 6

’
i
Signals and systems 5
4 Network analysis and synthesis
|Afco)

—.
CO

.
FIG 1.5a Continuous amplitude spectrum.
CO

6( oi )

. .
FIG 1.7 Rotating phasor .
- CO o CO
when S(f ) is represented as a rotating phasor,1 as shown in Fig. 1.7.
The angular frequency co of the phasor can then be thought of as a velocity
at the end of the phasor. In particular the velocity co is always at right
angles to the phasor, as shown in Fig. 1.7. However, consider the general
. .
FIG 1.5 b Continuous phase spectrum .
case when the velocity is inclined at any arbitrary angle y> as given in
Figs. 1.8a and 1.86. In this case, if the velocity is given by the symbol s ,
pulse in Fig. 1.6 can only be described in terms of continuous spectra we see that s is composed of a component co at right angle to the phasor S
through the Fourier integral transform. as well as a component a, which is parallel to S. In Fig. 1.8a, s has a
—
component a toward the origin. As the phasor S spins in a counter-
clockwise fashion, the phasor decreases in amplitude. The resulting wave
1.2 COMPLEX FREQUENCY
for the real and imaginary parts of S( t ) are damped sinusoids as given by
In this section, we will consider the concept of complex frequency . As Re S(f ) = Ae at cos cof ~

we shall see, the complex frequency variable (1.7)

Im S( f ) = Ae at sin cot ~

s = a + jm ( 1.5) A,

is a generalized frequency variable whose real part a describes growth and o'
decay of the amplitudes of signals, and whose imaginary part jco is angular
frequency in the usual sense. The idea of complex frequency is developed
by examining the cisoidal signal 'S

S( f ) = Aeiat ( 1.6)
»( t )

( b)
. .
FIG 1.8 (a) Rotating phasor with exponentially decreasing amplitude. (6) Rotating
phasor with exponentially increasing amplitude.

0| T t 1A
phasor S is a complex number characterized by a magnitude and a phase angle
.
FIG. 1.6 Triangular signal.
I
:
(see Appendix C).
i

- -
 Signals and systems 7
6 Network analysis and synthesis
S(t )
Re S|
A

u: Envelope = Ae ~ ot

t
Ol
A

0 t
Im S
A
Envelope = Ae ~ at . .
FIG 1.11 Exponential signals.

which are shown in Fig. 1.9. Note that the damped sinusoid has an
exponential envelope decay, Ae at . In Fig. 1.86, the phasor is shown with
~

7 a positive real component of velocity + cr. Therefore, as the phasor spins,

0|
the amplitudes of the real and imaginary parts increase exponentially
with an envelope Ae+at , as shown by Im S(/ ) in Fig. 1.10 .
From this discussion, it is apparent that the generalized cisoidal signal
.. S(0 = AeH = Ae( a+ia>u (1.8)
.
FIG. 1.9 Damped sinusoids.
describes the growth and decay of the amplitudes in addition to angular
Im S( t ) / frequency in the usual sense. When a = 0, the sinusoid is undamped, and
/ Envelope = Aeat when jco = 0, the signal is an exponential signal

S( t ) = Ae**1 (1.9)
/
/
m asA.shown in Fig. 1.11. Finally, if a = jcv = 0, then the signal is a constant
Thus we see the versatility of a complex frequency description.
A
t 1.3 NETWORK ANALYSIS
- A^
0

As mentioned before, the characterization of the excitation and response

x , signals in time and frequency makes up only part of the analysis problem .
The other part consists of characterizing the network itself in terms of
i . \ -
x
time and frequency, and determining how the network behaves as a signal
processer. Let us turn our attention now to a brief study of the properties
\
\ of linear networks and the general characteristics of signal processing by
1 a linear system.
FIG 1.10. . Exponentially increasing sinusoid.

BL X.

Signals and systems 9

8 Network analysis and synthesis e( t )
r( t ) I

BASIC DEFINITIONS
and
if (a) the principle of superposition
Linear A
B
A system (network) is linear hold.
System

(6) the principle of proportionality , ( ), ( t )] and

if, for a given network, [ex i rx were 0l t 7
By the superposition principle excitation
[e2( t), r2( t )] are excitation-
response pairs, then if the r2( t ) By the , .
would be r( t ) = r ( t ) +
e( t ) = ex + 2
( t ) e ( t ) , the response
if the excitation were Cxex
( t ), where Cx is a e( t - Till r( t - TO \
proportionality principle ,
would be Cjr t ),i.e., the constant
of propor -
constant , then the response
tionality Cx is preserved by the
^
linear network. The two conditions
of A
System B
linearity are summarized in Fig
. 1.12. response
definition of a linear network is that the excitation and . We shall T\ T + Ti t Ti t
Another by a linear differential equation
of the network are related .
Chapter 4 on differential equations .
FIG 1.13. Time-invariant system .
discuss this definition in
Passive the network is excitation and response remains the same. Thus must be true for any
2 if (a) the energy delivered to
network is passive choice of points of excitation and response.
A linear , and (6) if no voltages or currents
nonnegative for any arbitrary excitation before an excitation is applied
. Causal
appear between any two terminals We say a system is causal if its response is nonanticipatory, i.e., if
Reciprocal of excitation and
A network is said to be
reciprocal if when the points
relationship between
e( t ) =0 t<T
(1.10)
interchanged, the
measurement of response are then r( t ) = 0 t<T
In other words, a system is causal if before an excitation is applied at
Ci e\( t ) System
Ciri( t ) t = T , the respdnse is zero for oo < t < T. —
Time invariant
A system is time invariant if e( t ) -> r( t ) implies that e( t ± T ) -* r{ t ± T ),
i —
where the symbol > means “gives rise to.” To understand the concept
of time invariance in a linear system, let us suppose that initially the
C 2 r2 h )
C 2 e2 ( t ) System excitation is introduced at / = 0, which gives rise to a response r( t ) If .
the excitation were introduced at t = T , and if the shape of the response
waveform were the same as in the first case, but delayed by a time T
(Fig. 1.13), then we could say the system is time invariant . Another way
of looking at this concept is through the fact that time invariant systems -
Ciei(t) + C2 e2(0 System
Cm( t ) 4 (W0 - contain only elements that do not vary with time. It should be mentioned
here that linear systems need not be time invariant.

FIG. 1.12. Linear system . Derivative property

i From the time-invariant property we can show that, if e( t ) at the input
in Terms of Time and gives rise to r( t ) at the output (Fig. 1.14), then, if the input were e' ( t ),
of Passive Linear Networks
2G. Raisbeck, “A Definition 1954), 1510 1514. -
. (Dec.
.
Energy,” J. Appl Phys , 25

Li i
 10 Network analysis and synthesis Signals and systems 11
eft )
Amplifier Ke( t )
i

.
FIG 1.15. Amplifier.

deft ) dr( t )
System f( t )
Kdfio
IF
FIG. 1.16 . Differentiator.
d 2 e( t ) d 2 r( t )
System 1S
dt 2
I' f M d r
.
FIG. 1.17 Integrator .
J‘e( ) dr-
T System j‘r( r ) dr
0

. D( T )
f( t -T )
FIG 1.14 Some implications of linear time-invariant systems
. .
i.e , the derivative of e( t ), the response would be r ( t .
' ) The proof is quite
. ) where e is a real quantity. By the
Time-delay network.
simple. Consider an excitation e( t + e
time-invariant property, the response would be r( t + e ). Now suppose the
excitation were
«i(0 = [ e( t + «) - e( t )J
“
(1- 11)

, the response
then according to the linearity and time-invariant properties
would be
-
(1.12) £ T 2
ri(0 = - [r( t + ) r( 0]
. .
FIG 1.19 Excitation function.
Taking the limit as e -* 0, we see that 1. Amplifier: An amplifier scales up the
r( t ) = Ke( t ), where K is a constant (Fig. 1.15magnitude of the input, i.e.,
lim Ci(0 = ~ j e( 0 1 ).
(1.13) 2. Differentiator: The input signal is
or down (Fig. 1.16).
differentiated and possibly scaled
lim ra(0 = j r( t ) - |£ up
*o at- r
¥> 3. Integrator: The output is the
integral of the input, as shown in
We can extend this idea to higher derivatives as well as for
the integrals of Fig. 1.17.
e{ t ) and r( t ), as shown in Fig. 1.14.
& .
4 Time delayer: The output is
same wave shape as the input (Fig. 1.18). by an amount T, but retains the
1 delayed
-
j
Ideal models
Let us now examine some idealized models of linear
systems. The Suppose we take the triangular pulse in Fig. 1.19 as
make them very Then the outputs for each of the four systems the input signal.
systems given in the following all have properties which just described are shown
useful in signal processing. in Figs. 1.20a-l.20
^.

Signals and systems

12 Network analysis and synthesis 13
r( t )

T w . .
FIG 1.22a Voltage source. . .
FIG 1.22b Current source.

indicated for the voltage source and the direction

of flow for the current
source are arbitrarily assumed for reference
(b)
purposes
voltage source is an energy source that provides, at only. An ideal
a given port, a voltage
signal that is independent of the current at that port.
If we interchange the
words “current” and “voltage” in the last
definition , we then define an
rff) i ideal current source.
In network analysis, the principal problem is to
that exist between the currents and voltages at
find the relationships
the ports of the network.
Certain simple voltage-current relationships for the
network elements also
serve as defining equations for the elements
when the currents and voltages are expressed as
themselves . For example,
functions
R, L, and C elements, shown in Fig. 1.23, are defined of time, then the
by the equations
0 i
v( t ) = Ri( t ) or i(0 = ~R v( t )
output , (d)
. Id) Amplifier output ( b) Differentiator output. (c) Integrator
FIG. 1.20
,

Delayed output. v( t ) = L
dt
or l( 0 =
J2 v( x ) dx + !(0) (1.14)

—
0

Ideal elements
, we use idealized linear mathematical v(t) = I
In the analysis of electric networks
elements most often encountered
models of physical circuit elements. Thecapacitor C, given in farads, and
C vo
i( x ) dx + v( 0 ) or i( 0 = cd-
^ dt
are the resistor R, given in ohms, theThe endpoints of the elements are where the constants of integration /(0) and i;(0)
are initial conditions to be
the inductor L, expressed in henrys
. discussed in detail later.
called terminals. A port is defined as any pair of two terminals into which Expressed as a function of the complex frequency
network variables may be variable s, the equations
energy is supplied or withdrawn or where an example of a two port -
measured or observed. In Fig. 1.21 we have
network. the excitation functions are ideal I
C
h°
Energy
source
£ .
Two-port
network

. -
FIG 1.21 Two port network.
Response
measurement

. .
FIG 1.23 ( a) Resistor. (b ) Inductor, (c)
Capacitor.

m
 Signals and systems 15
14 Network analysis and synthesis
I( s ) +
CK
I( s ) I( s ) o >- I( s )

m V( s )
V(s) b'
V( s ) R V( s ) sL V( s ) ~ sC l’ Z -
.
FIG 1.25. Driving point impedance
- .
FIG. 1.26 Black box.
- Z( s) = R.
o - 5- o

( a) (b) <0
, (c) Capacitor.
system function
FIG. 1.24. (a) Resistor. (6) Inductor
H( s ) = (1-17)
C elements, shown in Fig. 1.24, are
(ignoring E( s )
defining the R, L, and
initial conditions for the moment) Since R( s) and E( s ) are voltages or currents, then H( s) is denoted generally
1 as an immittance if R( s ) is a voltage and E( s ) is a current, or vice versa.
F(s) = RI ( s ) or I( s) =|V( s) -
A driving point immittance3 is defined to be a function for which the
variables are measured at the same port. Thus a driving point impedance -
Z( s ) at a given port is the function
K(s) = sLI ( s ) or J(s) =1
. F(s) -
(1 15)

F(s) =
^
1
sC,
/(s) or
sL

7(s) = sCK(s)
Z(s) =
^I( s )
where the excitation is a current I( s) and the response is a voltage V(s), as
.
shown in Fig. 1.25 When we interchange the words “current” and
(1.18)

the independent variable is t , “voltage” in the last definition, we then have a driving point admittance. -
We see that in the time domain, i.e., where
the voltage-current relationships are given
in terms of differential equations. -
An example of a driving point impedance is the network in Fig. 1.25,
domain , the voltage-current
On the other hand, in the complex-frequency in algebraic equations.
where
relationships for the elements are expressed solved than differential
Algebraic equations are, in most cases, more
easily
signals and networks
Z( s ) =
I( s )
=R ^ (1.19)

equations. Herein lies the raison d' etre for describing Now suppose the resistor in Fig 1.25 were enclosed in a “ black box.”
.
the time domain. We have no access to this black box, except at the terminals 1-1' in Fig.
in the frequency domain as well as in
When a network is made up of an interconnection of linear circuit
- .
1.26 Our task is to determine the network in the black box. Suppose we
system or transfer function H( s) . are given the information that, for a given excitation I( s ) , the voltage
elements, the network is described by its
( s ) are related by the equation
The response R( s) and the excitation E response K(s) is proportional to I( s) by the equation
R( s ) = H( s ) E( s ).
( 1.16) V( s ) = KJ( s ) (1.20)

( ) , and we can obtain H( s ) directly An obvious solution, though not unique, is that the network consists of a
In network analysis, we are given E s resistor of value R = K £7. Suppose next that the excitation is a voltage
R( s ).
from the network. Our task is to determine V( s ), the response is a current I( s ) , and that

1.4 NETWORK SYNTHESIS y(s) = M = 3 + 45 (1.21)

F(s)
problems germane to network
We will now briefly introduce some of the the IRE Standards on Circuits “Linear Passive Networks,” Proc IRE , 48, No 9 . .
. In network synthesis, we are given the response R( s ) and
3

synthesis synthesize the network from the (Sept. I 960), 1608-1610.

excitation E( s), and we are required to
I

1 tL >
Signals and systems 17
16 Network analysis and synthesis
h( s ) \ H< jo> )\
h( s - O
lo-
Y( s ) -
v L +
Vi(s)
.Two-port V 2( s )

l'o T J
network

FIG .1.27. Network realiza - . .

FIG 1.28 Two- port network .
tion for Y( s) .
0 uc V}
Our task is to synthesize a network equivalent to the network in the black . .
FIG 1.29 Ideal amplitude spectrum for low pass filter. -
box From a close scrutiny of the driving- point admittance Y( s ), we see
.
that a possible solution might consist of a resistor of J Q in parallel with As for driving-point functions, there are certain properties which a
a capacitor of 4 farads, as seen in Fig. 1.27. transfer function must satisfy in order to be realizable. We shall study
The problem of driving-point synthesis, as shown from the examples these realizability conditions and then proceed to the synthesis of some
just given, consists of decomposing a given immittance function into simple transfer functions.
basic recognizable parts (such as 3 + 4J). Before we proceed with the The most important aspect of transfer function synthesis is filter design.
mechanics of decomposition, we must first determine whether the function A filter is defined as a network which passes a certain portion of a fre-
is realizable, i.e., can it be synthesized in terms of positive resistances, quency spectrum and blocks the remainder of the spectrum By the term .
inductances, and capacitances ? It will be shown that realizable driving- “blocking,” we imply that the magnitude response \H( j(d)\ of the filter is
,
point immittances belong to a class of functions known as positive real or approximately zero for that frequency range. Thus, an ideal low pass -
simply, p.r. functions. From the properties of p. r . functions , we can test a filter is a network which passes all frequencies up to a cutoff frequency
given driving point function for realizability. (The Appendices present a
- coc, and blocks all frequencies above wc, as shown in Fig. 1.29.
short introduction to complex variables as well as the proofs of some One aspect of filter design is to synthesize the network from the transfer
theorems on positive real functions.) With a knowledge of p. r. functions
, function H ( s ). The other aspect deals with the problem of obtaining a
we then go on to examine special driving- point functions . These include realizable transmittance H( s) given the specification of, for example, the
functions which can be realized with two kinds of elements only the L C , — - magnitude characteristic in Fig. 1.29. This part of the synthesis is generally
- -
R C, and R L immittances.
Next we proceed to the synthesis of transfer functions. According to
referred to as the approximation problem. Why the word “approxi-
mation ? ” Because frequency response characteristics of the R, L, and C
the IRE Standards on passive linear networks, a transfer function or
4

transmittance is a system function for which the variables are measured at

different ports. There are many different forms which a transfer function
might take. For example, consider the two-port network in Fig 1.28 . .
•
_ elements are continuous (with the exception of isolated points called
resonance points), a network containing these elements cannot be made to
cut off abruptly at a)c in Fig. 1.29. Instead, we can realize low-pass filters

If the excitation is Ifs ) and the response V2( s ), the transfer function is a
- which have the magnitude characteristics of Fig. 1.30. In connection with

transfer impedance \ H( j<o)l

h( s )
(1.22)
m
On the other hand , if KX(J) were the excitation and V f s ) the response, then
-
we would have a voltage ratio transfer function
(1.23)
Vi(*) toc 03 0 03C 03

. .
FIG 1.30 Realizable low-pass filter characteristics.
* Loc. cit.

:
 Signals and systems 19
18 Network analysis and synthesis
the filter design problems, we will discuss
certain problems in magnitude find the response R( s ) if the excitation is
designing a filter, we deal with
and frequency normalization so that, in instead of “ practical”
element values such as /? = 0.5 ohmandC =
2 farads s
element values of, for example, R =
(pico = 10
500,000 ohms and C 2 picofarads
~12). Also we will study a method whereby low-pass
—
filter 1.5 Given the driving-point functions find their simplest network realiz-
ations.
, band-pass, and band -
designs might be transformed into high-pass this method is called 1
elimination filters. The mathematical basis
of ( fl ) Z( s ) = 3 + 2s + -
frequency transformation.
We next discuss some aspects of analysis
excitation and response functions are given in
and synthesis in which the
terms of power rather than
(b)
*) - * + JT_2
.
of voltage and current We will examine -
the power transfer properties of
, -
linear networks, using scattering parameters

Finally, in Chapter 15, we will examine

, which describe the incident
and reflected power of the network at its ports
some
.
of the many uses of high-
design. In addition to a
speed digital computers in circuit analysis and some specific computer
( c)

(<0
Z( ) =3+

=
^ 2

37T 2 + 7T 4
general survey of the field, we will also study 1.6 For the network shown, write the mesh equation in terms of (a) differential
programs in circuit analysis. equations and (b) the complex-frequency variable s.
it L
Problems A/W —nmr' ,
:
1.1 Draw the line spectra for the signal

s( t ) = 3 sin (' + ?) + ( i) + 6
4 sin 2/ ~ sin 3/ 6
•
into a sampler that closes every
1.2 Find the response to the excitation sin t response for 0 t 2 IT.
KIT14 seconds where K = 0, 1 2 , Draw the
PROB. 1.6
in the figure when the network
1.3 Find the response to the excitation shown .
is (a) an ideal differentiator; (b) an ideal integrator 1.7 For the network shown, write the node equation in terms of (a) differential
: equations and (b) complex-frequency form.
: e( t )

i
I i R

0 1 1.5 2 t
PROB. 1.3 I PROB. 1.7

1.4 If the system function of a network is given

as I 1.8 Suppose the response of a linear system to an excitation e( t ) were
1 4B KO - 3e~**. What would the response be to an excitation of e( t 2) ? —
i H ( s ) = ( s 2)(J 3)
+ +

r:
:llii Signals and waveforms 21
s( t ) \

— 2T \
^T\ 0 T \ 2T \ 3T
~
4 T \ 5T 7

chapter 2 .
FIG. 2.1 Square wave.

t.
Signals and waveforms where T is the period of the signal. The sine wave, sin t , is periodic with
period T = 2TT . Another example of a periodic signal is the square wave
given in Fig. 2.1. On the other hand, the signals given in Fig. 2.2 are
aperiodic, because the pulse patterns do not repeat after a certain finite
r- interval T. Alternatively, these signals may be considered “periodic” with
an infinite period .
Our main concern in this chapter is the characterization of signals as Next, consider the symmetry properties of a signal. The key adjectives
functions of time. In previous studies we have dealt with d -c signals
.
here are even and odd A signal function can be even or odd or neither.
that were constant with time, or a-c signals which were sinusoids
of An even function obeys the relation
-
constant amplitude, such as s( t ) = A sin ( cot -| 6 ). In engineerin g practice, !
in scope than s( t ) = s( -t ) (2.2)
the class of signals encountered is substantially broader
simple a-c or d -c signals. To attempt to characterize each member of For an odd function s( t ) = -s( -t ) (2.3)
the class is foolhardy in view of the almost infinite variety of signals ,
encountered. Instead, we will deal only with those signals that can be For example, the function sin t is odd, whereas cos t is even. The square
characterized in simple mathematical terms and which serve as building l pulse in Fig. 2.2a is even, whereas the triangular pulse is odd (Fig. 2.2b ).
blocks for a large number of other signals. We will concentrate on formu - Observe that a signal need not be even or odd . Two examples of signals
lating analytical tools to aid us in describing signals, rather than deal with of this =type are shown in Figs. 2.3a and 2.4a. It is significant to note,
the representation of specific signals. Because of time and space limita- however, that a'ny signal s( t ) can be resolved into an even component se( t )
,
tions, we will cover only signals which do not exhibit random behavior and an odd component s0( t) such that
i.e., signals which can be explicitly characteriz ed as functions of time .
These signals are often referred to as deterministic signals. Let us first s( t ) = Se( t ) + S0( t ) (2.4)
discuss certain qualitative aspects of signals in general.
For example, the signals in Figs. 2.3a and 2.4a can be decomposed into
2.1 GENERAL CHARACTERISTICS OF SIGNALS
odd and even components, as indicated in Figs. 2.3b, 2.3c, 2.4b, and 2.4c .
In this section we will examine certain behavior patterns of signals. 7 s( t )
,
Once these patterns are established , signals can be classified accordingly s( t )
and some simplifications result. The adjectives which give a general
qualitative description of a signal are periodic , symmetrical, and continuous. :
7
/1
Let us discuss these terms in the given order.
. T
rT i t

First, signals are either periodic or aperiodic. If a signal is periodic, then (a) ( b)
it is described by the equation
s( t ) = s( t ± kT ) k = 0, 1, 2, . . . (2.1) . .
FIG 2.2 (a) Even function (6) Odd function. .
i
! 20

m
Hi! 1
 Signals and waveforms 23
22 Network analysis and synthesis
sft ) sft ) sf -t) I
s( t ) \
1
A 0, t
1
-1 0 t

(a)
fa ) 0
fa )
1 t 0
(b)
^ t

sM \
Se( t ) soft )

se( t )
i
il- t £ I
:
-1 0 i t 0 1 t 0 1 t

( b)
( b) -i

fc ) (d )
Soft ) I
soft ) .
FIG 2.5. Decomposition into odd and even components from sft ) and

* *ol1 Now let us turn our attention to the continuity property of signals.
0| t
"
7
t
I Consider the signal shown in Fig. 2.6. At t T , the signal is discontinuous .
The height of the discontinuity is
—
fc )
f (T + ) f ( T — ) = A — (2.7)
(c )
i
. .
FIG 2.3 Decomposition into . .
FIG 2.4 Decomposition into where /(T + ) = lim /(T + e)
odd and even components , even and odd components. e- 0 »

fa ) Unit step function , fb ) (2.8)

(a) Original function fb )
— e)
f ( T ~ ) = lim /(T
-
,

Even part (c) Odd part .

, Even part of unit step , fc ) 0
Odd part of unit step.
and « is a real positive quantity. In particular, we are concerned with
From Eq . 2.4 we observe that discontinuities in the neighborhood of / = 0. From Eq . 2.8, the points
s(-0 = s.( -t ) + —
S0( t)
(2.5)
&
&
fft )
= S,( t ) - So ) 0
as
Consequently, the odd and even parts of the signal can be expressed w
s,( t ) = + -Q ] mo *- (2.6)
r 3:
A

s0( t ) = two - «( /)]

Consider the signal s( t ), shown in Fig . 2.5a. The function s( t ) is equal
We then
— Ol T t
to s( t ) reflected about the 1 = 0 axis and is given in Fig. 2.5A.
obtain s,(t ) and s0( t ) as shown in Figs. 2.5c and d , respectiv ely . . .
FIG 2.6 Signal with discontinuity.

II i
Signals and waveforms 25
24 Network analysis and synthesis
sft )

0 rT T2 t

FIG 2.7 . . Signal with two discontinuities.

/(0+) and /(0 — ) are !

/(0+ ) = lim /(«) i-
-0 »

(2.9) . .
FIG 2.8 Normalized curve for time constant T = 1 .
/(0 —) = lim / ( — e)
-* 0 Observe that the larger the time constant, the longer it requires for the
Fig. 2.7 has two discontinuities, at Tx and
I waveform to reach 37 % of its peak value. In circuit analysis, common
For example, the square pulse in time constants are the factors RC and RjL.
r2. The height of the discontinuity at 7\ is
RMS Value
KTi + ) - s(7\ ) = K — (2.10)
The rms or root mean square value of a periodic waveform e( t ) is defined
— K. as
[ />>*]
Similarly, the height of the discontinuity at T2

2.2 GENERA L DESCRI PTIONS

OF SIGNALS
is
firms
-}
where T is the period. If the waveform is not periodic, the term rms does
Vi
(2.14)

not apply. As an example, let us calculate the rms voltage for the periodic
descriptions of signals .
In this section we consider various time domain - waveform in Fig. 2.9.
the following terms : time
In particular, we examine the meanings of
constant , rms value, d- c value , duty cycle , and crest
factor . The term , time
; the remaining four
-a firms
constant , refers only to exponential waveforms
terms describe only periodic waveforms. 4AV t>2 IT | H
T 2
3 o
+ A' l IT/ I
Time constant

-(f r
know how quickly a (2.15)
In many physical problems, it is important to
Im
the decay of an exponential is the
waveform decays. A useful measure of
described by *
time constant T . Consider the exponential waveform
I = 72/ 3 A v
r( t ) = Ke~‘ IT u( t ) (2.11) M -
D C Value
t T, The d-c value of a waveform has meaning only when the waveform is
From a plot of r( t ) in Fig. 2.8, we see that when = periodic . It is the average value of the waveform over one period
r( T ) = 0.37r(0) (2.12)

Also r( 4T ) = 0.02r(0) (2.13) fid - c

iJ/0* (2.16)

l il
 Signals and waveforms 27
26 Network analysis and synthesis
I <40

A
epp I i
w w ' 2T t

T [ «— —
t eb

-A
FIG. 2.11 . Periodic waveform with zero d-c and small duty cycle.

We see that the smaller the duty cycle, the smaller the rms voltage. The
square wave in Fig. 2.1 has a 50 % duty cycle.
FIG . 2.9. Periodic waveform.
rm in Crest factor
d-c value, whereas the wavefo Crest factor1 is defined as the ratio of the peak voltage of a periodic
The square wave in Fig. 2.1 has zero \
Fig. 2.9 has a d-c value of waveform to the rms value (with the d-c component removed). Explicitly,
for any waveform with zero d-c such as the one shown in Fig. 2.11—crest
ir - —2 J1 =
"
(2.17)
fid- c =
TL 4 ^ 4 factor, CF, is defined as
2- or
_
fi 6
( 2.20)
Duty cycle the CF =~
as the ratio of the time duration of firms firms
The term duty cycle , D, is defined is
rm to the period, T , that ,
positive cycle t0 of a periodic wavefo whichever is greater. For the waveform in Fig. 2.11 , the peak -to-peak
(2.18) voltage is defined as
D= —T • fijin fi0 -F fi& (2.21)

important in dealing with wave- Since the waveform has zero d-c value
The duty cycle of a pulse train becomes R
, where most of the energy is con-
! forms of the type shown in Fig. 2.10 . The rms voltage of the waveform eJo = eb( T - t0 ) (2.22)
I centrated in a narrow pulse of width t0
in Fig. 2.10 is Also , ft = evvD (2.23)

i firms
( 2.19)
and .
« = ePP( l - D) (2.24)

= Aft 0IT 1/ The rms value of the waveform is

= AfD - Df t0 + e j D\T - <„) \ *
firms
_
; A A T (2.25)

'0 M = e, jD( 1 - D)
t 1 , G- Justice, “The Significance of Crest Factor,” Hewlett-Packard Journal
, 15, No. 5
T
FIG . 2.10. Periodic waveform with small
duty cycle. H 1

BBT (Jan., 1964), 4-5.

1 Signals and waveforms 29

28 Network analysis and synthesis s( t ) \
Since crest factor CF = ejerms, we have S(t ) I

CF =
fi-np( l ~
e , jD( l -
D)
D) (2.26)
u (t
— a) 4

= I
For example, if D =
1
—, w t 0] T t
FIG . 2.14. Shifted step function. FIG . 2.15. Square pulse.
CF “
Vi s - 1 (2.27)
argument ( t ) within the parentheses is negative, and is unity when the
= 7100 - 1 ~ 10 argument (t ) is greater than zero. Thus the function u{ t — a ) , where
1 a > 0, is defined by
If D
10,000 ’ (2.28)
u( t a ) = 0
=1
t<a
t> a
— (2.31)
CF = V 10,000 — 100
to read accurately rms values of and is shown in Fig. 2.14. Note that the jump discontinuity of the step
A voltmeter with high crest factor is able
, in particular, signals with occurs when the argument within the parentheses is zero. This forms the
signals whose waveforms differ from sinusoids of crest factor occurs for basis of the shifting property of the step function. Also, the height of the
low duty factor. Note that the smallest
value
of D , that is , Dmax = 0.5, jump discontinuity of the step can be scaled up or down by the multiplica-
the maximum value tion of a constant K .
CFmin = V 1 / 2W - 1 (2.29) With the use of the change of amplitude and the shifting properties of
the step function, we can proceed to construct a family of pulse waveforms.
=1 For example, the square pulse in Fig. 2.15 can be constructed by the sum
of two step functions
ASSOCIATED WAVEFORMS
2.3 THE STEP FUNCTION AND
2.12 is defined as
; s( t ) = 4u( t - 1 ) + ( - 4) u( t - 2) (2.32)
The unit step function u( t ) shown in Fig.
u( t ) = 0 /<0 K as given in Fig. 2.16 . The “staircase” function, shown in Fig. 2.17, is
I (2.30) characterized by the equation
= 1 t 0
corresponds to a switch S , s( t ) = Iu( t - IcT ) ( 2.33)
The physical analogy of a unit step excitation
which closes at t = 0 and connects a d-c battery
of 1 volt to a given circuit, -
Ar O

step is zero whenever the 8(0 \

as shown in Fig. 2.13. Note that the unit V 4|
<401

\
\ 1v -
+
=-
Switch
Network
0
^ t

-T -41
0 t FIG . 2.16 . Construction of square pulse by step function .
FIG. 2.12. Unit step function. FIG . 2.13. Network analog of unit step.

!
I
tilli ! mm .
 M Signals and waveforms 31
30 Network analysis and synthesis
s( t )
s( t )

3
2 . -! - <
' T ) 1

t
ol T 2T
. .
FIG 2.17 Staircase function.
0] 2T 3T t
FIG. 2.19. Sine pulse.
Fig. 2.1. Using the shifting
Finally, let us construct the square wave in
by (for / 0)
property, we see that the square wave is given
- 2 ( / 3T ) + - (2.34)
- - I Thus the square wave in Fig. 2.1 is simply expressed as
j(/)
= u( t ) 2u( t T ) + 2u( t 2T ) « using the property
- \
A simpler way to represent the square
that the step function is zero whenever its
wave is by
argumen t is negative. Restricting s( l) = sgn (* ?) ( 2.38)

ourselves to the interval t > 0, the function

Returning to the shifting property of the step function, we see that the
/ . vt
« sm -s(v.l,) = 1
seen by the waveform in Fig.
—
(2.35)
T
single sine pulse in Fig. 2.19 can be represented as

is zero whenever sin ( nt/ T ) is negative, as

Fig. 2.1 can be represented
s(f ) = sin ~ [u ( f - 2T ) - u( t - 3T)J (2.39)
2.18. It is now apparent that the square wave in
as (2.36) The step function is also extremely useful in representing the shifted or
s( t ) & delayed version of any given signal. For example, consider the unit ramp
- £ function
Another method of describing the square wave is to consider a generaliz p{ t ) = t u( t ) (2.40)
i ; (pronounced signum).
ation of the step function known as the sgn function
The sgn function is defined as E shown in Fig. 2.20. Suppose the ramp is delayed by an amount t = a, as
% shown in Fig. 2.21. How do we represent the delayed version of ramp ?
l!
sgn [/(f )] = 1 /(0 > o
= 0 f { i) = o (2.37) K '

IS = -l /(0 < o I $ s( t ) \

sfol
:I 1
i

1 I

2T 3T 4T t
I
.
FIG 2.20. Ramp function with zero time shift.
. .
FIG 2.18 The signal u(sin nt / T ).
: .

, iil
Signals and waveforms 33
32 Network analysis and synthesis
s( t ) \
s(t )
2

1 1

0| 1 t
a +1 t
0 0
. .
FIG 2.23 Triangular pulse .
FIG. 2.21. Ramp function with time shift = a.
As a final example, consider the waveform in Fig. 2.23, whose corn -
First, let us replace the variable t by a new variable t
' =t —. a Then ponent parts are given in Fig. 2.24. For increasing /, the first nonzero
P( t' ) = t' u( 0
(2.41) — —
component is the function 2(f 1) u{ t 1), which represents the straight
line of slope 2 at t = 1. At t = 2, the rise of the straight line is to be
When />( / ') is plotted against the resulting curve is identical to the plot
a t' in p( t' ), — arrested , so weadd to the firstcomponenta term equal to 2(f 2 ) u( t 2) — — —
— ——
of s( t ) versus t in Fig. 2.20. If, however, we substitute t = with a slope of 2. The sum is then a constant equal to 2. We then add a
we then have term 2u( t 2) to bring the level down to zero. Thus,
p( t' ) = ( f a ) u( t a ) - -
(2.42)

p{ t ) shown in
s( t ) = 2( t - 1) u( t - 1) - 2( - 2) u( t - 2 ) - 2w(f - 2)
f (2.45)
When we plot p(f ') against f , we have the delayed version of
Fig. 2.21. ( t ) u( l ) is 2.4 THE UNIT IMPUL SE
From the preceding discussion, it is clear that if any signal f
,

delayed by a time T , the delayed or shifted signal is given by

The unit impulse, or delta function, is a mathematical anomaly. P. A. M.
/(f ') = /(f T ) u( t T ) —
For example, let us delay the function (sin n-f / J) u( t) by a period
(2.43)
T . Then
— Dirac first used it in his writings on quantum mechanics.2 He defined the

the delayed function s( t' ), shown in Fig. 2.22, is s(t )\

h - 1) u(t - 1)
[ ] ( 2.44)

s( t' )

1
s( f') = sin

^ (f - T) « ( f - T)

1 t

-2u(t - 2 ) - -2 ) u( t 2)

; t

-1
m . .
FIG 2.24 Decomposition of the
\
triangular pulse in Fig. 2.23.

P. A. M. Dirac, The Principles of Quantum Mechanics Oxford University Press, 1930.

.
FIG 2.22. Shifted sine wave.

.
;

m
it
 5}
Signals and waveforms 35
34 Network analysis and synthesis
delta function d( t ) by the equations h
f—
" 5( dt 1
< 0 =
J 00
(2.46)
(2.47 )
<5( /) = 0 for t 9& 0
Its most important property is the sifting properly , expressed
by

J “ ( 5( d t
/ < 0 o =m
a nonrigorous
( 2.48)

In this section we will examine the unit impulse from refer to

approach. Those who prefer a rigorous treatment should in that 63 *«',(t>
material
Appendix B for development of this discussion. The
d functions originated by
of generalize
*« ,w
the theory 1
appendix is based on 62
G. Temple.3 In Appendix B it is shown that the unit
impulse is the
derivative of the unit step <5(0 = u\t )
(2.49) <T .
i * y»
At first glance this statement is doubtful . After all , the
derivative of the
unit step is zero everywhere except at the jump discontinu ity, and it does
not even exist at that point ! However, consider the function gft ) in Fig.
a unit step, that is, «i «3 62 t
2.25. It is clear that as e goes to zero, gc( t ) approaches
£l

FIG . 2.27. The sequence { gu{ t ) }.

lim ge( t ) = w (0 (2.50)
'
-* 0
by the values of eit as shown in Fig . 2.27. The sequence has the following
Taking the derivative of gc( 0 > we obtain g' c( t ), which is defined property :
equations
g\( 0 = - ;
1
0 t (2.51) lim f h> 0

-o Jti
g'</0 dt =1 (2.52)
if :
= 0; t < o, t >e
<( <o
where tx and /2 are arbitrary real numbers. For every nonzero value of e,
as shown in Fig. 2.26. Now let e take on a sequence of values e t such that there corresponds a well- behaved function (i .e., it does not “ blow up”)
decreasing g' tft ). As e{ approaches zero,
ei > ei+i - Consider the sequence of functions {g'tj(0} f°r
g' d o + ) co (2.53)
11 g( U )
g\( D
-
t< * 0

!
i
I so that the limit of the sequence is not defined in the classical sense.
l Another sequence of functions which obeys the property given in Eq .
f 2.52 is the sequence {/,/0} in Fig- 2.28. We now define the unit impulse
<5(/) as the class of all sequences of functions which obey Eq. 2.52. In
o t mSt particular, we define
0 -H t
Ch > o rtt > o
FIG. 2.25. Unit step when e * 0 - . FIG . 2.26. Derivative of ge( t ) in
Fig. 2.25. Jti <0
<5(0 d / = lim
i - 0 Jti < 0
>
g' ti{ i) dt
( 2.54)
*2 > 0
!
: 3
G. Temple, “The Theory of Generalized Functions,” Proc Royal
Society , A, 228, . = lim f <0
Ut ) dt
1955, 175-190.

Signals and waveforms 37

36 Network analysis and synthesis
Continuing with this heuristic treatment , we say that the area “under”
I
the impulse is unity, and, since the impulse is zero for / 0, we have
^
J ”
<5(0 dt =
°+d )
J
(l dt = 1 (2.56)

Thus the entire area of the impulse is “concentrated” at t = 0. Con-

sequently, any integral that does not integrate through / = 0 is zero,
r o- r+ oo
as seen by
J <5(0 dt = <5(0 dt =0
(2.57)
m|
s( t ) = Au( t
— a)

The change of scale and time shift properties A

discussed earlier also apply for the impulse
function. The derivative of a step function o

—n l
s( t ) = A u( t
yields an impulse function
— a) (2.58)

s’( t ) |
s'(t) = A d( t a) - (2.59) \A
which is shown . in Fig . 2.29. Graphically, we
represent an impulse function by an arrow- o t
head pointing upward, with the constant
multiplier A written next to the arrowhead.
s’( t ) = A 6 ( t a )
FIG . 2.29
—
Note that A is the area under the impulse
k -
A d( t a ).
Consider the implications of Eqs. 2.58 and 2.59 . From these equations
r we see that the derivative of the step at the jump discontinuity of height A
i yields an impulse of area A at that same point t T. Generalizing on this
argument, consider any function /(/ ) with a jump discontinuity at t = T.
—
Then the derivative, /'(/) must have an impulse at t = T. As an example,
-
consider /( / ) in Fig . 2.30. At t = T, f ( t ) has a discontinuity of height A ,
- £i
—-
«; en a + 61
6/i

FIG . 2.28. The sequence {/ti(01

t
m
definition (which, as
I! It should be stressed that this is not a rigorousmerely a heuristic one.
stated previously, is found in Appendix B) but the delta “function” as
of
From the previous definition we can think
having the additional properties ,
of T t
<5(0) = oo (2.55)
FIG . 2.30. Function with discontinuity at T.
<5(f ) =0 for t 9* 0

1 m
 38 Network analysis and synthesis
Signals and waveforms 39
which is given as As a second example, consider the
function git ) shown in Fig. 2.32a.
A = /(r+ ) - f i T - ) (2.60) We obtain g'( 0 by inspection, and note
that the discontinuity at t 1
produces the impulse in g\t) of area =
Let us define fft ) as being equal to /(f) for t < T , and having the
same shape as fit ), but without the discontinuity for t > T , that is,

m = /(0 - A u( t - T ) (2.61)
as .given in Fig. 2.32b.
S(l +) £(! ) = 2,
Another interesting property of the impulse
integral
- (2.66)

function is expressed by the

-
The derivative fit ) is then
fit ) = f\it ) + A Sit - T) (2.62) •/
This integral is easily evaluated if we
r+eo
— CO
fit ) 8( t
— T ) dt
(2.67)
consider that <5(r - T ) = 0 for all
= f (T )
The following example illustrates this point more clearly. In Fig. 2.31a,
the function fit ) is ^ .
t T Therefore, the product
f i t ) Sit - D = 0 all t
fit ) = A u( t a ) A uit b )- - -
(2.63)
Iffit ) is single-valued at r = T, fiT ) can be
T (2.68)
factored from the integral so
^
Its derivative is fit ) = Adit - a ) - A d( t - b ) (2.64) that we obtain

and is shown in Fig. 2.31*. Since f t ) has two discontinuities, at t = a and

l
fiT )
J oof—
"
dit - T ) dt = fiT ) (2.69)
t = b , its derivative must have impulses at those points. The coefficient of
the impulse at / = b is negative because
Figure 2.33 shows f t ) and S( t T ), where
If /(f ) has a discontinuity at t T, the
= integral
—
fit ) is continuous at t = T.
r + co
fb+ ) - fib- ) = - A
f( t ) I
(2.65)

is not defined because the value of fT )

J jiO Sit - T) dt

is not uniquely given. Consider

A the following examples.
g( t )
4 Example 2.1

/•+ CO
ft) =
i
a
( a)
b t 2 I eimi6it - T )dt = eimT (2.70)
Example 2.2
!l - m ft) = sin t
°i (a ) 1 (2.71)
| g'( t ) V2
M A
|2
2
A
a t

u ( b)
-A
.
( b)
.
t

. .
FIG 2.31 (a) Square pulse. FIG 2.32 (a) Signal . 0
(b) Derivative of square pulse. (6) Derivative. .
FIG 2.33
*
l:
!
!!

I
40 Network analysis and synthesis
Signals and waveforms 41
m ; where fiT ) is the derivative of f t )
evaluated at t = T where, again, we
assume thatft ) is continuous Equation 2.74
by parts. Thus,
.
can be proved by integration
i f(t )

mm
0
^ T t
J” f t ) d' it - T ) dt = f t ) S( t — T ) I
— f—
°0
~
J oo
fit ) bit - T ) dt (2.75)
. .
FIG 2.34 Impulse scanning. = -fiT ).
Consider next the case where f it ) is continuous for
us direct our attention to the integral
— oo < t < oo. Let
It can be shown in general that

£ “ /(0 <5(f ~ T ) dt
— fiT ) (2.72) l JO ) <5<\
n
t
— —
T ) dt = i l ) f <n>( T )
” (2.76)

which holds for all t in this case. If T were varied from co to + co,
then ft ) would be reproduced in its entirety. An operation of this sort
—
corresponds to scanning the function fit ) by moving a sheet of paper with Problems
a thin slit across a plot of the function, as shown in Fig. 2.34.
Let us now examine higher order derivatives of the unit step function. 2.1 Resolve the waveforms in the
figure into odd and even
Here we represent the unit impulse by the function fft ) in Fig. 2.28, components.
which, as e -+ 0, becomes the unit impulse. The derivative offft ) is given m
in Fig. 2.35. As e approaches zero, /' ft ) approaches the derivative of the

-r l^i
unit impulse S' it ), which consists of a pair of impulses as seen in Fig. 2.36. ~
The area under d\t ), which is sometimes called a doublet , is equal to zero.
ij Thus, 1 0 it
”
f t ) dt = 0

The other significant property of the doublet is

J (2.73)
(a )

'
!
J> ft — T ) d t — -f i T ) (2.74) s( t )
2
fit )
2|
I MO
!i h
i
m
I o t 0 t t 0 1 2. t -1 0 1 t

~ h
.
FIG 2.35. Unit doubles as t -> 0 . . .
FIG 2.36 The doublet (c )
ft) . (d )
PROB. 2.1
 Signals and waveforms 43
2.5 For the waveform /(/), shown in the figure, determine what value K
42 Network analysis and synthesis must be so that
using shifted step
the waveforms in the figure
2.2 Write
functions.
the equatio n for
ia ) /(/) dt = 0
J
f— oo
m (b) ( “ fiOdt
=0
Jo+
m 2 2.6 For the waveforms shown in the figure, express in terms of elementary
+

E
1 time functions, that is, tn, u{ t —„
/ ), 6{ t - h ), ( 0 ) fit ) ib ) f' it ) ic ) fir ) dr f
J- 00

0 2 t Sketch the waveforms for (6) and (c) neatly.

fo T 2\
T 3T /4T t
fit )
-E <a )
(b) fit )

20
fit )
fit )
2 4
2
1.5 1
0 4 l W 1 t

(i) (ii) Y 5 -
0
t
f0 1 2 t PROB. 2.6
id )
ic ) 2.7 Prove that
PROB 2.2 . (a ) = -d' i -x )
S' ix )

ib ) — Six ) = x d'ix )

for the derivatives, using shifted

waveforms in Prob. 2.2 and
2.3 Find the derivative of the step and/or impulse functions.
write the equations

in the figure, plot carefully

ic ) :
J
2.8 The waveform fit ) in the figure is defined as
x Six ) dx =0

2.4 For the waveform fit given

)
fit ) = - <\
) 0 t (

fir ) dr
= 0, elsewhere
for a value of t > T.
Show that as e -+ 0, fit ) becomes a unit impulse.

fit ) fit )

r *
Edit )
*E T
0
W t o f

Edit — T)
\K 6{t - T) PROB. 2.8
PROB. 2.5
PROB 2.4 .

it ’

M
The frequency domain: Fourier analysis 47
of its harmonic terms, a0 alt a2 , , , blt b2, ... ....
These coefficients con-
stitute frequen domain description of the signal. Our task now is to
a cy
derive the equations for the coefficients at , b{ in terms of the given signal
.
function s( t ) Let us first discuss the mathematical basis of Fourier series,
the theory of orthogonal sets.

chapter 3 3.2 ORTH OGON AL FUNC TIONS

Consider any two functions fft ) and f 2( t ) that are not identically zero.
The frequency domain: Then if

Fourier analysis rW oa - o
JTi
(3.4)

we say that fft ) and fft ) are orthogonal over the interval [T1 7’2] For ( .
example, the functions sin t and cos t are orthogonal over the interval I

nlrr < ? < (« + l )27r. Consider next a set of real functions

<f>2( t), . . . , </> „(<)} If the functions obey the condition
;

3.1 INTRO DUCT ION

One of the most common classes of signals encoun

tered are periodic <f> ,) = f fait ) tfxfi) dt = 0, i
*j (3.5)
JTi
signals. If T is the period of the signal, then
(3.1)
then the set {$J forms an orthogonal set over the interval [ Tx , T 2 ). In
s( t ) = s( t ± nT ) « = 0, 1, 2, . . . Eq. 3.5 the integral is denoted by the inner product ( fa, <f> ) For conven- ,.
number of discon- ience here, we use the inner product notation in our discussions.
In addition to being periodic, if s( t ) has only a finite ,
The set {<£ } is orthonormal over [7 , T2 ] if
tinuities in any finite period and if the integral
r <*+T ( 4>i , <f> j )
^
= 0 i
^ j
£ KOI dt
s( t ) can be expanded
i
=1 i j =
(3.6)

is finite (where a is an arbitrary real number), then

series The norm of an element fa in the set { fa } is defined as
into the infinite trigonometric

m = ( fa , fa ) , = ( j
A
dtj'
—2 + a , cos cot + a cos 2cot +
^
s(0 =
• A faXt ) (3.7)
2 ( 3.2)
+ bx sin cot + b2 sin 2cot + • • • We can normalize any orthogonal set { fa, f a, ... , </>„} by dividing each
. referred to as the
where co = 2njT This trigonometric series is generally term <f> k by its norm ||</>J .
.
Fourier series In compact form, the Fourier series is
Example 3.1. The Laguerre set,1 which has been shown to be very useful in
s(t) = — ncot + bn sin (3.3) time domain approximation, is orthogonal over [0, co ]. The first four terms
2 + n2= (°l
n COS

ed in a W. H. Kautz, “Transient Synthesis in the Time Domain,” Trans. IRE on Circuit

It is apparent from Eqs. 3.2 and 3.3 that, when 5(0 expand ents
is 1
1 Fourier series, we can describe 5(0 completely in terms of the coeffici .
Theory, CT-1, No. 3 (Sept 1954), 29-39.
46

!
;
 48 Network analysis and synthesis The frequency domain : Fourier analysis 49
of the Laguerre set are . When Eq. 3.13 is satisfied, we say that {/„(/)} converges in the mean to
uo = .
fit )
To examine the concept of convergence in the mean more closely, we
UO = e-at [ l - 2(.at )] must first consider the following definitions:
Definition 3.1 Given a function/(f ) and constantp 0 for which
UO = - 4(at ) + 2( atf ] (3.8) >
UO = e~a‘ [ l - 6at + 6(af )2 - f («f )3] PiKt \
JTI
) vdt < oo,
we say that /(f ) is integrable L* in [7V T,], and we write (f ) e Lv in
[ T » T2 ] .
/
To show that the set is orthogonal, let us consider the integral
Definition 3.2 If /(f ) e Lp in [7 , Tt ] , and {/„(/ )}is a sequence of func
I tions integrable L* in [Tlt J2], we say that if ^ -
/ "
UO UO dt = - 4(°0 + 2(at?] dt (3 9)-
f *
Letting r = at, we have !im
n -* co JTi
!/(/) - fn( t )\ Pdt = 0
1 f 00
(U U = -a
Jo
e~ 2r( l - 4r + 2 ra) dr then {/„(0} converges in the mean of order p to /(f ) Specifically, when
„ , .
~ 2 we say that { „(f )} converges in the mean to ( )
(3.10) P / /f.
The principle of least squares

-
Now let us consider the case when /„(f ) consists of a linear combination
The norms of UO and UO are of orthonormal functions <f>lt <f>2, . . . , <f>„.

m - (f r- V 2a
i
-
(3 11)
fn( 0 = 2 «, UO
» =»1
(3.14)

l &l = {/ V *! - 2 ! 4( at ) + 4( aty ] dtjA (3.12)

Our problem is to determine the constants a, such that the integral
squared error
1
V 2a 11 / ~ /»11 * =
JTi
- /„( )]* dt
! (3.15)
It is not difficult to verify that the norms of all the elements in the set are also I is a minimum. The principle of least squares states that in order to
attain
equal to 1/ V2a. Therefore, to render the Laguerre set orthonormal, we divide minimum squared error, the constants at must have the values
each element 4>i by 1/ 2«.
^ f
3.3 APPROXIMATION USING ORTHOGONAL FUNCTIONS
\
a =JTI'mmdt
T
(3.16)

In this section we explore some of the uses of orthogonal functions in

the linear approximation of functions. The principal problem is that of
.
Proof We shall show that in order for \\ f
must set a( = ct for every i = 1, 2, . . . , n . —/ „||2 to be minimum, w e

approximating a function /(f ) by a sequence of functions/„(f ) such that the II / - /J2 = (/ /) - 2(/, /„) + (/„, /„)
mean squared error
(3.17)
PVM - AO)] df = 0 = 11 / 112 - 22 /,
-^ ^ )
-
2
e = lim J T i (3.13)
< r i i

50 Network analysis and synthesis The frequency domain : Fourier analysis 51

H JI = 1, and by definition, c{ = * t)I<
Since the set {/J is orthonormal,
^
2

.
( f , ({> ,) We thus have
.A
11 / - All = ll / l — 2
Adding and subtracting 2 c<2 gives
« =i
2 2
2
i= l
a( Ci + 2U
-
* l
(3.18)
rwtw
—
3JT
\ 2W IT

. .
0 7T 27T
FIG 3.1 Rectified sine wave
37T
.
t

ll / - /„ll 2 = ll / ll — 2 2=1aiC + <2=1c< + 21‘J* 2 ci2

2 2 2
{ ~
* '
** <! In explicit form the Fourier coefficients, according to the definition given
1
( 3.19)
earlier, are obtained from the equations
= ll / ll + 2= (c< - aiY - <2= i c< 2
2
2 C*+T
° = TJa SW dt
1 a

-
1
(3.23)
We see that in order to attain minimum integral squared error, we must
set at = c{ . The coefficients c,-, defined in Eq. 3.16, are called the Fourier 2 f «+ rs(f ) cos
coefficients of /(f ) with respect to the orthonormal set {<f>i( 0 } - ° k
TL ot+T
kcot dt (3.24)

Parseval ’s equality
Consider fjf ) given in Eq. 3.14. We see that II s(f ) sin kcot dt (3.25)

rw
JTi * = i, i=l
2 (3.20)
We should note that because the Fourier series sn( t) only converges to
s(f ) in the mean; when s( f ) contains a jump discontinuity, for example,
at f 0
since </> f are orthonormal functions. This result is known as Parseval’s
equality, and is important in determining the energy of a periodic signal.
s„(f 0) = L° + -+ s( o )
)
2
~
^ ‘
(3.26)
At any point t1 that s( t ) is differentiable (thus naturally continuous)
3.4 FOURIER SERIES
UO ) converges to sfo).2
As an example, let us determine the Fourier coefficients of the fully
Let us return to the Fourier series as defined earlier in this chapter, rectified sine wave in Fig. 3.1. As we observe, the period is T = 7T so
that the fundamental frequency is co = 2. The signal is given as
s( f ) = — + 2 an
2
(
n=l
cos ncot + si
° nCO0 (3.21) s( t ) = A |sin f| (3.27)
f. Let us take a = 0 and evaluate between 0 and TT. Using the formula just
From our discussion of approximation by orthonormal functions, we can BKl derived, we have
see that the periodic function s( t ) with period T can be approximated by a
Fourier series s„( t ) such that s„( t ) converges in the mean to s( t ), that is,
bn —
s( t ) sin 2nt dt = 0
TT JO
(3.28)

lim
00 r [s(f ) - s„( f )]2 dt -^ 0

where a is any real number. We know, moreover, that if n is finite, the

(3.22)
I
a0 =

a„
-
— Jo sin t dt = 4—A
2 Afr
7T
!

- / Vo cos 2nt dt
7T
(3.29)

—
mean squared error ||s(f ) J„(f )||2 is minimized when the constants bi
are the Fourier coefficients of s( t ) with respect to the orthonormal set m
TT JO
1 4A
(3.30)
1 - 4« 2 TT
jcos kcot
1 (772)* ’
sin kcot
(r/2)*
, j * = 0, 1, 2, . . . g * For a proof see H. F. Davis Fourier Series and Orthogonal Functions, Allyn and
HIT Bacon, Boston, 1963, pp. 92-95.

m 1 X
 The frequency domain : Fourier analysis 53
52 Network analysis and synthesis
s( t )
Thus the Fourier series of the rectified sine
wave is

s(0 =
2TA ——
+ ny= i 1 -14n-2 cos 2« / )/
‘ (3.31) A

3.5 EVALUATION OF FOURIER

COEFFICIENTS — T -I
"
0

— A
r
2
\T t

forms of Fourier series.

In this section we will consider two other useful
to simplify the evaluation
In addition, we will discuss a number of methods
how the evaluation of .
of Fourier coefficients. First, let us examine .
coefficients is simplified by symmetry considerations
. From Eqs 3.23-3.25 FIG 3.2

coefficie nts, let us take

which give the general formulas for the Fourier separate parts, Suppose next, the function s{ t ) obeys the condition
—
a = 7)2 and represent the integrals as the
sum of two

Kr )
-
that is, * = -s(0 (3.36)

|[i
T/2
s( 0 cos ncot dt +
J -T / 2
f
s(0 cos ncot dt
J . .
as given by the example in Fig 3.2 Then we can show that s( t ) contains
(3.32)
only odd harmonic terms, that is,
bn =|
T/ 2
[| s(0 sin n (0 t dt + J T/2
s(0 sin n(ot
an = bn = 0 ; n even
dummy variable, let us T/2
Since the variable (0 in the above integrals is a
substitute x = t in the integrals with limits (0 ; T/2
), and let x / in the
—— and s( t) cos ncot dt
(3.37)
—
integrals with limits ( T/2; 0). Then we have
bn =
4 RT / 2
—
T Jo
s( / ) sin ncot dt , n odd
rr / 2
an = -
9

T Jo
[s(:c) + s( *)] cos ncox dx — (3.33) With this knowledge of symmetry conditions, let us examine how wc
can approximate an arbitrary time function 5(/) by a Fourier series within
9 fT/2
bn = - [s(a:) - s(-a:)] sin ncox dx K an interval [0, T]. Outside this interval, the Fourier series sn( l ) is not
T Jo required to fit s( t ). Consider the signal s( t ) in Fig. 3.3. We can approxi-
Suppose now the function is odd, that is, s x) =
( s( — — x), then we see that | mate s( t ) by any of the periodic functions shown in Fig. 3.4. Observe
n that each periodic waveform exhibits some sort of symmetry.
an = 0 for all n, and T/2 Hr Now let us consider two other useful forms of Fourier series. The first
s( x ) sin ncox dx ( 3.34) Hi is the Fourier cosine series, which is based upon the trigonometric identity,

This implies that, if a function is odd , its Fourier

series will contain only
function is even, that is,
- Cn cos ( ncot + 6n ) = C„ cos ncot cos 0„ Cn sin ncot sin 6 „ — (3.38)

sine terms. On the other hand, suppose the

—
S(t )

s( x ) = s( x ), then bn = 0 and
,
4 rT i x ) cos ncox dx A
an T Jo s(
= — (3.35)

Consequently, the Fourier series of an even function

will contain only
. .
FIG 3.3 Signal to be approximated.
0
^ T t

cosine terms.

I
Ui
Network analysis and synthesis The frequency domain: Fourier analysis 55
54
s( t ) It should be noted that the coefficients C„are usually taken to be positive.
A
use the equivalent form
—
If however, a term such as 3 cos 2cot carries a negative sign, then we can

-T o
T
t —
3 cos 2cot = 3 cos (2cot + IT ) (3.43)
For example the Fourier series of the fully rectified sine wave in Fig. 3.1
,
(a)
was shown to be
s( t )
/1
cos 2 nt ) (3.44)

Expressed as a Fourier cosine series, s( t ) is

T t

lb )
— A
1
cos (2nt + TT) J. (3.45)

Next we consider the complex form of a Fourier series If we express

s( t )
cos ncot and sin ncot in terms of complex exponentials, then the Fourier
series can be written as
A
gjjKot g — jntot

+ bn
eino>t — )
e jno>t

-TT
- A
(O
L
- t

(
±i \ -
= ~2 +nl n ~
2
jKe
2

+ an + e
jbn
2
~
2 j
inai
)
(3.46)

. .
FIG 3.4 ( a ) Even function cosine terms only. (b) Odd function sine terms only
, (c)
If we define
Odd harmonics only with both sine and cosine terms.
P-n = an +2
~ jbn jbn
Pn = 2
> = (3.47)
We can derive the form of the Fourier cosine series by setting
then the complex form of the Fourier series is
an = Cn cos 0 n (3.39)

and —
bn = C„ sin 6 n (3.40) <0 = Po + 'i { Pneinmt + p_ne- inat )

(3.48)
We then obtain Cn and 6 „ in terms of an and bn, as
= I fin*
C„ = ( af + bf )'
A = 71
— 00

C.f = (3.41)
If
We can readily express the coefficient /?„ as a function of s( t ), since

Pn = an 2
~ jbn

0 n = tan-1 (- 5D I
—— I1 CT
T Jo
s(/)(cos ncot
— j sin ncot) dt (3.49)

If we combine the cosine and sine terms of each harmonic in the original
series, we readily obtain from Eqs. 3.38-3.41 the Fourier cosine series
fit ) = C0 + Cj cos ( cot + 0X) + C2 cos ( 2cot + 02) : .
4J>- / net d(

Equation 3.49 is sometimes called the discrete Fourier transform of s( t)

+ C3 cos (3tu / + 03 ) + + C„ cos ( ncot + 0„) + • • • (3.42) and Eq. 3.48 is the inverse transform of Pfjnco ) = fin .
 The frequency domain : Fourier analysis 57
56 Network analysis and synthesis
Observe that /?„ is usually complex and can be represented as
fin = Re /?„ + j Im fin (3.50)
The real part of /9n, Re fin , is obtained from Eq. 3.49 as
1 fr
Re =—
T Jo
s( t ) cos nwt dt (3.51)
Co
and the imaginary part of fin is
j RT
—~
= T- JoI s( t ) sin nwt dt
j Im fin (3.52)

Cil Ci It is clear that Re is an even function in n , whereas Im fin is an odd

.
function in n The amplitude spectrum of the Fourier series is defined as
\ PJ = ( Re2 |8„ + Im2 /?n)“ (3.53)

and the phase spectrum is defined as

c2 | I c2 <f>n = arctan Re ( 3.54)
/?n
It is easily seen that the amplitude spectrum is an even function and the
!1
———
Ci .
4 oi 3oi Zoi
— tn m w 2w 3« 4w .
phase spectrum is an odd function in n The amplitude spectrum provides
us with valuable insight as to where to truncate the infinite series and still
maintain a good approximation to the original waveform. From a plot of
the amplitude spectrum, we can almost pick out by inspection the non-
trivial terms in the series. For the amplitude spectrum in Fig. 3.5, we see
. .
FIG 3.5 Amplitude spectrum. that a good approximation can be obtained if we disregard any harmonic
above the third. '
As an example, let us obtain the complex Fourier coefficients for the
square wave in Fig. 3.6. Let us also find the amplitude and phase spectra
of the square wave. From Fig. 3.6, we note that s( t ) is an odd function.
A
— —
Moreover, since s( t T/ 2 ) = s( t ), the series has only odd harmonics.
From Eq. 3.49 we obtain the coefficients of the complex Fourier series as
— 2T|
_ ±1
! r7l 0
1 71 2T \ t
Ae-iruot d t
rf
T JTI 2
Ae~ inmt dt
(3.55)
__
— jnwT
A
—
2g iinoT / 2 ) | g — jnaiT'
)

J . .
FIG 3.6 Square wave. Since nwT = n2n, fi„ can be simplified to
1 (3.56)

N m .

The frequency domain : Fourier analysis 59

58 Network analysis and synthesis
«W
Amplitude

M |
A A

i
1 1 To TT 3T 2T W t
-5 -3 3 5 oi I 2
. .
FIG 3.8 Impulse train.
2 2

( a)

Phase We see that the complex Fourier coefficients for impulse functions are
obtained by simply substituting the time at which the impulses occur into
the expression, e~ina .
“
In the evaluation of Fourier coefficients, we must remember that the
— 5 ” 3 “ 1

-!
101
0)
limits for the /?„ integral are taken over one period only, i e., we consider
only a single period of the signal in the analysis. Consider, as an example,
the square wave in Fig. 3.6. To evaluate /?„, we consider only a single
.

period of the square wave, say, from t = 0 to

(b ) t = T , as shown in Fig. 3.9a. Since the square \ s( t )
. (b) Phase . A;
.
FIG 3.7. Discrete spectra of square wave (a) Amplitude
,
wave is not m de up of impulses, let us
^
differentiate the single period of the square
Simplifying one step further, we obtain
'
wave to give \ s t ), as shown in Fig. 3.9b. We
1 T
2A can now evaluate the complex Fourier coeffi- 2
fin = n odd . cients for the derivative s' ( t ) , which clearly is
^
jmr (3.57)
made up of impulses alone. Analytically, if A -
n even
=0 s( t ) is given as fa)
given in Fig. 3.7.
The amplitude and phase spectra of the square wave are fine 1not
s( t ) = 2 (3.60)

3.6 EVALUATION OF FOURIER COEFFICIENTS

USING f
•
then the derivative of s( t ) is
Is'ft)
UNIT IMPULSES
functions to
m s'( t ) = i jncofi„einat (3.61)
M
In this section we make use of a basic property of impulse o
, This method is
simplify the calculation of complex Fourier coefficients
only. Here, we define a new complex coefficient i
restricted to functions which are made up of straight line components -
’ Thus the method applies for the square wave in Fig. 3.6 . The method is
Yn = Jna> fin (3.62)
:; _
based on the relation
°°
J
( <5( t - T ) dt = / (T1)
/0 1
(3.58)
or fin = Yn
o
—jnw
(3.63)
- 2A
If the derivative s' ( t ) is a function which (b)
for
Let us use this equation to evaluate the complex Fourier coefficients
~ lnat, we have consists of impulse components alone, then (a ) S uare wave
the impulse train in Fig. 3.8. Using Eq. 3.58 with f = ' 3’9
^
^
( t ) e we simply evaluate y„ first and then obtain jSw
over
| fin from Eq. 3.63. For example, the derivative
d jTd ( t - = d e-" (3.59) P ative of square wave over
i fin = dt
m’ of the square wave yields the impulse train period [0, T ].

V.

1:
 The frequency domain: Fourier analysis 61
60 Network analysis and synthesis
s( t )
.
in Fig. 3.9b In the interval [0, T ] , the signal s'( t ) is given as A

s'( 0 = A <5(0 - 2A 6 + A d( t - T) (3.64)

01 I T
¥ t
Then the complex coefficients are (a )

Yn = irsV
T Jo
)e-inatdt
*1
s’( t )

_ > _ _ e-
(3.65)

=^ T
(1 2g~ li«“272 | inmT )
0 , I T :

The Fourier coefficients of the square wave are T ( b)

2A s"( t ) 2.4

_ jna>

^
jncoT
-
2e~‘t ina>Tli p e~ ina>T )
)
(3.66 )
T

° l
2 \
T

T
I t

which checks with the solution obtained in the standard way in Eq. 3.55
.
If the first derivative, s' ( t ), does not contain impulses, then we must
differentiate again to yield
s"(0 = heincut 2 ' (3.67) (c )
— «= 00 .
FIG 3.10. The triangular wave and its derivatives.
where h = jncoy„= ( jnco? pn (3.68)
From An we obtain
For the triangular pulse in Fig. 3.10, the second derivative over the
period [0, T ] is .
'
h
Pn = ( jeon )2
[ [- ) ]
^UTs ^ -
s"( 0 = d ( t ) 2d t + d( t ~ T ) (3.69) 2A
n odd (3.72)
nV
The coefficients A„ are now obtained as =0n even
A slight difficulty arises if the expression for s’( t ) contains an impulse
"( t )e- inmtdt in addition to other straight-line terms. Because of these straight line
h = T Jo -
!
_M _ (i 2e~UnaTI 2
)
+ e~inaT )
(3.70) terms we must differentiate once more. However, from this additional
differentiation, we obtain the derivative of the impulse as well. This
presents no difficulty, however, because we know that
T*
: which simplifies to give
8A
h = T 2, ~ n <> dd
(3.71)
£% (/) d' ( t - T) = — s' ( T ) (3.73)

=0 n even
so that |
—
J 00
<5'(< - T )e~ inat dt = jnwe~inaT (3.74)
1

: I >
.•
The frequency domain : Fourier analysis 63
62 Network analysis and synthesis
The complex coefficients are now obtained as
derivatives of impulses
We can therefore tolerate doublets or even higher . 3.1 la. Its derivative
in Fig
in the analysis. Consider the signal s( t ) given h
, expressed as
s' ( t ), shown in Fig. 3.11Z> can be ( jeon?
,,)
(3.78)
(3.75)
s' ( 0

^ 2e~Untt T i)
=0 7?
(1 " + ~

of impulses and a pair of

The second derivative s"( t ) consists of a pair

- .
Simplifying, we have
doublets as given by

§[«0 - a( - f )] + no - 2*(, -f ) (3.76) o

Pn
1
+
, 3
n odd
s"( 0 nh2 j 2nn
(3.79)
as shown in Fig. 3.11c. We therefore evaluate An as 1
n even
j 2nn
-
, inut dt
h
_ 2. — 2
(l e-UnoTI 2 \
) i jj£R ( J
T
_ 2e-
( M(0 T/ «)
(3.77) In conclusion, it must be pointed out that the method of using impulses
to evaluate Fourier coefficients does not give the d c coefficient, a0/ 2 or /?0.
We obtain this coefficient through standard methods as given by Eq. 3.23.
-
T
s’ ( t )
s(t ) 3.7 THE FOURIER INTEGRAL
2 S (t )
In this section we extend our analysis of signals to the aperiodic case.
l T
We show through a plausibility argument that generally, aperiodic signals
t
have continuous amplitude and phase spectra . In our discussion of Fourier
0
I T t
° series, the complex coefficient /?n for periodic signals was also called the
discrete Fourier transform
i
i m
s i rr / i
P( nfo) = -1 J T / 2s{ t )e-in2^ dt
( a) ( b) ' -* H ) -
(3.80)

f) and the inverse (discrete) transform was

i s" ( t )
\$6 ( t ) S( t ) x—
= «= oo£("/oy (3.81)

f i From the discrete Fourier transform we obtain amplitude and phase

o | spectra which consist of discrete lines. The spacing between adjacent lines
| in the spectrum is
l
A/ = ( n + l )/0 - w/„ = - (3.82)

As the period T becomes larger, the spacing between the harmonic lines in
the spectrum becomes smaller. For aperiodic signals, we let T approach
(c)
infinity so that , in the limit, the discrete spectrum becomes continuous.
FIG 3.11. :
 The frequency domain : Fourier analysis 65
Network analysis and synthesis
64
We now define the Fourier
integral or transform as 1 A( f ) MO

(3.83) [A
S( f ) = lim
/o
The inverse transform is
- f o +/ -7
s{ t ) = r—
J 00
S ( f )e*' ftdf
pair .
(3.84)

called the Fourier transform

Equations 3.83 and 3.84 are sometimes
of Fourier transformation and IF-1 denote

If we let 3r denote the operation

(a) (b)

inverse transformation, then

S ( f ) = lF s( t ) -
(3.85)
. .
FIG 3.12 Amplitude and phase spectrum of A 6( t
— /0).

Example 3.2.3
s( t ) = $"-» S( f ) s( t ) = A S( t — t0 )

In general , the Fourier transform S( f ) is complex and can be

)
denoted as
(3.86)
S( f ) =
J
r— A <5(/ - t )e-^‘ dt
0

S( f ) = Re S(J ) + j Im S( f
CO

= Ae~ii,rfto (3.92)

The real part of S( f ) is

obtained through the formula Its amplitude spectrum is
A( f ) =A (3.93)
Re S(/) = US( f ) + S( /)
]
(3.87)
- while its phase spectrum is
4>( f ) = 2 TT/?0 (3.94)
(* 5(0 cos 27rf t dt
J 00 —— as shown in Fig. 3.12.
Example 3.3 .
Next consider the rectangular function plotted in Fig. 3.13 .
-
:
and the imaginary part
through Formally, we define the function as the rect function .
W
Im S(/) = T2 j. PC/) - S( ~/)] (3.88)
( =1 N T
(3.95)
rcct xl

! S( f ) is defined as
= —J s(0 sin lirft dt =0
The inverse transform of rect / is defined as sine t (pronounced sink ),
1*1 > T
W

The amplitude spectrum of (3.89)

)* ]'/i afreet / ] = sine t
A( f ) = [Re S( f ) + Im S( f
2
: : rwi2
12irft df (3.96)
and the phase spectrum is j J -w/ 2
Im S (J ) (3.90)
/ ^arctani 7
) I sin nWt
^ ^
and phase definition of
)
the Fourier transform, the
7rt

Using the amplitude It should be noted here that the Fourier transform of a generalized function is also
—
3

inverse transform can be

expressed as a generalized function. In other words, if <f> e C, ( tF • <f> ) e C. For example, F • 6 ft )
(3.91) 1, where 1 is described by a generalized function 1„(/). We will not go into the formal
s( t ) = f " A( f ) cos [2nft - df details of Fourier transforms of generalized functions here. For an excellent treatment
..
J — 00 of the subject see M J Lighthill, Fourier Analysis and Generalized Functions, England,
Cambridge University Press, 1955 .
Let us examine some examples
.
.
I

66 Network analysis and synthesis

The frequency domain : Fourier analysis
From the plot of sine t in Fig. 3.14 we see that sine t falls as does |/| 1
_,
67

rect x with zeros at t = n / W , n = 1, 2, 3, . . . We also note that most of the

energy of the signal is concentrated between the points l / W < t < l / W.
Let us define the time duration of a signal as that point, t0 , beyond which
—
the amplitude is never greater than a specified value, for example, e0. We
can effectively regard the time duration of the sine function as /0 = ± 1/ W.
The value W , as we see from Fig. 3.13, is the spectral bandwidth of the
rect function. We see that if W increases, t0 decreases. The preceding
-f example illustrates the reciprocal relationship between the time duration
of a signal and the spectral bandwidth of its Fourier transform. This
FIG. 3.13 Plot of rect function.
. concept is quite fundamental. It illustrates why in pulse transmission,
narrow pulses, i.e., those with small time durations, can only be trans-
1.0 7T\ mitted through filters with large bandwidths ; whereas pulses with longer
0.8 time durations do not require such wide bandwidths, as illustrated in
0.6 Fig. 3.15.
sine f 0.4
0.2 3.8 PROPERTIES OF FOURIER TRANSFORMS
0.0 KQ A In this section we consider some important properties of Fourier
-0.2 transforms.
-0.3- - # -b °t w w WWW
I -w "
W
Linearity
.
FIG. 3.14 The sine t curve. The linearity property of Fourier transforms states that the Fourier
transform of a sum of two signals is the sum of their individual Fourier
r(t)
eft )
System Response transforms, that is,
Excitation
Ti sft ) + c2 J2(/ )] = Ci S f f ) + c2 S f f ) (3.97)
BW = f Differentiation
t This property states that the Fourier transform of the derivative of a
t
Narrow pulse signal is jlirf times the Fourier transform of the signal itself :
Narrow pulse Wide band %
m or more generally,
3r \ -
s t ) = jlirf S(J ) (3.98) i

r (0
* Sr • sln )( t ) = ( jlirf )n S(Jj (3.99)
eft ) Response
Excitation System
The proof is obtained by taking the derivative of both sides of the inverse
10 Tj transform definition,
BW = ±
i2* ft
dJ

.
FIG 3.15
— J 10 T\ . -
Wide pulse
. Illustration
t
Narrower band
of the reciprocity relationships between
Wide pulse
time duration and
1
t

= j>/^ i2 „tdf (3.100)

bandwidth.

i:
 The frequency domain : Fourier analysis 69
68 Network analysis and synthesis 8(f )
If(t )
transform of the integral of s( t is
)
Similarly, it is easily shown that the TO

*[ LK T ] - ]b ) dT
/
SU) '
(3.101)

-t t -f o f

Consider the following example

(3.102)
s( t ) = e~at u( t )
Its Fourier transform is
FIG . 3.16. Fourier transform off ( t ) = 1.0.
,,
S(/) =
J °° ~at u( t )e~ an t dt
e
_ (3.103)
Scale change
The scale-change property describes the time-duration and bandwidth

-f d(
1
a + ;2TT/
reciprocity relationship. It states that

The derivative of s( t ) is
s' ( t ) = <5(0 - ae at u( t )
~ (3.104)
* Hi )] = w s ( af ) (3.111)

Its Fourier transform is Proof . We prove this property most easily through the inverse trans-
a j 2uf form
nm] = i - a + j2vf a + j 2irf (3.105)
= laljT
*
S( af )ei2* fl df (3.112)
= j 2 irfS( f )
Letf' = af ; then
Symmetry
The symmetry property of Fourier
transforms states that if
'

= Ia|Jf-oo S(/'y "’ Vo>

" 2 ,
(3.106) a
( f • x( t ) = X (J ) ( 3.113)

then
This property follows directly from the
ir • m = *(-/)
symmetrical nature of the Fourier
(3.107)

As an example, consider
-0
transform pair in Eqs. 3.83 and 3.84 . 1
.
Example 3.4 From the preceding section
, we know that ne ~al
«(«)] =
;2 TT/ + a
(3.114)
(3.108) then
7 sine t = rect f
It is then simple to show that
(3.109 )
7 • rect t = sine ( -f ) = sine / j 2 naf +a (3.115)
symmetry property, Consider next 1
which conforms to the statement of the , 7 8( f ) 1. From the symmetry
the Fourier transform of the unit impulse = j 2nf + l
property we can show that (3.110) if a > 0.
f • 1 = 6( f )
Folding
example is also an extreme illustration
as shown in Fig. 3.16. The foregoing The folding property states that
of the time-duration and bandwidth
reciprocity relationship. It says that zero;
time duration , 6( t ), gives rise to infinite
bandwidth in the frequency domain
to infinite time duration . m-»] = s( -f ) (3.116)
while zero bandwidth, 8( f ) corresponds
.T4 ;
;

The frequency domain: Fourier analysis 71

= eiu,° s ) + e- ’to* s( t) = 2s(t ) cos 2nf t
Network analysis and synthesis
70
definition of the Fourier transform .
Then 7 - sff )
1 t (t V 0 (3.122)
The proof follows directly from the
An example is Thus we see that multiplying a signal by a cosine or sine wave in the time domain
corresponds to shifting its spectrum by an amount ±/0. In transmission termi-
F [ e u( f )] =
1 ‘
1
j2 nf— (3.117)
- nology /0 is the carrier frequency , and the process of multiplying s( t ) by cos 2 wf0t
is called amplitude modulation.
Delay
t in the time domain, the corre-
If a signal is delayed by an amount 0 m of the Parseval ’s theorem
sponding effect in the ~frequen cy domain is to multiply the transfor An important theorem which relates energy in the time and frequency
i 2 ,rfto , that is,
undelayed signal by e domains is ParsevaVs theorem, which states that
nsO / )] = e- -„jiaftn
SCO (3.118)
For example,
- )
(F [ e aU~to u( t - t0)] = a + j2rrf
jVvfto
( 3.119) J_ ro
Si(0 s*(0 it = S f f ) S2( -f ) df (3.123)

The proof is obtained very simply as follows :

Modulation
The modulation or frequency shift propert
y of Fourier transforms
states that if a Fourier transform is shifted
the corresponding effect in time is describ
,
in frequency by an amount ,
ed by multiplying the original ^ f ” si(0 st(0 dt =J dtj*
s2( t ) “ Sff )ei2' ft df

signal by e w®‘, that is,

^ Hsa -Zo)] = «i s(0
"

Example 3.5. Given S( f ) in Fig. 3.17a, let

in terms of s( t ) = f
2v f o t (3.120)
us find the inverse transform of
). We know that
>
=/_ <« 41 *
s2( ty2^ dt ( 3.124)

S f f ) in Fig. 3.176
S(

Sx( f ) = S( f - f0 ) + S( f + f0 ) ( 3.121) =J f—° S f f ) S (

°
CO
2 -f ) d f

S( D
In particular, when sft )
as eral' s .
— s ( ),
2 t we have a corollary of Parseval’s theorem
known Planch theorem

fy» dt f j s w d f ^ (3.125)

o +f If s( t ) is equal to the current through, or the voltage across a 1 ohm -

-/ resistor, the total energy is
(a)
sdo f_y« n )

We see from Eq. 3.125 that the total energy is also equal to the area under
,
.
the curve of |S(/)|2. Thus |S(/)|2 is sometimes called an energy density
or energy spectrum.

—f -fo
ft
o ^ fo +.f Problems
3.1 Show that the set (1, sin nirt/T , cos nnt/T }, n = 1, 2, 3 forms an
( b) orthogonal set over an interval [a, a + IT ], where a is any real number. Find
. the norms for the members of the set and normalize the set.
. .
FIG 3.17 Demonstration of amplitude modulation

£ I
 74 Network analysis and synthesis
3.9 Prove that (a) if f ( t ) is even, its Fourier transform F( j<o) is also an even
function ; (6) if (/) is odd, its Fourier transform is odd and pure imaginary.
/
F( jo>)
B *

chapter 4
— COO
PROB. 3.10
+ «0 CO

Differential equations
3.10 Find the inverse transform of
F( ja>) = ft <5( co a>0 )
— + P <5( co + <o0)
as shown in the figure. What can you say about line spectra in the frequency
4.1 INTRODU CTION
domain ?
This chapter is devoted to a brief study of ordinary linear differential
equations. We will concentrate on the mathematical aspects of differential
equations and leave the physical applications for Chapter 5. The differen-
tial equations considered herein have the general form
/] = 0 (4.1)
where t is the independent variable and x( t ) is a function dependent upon
.
t The superscripted terms xV )( t ) indicate the zth derivative of x( t ) with
respect to t , namely,
x »( t ) =' *(!)
dt*
*' - ( 4.2 )
The solution of F = 0 in Eq. 4.1 is x( t ) and must be obtained as an
I .
explicit function of t When we substitute the explicit solution x( t ) into F,
the equation must equal zero. If Fin Eq. 4.1 is an ordinary linear differ
ential equation, it is given by the general equation
-
* M (0 + + • • • + «! At ) + «0 <t ) = /(0
(4.3)
The order of the equation is n, the order of the highest derivative term.
The term f ( t ) on the right-hand side of the equation is the forcing function
.
or driver, and is independent of x( t ) When /(f ) is identically zero, the
equation is said to be homogeneous ; otherwise, the equation is non-
homogeneous.
In this chapter we will restrict our study to ordinary, linear differential
equations with constant coefficients. Let us now examine the meanings of
these terms .
75

Differential equations 77
76 Network analysis and synthesis
in which there is only It can be shown, in general, that the solutions of homogeneous, linear
Ordinary. An ordinary differential equation is one differential equations consist of exponential terms of the form Ciev t. To
one independent variable (in our case, t ). As a result there is no need for ‘
partial derivatives.
Constant coefficients. The coefficients an, a„ lt
_ .
. . , a2 , au a0 are con -
obtain the solution of any differential equation, we substitute Cevt for x( t )
in the equation and determine those values of p for which the equation is
zero. In other words, given the general equation
stant, independent of the variable f. only terms of the
Linear. A differential equation is linear if it contains an *( n)(f ) + • • • + «!\
x t ) + a0 x( t ) = 0 (4.11)
derivatives , as given by Eq. 4.3. For
first degree in x( t ) and all its higher we let x( t ) = Cevi , so that Eq. 4.11 becomes
example, the equation (4.4)
3x t ) 2x( t ) sin t
\ + = Cevt( anpn + + • • • + alP + a0 ) = 0 (4.12)

is a linear differential equation. On the other

hand,
for Eq. 4.12 occur when the polynomial
—
Since evt cannot be zero except at p = oo, the only nontrivial solutions
(4.5)
3 [ x' ( t )]2 + 2x( t ) x' ( t ) + 4 x( t ) = 5/
H( p ) = anpn + fln-ip"-1 + 1- «i/> + a0 = 0 (4.13)
x( a:'(/) are nonlinear by the
is nonlinear, because the terms [x'(f )] and t )
2
Equation 4.13 is often referred to as the characteristic equation, and is
definition just given. denoted symbolically in this discussion as H( p ). The characteristic equation
is the superposition
An important implication of the linearity property is zero only at its roots. Therefore, let us factor H( p ) to give
, if xft ) and x2( t ) are
.
property According to the superposition property
solutions of a given differential equation for forcing
functions f { t ) and/2(/),
combination of = an{ p - p0 )( p - />,) • • (/> - />„_,)
H (P ) (4.14)
respectively, then , if the forcing function were any linear
From Eq. 4.14, we note that C0e*•* , Cxe*d, . .. , are all solutions
fi( t ) and f f t ) such as of Eq. 4.11. By the superposition principle, the total solution is a linear
no = am + bm (4.6)
combination of all the individual solutions. Therefore, the total solution
the solution would be (4.7) of the differential equation is
x( t ) = a x f t ) + b x f t)
be emphasized that the
x( t ) = + Cxe*d + + Cn_xe*^ * (4.15)
where a and b are arbitrary constants. It should _
superposition property is extremely important and should
be kept in .
where C0, Cu . C„ x are generally complex. The solution x( t ) in Eq.
mind in any discussion of linear differential equations
. ..
4.15 is not unique unless the constants C0 , C l t . , C„_ x are uniquely
specified. In order to determine the constants Cit we need n additional
EQUATIONS pieces of information about the equation. These pieces of information are
4.2 HOMOGEN EOUS LINEAR DIFFERENT IAL usually specified in terms of values of x( t ) and its derivatives at / = 0 +,
of homogeneous, and are therefore referred to as initial conditions. To obtain n coefficients,
This section deals with some methods for the solution
linear differential equations with constant coefficients
First, let us find . .
we must be given the values x(0 + ), z'(0 +), . . . *< n-1)(0 -l-). In a number
the solution to the equation —
of special cases, the values at / = 0 are not equal to the values at
-
x\t ) 2x(/ ) = 0 (4.8) f = 0 + . If the _ initial specifications are given in terms of z(0 ), —
Now, with a little prestidigitation, we assume the solution
to be of the form — —
x' (0 ), . . . , *( B 1>(0 ), we must determine the values at t = 0 + in order
to solve for the constants Ct. This problem arises when the forcing
x( t ) = Ce2t (4.9) function f ( t ) is an impulse function or any of its derivatives. We will
discuss this problem in detail in Section 4.4.
x( t ) Ce2 >
where C is any arbitrary constant. Let us check to see whether in = . 4.8, For example, in Eq. 4.9 if we are given that ar(0 + ) = 4, then we obtain
is truly a solution of Eq. 4.8. Substituting the assumed
solution Eq the constant from the equation
we obtain (4.16)
2 Ce2 - 2Ce2* = 0
‘ (4.10) *(0 + ) = C e° = C
 78 Network analysis and synthesis Differential equations 79
Example 4.2 Solve the equation
so that x( t ) is uniquely determined to be
- 8*'(/) + 16 x( t ) = 0
x’( t ) (4.27)
x( t ) = 4e 21

with x( 0 +) = 2 and *'(0 +) = 4

Example 4.1 Find the solution for
(4.17) Solution. The characteristic equation is
x’( t ) + 5x'( t ) + 4x( t ) = 0
given the initial conditions
H ( p ) = p2 - Sp + 16 = ( p - 4) 2 (4.28)

x( 0 + ) = 2 x'( 0 + ) = - l Since H( p ) has a double root at p = 4, the solution must take the form
Solution. From the given equation, we first obtain the characteristic equation
x( t ) = Cxe" + Cite" (4.29)

(4.18) In order to determine Cx and C2, we evaluate x( t ) and x' ( t ) at / = 0 + to give

H ( p ) = p2 + 5p + 4 = 0
which factors into (p + 4 )( p + 1) = 0 (4.19) *(0 +) = Ci = 2 (4.30)
The roots of the characteristic equation (referred to here as characteristic
values ) *'(0 +) = 4CX + Ci = 4
— —
arep = 1 ; p = 4. Then x( t ) takes the form
~ -41 (4.20)
Thus the final solution is x( t ) = 2e" - 4teu (4.31)
*(/) = Cxe ' + Cie Another interesting case arises when H( p ) has complex conjugate roots.
Consider the equation
From the initial condition a:(0 +) = 2, we obtain the equation
x( 0 + ) = 2 = CX + Ci (4.21) H (P ) = a2( p Pl )( p p2 ) -(4.32) -
where px and p2 are complex conjugate roots, that is,
In order to solve for Cx and C2 explicitly, we need the additional initial condition
—
*'(0 +) = 1. Taking the derivative of
x( t ) in Eq. 4.20, we have
Pi ,Pi = a ± j<o (4.33)
x' ( 0 = Cxe t - 4 C2e- - - (4.22)
" The solution x{ t ) then takes the form
At / = 0 +, tx' ( ) is
*'(0 + ) = -1 = - Cx 4Ct
(4.23)- x( t ) = Cxe< a+ia ) l + C2etff-^‘"
,, (4.34) '

Solving Eqs. 4.21 and 4.23 simultaneously, we find that Expanding the term eM by Euler’s equation, x( t ) can be expressed as

Thus the final solution is

Q
—
ait = )
3 cz = — i
(4.24)
x( t ) =
which reduces to
Cxe°\cos cot + j sin cot ) + C2eat( cos cot — j sin cot ) (4.35)

Next, we examine the case when the characteristic equation H( p) has *(0 = ( Cx + Ci )e" cos cot + j( Cx - Ci )eat sin ait ( 4.36)
multiple roots. Specifically, let us consider the case where H( p ) has a
root p = p0 of multiplicity k as given by Let us introduce two new constants, Mx and M 2 , so that x( t ) may be
expressed in the more convenient form
H( p ) = an( p p0 f ( p - p x ) -
( p P n) (4.25) --- - x( t ) = Mxeat cos cot + Mieat sin cot (4.37)
It will be left to the reader to show that the solution must then contain k
terms involving e®0' of the form where Mx and M2 are related to the constants Cx and C2 by the equations

x( t ) Cwe + Coxte*> + C0it 2eVot + • + C t ^ e

1
" Mx = Cx + Ci
= ^ ‘ ^ (4.38)
+ Cxe*i* + C2e*« ‘ + • • • + Cne**
1 (4.26) Mi = j( Cx - Ci )
-
where the double scripted terms in Eq. 4.26 denote the terms in the solution The constants Mx and Mt are determined in the usual manner from
due to the multiple root, ( p p0 )
k
— . initial conditions.

ill .

Differential equations 81
80 Network analysis and synthesis
x( t ) can be obtained if we Solution. The characteristic equation is
Another convenient form for the solution f , defined by the equations
, and
introduce still another pair of constants M
<>
H( p) = pi + 9p* + 32 p3 + 58 p2 + 56p + 24 = 0 (4.50)
Mx = M sin
M2 M cos <f> — ^
(4.39) which factors into

form of x( t ), namely,
H( p ) = ( p + 1 + jl )( p + 1 - jl )( p + 2\
) p + 3) = 0 (4.51)
With the constants M and <j> we obtain another From H ( p ) we immediately write x( t ) as
(4.40)
x( t ) = Meal sin (cot + <f> )
x( t ) = Mxe ~l
cos t + M2e~‘ sin t + C0e~2t + Cxte~il + C2e~2t (4.52)
Example 4.3 Solve the equation
x\t ) + 2x\t ) + 5 x( t ) = 0
(4.41) Since there are five coefficients, we need a corresponding number of equations
to evaluate the unknowns. These are
with the initial conditions
z(0 +) = 1 *'(0 +) = 0 x( 0 + ) = Mx + C0 + C2 = 1
. The characteristic equation H( p) is
Solution -
a:'(0 +) = Mx + Mi 2C0 3C2 + Cx = 0 - -
H(P ) = p + 2p + 5 = ( p + 1 + j2 )( p +
2 1 y 2) (4.42)
— '
-
x'(0 + ) = 2 Mi + 4C0 4 Cx + 9C2 = 1 - - (4.53)

so that, assuming the form of solution in Eq

Then x( t ) is
. 4.37, we have o
~l
1 and co = 2.

(4.43)
— a:< 4 (0 + ) = — MX +
4
- „
a;(3)(o +) = 2Mx + 2 M 2 8C + 12Cx 27C2 = 1
, 16C0 + 81C2 - 32Cx
-
=0
x{ t ) = Mxe~ cos 21 + M2e sin 21
‘ (4.44) Solving these five equations simultaneously, we obtain
At t = 0 +
The derivative of x( t ) is
x(0 +) = 1 = MX

—
Mx = 0 Mi = § C0
- 1 Cx =\ Ci =0
— ~
x'( t ) = Mx{ e~l cos 21 2e * sin 2t ) — + M2{ e~l sin 2t + 2e~l cos 2t ) (4.45) so that the final solution is

At t = 0 +, we obtain the equation x{ t ) = %e ~l

sin t + e~2t +\te~u (4.54)
(4.46)
x' ( 0 +) = 0 = - Mx + 2Mi We have seen that the solution of a homogeneous, differential equation
, we find Mx = 1 and Mi = + £.
Solving Eqs. 4.44 and 4.46 simultaneously may take different forms depending upon the roots of its characteristic
Thus the final solution is equation. Table 4.1 should be useful in determining the particular form of
x( t ) = e~((cos 2t + J sin 21 )
(4.47) solution.
TABLE 4.1
in . 4.40, we would have obtained the
If we had used the form of x( t ) given Eq
solution Roots of H ( p ) Forms of Solution
x( t ) = sin (2/ + tan-1 (2)] (4.48)

that illustrates everything we

.
1 Single real root, p = p0 evo *
Now let us consider a differential equation
have discussed concerning characteristic
values . .
2 Root of multiplicity, k , ( p
3. Complex roots at p2 3 = o
—px )k
± jco
C0epi
‘
Mxevl
+ Cxtevi* +
cos o t
+ Cw/^-V***
+ Mieat cot
< sin
•••

or
Example 4.4 The differential equation is
Meat sin (cot + <f )
x t ) + 56 x' ( t ) + 24x( t ) = 0 (4.49)
x< 5>(/) + 9x< 4 >( t ) + 22xi ( t ) + 58\
The initial conditions are
3)
.
4 Complex roots of multiplicity k
a* />4 , s = a ± j<°
_
M0eat cos cot + Mxteal cos cot +
+ Mk xtk1 1eal cos cot + N^e" 1 sin cot
~

, , l + N e* sin cot + • • •
a:<4 (0 +) = 0 2<3 (0 + ) = 1
+ ^ sin cot
*'(0 +) = 0 *(0 + =
x’(0 + ) = -1 ) 1
 82 Network analysis and synthesis
Differential equations 83
We see that the polynomial within the parentheses
4.3 NONHOMOGENEOUS E Q U A T I O N S is the characteristic
.
equation H( p ) with p = p Consequently, the unknown
coefficient is
obtained as
As we mentioned in the introduction to this chapter, a‘nonhomogeneous
differential equation is one in which the forcing function /(f ) is not
A = -Z-
identically zero for all f. In this section, we will discuss methods for (4.60)
provided that H( fi ) 0.
obtaining the solution x( t ) of an equation with constant coefficients
^
Example 4.5 Determine the solution of the
*< (0 + *„-1 -" ( O + • • • + «« *(0 = /(0
n)
^ (4.55) equation
Let a: ft ) be a particular solution for Eq. 4.55, and let *,(/) be the solution *'(/) + 3 x'(t ) + 2x(t) = 4 e‘ (4.61 )
of the homogeneous equation obtained by letting /(f ) = 0 in Eq. 4.55. with the initial conditions, z(0 + ) = 1, x' ( 0 )
It is readily seen that ~' + = -1 .
Solution . The characteristic equation is
*(0 = xf 0 + « ft ) (4.56)
H ( p ) = p2 + 3p + 2 = ( p + 2 )( p + 1)
is also a solution of Eq. 4.55. According to the uniqueness theorem , the
solution x( t ) in Eq. 4.56 is the unique solution for the nonhomogeneous so that the complementary function is
differential equation if it satisfies the specified initial conditions at I-
t = 0 + .1 In Eq. 4.56, xft ) is the particular integral ; xft ) is the comple
mentary function ; and x( t ) is the total solution.
- *c(0 = Cie ~
‘ + C2e~2t
For the forcing function /(f ) = 4e\ the constants in . 4.60
Since we already know how to find the complementary function xft ), Eq are a = 4, 0 = 1.
we now have to find the particular integral xft ). In solving for xft ), a
Then A=
_i_ 2
very reliable rule of thumb is that xft ) usually takes the same form as the tf( D = 3
forcing function iffit ) can be expressed as a sum of exponential functions.
Specifically, xP( i ) assumes the form of /(f ) plus all its derivatives. For Thus we obtain xfi ) = § e‘
example, iff ( t ) = a sin cot , then xft ) takes the form The total solution is
xP( t ) = A sin cot + B cos cot
*( ) = xfO + xp( t ) =
' + C2e~2i + (4.62)
The only unknowns that must be determined are the coefficients A and B
of the terms in xft ). The method for obtaining xft ) is appropriately To evaluate the constants C1 and C2, we substitute the given
initial conditions,
called the method of undetermined coefficients or unknown coefficients.
In illustrating the method of unknown coefficients, let us take /(/ ) to be *(0 +) = 1 = Cx + C2 + f
(4.63)
*'(0 +) = -1 = -Ci - 2C2 + f
/(0 = «'* (4.57)
Solving Eq . 4.63, we find that Ca = - 1, C2 = Consequently,
where a and /3 are arbitrary constants. We then assume xft ) to have a f
similar form, that is, x( t ) = - e-* + $e-2< + § e‘ (4.64)
xp{ t ) = Ae?1 (4.58) 1 r
It should be pointed out that we solve for the constants
and A is the unknown coefficient. To determine A , we simply substitute Ck and C2 from the
initial conditions for the total solution. This is because initial !
the assumed solution xff ) into the differential equation. Thus, not given for xc( t ) or xp( t ), but for the total solution .
conditions are

Ae? Xanffi + an _ 1
+ • • • + affi + a0 ) = ccefl (4.59) Next, let us consider an example of a constant forcing function (f )
/ = a. !
!
We may use Eq. 4.60 if we resort to the artifice
1 See, for
example, C. R. Wylie, Advanced Engineering Mathematics (2nd ed .),
McGraw-Hill Book Company, New York , 1960, pp. 83-84. /(f ) = a = oceto (4.65)
:
j

46 !
*
Differential equations 85
84 Network analysis and synthesis
4, and the particular integral xft ) for the original driver /(f ) = 2 sin 3f is
that is, /3 = 0. For the differential equation in Example 4.5 with / f ) =
(
we see that 2
xff ) = Im xPl ( t ) = sin [3 f + tan-1 (3) - n ] (4.75)
xM =
A =2 ~m (4.66) 5 Vl 0
There are certain limitations to the applicability of the method of
x( t ) = Cier* + Ca<r 2‘ + 2 (4.67)
and undetermined coefficients. If /(f ) were, for example, a Bessel function
/0(f ), we could not assume xft ) to be a Bessel function of the same form
When the forcing function is a sine or cosine function, we can
still
form and make use of (if it is a Bessel function at all). However, we may apply the method to
consider the forcing function to be of exponential
the method of undetermined coefficients and Eq . 4.60. Suppose forcing functions of the following types:
(4.68) 1. /(f ) = A ; constant.
( f ) aeimt a(cos cot + j sin cot )
/ = = _
then the particular integral xpft ) can be written as
2. /(f ) = A( tn + bn 1tn~1 + - -
1 bf f b0 ) ; n, integer.
3. /(f ) = evt ; p real or complex.
(4.69) 4. Any function formed by multiplying terms of type 1, 2, or 3.
*»1(0 = ReM0 + ylm xpl ( t )
For the purposes of linear network analysis, the method is more than
From the superposition principle, we can show that adequate.
if /(f ) = « cos cot then x f t ) = Re xpl( t ) Suppose the forcing function were
if /(/) = a sin cot then x f t ) = Im xpl( t ) /(f ) = Atkev‘ p = 0 + jco
Consequently, whether the excitation is a cosine function a cos cot or a The particular integral can be written as
;
sine function a sin cot , we can use an exponential driver /(f ) = ae "
part of the resulting particular integral . *»(0 = ( Akt* + + • • + Art + A0 )e”‘
• (4.76)
then we take the real or imaginary
Example 4.6 Find the particular integral for the equation
where the coefficients Ak , Ak _lt . . . , AltA0 are to be determined.
(4.70)
xft ) + 5 x' ( 0 + 4 x( f ) = 2 sin 3f 4.4 STEP AND IMPULSE RESPONSE
Solution. First, let us take the excitation to be In this section we will discuss solutions of differential equations with
fft ) = 2ei3t (4.71 ) step or impulse forcing functions. In physical applications these solutions
are called , respectively, step responses and impulse responses. As physical
so that the particular integral xpl( t ) takes the form quantities, the step and impulse responses of a linear system are highly
= Aei3t (4.72) significant measures of system performance. In Chapter 7 it will be shown
xPl ( t )
that a precise mathematical description of a linear system is given by
From the characteristic equation \i its impulse response. Moreover, a reliable measure of the transient
behavior of the system is given by its step and impulse response. In this
H{ p ) = + 5p + 4 section , we will be concerned with the mathematical problem of solving
we determine the coefficient A to be for the impulse and step response, given a linear differential equation with

A
2 2 _A_ - -l
ej[ tan »(3)
- (4.73)
initial conditions at f = 0 . —
From Chapter 2, recall that the definition of the unit step function was
Hij3 ) — 5 + /15 5 V1O
u( t ) = 1 f >0
Then xPl( t ) is *pl(f ) = 5 VI0 -
2 - -
j[ tan l( 3) + 3f »] (4.74)
=0 t <0
.
 Differential equations 87
86 Network analysis and synthesis
Fora step forcing function, only the highest derivative term is discontinuous
and the unit impulse was shown to have the properties : at t = 0.
<5(/ ) = oo / =0
t 0
Since initial conditions are usually given at t 0
, our task is to
determine the values z n ) (0 + ) and a:( B-1 (0 + ) for an impulse forcing
, —
=0 * function. Referring to Eq. 4.77, let us integrate the equation between
r*+ — —
t 0 and t 0 +, namely, —
J
r ,_
and <5( 0 dt =1
fo+ fo+
In addition, we have the relationship a„ fV’O) dt +
Jo-
°+
J 0- *
B 1)
(0 dt + -- '
+ a0 Jo x( t ) dt = A JI o- d( t ) dt
-
<5( 0
_ du{ t ) (4.78)
dt After integrating, we obtain
As the definitions of <5(0 and u( t ) indicate, both functions have dis-
_, _ >(0
an[*( n 1 (0 + ) - z < n-1) (0 — )] + fl„_ [x< 1
B-2 >(0 )
+ -* (B 2
— )] + • • •
=A
continuities at t = 0. In dealing with initial conditions for step and impulse
drivers, we must then recognize that the solution x( t ) and its derivatives i r (4.79)
x' ( t ), x"( t ), etc., may not be continuous at t = 0. In other words, it may be
that
( B,
We know that all derivative terms below (n
Consequently, Eq. 4.79 simplifies to
— 1) are continuous at t = 0.
*<"> (0 -) # * (0 + )
a;< »-i> (0 )—^ x < n-i ) (0 + ) an [ x n~1 (0+ ) - *<"- >(0-)] =
( ) i

_ l)
A (4.80)

x( 0 ) T* z(0 +) so that *
(B 1 >(0+ ) = — + *(n —
(0 ) (4.81)
In many physical problems, the initial conditions are given at t = 0 — .
However, to evaluate the unknown constants of the total solution, we must We must next determine z ( B)(0 +). At t
in Eq. 4.77 is
— 0 +, the differential equation
have the initial conditions at / = 0 +. Our task , then, is to determine the
+ *„_!*<"-»(0 + ) +
conditions at t = 0 + , given the initial conditions at t = 0 The method
discussed here is borrowed from electromagnetic theory and is often an * ( B)
(0 + ) • •
+ a0 z(0 + ) = 0 (4.82)
’s function .” 2
referred to as “integrating through a Green
Consider the differential equation with an impulse forcing function
—
Since all derivative terms below ( n 1) are continuous, and since we
have already solved for *( B-1) (0 + ), we find that
an xln )(t ) + an !*<n-1(/) + _ 1- a0 x( t ) = A d( t ) (4.77)
- , = - — [a„_1a:< - >(0 + ) +
1
, _
To insure that the right-hand side of Eq. 4.77 will equal the left hand
.
a;( B (0 + ) B 1 •
+ fll*'(0 + ) + a0 x( 0 + )]
side, one of the terms a;( B (0> a:< n 1 (0 • • • » x( 0 must contain an impulse.
)
(4.83 )
The question is, “Which term contains the impulse ?” A close examination ,
shows that the _ highest derivative term £ ( B (f ) must contain the impulse, For a step forcing function A «(/), all derivative terms except x ( n )(t), are
because if *( B 1)(<) contained the impulse, x >( t ) would contain a doublet
(n

C <5'(f ). This argument holds, similarly, for all lower derivative terms of
*

: —
continuous at t 0. To determine a:( B)(0 +), we derive in a manner
similar to Eq . 4.83, the expression
( n~v ) would contain a
x( t ). If the term x ( t ) contains the impulse
( n) , then x (t
step and x ( t ), a ramp. We conclude
( n~2> therefore that , for an impulse (n , - - - [an_ *
, B^1)
+ • + a 0z(0 + )] (4.84)
forcing function, the two highest derivative terms are discontinuous at t = 0. * (0 + ) = 1 (0 + ) ••

an an
2 The Green’s function is another name for impulse response; see, for example, The process of determining initial conditions when the forcing function
Morse and Feshbach, Methods of Theoretical Physics, McGraw-Hill Book Company
, is an impulse or one of its higher derivatives can be simplified by the visual
New York , 1952, Chapter .
7

&

Differential equations 89
88 Network analysis and synthesis
process shown in Eqs. 4.85 and 4.86. Above each derivative term we
draw from which we obtain B = -12 and .
C = 28 Therefore, at / = 0, it is
true that
its associated highest-order singularity. Note that we need only
go as low
*"(0 = 4<5'(/) - 12<5(r) + 28K(0
as a step in this visual aid.
x' ( t ) = 4 <5(0 - 12«(0 (4.92)
x( t ) = 4u( t )
The u( t ) terms in Eq. 4.92 gives rise to the discontinuities in the initial
anx< n>(0 + nn _1a:< »-1>(r) + an_2x^( t ) + + a0x( t ) = <5'(0 (4.85)
conditions at t = 0. We are given the initial conditions at t 0 . Once
we evaluate the A , B , C coefficients in Eq . 4.88, we can obtain the initial
— —
conditions at t = 0 + by referring to the coefficients of the step terms. For
example, if
, (4.86)
*(0 ) = -2 —
a»*< B (0 + + an-2X <n 2 )( t ) +
~ • + a0 x( t ) = <5(0
* <> — ) = - 1
'(

It should be noted that if a derivative term contains a certain singularity — *"(0 )— =7

for example, a doublet it also contains all lower derivative terms.
— For
example, in the equation Then from Eq. 4.92 we obtain

*(0 + ) = -2 + 4 = 2
z'(0 + ) = -1 - 12 = -13 (4.93)

x’( t ) + 3x' ( t ) + , 2x( t ) = «

4 <5' (4.87) a;"(0 + ) = 7 + 28 = 35

we assume the following forms for the derivative terms at / = 0: The total solution of Eq. 4.87 is obtained as though it were a homogeneous
equation, since d' ( t ) = 0 for / ?£ 0. The only influence the doublet driver
x"( f ) = A <5'(0 + B <5(f ) + C u(t ) has is to produce- discontinuities in the initial conditions at / = 0. Having
(4.88) evaluated the initial conditions at t = 0 + , we can obtain the total
x' ( t ) = A <5(0 + Bu( t )
solution with ease. Thus,
x{t ) = + u( t ) x( t ) = Cie-' + C e 21 2
~
(4.94)
Substituting Eq. 4.88 into Eq. 4.87, we obtain
From Eq. 4.93 we readily obtain
+ B <5(0 + C u( t ) + 3 A <5(0 + 3B u( t ) + 2 A u( t ) = 4<5 (0 (4.89)
'
A <5'(0
Or in a more convenient form, we have
x( t ) = ( 9e - ‘ + Ue- ‘ ) u( t )
~ 2 (4.95)

The total solution of a differential equation with a step or impulse forcing

2 A ) u( t ) = 4<3'(0 (4.90) function is obtained in an equally straightforward manner. For a step
A <5'(0 + ( B + 3A ) <5(0 + ( C + 3 B +
forcing function, only the highest derivative term has a discontinuity at
Equating like coefficients on both sides of the Eq. 3.90 gives t = 0. Since we do not need the initial condition of the highest derivative
A= 4
term for our solution, we proceed as if we were solving a standard non -
homogeneous equation with a constant forcing function. For an impulse
B + 3A = 0 (4.91) driver, once we determine the initial conditions at t = 0 + , the equation is
: solved in the same manner as a homogeneous equation .
C + 3B + 2A = 0

A: I
 90 Network analysis and synthesis Differential equations 91
Example 4.7 Find the step and impulse response for the equation Substituting these initial conditions into x( t ) d/ dt d/ dt
and x'( t ), we find that Ax = -0.05 A2 = -0.1. XP(0 *»(0 ,
* (/ )

are
KM/) = /(/)
2*'(0 + 4 x' ( t ) +
where /(/) = <5(0 and /(f ) = u( t ), respectively. The initial conditions at t = 0

x(0 -) = *'(0-) = *'(0 -) = 0

—
Therefore, the step response is
S „(0 = 0.1[1 - e-<(0.5 sin 2/ +
cos 20] u( t ) (4.105) xs0 )
:J . :/ * (0
. .
FIG 4.1
/
x t)

Note that the impulse response and the

Solution. Let us first find the impulse response. We note that the x" term
step response are related by the equation
contains an impulse and a step; the x’ term contains a step ; the x term contains
a ramp, and is therefore continuous at / = 0. Thus a(0 + ) = *(0 ) = 0. To
obtain »'(0 + ), we use Eq . 4.81
— x» 0 ) =
J
at
xu( 0 (4.106)
K 1 1
(4.96) We can demonstrate Eq. 4.106 by the following procedure. Let us substitute
*'( + ) = - + *'(0 -) = 2 + 0 = j
° xu( t ) into the original equation
Note that we actually need only z(0 + ) and x' ( 0 + ) to evaluate the constants
-
for the second order differential equation. Next, we proceed to the comple- 2 + 4 ~dt + 10 Xu® = (4.107)
mentary function xe( t ). The characteristic equation is ?
d
H ( p ) = 2(/ + 2p + 5) = 2( p + 1 + j2 )( p + 1 j2 ) (4.97)

Since H( p ) has a pair of complex conjugate roots, we use a standard form for
— Differentiating both sides, we have

xc( 0- xe( t) = Me-* sin ( It + 4> ) (4.98)

2
£l
from which Eq. 4.106 follows.
=W (4.108)

Substituting the initial conditions at t = 0 +, we obtain Generalizing, we see that, if we have the step response for a differential
*(0 +) = 0 = M sin </> '

(4.99) equation, we can obtain the impulse response by differentiating the step
=\ = 2 M cos <f> - M sin <f>
x' ( 0 + ) response. We can also obtain the response to a ramp function /(t) =
A p( t ) (where A is the height of the step) by integrating the step response.
from which we find </> = 0 and M = J. Thus the impulse response, which we
The relationships discussed here are summarized in Fig. 4.1.
denote here as ( t ), is
xs x6( t ) =\e~l sin 21 u( t ) (4.100)
4.5 1 NTEGRODIFFERENTIAL E Q U A T I O N S
Next we must solve for the step response xjjt ). For convenience, let us write the
complementary function as In this section , we will consider an integrodifferential equation of the
xc( t ) = e- sin 21 + A2 cos 2t ) (4.101) form
^
The particular integral is evaluated by considering the forcing function a an xn( t ) + an _ x xn\
~ t)
+ • • • + a0 x( t ) + O- ) dr = f ( t) (4.109)xjVr
constant f ( t ) = 1 so that ‘
1 1
(4.102) > where the coefficients { an , an_u .. . , a_1} are constants. In solving an
xvV ) (o) 10 equation of the form of Eq. 4.109 we use two very similar methods. The
The total solution is then
^ first method is to differentiate both sides of Eq. 4.109 to give
A2 cos 2t )e-* (4.103) ; t ) + an_ x x n>( t ) +
+ a0 x’( t ) + a_ x *(0 = /'(<) (4.110)
x( t ) = ( Ai sin It +
Since x' ( t ) and x( t ) must be continuous for a step forcing function,
+
^ an x( n+\
1 (

The second method consists of a change of variables. We let \ y t ) = x( t ) ;

i Eq. 4.109 then becomes
*(0 + )= x(0 ~ ) = 0 (4.104) •
f-
2/ < + (0 + «» i V \t ) +
n 1) {n
*'(0 +) = x'(0 ~ ) = 0 - ••
+ a0 V\t ) + y( t ) = /(/) (4.111)

Differential equations 93
92 Network analysis and synthesis
Note that from Eq. 4.110 we obtain x( t ) directly. From Eq. 4.111, we Differentiating y( t ), we obtain
obtain y( t ), which we must then differentiate to obtain x( t ) An important . x( t ) = y'( t ) = 4e-‘ ~ 3e~ 2
point to keep in mind is that we might have to derive some additional ‘ (4.124)

initial conditions in order to have a sufficient number to evaluate the

unknown constants. 4.6 SIMULTANEOUS DIFFERENTIAL E Q U A T I O N S

Example 4.8 Solve the integrodifferential equation Up to this point, we have considered only differential equations with a
single dependent variable x( t ). In this section, we will discuss equations
x' ( t ) + 3x( t ) + 2 X( T ) dr = 5u( t ) (4.112) with more than one dependent variable. We shall limit our discussion to
o
equations with two unknowns, x( t ) and y( t ). The methods described here,
-

The initial condition is <c(0 ) = 1. —

Solution. Since the characteristic equation of Eq. 4.112 is of second degree
,
however, are applicable to any number of unknowns. Consider first the
system of homogeneous equations
we need an additional initial condition ®'(0 + ). We obtain x' ( Q + ) from the

given equation at / = 0 + : *'(0 + + Pi y’( 0 + Po 2/(0 = 0

«o *(0
*'(0 +) + 3*(0 + ) + 2
Since z(/) is continuous at / = 0,
J x( r ) dr =5 (4.113)
Yi xX0 + Yox( 0 + 2/ (0 + <5« 2/(0 = 0
where <xit /9 yit are arbitrary constants. The complementary function is
(4.125)

+
/*0
(4.114) ^
obtained by assuming that
x( r ) dr =0
L
and = x(0 -) = 1
x(0 +) (4.115) x( t ) = Cxevl\ y( t) = C2e* ‘
Therefore, x' ( 0 + ) = 5 - 3cc(0 + ) = 2 (4.116)
so that the characteristic equation is given by the determinant
METHOD 1. Differentiating both sides of Eq. 4.112, we obtain
(4.117) H(P ) =
( «iP + ao) ( PiP + Po ) (4.126)
x" ( i) + 3 x' ( t ) + 2x( t ) = 5<5(/)
(YiP + Yo ) (<5iP + d0 )
The complementary function is then
The roots of H( p ) are found by setting the determinant equal to zero,
x „( t ) = Cie~‘ + C2<?-2‘ (4.118)
that is,
Using the initial conditions for x( 0 +) and x'(0 + ), we obtain the total solution „
( OLIP + «o)(<5iP + <5 ) - ( PiP + Po )(YiP + Yo ) = 0 (4.127)
x( t ) = 4e~ - 3e 2 ‘ (4.119)
‘
'

It is seen that a nontrivial solution of H( p ) = 0 exists only if

METHOD 2 . Letting y'( t ) = x( t ), the original differential equation then
becomes
(4.120)
: ( « iP + «0)(V + <5«) ^ ( PiP + PoXYiP + Yo ) (4.128)
y"( t ) + + 22/(0 = 5H(0
32/ (0
Assuming that the preceding condition holds, we see that H( p ) is a second-
(0 + ) = a:(0 + ) = 1 v
We know that
^
2/'(0 + ) = *'(0 + ) = 2
(4.121) degree polynomial in p and can be expressed in factored form as
H( p ) = C( p - p )( p - px )
0 (4.129)
From Eq. 4.120, at / = 0 +, we obtain
where C is a constant multiplier. The complementary functions are
2/(0 + ) = £[5 - 2/'(0 + ) - 32/ XQ + )] = 0 (4.122)
x( t ) = Kxevo* + K2e*d
Without going into details, the total solution can be determined as (4.130)
y( t ) = K3efo* + Kxe* « •
2/(0 —
= 4e~‘ + fe ‘ + f
_2 (4.123)

&
k
-
 Differential equations 95
94 Network analysis and synthesis
initial conditions. Next, let us determine the solutions for a set of nonhomogeneous
and the constants Ky , K2 , K 3 , Kx are determined from roots ; i .e. , differential equations. We use the method of undetermined coefficients
of double
As in the case of a single unknown, if H( p ) has a pair here. Consider first an exponential forcing function given by the set of
if po — pit then equations
x( t ) = ( Ky + K2 t )e*°‘ + a0a; + &J/' + poy = Neet
(4.131) (4.139)
y( t ) = ( K , + Kxt )e*°' Yix' + Yox + +\y = 0
If H( p ) has a pair of conjugate roots,
We first assume that xv( t ) = Aeet (4.140)
V .{ t ) = Be°‘
Pi z Then Eq . 4.139 becomes
=a ± ju>
Pi* .
M+ + (M + fto )B = N
(4.141)
then x( t ) = Mle° i sin ( cat + f> y ) (

(4.132) ( yiB + y0 ) A + ( did + S0 )B = 0

y( t ) = M 2eal sin ( cot + <f> 2 ) The determinant for the set of equations 4.141 is
«i6 + a0 ft$ + /?„
Example 4.9. Consider the system of equations H ( d ) = A( d ) = (4.142)
2x'( i) + 4x( t ) + y'( t ) - y( t ) = 0 ViB + Y o W + <5o
(4.133)
where H ( d ) is the, characteristic equation with p = d . We now determine
x'( t ) + 2x( t ) + y'( t ) + y(0 = 0 the undetermined coefficients A and B from A(0) and its cofactors, namely,
with the initial conditions
x'(0 + ) = 2 y'( a +) = - 3
(4.134)
m (4.143)
a:(0 +) =0 2/(0 +) =1 NA12( d )
B=
A (0)
Solution . The characteristic equation is where Ai 3 is the ijih cofactor of A(0) .

H( p) =
|2p + 4 p 11 = 0 — (4.135) Example 4.10. Solve the set of equations
|
p +2 p +1
+ 4 x + y' - y = 3e"
2x'
(4.144)
Evaluating the determinant , we find that x’ + 2x + y' + y = 0

H( p ) = p3 + 5p + 6 = ( p + 2 )( p + 3) (4.136) given the conditions z'(0 + ) = 1 , a;(0 +- ) = 0, 2/'(0 + ) = 0, 2/(0 + ) = 1 . —

Solution . The complementary functions xc( t ) and yc( t ), as well as the charac-
so that y( t ) = Kie-2t + K2e~3t teristic equation H( p ), were determined in Example 4.9. Now we must find
(4.137) A and B in the equations
x( t ) = K 3e-2t + Kxe-3t
*„(/) = Aeu (4.145)
With the initial conditions, *'(0 + ) = 2, a (0 + ) = 0, we obtain K 3 = KX = - 1 .
; 2, 2. Vp( t ) = Be"
From the conditions y'(0 + ) = — 3, y(0 + ) = 1 , we obtain Kx = Kt =
0 ,
The characteristic equation with p = 4 is
Thus the final solutions are
x( t ) = 2e ~2t - 2e ~31 2(4) + 4 (4) - 1
(4.138) Hi4 ) = = 42 (4.146)
~ st (4) + 2- (4) + 1
2M = e

Differential equations 97
96 Network analysis and synthesis
5 A„(0) 5(3)
Then we obtain from Eq . 4.143 the constants and Ci = A(0) 5
B = -f (4.155)
A- A (5 ) 1
Ci = “
5 = -i
The incomplete solutions are
u The general solution is then
x( t ) = Kie ~
+ K2e~zt + Ae*‘ (4.147)
-3t
+ Kxe 3t - je4‘~ Axe l cos 2t + A2e l sin 2 t + 3
x( t ) = ~ ~

2/(0 = K 3e (4.156)
~
2/( /) = Bie l cos 2t + B2e l sin 21 - 1
~

Substituting for the initial conditions, we finally obtain

In order to find Av A2 , Bu B2 , we need the values x(0 + ), x’( 0 + ), y( 0 + ),
2/(0 = }( - 4e-3t - 3c" ) (4.148) 2/ 0 + ). The values for *(0 + ) and y( 0 + ) are first obtained by integrating the
'(
- 6e-2i + e 3t + 5e" ) ~ original differential equations between t = 0 — and / = 0 + . Thus,
x( t ) =
^
Example 4.11. Solve the system of equations
/» o+
( 2x' + 4 x + y' + 7 y ) dt =
r 0+
5 u( t ) dt
(4.157)
2 x' + 4x + y’ + ly = 5u( t ) (4.149)
’0 +
| ro+
x' + x + y' + 3 y = 5.5(0 J-
0
( x' +x+ ?/' + 3 y ) dt =
J
5<5(0 dt

given the initial conditions We know that only the highest derivative terms in both equations contain
impulses at / = 0. Moreover, both a;(0 and 2/(0 contain, at most, step dis-
z(0 - ) = x'(O -) = 2/(0 -) = 2/'(0 -) = 0 continuities at / = 0. Therefore, in the integration
Solution. First we find the characteristic equation ro i
(4a; + 7 y ) dt = 0
Jo—
r
2p + 4 p + 7 (4.158)
H ( p ) = A( p) = (4.150)
p+1 p+3 (x + 3 y ) dt =0
which simplifies to give
After integrating, we obtain
(4.151 )
H( p ) = ( p3 + 2p + 5) = ( p + 1 + j 2 )( p + 1 - j 2 ) 2x(0 + ) + 2/ (0 + ) = 0
The complementary functions xe( t ) and ye( t ) are then
i a;(0 + ) + 2/(0 +) = 5
(4.159)

Solving, we find
xc( t ) = Aye-* cos 21 + A2e~l sin 2t (4.152) x( 0 + ) = —5 2/(0 + ) = 10 (4.160)
yc( t ) = Byecos 2t + B2e~* sin 2t
To find z'(0 + ) and 2/ '(0 + ), we substitute the values for ®(0 + ) and 2/(0 + ) into
with t > 0,
The particular solutions are obtained for the set of equations the original equations at / = 0 + . Thus,
namely, 2a;'(0 + ) - 20 + 3/ (0 + ) + 70 = 5
2a;' + 4* + y' + 7 y = 5 (4.161 )
(4.153)
3y =0 a;'(0 + ) - 5 + 2/ (0 + ) + 30 = 0
x' + + x 2/' +
Using the method of undetermined coefficients, we assume that xv
and yp are so that *'(0 + ) = -20 (4.162)
constants: xv = Q; yt = C2. Since the forcing function
is 5 = 5eot
5
,
can
we
be
can
regarded
solve for
i 2/ (0 + ) = -5
as an exponential term with zero exponent , that ,
Thus Substituting these values into Eq . 4.156, we eventually obtain the final solutions
Cy and C2 with the use of the characteristic equation H( p ) with p = 0. ,

14 7
x( t ) —
= ( 8e ‘ cos 21 — 14e~* sin 2/ + 3) u( t )
~

(4.163)
H ( 0) = A(0) = =5 (4.154) y( t ) = ( lie-* cos 21 + 3e-‘ sin 2/ - 1 ) u( t )
1 3

..

i i
 Differential equations 99
Network analysis and synthesis
98
with the intial conditions x(0 — ) = 12,
x<3>(0 +), *'(0 +), *'(0 + ), and x(0 +).
*'(0 — ) = 6, and x"( Q — ) = — 7, find
Problems 4.7 Given the initial conditions x'(0 — ) = x'(0 — ) = 0, find the solutions
4.1 Show that for the equations
= M1e t cos 2/
~
x2( t )
(«) x’( t ) + 2x’( t ) + 2x( t ) = 26( 1 )
and x2( t ) = M2e~* sin 2/ ( b) x’ ( t ) + 7x'( t ) + 12 x( t ) = 5u( t )
are solutions for the equation 4.8 Given the initial condition *(0 — ) = — 2, solve the integrodifferential
x’( t ) + 2x'(0 + 5x( t ) = 0 equations

Show that x 2 + x 2 is also a solution .

4.2 Determine only the form of the solution
for the equation
I
(«) x' ( t )
*
j
+ 5 x( t ) + 4 x( r ) dr - 2 sin t

x"( t ) + 4 x' ( t ) + 2 x( t ) = 0
(«)
( b) x’( t ) + 8*'(/) + 5*(/) = 0
_
*'(/) 5 x( t ) = 0
(b) x'( 0 + 2x ( t ) + 2
J X( T ) dr = 1e * ~

(c)
(d )
( e)
x"( t ) + 5 x( t ) = 0
x’( t ) + 6 x ( 0 + 25 x( t ) = 0
'
(c) x’( t )

4.9 Given the set of equations

J
+ 6 x ( t ) + 9 x( r ) dr = 2 u( t )

(/ ) x' ( t ) + 6 x' ( t ) + 9 x( t ) = 0
»'(0 +) = — 1 > determine the
*'< / ) + x( t ) + y( t ) = u( t )
4.3 Given the initial conditions x(0 +) = 1 , •W + y’( t ) + 2 y( t ) = 6( t )
solutions for
x’( t ) + 6x'( t ) + 25x( t ) = 0 with x(O -) = x'(0 — ) = 2/(0 - ) = y' ( 0 - ) = 0, find x(0 +), x'(0 + ), 2/(0 + ), and
(«)
2/'(0 + ).
( b) x’( t ) + 8x' ( t ) + 16a;(/) = 0
4.10 Solve the set of equations
(c) x"( t ) + 4.81a;'(0 + 5.76 x( t ) = 0
2*V ) + 3 x( t ) + y’( t ) + 62/(0 = <K 0
4.4 Find only the particular integrals for the equations + 62/(0 = «(0
3t *'(0 + x( 0 + y' ( 0
x' ( t ) + 7 x'( t ) + 12a:(/) = e
~

(«)
All initial conditions at / = 0 - are zero.
0b ) x’( i) + 2 x' ( t ) + 2x( 0 = 2 sin 3/
4.11 Solve the set of equations
*'(0 + 2 x' ( t ) + 5a ( /) = ~‘st
(c)
; e~ sin 2 /
x'( t ) + 2 x'( t ) + 5 x( t ) = e / 2 2x’( t ) + 2x( t ) + 2/'(0 - 2/(0 = 2 5( 0
(

id )
W x’( t ) + 5.0x'( t ) + 6.25x( / ) = 6 ®'( 0 + *(0 + 2/'(0 + 22/(0 = 3e-< «( /)
~x ~3 t
x’( t ) + 6x'(0 + 5x( /) = 2e + 2 e The initial conditions are
V)
x'(0 +) = 0, determine the
i
4.5 Given the initial conditions x(0 + ) = 1 , -
x(0 ) = 1 2/(0 -) = 0
solutions for x'(O -) = - 1 2/'(0 -) =0 1
(«) x’( t ) + 4x'( t ) + 3x(0 = 5e * sin 2/
x"( t ) + 6 x'( t ) + 25 x( t ) = 2 cos t
x’( t ) + 8x'(0 + 16x(0 = 2
,
( 0
4.6 ( a ) A system is described by the differential equatio
n
y’( t ) + 3 y'( t ) + 22/(0 = <5(0
— 2. Find 2/(0 + ) and y’( 0 + ).
The initial conditions are y(0 ) = 1 ; y’(0 - ) =
( b) Given the differential equation
x*\t ) + Ux\t ) + 8a;( / ) = 6<5(0
.

V,
I
Network analysis: I 101

Energy
source Network

FIG. 5.1. Switching action.

chapter 5 switch is closed, the currents and voltages in the network have known
values. These values at / = 0 — are the initial conditions. We must then
determine the values of the currents and voltages just after the switch
Network analysis: I closes (at t = 0 + ) to solve the network equations. If the excitation is not
an impulse function or any of its derivatives, the current and voltage
variables are continuous at t = 0. For an impulse driver the values at
0 + can be determined from methods given in Chapter 4. Having
obtained the initial conditions, we then go on to solve the network
differential equations. Unless otherwise stated , all the solutions are valid
5.1 INTRO DUCT ION only for t 0 + .
Since we are dealing only with linear circuits, it is essential that we bear
dge of differential equations to
In this chapter we will apply our knowle networks. We will assume in mind the all-important principle of superposition. According to the
the analysis of linear, passive, time-invariant superposition principle, the current through any element in a linear circuit
Kirchhoff’s current and voltage
that the reader is already familiar with with n voltage and m current sources is equal to the algebraic sum of
laws, and with methods for writing mesh
and node equations for a-c or d-c
briefly the problem of writing currents through the same element resulting from the sources taken one
circuits.1 We will, therefore, consider only variable is time t . The at a time, the other sources having been suppressed . Consider the linear
mesh and node equations when the independent : Given an excitation network depicted in Fig. 5.2a with n voltage and m current sources.
format
problems in this chapter have the following , a specified response that
source and the networ k 2+
signal from an energy determined. When relating V2 L
in the networ k is to be +
is a current or voltage
Chapter 4, we shall see that,
these problems to the mathematics in
to the excitation ; the network S>n
physically, the forcing function corresponds; and the unknown variable x( t ) +
is described by the differential equation +/
is the response. *>11
First, we must write the
The problems encountered will be twofold
,

Kirchhoff ’s current and voltage

differential problems of the network using Linear
for a specified current or
laws. Next, we must solve these equations network
are equally important , It is
voltage in the network . Both problems
equation which is set up in-
useless , for example, to solve a differential '

ctly specified .
correctly, or whose initial conditions are incorre
chapter might generally be
The usual type of problem presented in this
0, which connects an energy
described as follows. A switch is closed at / =5.1). The analog of a switch
(voltage or current) source to a network
(Fig. S
the energy source whose output is e( t ) u( t ). Before the <2 !
closing at t = 0 is Tk
1For a comprehensive treatment , see H. H
. Skilling, Electrical Engineering
Circuits .
FIG 5.2a
( 2nd Edition), John Wiley and Sons, New
York, 1965 .
100

I
I
 102 Network analysis and synthesis Network analysis: I 103
principle, the total current iT ( t ) due to all of the sources is equal to the
VJ
+ algebraic sum
h =\ + 4, + h 4„ + 4, + *«, + •• + hm (5.1)

5.2 NETWORK ELEMENTS

In this section, we will discuss the voltage current ( v i ) relationships - -

5)E
that exist for the basic network elements. Before we examine these
Linear
network relationships, it is important to assign first arbitrary reference polarities for
the voltage across an element, and a reference direction of flow for the
current through the element. For the purposes of our discussion, we
assume that the positive polarity for voltage is at the tail of the current
arrow, as shown by the resistor in Fig. 5.3. Now, let us review the voltage -
current relationships for the resistor, the inductor, and the capacitor,
which we first discussed in Chapter 1 .
Resistor
FIG. 5.2 b
The resistor shown in Fig. 5.3 defines a linear proportionality relation-
a element Z, ship between v( t ) and i( t ), namely,
Suppose we are interested in the current iT( t ) through given
as shown. Let us open-circuit all the current sources and short circuit -
—
n 1 voltage sources, leaving only v / jt
to
)
the
shown
voltage
in Fig .
source
5.26
v ,( t
.
)
By
alone
4/0,
. In
*(00 == R v(0
i G (t )
i
G = -R
1 (5.2)
we denote the current through Z due due to the
the current through Z where R is given in ohms and G in mhos.
similar fashion, by 4/0 we denote
current source ik( t ) alone, as depicted in Fig . 5.2c . By the superposition
Capacitor
.
For the capacitor shown in Fig 5.4a the v i relationships are -
i(0 = C
^>
dt
(5.3)

Linear
network 5>E ' i( t )
"
-o
(0 = ~

^
Jj (T) dr + %(0 — )
-o

A M°
-o

-t
)

:R V (t)
i(t ) c v( t )
Hr v( )

(a) ( b)
ik . .
FIG 5.3 Resistor. .
FIG 5.4. (a) Capacitor, ( b ) Capacitor
FIG. 5.2c. Superposition in linear circuits. with initial voltage.

104 Network analysis and synthesis Network analysis: I 105

mo We must, in addition, choose the mesh currents such that at least one mesh
current passes through every element in the network.

L m —
»(0 )
v( t )
Example 5.1. In Fig. 5.6, a network is given with seven branches and five nodes.
—
We therefore need 7 5 + 1 = 3 independent mesh equations. The directions
of the mesh currents 4» 4, i3 are chosen as indicated . We also note that the

( b)

.
FIG. 5.5 ( a ) Inductor , ( b ) Inductor with initial current.

where C is given in farads. The initial value cc(0 ) is the voltage across
the capacitor just before the switching action. It can be regarded
as an
—
independent voltage source , as shown in Fig. 5.46 . We should point out
—
also that i>c(0 ) = fc(0 +
derivatives of impulses.
) for all excitations except impulses and
FIG 5.6 .

Inductor capacitors in the circuit have associated initial voltages . These initial voltages
are assigned reference polarities, as shown in the figure. Now we proceed to
The inductor in Fig. 5.5a describes a dual relationship between voltage
and current when compared to a capacitor. The v i relationships are - write the mesh equations.
Mesh 4 :
,,(0. = L-di
— J
*
T

(5.4)
vi( 0 —
vCi( 0 ) = 7?! 4(0 +— 4(r ) d T ~ cf 4(t)

- \
7L J o‘-cir ) dr + iL(0 ) -- Mesh i2 :
1(0
. jjl( )
where L is given in henrys. The initial current 4,(0 ) can be regarded
as
—
®C ,(? -) - ®C (0 “) = - ~ r dr +L
^
-k
an independent current source, as shown in Fig. 5.56. As +( + )
k kS
^
is true for the

^\J TUr
dr
voltage across the capacitor, the current through the inductor is similarly
continuous for all t , except in the case of impulse excitations.
Mesh 4 :
- j
When the network elements are interconnected, the resulting i v equa-
tions are integrodifferential equations relating the excitation (voltage
current sources) to the response (the voltages and currents of the
There are basically two ways to write these network equations
way is to use mesh equations and, the second , node equations.
.
elements
The
or
).
first
' + "c/°
() ) = jnT "
\ ( r ) dr +

.
After we find the three unknowns, 4» 4 and j8, we can determine the branch
currents and the voltages across the elements. For example, if we were required
to find the branch currents ta and iCt through the capacitors, we would use the
^ ( r ) dr + R2 i3( t ) (5.5)

Mesh equations are based upon Kirchhoff ’s voltage law. On the mesh following relationships
basis, we establish a fictitious set of loop currents with a given reference 'c, = 4 - 4 (5.6)
direction, and write the equations for the sum of the voltages around the
loops. As the reader might recall from his previous studies, if the number ‘ — —4 c, 4
, the Alternatively if,the voltage v ( t ) in Fig. 5.6 is our objective we see that ,
of branches in the network is B, and if the number of nodes is N
3

number of independent loop equations for the network is B N + 1.® — v3(0 = 4(0^2 (5.7)
Network equations can also be written in terms of node equations,
a See Skilling, op. cit . which are based upon Kirchhoff ’s current law. If the number of nodes in
 Network analysis: I 107
106 Network analysis and synthesis
There are two ways to obtain initial conditions at / = 0 + for a network:
Vl
JT57TIP- ~
-
id0 ) l>2 (a) through the differential equations describing the network, (b) through
knowledge of the physical behavior of the R, L, and C elements in the
network.

'*(/ ) ~c
Gl Gz Initial conditions for a capacitor
> > For a capacitor, the voltage-current relationship at / = 0 + is

T
.
FIG 5.7
Datum

°— ^
^
vc(°+ ) = j _ >W dr + vc( 0 — )
If /(/) does not contain impulses or derivatives of impulses, fc( + ) =
wc( )• <1 is the charge on the capacitor at t = 0 , the initial voltage is — °
(5.9)

the network is N , the number of independent node equations required

for the complete solution of the network is N l .3 We can therefore
select one datum node in the network. All the node voltages will be
— yc(0 + ) = vc( Q ~ )
When there is no initial charge on the capacitor, uc(0 +) = 0. We con-
— ~ (5.10)

positive with respect to this datum node. clude that when there is no stored energy on a capacitor, its equivalent
Consider the network in Fig. 5.7. Let us write a set of node equations circuit at t = 0 + is a short circuit. This analogy is confirmed by examining
for the network with the datum node shown. Since the number of nodes

These are written for nodes v1 and v2 , as given below.

—
in the network is N = 3, we need N 1 = 2 independent node equations.
the physical behavior of the capacitor. As a result of the conservation of
charge principle, an instantaneous change in voltage across a capacitor
implies instantaneous change in charge, which in turn means infinite
Node vx : current through the capacitor. Since we never encounter infinite current
in physical situations, the voltage across a capacitor cannot change
i/ 0 - iL( 0 — ) = Gi »i ( 0 + - jJ ®i(r) dr ~
f Jj*( ) > T dr
instantaneously. Therefore at t = 0 + , we can replace the capacitor by a
voltage source if an initial charge exists, or by a short circuit if there is no
Node v2 : initial charge.

1,(0 -) = L Jo- —— f
V / T ) dr +
L"O
— f—
v 2( r ) dr + C -
dt
^ - + G v (t ) 2 2 (5.8) . -
Example 5.2 Consider the R C network in Fig. 5.8«. The switch is closed at
t = 0, and we assume there is no initial charge in the capacitor. Let us find
the initial conditions i(0 + ) and /'(0 + ) for the differential equation of the circuit
Further examples are given in the following sections.
jjiOdr
5.3 INITIAL AND FINAL CONDITIONS

In this section, we consider some methods for obtaining initial conditions

for circuit differential equations. We also examine ways to obtain particular
C
Vu( 0 = R i( t ) +

^ (5.11)

integrals for networks with constant (d-c) or sinusoidal (a c) excitations. -

In the solution of network differential equations, the complementary + R
function is called the transient solution or free response The particular .
integral is known as the forced response. In the case of constant or periodic
excitations, the forced response at t = co is the steady state or final - R-C network
T
Equivalent circuit at < = 0+
solution. ( a) ( b)

. .
See Skilling, op cit
. - .
FIG 5.8, (a) R C network (b) Equivalent circuit at t = 0 +.

108 Network analysis and synthesis Network analysis: I 109

The equivalent circuit at t = 0 + is given in Fig. 5.86, from which we obtain From the equivalent circuit at / = 0 +, shown in Fig. 5.9b, we see that

i'(0
» > -£
+), we must refer to the differential equation
(5.12) /(0 + ) = 0
We then refer to the differential equation to obtain / '(0 + ) -
(5.18)

To obtain
/(/)
V — L i '(0 + ) + Ri(0 + ) (5.19)
vm — R i'iO + -Q (5.13) V R
L - L'
=-
Thus /'(o + ) ~ (0 + )

At t = 0 + we have 0 = R i' ( 0 + ) + —+
i (0 )
(5.14) V
L
(5.20)

We then obtain '< >

" 0+
- » (0 + )
~~
RC F?C
V
(5.15)
The steady state solution for the circuit in Fig. 5.9« is obtained through the
-
-
knowledge that for a d c source, an indicator is a short circuit.
The final condition, or steady-state solution, for the current in Fig. 5.8
« is
- V
obtained from our knowledge of d -c circuits. We know that for a d c excitation ,
i/ 0 = « ( °°) = (5.21)
a capacitor is an open circuit for -
d c current . Thus the steady state current is - ^
i/ 0 = /( oo) = 0 Example 5.4 .
In this example we consider the two-loop network of Fig. 5.10.
As in Examples 5.2 and 5.3, we use the equivalent circuit models at / = 0 +
Initial conditions for an inductor and I = oo to obtain the initial conditions and steady-state solutions. At
For an inductor, the voltage-current relationship at / = 0 + is t = 0 the switch ploses. The equivalent circuits at t = 0 + and / = oo are
shown in Fig. 5.11« and b, respectively. The initial currents are
ii(0
'
ff
+ ) = L Jo- v( r ) dr + ijf 0-) (5.16)
V
If v( t ) does not contain impulses, then /i(0 + ) = iL( 0 ). If there is no
initial current, i'L(0 +) = 0, which corresponds to an open circuit at
— hi0 + ) = ir
4(0 +) = 0
Ri (5.22)

t == 0 + . This analogy can also be obtained from the fact that the current
through an inductor cannot change instantaneously due to the conserva-
-
The steady state solutions are
V
tion of flux linkages. 4( °°) = i?i + R2 = 4( °° ) (5.23)
Example 5.3 In Fig. 5.9«, the switch closes at / = 0. Let us find the initial
Final conditions for sinusoidal excitations
conditions /(0 + ) and /'(0 +) for the differential equation
When the excitation is a pure sinusoid, the steady-state currents and
Vu( t ) = L ~ f ~ + RH0 (5.17) voltages in the circuit are also sinusoids of the same frequency as the
excitation. If the unknown is a voltage, for example, (f ), the steady-state
L solution would take the form ^
v1 / 0 = \V( jco0 )\ sin [co0t - f ( co )]
<> 0 (5.24)
where co0 is the frequency of the excitation, and |V( ja>0 )\ and
represent the magnitude and phase of vlv( t ). A similar expression would
hold if the unknown were a current.
To obtain the magnitude and phase, we follow standard procedures in
a-c circuit analysis. For example, consider the R C circuit in Fig. 5.12. -
(a) The current generator is
. . -
FIG 5.9 (a) R L network. (6) Equivalent circuit at t - 0+. i / 0 = (Jo sin (o0t ) u( t ) (5.25)

4
 NO Network analysis and synthesis Network analysis: I III
i?2 If the steady-state voltage takes the form shown in Eq. 5.24,
h
\V ( j°>o )\ =
inMOi (5.26)
C L /0
(G 2 + 002C 2 )'A
(

OQC
(
and <f>( co0 ) — tan G
(5.27)
.
FIG 5.10
A,
ao = (G
^ — tan 1
so that < sin eo0f
-
( 5.28)
2
+ a>0*cY G /
We refer to this problem in Section 5.5.

5.4 STEP AND IMPULSE RESPONSE

+
o.c.
As an introduction to the topic of solution of network differential
equations, let us consider the important problem of obtaining the step and
impulse responses for any voltage or
Equivalent circuit at t = 0+
( a) current in the network. As we shall see in
Chapter 7, the step and impulse responses T
* 1 *2 are precise time-domain characterizations + 1
vw VWNA
of the network. The problem of obtaining
+ the step and impulse response is stated as
follows : Given a network with zero initial
T' - I

energy, we are required to solve for a

specified response'(current or voltage) due FIG. 5.13
Equivalent circuit at t = to a given excitation function u( t ) or d( t ),
( b) which either can be a current or a voltage source. If the excitation is
. .
FIG 5.11 (a) Equivalent circuit at t = 0 +. (b) Equivalent circuit at / = co. a step of voltage, the physical analogy is that of a switch closing at
—
time l = 0 which connects a 1-v battery to a circuit. The physical
analogy of an impulse excitation is that of a very short (compared to the
—
v( t ) time constants of the circuit) pulse with large amplitude.
The problems involved can best be illustrated by means of examples.

R the circuit is
-
Consider the series R C circuit in Fig. 5.13. The differential equation of

v( t ) = <5( 0 = R i(0 + ~C Jo- (r) dr fli (5.29)

—
We assume vc(0 ) = 0. Since Eq . 5.29 contains an integral, we substitute
x’( t ) for /(/) in the equation, giving us
.
FIG 5.12
<5(0 = K *'(0 + ~ x( 0 (5.30)

t
112 Network analysis and
Integrating both sides between 0
syndesis

* >
—
0+
_
and 0 + gives
1
- (5.31) JoR
vu(0
Network analysis: I 113

The characteristic equation is

H ( p ) = Rp + —
1
C
(5.32)
0
( a)
t

and with little effort we have

,
x( t ) = - e-l RCu( t ) (5.33)
R Vt ( t )

so that
-
w ;[«« - Sc'*"«co]
which is shown in Fig. 5.14a. We thus arrive at the current impulse
( 5.34) b
c
.

response /(/) as the result of an impulse voltage excitation. In the process 0 t

we have also obtained the step response*(/) in Fig. 5.146 since, by definition, l
the derivative of the step response is the impulse response. The reader (b)
should check this result using a step excitation of voltage. . .
FIG 5.15 ( a ) Step response of parallel R-C circuit. (6) Impulse response of parallel
In the second example consider the parallel R-C circuit in Fig. 5.12 -
R C circuit .
driven by a step current source i( t ) = /0 u( t ), where 70 is a constant.
Assuming zero initial conditions, the differential equation is
i( t )
' 5( t )

t
I 0 u( t ) = Gv( t ) + C

from which we obtain the characteristic equation

H( p ) = Cp + G
^ dt
- (5.35)

(5.36)

The steady-state value of v( t ) is

R2C * =
vJt )
— =-
H ( 0) G
= /„« (5.37)

i( 0

Io
h

0 t

. . -
FIG 5.14 (a) Impulse response of R C circuit. (6) Step response of R C circuit.
- . .
FIG 5.16 Pulse excitation.
 114 Network analysis and synthesis Network analysis : I 115
\ v( t ) instance, if the unknown quantity is a branch current, it is preferable to
write mesh equations. On the other hand, if we wish to find a voltage
IoR across a certain element, then node equations are better. In many cases
the choice is quite arbitrary. If, for example, we wish to find the voltage v
across a resistor R , we can either find v directly by node equations or find
the branch current through the resistor and then multiply by R .
0 T\ t
\ Example 5.5. Find the current /(f ) for the
\ network in Fig. 5.18, when the voltage R= i a
- -AA/V-
- IoR
source is e( t ) = 2e~° M u( t ) and VQ( 0 ) = 0.
Solution. The differential equation is
-
.
FIG 5.17. Response to pulse excitation.
e{ t ) = Ri ( t ) +—
1
j / ( T) dr + vc(0 — )
0
Thus the complete solution for the voltage step response is
v( t ) = ( Ke~t / RC + hR ) u( t) (5.38 )
Or, in terms of the numerical values, we
(5.43)
^ C = if
:

have
From the initial condition u(0 + ) = 0, we obtain K = — T R so that
0
2e-o.st u( t ) = /(/) + 2 j* ( )
i r dr (5.44) .
FIG 5.18
v( t ) = I 0 R( l - e-i / RC ) u( t ) (5.39) *
Differentiating both sides of Eq . 5.44, we obtain
Differentiating Eq . 5.39 gives us the voltage impulse response
23(f ) - e~o st u( t ) = + 2/(f ) (5.45)
dt
v( t ) = -° e -tlRCu( t ) ( 5.40) 1'
C To obtain the initial condition /(0 + ), we must integrate Eq. 5.45 between
the limitsf = 0 — and f = 0 + to give /(0 + ) = 2. From the characteristic
The step and impulse responses are plotted on Figs. 5.15a and b . Suppose
equation
the excitation in Fig . 5.12 were a pulse
H( p ) =p +2=0 (5.46)
/( /) = /„[«( / ) - u( t - T )] (5.41 ) we obtain the complementary function as
shown in Fig. 5.16 . Then, according to the superposition and time- /c(f ) = Ke u ~
(5.47)
invariance postulates of linear systems, the response would be If we assume the particular integral to be ijj) = Ae ° - 5t ,
~
then we obtain
v( t ) = 70R [(1 - e ~t ,RC ) u( t ) - (1 - e- t-T ,RC ) u( t - T)]
{ )
(5.42) 1 2
A= - (5.48)
which is shown in Fig. 5.17 . H ( - 0.5) 3
The incomplete solution is
/(f ) = Ke~2‘ - fe-° - 5< (5.49)
5.5 SOLUTION OF NETWORK E Q U A T I O N S -
From the initial condition /(0 +) = 2, we obtain the final solution,
In this section we will apply our knowledge of differential equations /(0 = (i<r2( - §<r°-5t) «(f ) (5.50)
to the analysis of linear networks. There are two important points in
network analysis : the writing of network equations and the solution of As we already noted in Section 5.3 , in the solution of network differential
these same equations. Network equations can be written on a mesh, node, equations, the complementary function is called the free response, whereas
or mixed basis. The choice between mesh and node equations depends the forced response is a particular integral , and in the case of constant or
periodic excitation, the forced response at t = oo is the steady-state
largely upon the unknown quantities for which we must solve. For
solution. Note that the free response is a function of the network elements

116 Network analysis and synthesis

l
Network analysis : I 117
1
v( t )
alone and is independent of excitation. On the other hand , the forced
response depends on both the network and the excitation. Im
It is significant that, for networks which have only positive elements,
the free response is made up of only damped exponential and/ or sinusoids
with constant peak amplitudes. In other words, the roots of the charac-
teristic equation H( p ) all have negative or zero real parts. For example, .
ttW ® iR c
toC

if p1 is a root of H( p ) written as p^ = a ± j(o , then Re ( pj = a < 0.

This fact is intuitively reasonable because, if a bounded excitation produces
a response that is exponentially increasing, then conservation of energy is
not preserved . This is one of the most important properties of a passive
Datum
0 G —1 * Re

FIG 5.19 . FIG. 5.20

network . If a characteristic equation contains only roots whose real parts
are zero or negative, and if the jco axis roots are simple, then the network From the complementary function in Eq . 5.53, we see that the time constant of
which it describes is said to be stable ; otherwise, the network is unstable * the circuit is T = C / G = RC .
Stability is an important property of passive networks and will be discussed
in greater detail later. Next, let us examine an example of the different kinds of free responses
of a second-order network equation that depend on relative values of the
Example 5.6. For the R-C network in Fig. 5.19 with the excitation given by
network elements. Suppose we are given the network in Fig. 5.21 ; let us
Eq . 5.25, find the voltage v( t ) across the capacitor ; it is given that u(0 - ) =
find the free response vc( t ) for the differential equation
vc(0 -) = 0.
Solution . We have already obtained the particular integral in Eq . 5.28 .
Now let us find the complementary function . The differential equation on a
node basis is
dv
i „( 0 =C
dv
dt
+ Gt > + iL fJo-V ( T ) dr + ijr (0 )
— (5.58)

C + Gv - /Q sin mt (0 — (5.51 ) Differentiating both sides of Eq . 5.58, we have

“ 1
i' „( t ) = Cv”( t ) + Gv\t ) + - v( t ) . (5.59)
from which we obtain the characteristic equation as
H( p ) = Cp + G ( 5.52) The characteristic equation is then

- ( )
C‘ + G P + i = c P' +
^^
i. (5.60)
so that vc( t ) = Ke-Gt>G u( t ) ( 5.53) f l(p) +
and the incomplete solution is In factored form , H( p ) is
v( t ) = Ke ~ Gt / cu( t ) + (G 2
°
+ o)2OfA
sin ( iot — tan — j u( t ) 1 (5.54) H ( p ) = C( p - pi )( p - p2 ) (5.61)
^
From the initial condition t>(0 ) = 0, we obtain —
t> (0 + ) =
Io
K0 — ) = K - (G2 + <o2C2)!4 sin (5.55 ) sJK
From the argand diagram in Fig. 5.20 we see that ,
t _ (5.56)
| G L

) = (G + e^C )^
i
sin tan ~ 2 2
\ G
Consequently , iL(0 ~ )
I„ u( t ) T a>Ce-G‘lc
— tan — j J
«
^ (5.57)
1
+ sin
^L
v )= 2 <o /
(G + £O C ) (G2 + to2C2)!4
2 2

FIG. 5.21
4
This is not a formal definition of stability, but it suffices for the moment .
 118 Network analysis and synthesis Network analysis : I 119
v(t )

K2
KI + K2

0
0 t

.
FIG. 5.22 Overdamped response.

~ K2
PI
where = —A ± B (5.62)
P2 ) 2C 2 L \ C/ LCJ .
FIG . 5.24 Underdamped response.
There are three different kinds of responses depending upon whether B is
real, zero, or imaginary. CASE 3 . B is imaginary , that is,
CASE 1 . B is real, that is,

\(c)) > LC
2' -
Letting B = y/?, ' we have
I
\( cl < LC
£ -
then the free response is

vG( t ) = Kie
-u-BU . K e-( A+ BU
+ 2 (5.63) vc( 0 = e At
( K 1 sin /92 + K 2 cos /Sf ) (5.65)
which is a sum of damped exponentials. In this case, the response is said In this case, the response is said to be underdamped, and is shown by the
to be overdamped. An example of an overdamped response is shown in damped oscillatory curve in Fig. 5.24.
Fig. 5.22. Example 5.7. In this example we discuss the solution of a set of simultaneous
CASE 2. B = 0, that is, network equations. As in the previous examples, we rely upon physical reasoning

then Pi —
-
( g)' ±
\C / LC
Pi A
rather than formal mathematical operations to obtain the initial currents and
voltages as well as the steady-state solutions. In the network of Fig. 5.25, the
switch S is thrown from position 1 to position 2 at 2 = 0. It is known that
prior to t = 0, the circuit had been in steady state . We make the idealized
~At assumption that the switch closes instantaneously at 2 = 0. Our task is to
so that *o(0 = ( Ki + K* t )e (5.64) find 2,(2) and 4(0 after the switch position changes . The values of the batteries
When B = 0, the response is critically damped, as shown in Fig. 5.23. —
V 1 and V2 are V 1 = 2 v , V2 3 v ; and the element values are given as

v(t) L = 1 h, 1?! = 0.5 Q

c=K R2 = 2.0 n

The mesh equations for 4(0 and 4(0 after 2 = 0 are

Ki
V2 = L i\( t ) + 25, 4(0 - Ri 4(0 (5.66)
0 t
1 r
. .
FIG 5.23 Critically damped response. 0-) = — Ri 4(0 + gj ‘ 4( ) dT t + ( Ri + B2 ) 4(0 (5.67)

120 Network analysis and synthesis Network analysis : I 121

impulses . Before the switch is thrown at t = 0, the circuit with V 1 as the voltage
source was at steady state . Consequently,
rc(0-) = k1 = 2 v
Vi
4(0 -) = 24 4 amP (5.73)

4(0 -) = 0
We next find i \ (0 + ) from Eq . 5.66 at t = 0 + .
V2 = Li\( 0 + ) + 244(0 + ) - 25, 4(0 + ) (5.74)

Since Eq . 5.67 contains an integral, we differentiate it to give Substituting numerical values into Eq . 5.74, we find
24(0 + ) = 1 amp/sec
0 = — 2424(0 +
^ *2(0 + (24 + 24) 24(2) (5.68)
From Eq. 5.68 at t = 0 + , we obtain similarly

Using Eqs. 5.66 and 5.68 as our system of equations , we obtain the character- Ri
24(0 + ) = (0 ) 0.2 amp/sec
24 + 24 24 + =
(5.75)
istic equation .
Lp + 25, - 74 With these initial values of 2,(0 and i2( t ), we can quickly arrive at the final
H( p ) = 1 solutions
— R\P + (25, + 252)/> — 2,(0 = (0.5e~‘ - 2.5e-° - 6 t + 6) «(0
(5.76)
—
4(0 = ( 0.5e-‘ + 0.5e-° et ) u( t )
(§ + ,252) p +
= L( R1 + 752)/>2 +

Substituting the element values into H ( p ) , we have

H( p) = 2.5 p* + 4p + 1.5 = 2.5( p + 1 ){ p + 0.6)
75

^ (5.69)

(5.70)
which are plotted in Fig. 5.26.

6.0

5.6

The free responses are then 5.0

4c(0 = ( Kie-o *‘ + K2e ‘ ) t ( u )~

(5.71 )
4.5

4c(0 = ( K 3e 0 6 t + K ,e-‘ ) u( t )
~ 4.0
§ ° loT "

. The steady-state solutions for the mesh currents are obtained at t = 00 by

0.08
considering the circuit from a d-c viewpoint . The inductor is then a short
circuit and the capacitor is an open circuit ; thus we have 0.06 hit )
2 V 0.04
4,(0 = -Ri
=- = 6 amp (5.72) . 0.02
4,(0 = 0 1
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
Now let us determine the initial currents and voltages, which, incidentally, have
—
the same values at t = 0 and 2 = 0 + because the voltage sources are not .
FIG 5.26

I
 122 Network analysis and synthesis Network analysis: I 123

5.6 ANALYSIS OF TRANSFORMERS

1 h
According to Faraday' s law of induction, a current ix flowing in a coil Lx Voltage /"
source
may induce a current 4 in a closed loop containing a second coil Lz. The Voltmeter
sufficient conditions for inducing the Iron - core transformer $, I
o il
current 4 are: (a) part of the flux in
FIG. 5.28. An experiment to determine dot references.
. .
A
M + the coil Lj must be coupled magnetically
to the coil L2 ; (b ) the flux C4 must be
changing with time.
»1 »2
which we connect the positive lead of the voltage source. The dot reference
Li L2 In this section we will analyze circuits is placed on the secondary terminal at which the voltmeter indicates a
containing a device made up of two positive voltage. In terms of the primary current 4, the positive voltage at
magnetically coupled coils known as a the secondary dot is due to the current 4 flowing into the dot on the
° transformer. In Fig. 5.27, the schematic
"

primary side. Since the positive voltage at the secondary dot corresponds
FIG. 5.27. Transformer. of a transformer is given . The Lx side to the current i2 flowing into that dot, we can think of the dot references
of the transformer is usually referred to in the following way. If both currents are flowing into the dots or away
as the primary coil and the L2 side as the secondary coil. The only distinction from the dots, then the sign of the mutual voltage term M dijdt is positive.
between primary and secondary is that the energy source is generally at When one current flows into a dot and the other away from the second dot,
the primary side. the sign of M dijdt is negative.
The transformer in Fig. 5.27 is described mathematically by the If Nx and N 2 are the number of turns of coils Lx and L2 , then the flux
equations linkages of Lx and L2 are given by Nfb1 and Nfb2, respectively. If both 4
= Lj dJl + M dJl
2
vft ) and i2 flow into the dots, the sum of flux linkages of the transformer is
dt dt
(5.77)
2 d> linkages = Nft>x + Nf £> (5.79)
v2( t ) = M
dt —
+ L2 dJl
dt
2

If, however, one of the currents, for example, 4, flows into a dot, and the
other i2 flows out of the other dot, then
where M is the mutual inductance associated with the flux linking Lx to L2 ,
and is related to Lx and L2 by the relationship 2 $ linkages = Nfbx JV20, (5.80)
M = Ks[LxL2 (5.78) An important rule governing the behavior of a transformer is that the
The constant K in Eq. 5.78 is called the coefficient of coupling. It is sum of flux linkages is continuous with time.
bounded by the limits 0 |JK] 1. If |/ST| = 1, then all of the flux <14 in The differential equations for the transformer in Fig. 5.29 are
coil Lx is linked magnetically to L2. In this case, the transformer is a unity
coupled transformer. If K = 0, the coils Lx and L2 may be regarded as
- Lx i’ ft ) + Rx ift ) + Mi' ft )
Vu{ t ) =
(5.81)
two separate coils having no effect upon one another. 0 = M / 4(0 + R2 i2( t ) + L2 i' 2( t )
'

For circuits with transformers, we must establish reference polarities

for the mutually induced voltages M dijdi. Usually, the references are
given by small dots painted on the input and output leads of a transformer,
as shown by the dots on the schematic in Fig. 5.28. The reference dots are
placed at the time of manufacture according to the procedure outlined
here. A voltage source v is connected to the primary Lx side of the
transformer, as shown in Fig. 5.28. A voltmeter is attached on the second -
ary. At the primary side, the terminal is assigned the dot reference to .
FIG 5.29

*M» -

Network analysis : I
124 Network analysis and synthesis 125

Integrating this set of equations between t = 0

the determinant
— and t = 0+ results in
Since the excitation does not contain an impulse (and since j4Z,2
M 2 0, as we shall see later), we can assume that 4(0 + ) and 4(0 + ) are
^
also zero. Then using Eq. 5.86 we can find i' f 0 + ) and /'4(0 + ).
—
APx(0 + ) - 4(0-)] M[4(0 + ) - 4(0-)] 0 6 =\
i ( 0+ ) + i' f 0+ )
M[4(0 + ) - 4(0-)] 7X4(0 + ) - 4(0-)]
By evaluating this determinant, we obtain
=

( LXL2 - M 2 ) [ if 0 + ) - 4(0-)][4(0 + ) - 4(0-)] = 0

(5.82)

(5.83)
Solving, we find
0
-
i ( 0 +) + 2/ '2(0 + )
\
With these initial con-
i' f 0 + ) = 12 and i '2(0 + ) = 6.
ditions, we obtain the final solutions of iff ) and iff )
(5.90)

If LXL2 > M , that is, K < 1, then the currents must be continuous at
2 4(0 = (2 - i-e~2t - ie~12‘ ) u( t )
(5.91)
t = 0 in order for the determinant in Eq . 5.82 to be equal to zero. Thus 4(0 = C -i-e-21 + f «-»‘) u( t )
4(0 + ) = 4(0-), K<l (5.84) Suppose, now, LXL2 = M 2 , that is, K = 1, then 4(0 and iff ) need not be
and 4(0 + ) = 4(0-), K<1 (5.85) —
continuous at t 0. In fact, we will show that the currents are discon-
tinuous at t = 0 for a unity-coupled transformer. Assuming that K = 1,
Example 5.8. For the transformer circuit in Fig. 5.29, Lx = 1h , L2 = 2h , consider the mesh equation for the secondary at t = 0 +
Rx = 3fJ, R2 = 8t2, and M = \h. The excitation is V = 6u( t ) Let us find .
— °
4(0 and 4(0, assuming that 4(0 ) = 4( ~ ) =
Solution . The differential equations °- circuit are
for the
R2 4(0 + ) = ~M
M
14(0 + ) - L2 i' fO+ ) (5.92)

= - - [la *4(0 + ) + M / 4(0 + )] (5.93)

6«(0 i 0 ) + 34(0 + i'2(0
=\ Li
(5.86) The mesh equation of the primary side then becomes
o = i’ ft ) + 2/ 4(0 + 84(0
The characteristic equation is given by the determinant
V = Rx 4(0 + ) + [Lj / 4(0 +) + M / 4(0 + )] = Rx 4(0 + ) - k
M
R2 4(0 + )

+3 p (5.94)
H( p) = P
p 2y? +8 We need an additional equation to solve for 4(0 + ) and 4(0 + ). This is
.
provided by the equation
= ( p + 3 )( 2p + 8 ) - p2
4X4(0 + ) - 4(0-)] + M [if 0+ ) - 4(0-)] = 0 (5.95)
= ( p2 + 14 p + 24)
= ( P + 2 )( p + 12) (5.87)
which we obtained from Eq. 5.82. Since 4(0 )
Eqs. 5.94 and 5.95 directly to give
— = 4 0 — = 0, we solve
( )

Thus the complementary functions are VL2

4(0 + ) =
.
4 (0 = Kxe- 2 t + K2 e -

(5.88)
RXL2 + R 2 LX
-VM
(5.96)
4 (0 = K2e 2 t + KAe ^
~ ~ 4(0 + ) =
C
RXL2 + R2 LX
To obtain the particular integral, or steady-state solution, we rely upon Consider the following example. For the transformer in Fig. 5.29
physical reasoning to arrive at the element values are
Lx = 4 h, L2 = lh
4X 0 = —=
V
2 amp
(5.89) Rx == 8 Q , R2 = 3 fi

4X0 = 0 amp
M 2 h,
— F = 10 v
 126 Network analysis and synthesis f Network analysis : I 12/

Assuming that the circuit is at steady state before the switch is closed at Problems
t = 0, let us find ift ) and i2( t ). The differential equations written on mesh
basis are 5.1 Write the mesh equations for the network shown.

10 —
= 4 7i + Sil + 2 ~dk
ft — njinp’
— *' '—r
22
- * 2 (0 — )

0=2

The characteristic equation is

— dt
+ —+
dt
3i2
(5.97)

e( t ) u( t )
* i <o

ii
)

3 R2

H( p ) =
4( p + 2) 2'P
=0 (5.98) PROB 5.1 .
2P (P + 3)
which yields Hip) = 20p + 24 = 20{ p + f ) (5.99) 5.2 Write a set of node equations to solve for the voltage v2( t ) shown in the
figure.
so that the complementary functions are
VW^
hoi0 =
(5.100)
hciO = K2e ' ~
Li Ci
The particular integrals that we obtain by inspection are i( t ) u( t )
© §*» Ip 0’ ~ I
R2 V 2( t )

vc0(° )

(5.101)
PROB. 5.2
W0 = 0 -
The initial conditions are 5.3 The network shown has reached steady state before the switch S is
opened at t = 0. Determine the initial conditions for the currents hit ) and
VL2 i2( l ) and their derivatives.
li(0 + ) =
RiL2 + R 2 LI
—
10
20
= 0.5 R R
—
h( 0+ ) = RyL —
2
VM
+ R2 L1
= -1.0
(5.102)

i
+
f 0 AAA

L
Wv
< 2 (0 , ±: C
We then find —
Kx = 0.75 and K2 = 1.0 so that — T
—
hit ) = ( 0.75e - * + 1.25) u( t )
-1 2

(5.103) PROB 5.3 .

i2{.t ) = uit )
5.4 The network shown has reached steady state before the switch moves
We see that as t approaches infinity /,(/) -> 0, while hit ) goes its steady - .
from a to b Determine the initial conditions for iL(i) and vc( t ) and their first
state value of 1.25. derivatives. Determine also the final values for iL( t ) and vc( t ).
i

The Laplace transform

Differential Algebraic
equation equation X( s ) x( t )
Transform Solve Invert

. .
FIG 6.1 Philosophy of transform methods.

where fit ) is the forcing function, x( t ) is the unknown, and y{ x{t )) is the
chapter 6 differential equation. Let us denote the transformation process by 7’( ),
and let j be the frequency variable. When we transform both sides of
The Laplace transform Eq. 6.4, we have 7M*(0)] = (6.5)nm
Since frequency domain functions are given by capital letters, let us write
Eq. 6.5 as Y( X( s ), s ) F( s ) (6.6)
=
where X( s ) = T [ x( t )\, F( s) = T[/(r )], and Y( X( s ), s ) is an algebraic
6.1 THE PHILOSOPHY OF TRANSFORM METHODS equation in s. The essence of the transformation process is that differential
equations in time are changed into algebraic equations in frequency. We
In Chapters 4 and 5, we discussed classical methods for solving differ- can then solve Eq. 6.6 algebraically to
ential equations. The solutions were obtained directly in the time domain .
obtain Xis) As a final step, we perform Differential
since, in the process of solving the differential equation, we deal with - an inverse transformation to obtain e( t ) equation r( t )
functions of time at every step. In this chapter, we will use Laplace xit ) = r-TO)] (6.7) E( s) System R( s )
transforms to transform the differential equation to the frequency domain, function
where the independent variable is complex frequency s. It will be shown In effecting the transition between the
FIG. 6.2. Linear system.
that differentiation and integration in the time domain are transformed " time and frequency domains, a table of
into algebraic operations. Thus, the solution is obtained by simple transform pairs (z(0, ATO ) can be very
algebraic operations in the frequency domain. helpful. A diagram outlining the use of transform methods is given
There is a striking analogy between the use of transform methods to in Fig. 6.1. Figure 6.2 shows the relationship between excitation and
solve differential equations and the use of logarithms for arithmetic response in both the time and frequency domains.
operations. Suppose we are given two real numbers a and b. Let us find
the product (6.1) 6.2 THE LAPLACE TRANSFORM
C= a X b ;
The Laplace transform of a function of time fit ) is defined as
by means of logarithms. Since the logarithm of a product is the sum of the
logarithms of the individual terms, we have
log C = log a X b = log a + log b (6.2)
£[/(01 = Hs ) = f “ f ( t )e-stdt (6.8)

so that C = log-1 (log a + log b ) (6.3) where s is the complex frequency variable
s = a + /ft) (6.9)
If a and b were two six-digit numbers, the use of logarithms would probably
facilitate the calculations, because logarithms transform multiplication This definition of the Laplace transform is different from the definition
into addition. given in most standard texts,1 in that the lower limit of integration is
An analogous process is the use of transform methods to solve integro-
differential equations. Consider the linear differential equation
1
M. E. Van Valkenburg, Network Analysis, 2nd Ed. Prentice Hall, Englewood Cliffs
, -
New Jersey, 1964.
yim = /(0 (6.4)
134
 136 Network analysis and synthesis The Laplace transform 137
—
t = 0 instead of t = 0 + . We thus take into account the possibility
that f ( t ) may be an impulse or one of its higher derivatives. It is clear that
complex integration known as a contour integration.4 Since it is beyond
the intended scope of this book to cover contour integration, we will use a
C[<5( t)] = 0 f r the + definition , whereas for the 0 definition, £[<5(/)] = 1.
° °
In the case when no impulses or higher derivatives of impulses are involved ,
— partial fraction expansion procedure to obtain the inverse transform. To
find inverse transforms by recognition, we must remember certain basic
—
it was shown in Chapter 4 that /(0 ) = /(0 + ). Therefore, all of the
“strong results” resulting from a rigorous treatment of Laplace trans -
transform pairs and also use a table of Laplace transforms. Two of the
most basic transform pairs are discussed here. Consider first the transform
forms2 obtained by using t = 0 + as a lower limit also apply for the 0
definition.
— of a unit step function u( t ).

In order for a function to possess a Laplace transform , it must obey the Example 6.1 . /(/ ) = «(/ ).
condition

J£ 1 /(01 * — at
dt < oo (6.10)
F( s ) = Jfo- u( t )e~al dt - — 'T—- o - f - V
s |o \ sj s
(6.13)

Next, let us find the transform of an exponential function of time.

for a real, positive a. Note that, for a function to have a Fourier trans- Example 6.2 . fit ) = eatu{ t ).
form, it must obey the condition
e-U -a )t too i

:/ 1/(01 dt < CO ( 6.11)

F( s ) eate~st dt = - s ~ a o-
With these two transform pairs, and with the use of the properties of Laplace
| s
— a
(6.14)

As a result, a ramp function or a step function will not possess a Fourier transforms, which we discuss in Section 6.3, we can build up an extensive table
transform,3 but will have a Laplace transform because of the added - of transform pairs.
.
convergence factor e~° l However, the function e*' will not even have a
Laplace transform . In transient problems, the Laplace transform is 6.3 PROPERTIES OF LAPLACE TRANSFORMS
preferred to the Fourier transform, not only because a larger class of
waveforms have Laplace transforms but also because the Laplace trans- In this section we will discuss a number of important properties of
form takes directly into account initial conditions at t = 0 because of
the lower limit of integration in the Laplace transform . In contrast, the
— Laplace transforms. Using these properties we will build up a table of
transforms. To facilitate this task, each property is illustrated by con-
Fourier transform has limits of integration ( oo, oo), and , in order to take
into account initial conditions due to a switch closing at / = 0, the forcing —— I
sidering the transforms of important signal waveforms. First let us
discuss the linearity property.
function must take a form as /(/) u( t ) + /(0 ) <5(0, where /(0 ) repre-
sents the initial condition . _ — '

Linearity
The transform of a finite sum of time functions is the sum of the trans -
The inverse transform C 1[F(J)] is forms of the individual functions, that is

/(0 = T ~ . J ai+ joo

ITT ] Jin- joo

F( s )est ds (6.12) c[l m] = ? c[/(0 i ] (6.15)

where o1 is a real positive quantity that is greater than the o convergence This property follows readily from the definition of the Laplace transform.
factor in Eq. 6.10. Note that the inverse transform, as defined , involves a Example 6.3 . fit ) = sin cut. Expanding sin eot by Euler’s identity, we have
D. V. Widder, The Laplace Transform, Princeton University Press, Princeton, 1941. 1
/(/) = -
2

y (**“' - e )
~iat (6.16)
3 These functions do not possess Fourier transforms in the strict sense, but many
possess a generalized Fourier transform containing impulses in frequency; see M. J .
Lighthill, Fourier Analysis and Generalized Functions, Cambridge University Press, New * For a lucid treatment, see S. Goldman, Transformation Calculus and Electrical
York, 1959 . -
Transients, Prentice Hall, Englewood Cliffs, New Jersey, 1949, Chapter 7 .

-
138 Network analysis and synthesis The Laplace transform 139
The Laplace transform oi fit ) is the sum of the transforms of the individual
s ( 2 °> 2\
so that C[cos cot] _ * (6.25)
cisoidal e ±m terms. Thus ( O s
\ + CO / J2 + CO 2
1 \
C[sin cot ]
1
2 j s - jeo {
s + jco J 52 + co2
(6.17) In this example, note that sin co(0 ) = 0.
.
—
Example 6.4b fit ) = «(/ ). Let us find C[/'(f )], which is the transform of the
Real differentiation
unit impulse. We know that
Given that £[/(/)] = F(s), then 1
C[«( /)] = - (6.26)
5
sF(s) - /(0-) (6.18)

—
where /(0 ) is the value of /(/ ) at t = 0
Proof . By definition,
— .
Then

—
since u( Q ) = 0.
£[<5(01 l (6.27)

C[/ '( 0] = f -V37 0

Jo
'( dt (6.19) Real integration
If C[/(0] = F( s ), then the Laplace transform of the integral of f ( t ) is
F( s ) divided by s, that is,
Integrating Eq. 6.19 by parts, we have
(6.28)
£[/'(/)] = <T* /(0 ‘ 0-
+ s
£7 > ^ (t - t (6.20)
Proof. By definition,
Since e~>l -* 0 as / ->- oo, and because the integral on the right hand side is -
£[/(/)] = F( s ), we have (6.29)
£[/'(/ )] = 5 F(5) - /(0-) (6.21)
Integrating by parts, we obtain
Similarly, we can show for the nth derivative
_ sn- / — - -7 — -
1
(0 ) s '(0 ) • -/
_ ,-
<n 1 (0 ) clLJo-/(r) drJ] = - —TJO- /(r) dJf- + - J(o—Me
"
~2 tf ( t ) dt (6.30)
= S lo
(6.22) Since e~s! 0 as t -> oo, and since
where fn-» is the (n - l )st derivative of /( / ) at t = 0-. We thus see
f- Kr ) drI
.£,„ -
(6.31)
that differentiating by t in the time domain is equivalent to multiplying
-=
0
Jo \t=0
by s in the complex frequency domain. In addition, the initial conditions
_
_arej taken into account by the terms / (,)(0 ). It is this property that — we then have
a
transforms differential equations in the time domain to algebraic equations
in the frequency domain.
[ *] (6.32)

Example 6.5 . Let us find the transform of the unit ramp function, p( t ) = t u( t ).
Example 6.4a. /(/ ) = sin cot. Let us find

C[cos cot ]
1d
co dt
sin cot ] (6.23) We know that
J_VT ) dr = p{ t ) (6.33)

By the real differentiation property, we have

Since £[«(/)] = ] (6.34)

(6.24) then
,„
£[r />(/)] = —— ^
£[«(/)]
j =
1
(6.35)
 140 Network analysis and synthesis The Laplace transform 141
Differentiation by s Real translation (shifting theorem )
Differentiation by s in the complex frequency domain corresponds to Here we consider the very important concept of the transform of a
multiplication by f in the domain, that is, shifted or delayed function of time. If £[/(f )] = F( s ) , then the transform
of the function delayed by time a is
mm = - im
ds
(6.36) £[/(/
— a ) u( t — a)] = e ’ F( s) ~a (6.46)

Proof. From the definition of the Laplace transform, we see that Proof. By definition,
d Fjs ) °
( - - f7
dt -
= Jo- 0 CIS
J
*
tf ( t )e- * dt = £[f /(f )] (6.37)
= P
Jo-
’ - l[ f (t - a) u( t - a )] = j e~stf ( t - a ) dt (6.47)
ds
Example 6.6. Given /( f ) = e at , whose transform is Introducing a new dummy variable T —— t a, we have

FW = — 1
«
(6.38) C[/(r) U(r)] =
£ £>
' s( r-l- a )
f ( r ) dr =e r )e~ T dr
,
= e~as F( s) (6.48)

let us find C [ te~ xt ]. By the preceding theorem , we find that In Eq. 6.48 /(r) w(r) is the shifted or delayed time function ; therefore
the theorem is proved .
s (rh) - ( s + af
1
C[nr**] = ~ (6.39) It is important to recognize that the term e~a> is a time-delay operator.
If we are given the function e~aa G( s ), and are required to find
Similarly, we can show that
n! Z-' [e-a> G(s)] = gft ),
(6.40)
is + a)"+! we can discard e~as for the moment, find the inverse transform
where n is a positive integer. t- Gis )] = g(f ),
Complex translation and then take into account the time delay by setting
^
By the complex translation property, if F( s ) = £[/(/)], then
F( s a) = £[«•* /(0] - (6.41) g{t - a) u( t - a) = gft ) (6.49)

where a is a complex number.

.
Example 6.8a Given the square pulse /(/ ) in Fig. 6.3, let us first find its trans -
form F( s ). Then let us determine the inverse transform of the square of F( s),
Proof . By definition, i.e., let us find
hi0 = c-1 )]
m/ m -JofV
-
/(oe-" dt = Jof -
" (
*—a ) t fit ) dt = F( s - a ) ( 6.42)
Solution. The square pulse is given in terms of step functions as
(6.50)
^
Example 6.7 . Given /(/ ) = sin cof, find £ [e at sin co / ]. Since
~
fit ) = uit ) — uit - a) (6.51)
CO
C[sin cot ] = (6.43) Its Laplace transform is then fit )
s* + oP 1
F( s) = - (1 - e-a° ) (6.52)
by the preceding theorem, we find that
Z [e~at sin <ot] = (s
CD
(6.44) Squaring F( s ), we obtain
+ a)2 + to2 1
- 2e as + e-20*)
^
~
F\s ) = (1 (6.53)
Similarly, we can show that
s +a (6.45) To find the inverse transform of F2(s), we 0 a
£[«-< cos tot ]
“ =
is + a)2 + to2 need only to determine the inverse transform FIG . 6.3. Square pulse.

The Laplace transform 143

142 Network analysis and synthesis
Since /(f ) is periodic, Eq. 6.59 reduces to
of the term with zero delay, which is
-,
c-1
&-] -uit - f «( f )

- -
(6.54) £[/(*)] = \Tfit
-
Jo
)e a dt + ( Tfit )e-’‘ dt +
J o-
Then C-1[F2( j)] = f uit ) - 2it
a) + (/ 2a) u( t 2a) (6.55)
a)
+ -‘ j
*
e aT fit )e- * dt’ + •••
so that the resulting waveform is shown in Fig. 6.4. From this example,
we see
nd to the square (6.60)
that the square of a transformed function does not correspo
of its inverse transform. = (1 +e ~*T
+ e~°*T +
CT
••
\jit
) )e~ sldt

1
hit )
, )e- stdt
fit ) 1 - e~ T )Jii
1' Kl *2
Ke
Example 6.8c . Given the periodic pulse train in Fig. 6.6 let us use Eq. 6.60 to
K0 , KP Ki determine its Laplace transform.

to a 2a 0
t
Ti 2Ti 3Ti 4Ti 5Ti 6Ti 7Ti
t Fis) = 1 -e
1
~ e~ sl dt

FIG . 6.4. Triangular pulse . FIG . 6.5 . Impulse train. -1 e~ 3t (6.61)

s(l - e ~sT ) |o —
Example 6.8b In Fig. 6.5, the output of an ideal sampler is shown. It consists
. 11 - e~as
of a train of impulses s1 - e~* T
,
fit ) = K0 <5(f ) + K <5(f - T:) + • • • + Kn 6( t nTJ
(6.56) - fit )
The Laplace transform of this impulse train is
£[/(/)] = K0 + Kie- T i + K2e- >Ti
z , + + Kne-n‘ Ti (6.57)
7’i
In dealing with sampled signals, the substitution z = e* is often
used Then .
we can represent the transform of the impulse train as
K „
C[ ( ] / O = ^, Y
+ + + - + -? (6.58)
5 - 0 a T T + a 2T 2 T + o
FIG. 6.6. Periodic pulse train.
.
The transform in Eq 6.58 is called the z transform of / f )
( This transform is
- .
widely used in connection with sampled data control systems . - Example 6.8d. In this example we calculate the Laplace transform of /(f ) in
Fig. 6.6 using summation of an infinite series. The periodic pulse train can be
represented as
PERIOD IC WAVEF ORMS

The Laplace transform of a periodic waveform can be obtained in two

fit ) - uit - a) + uit -+Tuit) --uit27—) iTuit+ a-)](2T + a )] + - -
= uit )
ways: (a) through summation of an infinite series as illustrated in Example Its Laplace transform is
- (6'62)

6.8d , and (ft) through the formula derived below.

rT r »T _, Fis) = -s
1
(1 - e~a )
+ e~ T
, - er< r + > «*
+ • • •)
~
+ JT /(f )e < dt +
£[/(0] =
Jo- ‘
fit )e ' dt
Hn+lir
••
(6.59) I ,
(6.63)

+ JnT fit )e- stdt + ••• = - [1 - e ~ai

+ e~ T { \ - e-aa ) + e~2 sT ( l - tr« ») + •• ]
•
 The Laplace transform 145
144 Network analysis and synthesis
Replacing s by 2, we have
which simplifies further to give
F( s ) = j (1 - e~a 3 )( l + e~ T‘ + e~2 T
, + )
• (6.64) T -
2* + 25 29
I
-
Perhaps a more subtle example is the evaluation of
5 1 (6.71)

form
We then see that F( s) can be given in the closed
1 1 e* ~ a
- (6.65) / =
First, we note that ? 2e-2|1l is an even function ; therefore
J Ve- ‘'
+
Sl dt (6.72)

in closed form. The reader is

Other periodic pulse trains also can be given .
= J \v
chapter
referred to the problems at the end of this / 2: 2<
dt ( 6.73)
Laplace transforms, Most of
At the end of the chapter is a table of of the properties just From the table of Laplace transforms we see that
the entries are obtained through simple applicationsin mind, because many
discussed. It is important to keep those properties can be obtained by using C[tV2<] = ——, (6.74)
transform pairs that are not given in the tableinverse transform of (* + 2)3
the
these properties. For example, let us find and the transform of
as (6.66) -
F(s) = rV2rdr (6.75)
(s + ff)2 + ft)2
in the time domain, we
Since the s in the numerator implies differentiation is £[/( <)] = (6.76)
2
can write cu s(s + 2)3
(6.67)
F( s ) = s 2 Later we will see that the partial-fraction expansion of £[/(?)] in Eq. 6.76
(s + of + ft)

consists of three terms for the multiple root ( s + 2)3, as given by

Using the differentiation property, we obtain 0.25 0.25 0.5 1
—
C~1[F(s)] —
= dt (e ~ at
sin at )

co cos at )e
-a t
(6.68)
£[/( <)] =
s
Taking the inverse transform of Eq . 6.77, we obtain
s+2 (s + 2)2 (s + 2)3
(6.77)

( u sin at = - + - e - 2te ‘ - 2t e
~2 t 2 ~il )

—
~2
f( t ) = 0.25(1 u( t ) (6.78)
Note that erai sin at at t = 0 is zero. Now observe that the definite integral in Eq. 6.73 is equal to 2/(f ) at
t = 1, that is,
6.4 USES OF LAPLACE TRANSFORMS t = 2/(1) = 2.5/r2 + 0.5 = 0.162 - (6.79)

Evaluation of definite integrals Solution of integrodifferential equations

the evaluation of definite
The Laplace transform is often useful in of
In Section 6.3 we said that the real differentiation and real .integration
integrals. An obvious example occurs in the evaluation properties of the Laplace transform change differential equations in time
to algebraic equations in frequency. Let us consider some examples

t
If we replace e~2 by e
' -J > ” sin 51 dt
the integral / then becomes the Laplace transform
(6.69)
using Laplace transforms in solving differential equations.
Example 6.9 . Let us solve the differential equation

*'(/) + 3*'(0 + 2*(0 = 4e* (6.80)

of sin 5/, which is
qsin 5<1 = |
J 25
(6.70)

-
—
given the initial conditions *(0 ) = 1, x' (0 ) = 1. — —

146 Network analysis and synthesis The Laplace transform 147

.
Solution We first proceed by taking the Laplace transform of the differential where A is the determinant of the set of equations in Eq. 6.87, and Aiy is the
yth cofactor of A . More explicitly, X ( s ) is
equation, which then becomes
[s22T(s) - s x 0 —) -*'(0-)] + 3
( (s X ( s) - *(0-)] + 2 X { s) = (6.81)
X ( s) = — 5s2
s( s2
- 30s + 15
+ 2s + 5)
(6.89)

Substituting the initial conditions into Eq. 6.81 and simplifying, we have Expanding X ( s) in partial fractions, we have
4
+s+2 (6.82) Kx K S + K
(s2 + 3s + 2) X ( s) = X ( s) = 7 + 7 - +2 2S~ +35 (6.90)
We then obtain X ( s ) explicitly as
s2 +s +2 (6.83)
Multiplying both sides of Eq. 6.90 by s and letting s = 0, we find
X ( s) = (s - l )(s + l )(s + 2)
= j 2f (s) |s=0 = 3.
1 -
To find the inverse transform »(/) = f [2f (s)], we expand A (s) into partial
"
K2 and Ks are then obtained from the equation
fractions
(6.84) 3 -85 - 36
+ 2, + 5
(6.91)

Solving for K
_lt Klt and K2 algebraically, we obtain A further simplification occurs by completing the square of the denominator
K-i = f K t - l K2 = -f
The final solution is the inverse transform of X ( s ) or
- of X ( s ) , that is,
52 + 25 + 5 = (5 + l )2 + 4 (6.92)
As a result of Eq. 6.92, we can rewrite
‘- ‘
~ -2( (6.85) AT0 as
*{ t ) = §e e + fe
3 8(5 + 1) + 14(2)
In order to compare the Laplace transform method to the classical method of X ( s) = -5 (5 + l )2 + (2)2
(6.93)
solving differential equations, the reader is referred to the example in Chapter 4, so that the inverse transform is
Eq. 4.64, where the differential equation'in Eq. 6.80 is solved classically. _
Example 6.10 . Given the set of simultaneous differential equations
a:(0 = (3 - 8e~l cos 2t
In similar fashion, we obtain Y( s ) as
— 14e !sin 20 «(0 - (6.94)

+ 4 x{t ) + l/'(0 + M 0 = 5«(0

2x’( t )
(6.86)
*'W + <0 + y’( 0 + 32/(0 = 55(0 T(5) = - - + 52 + 25+
1 115 17 1 ll (5 + l ) + 3(2)
+ ( S l )2 (2)2 (6.95)

.
-
Solution Transforming the set of equations, we obtain
—
with the initial conditions a:(0 ) = y( 0 ) = 0, let us find x( t ) and y( t ).
/ The inverse transform is then seen to be
+5 5 + +

2(s + 2) 2T(0 + ( s + 7) Y( s ) = - (6.87)

2/(0 =( — 1 + 1 le~l cos 2/ + 3e~ sin 20 «(0
This example is also solved by classical methods in Chapter 4, Eq. 4.163.
‘ (6.96)

(5 + 1) AT(5) + (5 + 3) Y( s ) = 5 We note one sharp point of contrast. While we had to find the initial
Solving for X ( s ) and 7(5) simultaneously, we have conditions at / = 0 + in order to solve the differential equations directly
in the time domain, the Laplace transform method works directly with
(5/QAn + 5 A21
m= the initial conditions at t = 0 In addition, we obtain both the comple-
A
(6.88) mentary function and the particular integral in a single operation when we
(5/5)A12 + 5 A22 use Laplace transforms. These are the reasons why the Laplace transform
7(5) = A method is so effective in the solution of differential equations.
 148 Network analysis and synthesis The Laplace transform 149
If we see that F( s ) can also be written as
6.5 PARTIAL- FRACTION EXPANSIONS
( s + 1 ) + ( s + 2)
F( s) = (6.103)
As we have seen, the ease with which we use transform methods depends ( s + 1 )(* + 2)
upon how quickly we are able to obtain the partial-fraction expansion of a then the partial-fraction expansion is trivially
given transform function. In this section we will elaborate on some simple
and effective methods for partial -fraction expansions, and we discuss
procedures for ( a ) simple roots, ( b) complex conjugate roots, and (c) *® - rn + TT 1 1
2
(6.104)
multiple roots. Example 6.12. Find the partial-fraction expansion of
It should be recalled that if the degree of the numerator is greater or
equal to the degree of the denominator, we can divide the numerator by s+5
F( s ) = (s (6.105)
the denominator such that the remainder can be expanded more easily + 2)2
into partial fractions. Consider the following example : We see that F( s ) can be rewritten as

Ms) ~ _ s3 + 3s2 + 3s + 2 ( )
F( s ) = - - - 2 +2 3
p(W
s\ ~
D( s ) s 2 + 2s + 2
(6.97)
Real roots
^
(s + 2)
= (— i
s + 2)
3
(s + 2 f
( 6.106)

Since the degree of N( s ) is greater than the degree of D( s ) , we divide D( s ) Now let us discuss some formal methods for partial-fraction expansions.
into N ( s ) to give First we examine a method for simple real roots. Consider the function
s
F( s ) = s + 1 - 2 (6.98) N ( s)
s + 2s + 2 F( s ) = (6.107)
Here we see the remainder term can be easily expanded into partial :
(s — s )( s -
0 Sj)(s - s2)
where s0 , sv and s2 are distinct, real roots, and the degree of N( s ) 3.
fractions. However, there is no real need at this point because the i Expanding F( s ) we have
<
denominator s2 + 2s + 2 can be written as
s2 + 2s + 2 = (s + l )2 + 1 (6.99) (6.108)
We can then write F( s) as
s —s 0 s - Si s - s2
Let us first obtain the constant K0 . We proceed by multiplying both sides
F( s ) = s +
(s 1) - 1
1- + 2 ( 6.100) —
of the equation by (s s0) to give
(*. + 1) + 1 (s - s0 ) K x (s - s0 ) K 2
-
_
(s - So)F(s) = K 0 + + (6.109)
so that the inverse transform can be obtained directly from the transform
tables, namely,
/ s - Si —
s s2
If we let s = s0 in Eq. 6.109, we obtain
C-1[F(s)] = d' ( t ) + <5(0 + e-‘(sin t - cos t ) (6.101) F0 = (s - s0) F(s)| , (6.110)

_
(

From this example, we see that intuition and a knowledge of the trans- Similarly, we see that the other constants can be evaluated through the
form table can often save considerable work . Consider some further general relation
examples in which intuition plays a dominant role. —
M = (s s,) F( s ) | s,- (6.111)
Example 6.11. Find the partial-fraction expansion of Example 6.13. Let us find the partial-fraction expansion for

F( s ) =
2s +3 (6.102) F( s ) =
s2 + 2s 2 — = *s0 *2 (6.112)
(s + (s + 2)
l ) s(s + 2)(s - 3) s +2 s —3

150 Network analysis and synthesis

The Laplace transform
151
Using Eq. 6.111 , we find A more convenient way to express the inverse transform Ht ) is to
K0 = sF(s) |3=0
s2 + 2s - 2
(s + 2)(s - 3) | o
_ 1
3
duce the variables M and <D defined by the equations
M sin 4> = 2 A
M cos <I> = 2B —
intro-

(6.119)
s2 + 2s - 2 1 (6.113)
where A and B are the real and imaginary parts of Fj in Eq . 6.116.
Ki = s(s - 3) |,=_ 2 5 terms of M and <b, the inverse transform is
In

s2
*2 = - s(s + 2) |s
+ 2s - 2
_ 3
13
15
ft( t ) = Meat sin ( fit
To obtain M and <I> from Klt we note that
+ d>) (6.120)
Complex roots
Equation 6.111 is also applicable to a function with complex roots in Me** = M cos <D + jM sin O = - 2 B + j 2 A = 2 jKx ( 6.121)
its denominator. Suppose F( s ) is given by When related to the original function F( s ) , we see from Eq. 6.115 that
N( s )
F( s ) =
T> i( s )( s - a - )/3)(s - a + jfi) Me* = (6.122)
+ jP )
K, (s)
+ Ni (s)
K2 Example 6.14. Let us find the inverse transform of
(6.114)
s — a - j@ '
s - a + jp Da
F( s ) = 2
s2
+3
(6.123)
where NjDi is the remainder term. Using Eq . 6.111 , we have :
(J + 2s + 5 )( s + 2)
For the simple root J = -2, the constant K is
N(* - + JP )
Kx = 2 jpD + jp ) X = (s + 2) F(s) I ,= _2 = f
^
N (* - jp ) .
(6.115)
For the complex conjugate roots
(6.124)

* = - jpD - jp)
2

^ zeros of 2
D s ).
. s2 + 2s + 5 = ( s + 1 + j 2 )( s + 1 ~ j2 ) (6.125)
where we assume that s = ± jP are
a not
It can be shown that the constants Kx and Kz associated with conjugate ^ We see that a = - 1 , /3 = 2, thus

roots are themselves conjugate. Therefore, if we denote

K1 = A + j B
as
(6.116)
Me** =
i2

2( s
+3
+ 2) =2
i- g
V5
—l( tan-12 + ir /2) (6.126)
/
The inverse transform is then
then K2 = A - J B = K * (6.117)
e-![F(r)] = 7- e~2t + -
If we denote the inverse transform of the complex conjugate terms as
/i(0> we see that
m {—
= £- Ls -
—
-+
a -- JP
Kx*
S - K + jP
]
7
= -5 e~2t
^ ^^
2
7
Vs
= e~* sin 2/

= e~* cos { It - tan-12)

- - tan-12 )
(6.127)

, Multiple roots
= eat { K 1ej t + K * e~ m ) Next let us consider the case in which the partial fraction involves
repeated or multiple roots. We will examine two methods. The first
= 2eat( A cos pt B sin pt ) — (6.118)
requires differentiation ; the second does not.
 152 Network analysis and synthesis The Laplace transform 153
Method A Example 6.15. Consider the function
Suppose we are given the function
m = -s4( s -+ 2l )
F(s) = (s m (6.128)
3 (6.137)

- s ) D,( s)
0
n which we represent in expanded form as

with multiple roots of degree n at s = J . The partial fraction expansion F( s ) = (J Ko Kx K2

+ l )3 + (s + l )2 + s + 1 +
A
(6.138)
of F(J) is j

F( s ) = Ko
(s - s0) n + (s
-s
Kx
0 )n 1 — (s - s0) -2
K2
+ B
• • •
+s
-
, NM i
s0 Dj[ s )
i
*
The constant A for the simple root at
A = s F( s)|J
=0 = 2
j
= 0 is
- (6.139)
To obtain the constants for the multiple roots
(6.129) , we first find Ft( s) .
where NM / DM represents the remaining terms of the expansion. The
problem is to obtain K0 , Kx
cited earlier for simple roots, that is,
_
Kn x. For the K0 term, we use the method
i> Using the general formula for the multiple root
—
F1( s) = ( s + lfF( s ) =
•S
'

expansion, we obtain
(6.140)

— (s - X d°
( ) L-
lS ^
K0 s0 )n F(J) |S= 3 (6.130) K -
Ko ~
Q
i Ud? 3 (6.141)
However, if we were to use the same formula to obtain the factors Ku l d
- ~
K2 Kn-i> we would invariably arrive at the indeterminate 0/0 con- K
dition. Instead, let us multiply both sides of Eq. 6.129 by ( s s0 )n and
define
— =2 (6.142)

Thus
Fi(s) = (s J0)” F(J) (6.131) — '
3= —1 =
2 (6.143)
;
so that m = (s + l) + (s
2 2 2
Fx( s ) = K0 + Kx( s - s0) + ‘
+ Kn^ ( s - Jo)"-1 + R( s )( s - J0)n
(6.132)
Method B
3
+ l )2 + * + 1 s (6.144)

The second method for arriving at the

where JR(J) indicates the remaining terms. If we differentiate Eq. 6.132 by
partial fraction expansion for
multiple roots requires no differentiation. It involves -
sy we obtain series expansion.5 Let us consider F( s ) and
a modified power
F J), defined in Eqs. 6.128 —
and 6.131 in Method A. We define a new
Then we can write Fx( s) as
variablep such thatp = s s0 ^ — .
— Fj(s) = K , + 2 Kz( s - *) + • + Kn- ( n ~ l)(s ~
ds
•
i ®o) j .-.
/
(6.133) F>i( p + s0) (6.145)
Dividing N( p + J0) by Dx( p + J0), with both
It is evident that Kx = ds Fx( s) 4 13= SO
(6.134) ascending powers ofp , we obtain
polynomials written in

On the same basis 1 d2

(6.135)
Fx( p + s0) = F0 + KlP + K 2 p2 + • • • + K p"-1 + Knpn
ls= S0 ^ F>iiP + s0)
and in general (6.146)

K< =
1
nZ< m ls=so . >
) = 0, 1, 2, . . . , n — 1 (6.136)
51 am indebted to the late Professor Leonard
Institute of Brooklyn for showing me this method O. Goldstone of the Polytechnic
.

c-
154 Network analysis and synthesis
The Laplace transform 155
The original function F( s ) is related to Fx( p

F{ p + s0)
_ Fxjp + Sp) _ Kp -Kl1 , . . . -^ n-
+ J0) by the equation
l
__
| Kn
Substituting j + 1 = p , we have

F( s )
2
= ( + l )3
2 2 2
pn pn p"- P DiiP + So) j (s + l )3 + ( s + 1) s+2 (6.154)

(6.147) Example 6.17 . As a second example, consider the function

- s = p in Eq. 6.147, we obtain
-
Substituting s 0 s+2
_ + ... + _, F( s ) = t?( s + l )2 (6.155)
F( s ) =
(j -
* 0

J )
n +
0 (s '
- s )"- 0
1
^
s - J0
+ £
Dx( s )
(6.148) Since we have two sets of multiple roots here (at s 0, and
choice of expanding Fx(,s) about s = 0 or s = 1. Let us
= s = 1) we have a
—
arbitrarily choose to
—
We have thus found the partial-fraction expansion for the multiple-root expand about s = 0 ; since/? = J here, we do not have to make
terms. The remaining terms AT„/[Z>x(s)] still must be expanded into partial
any substitutions.
Fx( s) is then s+2
fractions. Consider Example 6.16. Fyis ) =
2 + s
(s + l )2 " 1 + 2J + s2 (6.156)
Example 6.16 Expanding Fx( s ), we obtain
2
J2(3J
F( s) = ( s + l )3(s + 2) Fx(.s ) = 2 - 3J + (J +l )2
(6.149) 4)
(6.157)
+
Using the method just given, Fx( s ) is 2 3 3J + 4
i F( s ) is then F( s ) = 2 “ +

^ s (J + l )2 (6.158)
F1(S) = (S + 1)3 F(J) = 7 2
(6.150) J
We must now repeat this process for the term
Setting /? = + I, we then have
J
3J + 4
2
F P ~ 1)
* = (6.151 )

—
the numerator 2 by the denominator 1 + p, with both numerator and denomi
^
The expansion of Fx( p 1) into a series as given in Eq. 6.146 requires dividing
M

- 3J + 4 _
(s + l )2
Fortunately we see that the term can be written as
3(J + 1 ) + 1 3 1
nator arranged in ascending power of p. The division here is (s + l )2 (s + l )2 - J + 1 + (J + l )2 (6.159)
The final answer is then
2 - 2p + 2/
2 3 3 1
1 + p )2 + J 1 + (J (6.160)
J2 J + + l )2
2 + 2p
“ 2P

- -
2p 2p*
6.6 POLES AND ZEROS
2 / In this section we will discuss the many implications of a
^ + 2p
2p2 3
description of a given rational function with real coefficients F( j). - We
pole zero
- 2/ define the poles of F(J) to be the roots of the denominator of
F(J). The
Since the multiplicity of the root is TV = 3, we terminate the division after zeros of F(J) are defined as the roots of the numerator. In the complex
J
we have three terms in the quotient . We then have plane, a pole is denoted by a small cross, and a zero by a small circle
for the function
. Thus,
Flip — 1) =2 — 2p + 2/>2 — 2/
(6.152) J( J - 1 + jl )( s — 1 - jl )
The original function F( p
— 1) is
Flip 1) - 2 2
^2
TjTj

2
the poles are at
F(J) =
J =
(J + l)2(s + / 2)(s -;2)
1 (double)
(6.161)

Fip - 1) =
F* p3
p 2 p p +1
(6.153) J = -/ 2
J = +/ 2
 i ne Lapidct uan »iui in

Network analysis and synthesis

156
ju i
Iju F( s ) = ± F( s ) = + ff
v * «
s plane
^
J2 1 (a) ( b)
o I!
<S , ~
Double pole ju \ ju F( s ) J + <ro
V- F = (s
— — — —
i i i S2 + OI02 + <r0 ) 2 + u02
o 3 2 1 [o 1 2 3 a U»o I
o

-12 1 -cro a
* - juo - juo
— ju F( s ).
(0 (d )
FIG. 6.7. Pole-zero diagram of FIG. 6.8. Poles and zeros of various functions.

5 =0 From these four pole-zero diagrams, we note that the poles corre-
and the zeros are at —
sponding to decaying exponential waves are on the a axis and have zero
J = 1 + j\ imaginary parts. The poles and zeros corresponding to undamped
sinusoids are on the jco axis, and have zero real parts. Consequently, the
5 = 1 -;1 poles and zeros for damped sinusoids must have real and imaginary
5 = oo \ parts that are both nonzero.
~
Nowletus consider two exponential waves/ t ) = e ' and f2 ( t ) = e ai\
~a i

The poles and zeros of F( s ) are

shown in Fig. 6.7. rd where at > oq > 0, so that /2(t) decays faster than fx
' ( t ), as shown in Fig. ^
zero diagrams corresponding to standa x
Now let us consider some pole- the comple 6.9a. The transforms of the two functions are
step function is given in
signals. For example, the unit 1
frequency domain as Fits ) = s
1 (6.162) + oq (6.166)
=- mm s F*( s ) =
s + ai
shown in Fig. 6.8a. The exponential :
and has a pole at the origin, astransform as depicted by the pole-zero diagram in Fig. 6.9b. Note that the further
signal e "0*, where a0 > 0, has a
"

the pole is from the origin on the — a axis, the more rapid the exponential
1 (6.163) decay. Now consider two sine waves, sin cof and sin co 2 t , where co2 >
C[e-v°« ] = cox > 0. Their corresponding poles are shown in Fig . 6.10. We note here
S + ff 0

The cosine / m I ju
which has a single pole at s = 0 — o , as indicated in Fig. 6.86. 4 •

1' s plane
transfo rm is
function cos co0t, whose
s (6.164)
£[cos co0 t ] = -s j— 0> 2 Fi ( s )
+ Q
.
0 a
pair of conjugate poles at s = ± 0 ol
j(o , t
has a zero at the origin and a 6.8J shows the pole-zero diagram
as depicted in Fig. 6.8c. Figurewave, whose transform is (a) ( b)

corresponding to a damped cosine

s H- QQ (6.165) . .
FIG 6.9 Effect of pole location upon exponential decay .
cos m0 t ] = (s + a0Y + coy i

::
II
i
P
i The Laplace transform 159
158 Network analysis and synthesis
ju I ju
s plane
XjU 2 M-
X \ ju
jo)1 S2
X
Si C 1 <Fi <r
a
<r
0
- jUi X

X - jui
S
>
(b)
~ jU 2 .
FIG 6.13. Effect of right-half plane poles upon time response .

. .
FIG 6.10 Pole locations corre- FIG 6.11 . Let us examine more closely the effect of the positions of the poles in
sponding to sin oiit and sin eo, / . the complex frequency plane upon transient response. We denote a
frequency of . pole pt as a complex number pt = oq + ja>( . For a given function F( s )
that the distance from the origin on the jco axis represents with only first-order poles, consider the partial -fraction expansion
.
oscillation; the greater the distance, the higher the frequency
Using these rules of thumb , let us compa re the time respon ses /x( f ) and
( jj, Jj*} and { sz , J2* shown in Fig.
} F( s ) = +s~ (6.167)
/2( r) corresponding to the poleofpairs d sinusoids. s ~~ Po s — Pi Pn
6.11 . We see that both pairs poles corresponds to dampe
ion than /2(f ) The inverse transform of F( s ) is
The damped sinusoid /*(/) has a smaller frequency of oscillat
because the imaginary part of is less than the imaginary part The
of sz . Also,
than /i(f ) because Re sz > Re 2 J . time f { t ) = K 0e + •
+ Kne^
/2(/) decays more rapidly (6.168)
responses f t ) and f 2( t ) are shown in Fig. 6.12.
^ In fit ), we see that if the real part of a pole is positive, that is, at > 0, then
fi ( t ) = lme’2‘ the corresponding term in the partial-fraction expansion
,, ,
Kiea eia t
1 is an exponentially increasing sinusoid, as shown in Fig. 6.13a. We thus
see that poles in the right half of the s plane (Fig. 6.136) give rise to :
exponentially increasing transient responses. A system function that has
/ poles in the right-half plane is, therefore, unstable. Another unstable
situation arises if there is a pair of double poles on the jco axis, such as for ‘
h( t ) = Ime1!* the function
s
F(s) = ( 6.169)
t
(s 2
+ <o02)2
whose pole-zero diagram is shown in Fig. 6.14a. The inverse transform of
F(s) is :

.
(b )

FIG 6.12. Time responses for poles in Fig . 6.11 .

fit ) =
t
2 O0
(
sin (o0 t (6.170) i
 The Laplace transform 161
160 Network analysis and synthesis
f( t ) Then the response function is
HQEQ( S - z„)(s - z,)
17«
*00 = (s - p0 )( s - p,)(s - p ) 2
(6.174)

It is clear that R( s ) contains the poles and zeros of both H( s ) and E( s), ,
t
-
except in the case where a pole zero cancellation occurs. As an example, _[
o let us take the system function
2(5 + 1)
y
-
3 joio
H( s ) =
(s + 2 + y4)(s + 2 /4)
(6.175)

whose pole-zero diagram is shown in Fig. 6.15a. It is readily seen from the
— ^
pole-zero diagrams of the excitation signals in Fig. 6.18a that the step
fit ) = 2zr0 sin response of the system has the pole-zero diagram indicated in Fig. 6.156.
The response to an excitation signal 3 cos 21 has the pole-zero representa-
( b)
( a)
response.
tion of Fig. 6.15c. To specify completely the given function F( s ) on a pole-
. . the ju> axis upon time
FIG 6.14 Effect of double zeros on zero plot, we indicate the constant multiplier of the numerator on the
plot itself.
which is shown in Fig. 6.146. It is
apparent that a stable system function Let us examine the significance of a pole or zero at the origin. We
\ cannot also have multiple poles on the joy axis.
L Consider the system function H( s ) = N( s )/ D( s ). If we factor the _
know that dividing a given function H( s ) by s corresponds to integrating
.
the inverse transform h{ t ) = C 1[i/(s)] Since the division by J corresponds
, we obtain
numerator and denominator polynomials to a pole at the origin, we see that a pole at the origin implies an inte- <f-
Hn( s - Z„)(s - 2,) ' ~ ' (5 ~ *n ) (6.171) gration in the time domain. Because the inverse transform of H{ s ) in
H( s ) = - Fig. 6.15a is the impulse response, placing a pole at the origin must
( s Po )( s- Pi) • • • (s
- pm )
correspond to the step response of the system. In similar manner, we
specifie d in terms of its poles and zeros,
It is clear that H( s) is completely ( ) - deduce that a zero at the origin corresponds to a differentiation in the
H . From the pole zero plot of H s
and an additional constant multiplier 0 of information concerning the time domain. Suppose we consider the pole-zero diagram in Fig. 6.15c
we can obtain a substantial amount , we can determine whether the with the zero at the origin removed ; then the resulting pole-zero plot is
seen
behavior of the system. As we have half plane for poles and the ja> axis the response of the system to an excitation E0 sin 2t . Placing the zero at
0 system is stable by checking the right -
concerning its transient the origin must then give the response to an excitation £, cos 2 /, which, of
multiple poles. We obtain information
£ behavior from the positions of its poles and zeros; and, as we will discuss
for
\ j< ji
17« 17«
also gives us significant information
in Section 8.1, the pole-zero diagram
74 74 74
- and phase response. We thus / =2
concerning its steady state (/a>) amplitude .
K= 2 K K=6
description
see the importance of a pole-zero and the *
Suppose we are given the poles and zeros of an excitation E( s )
the pole-zero diagram of the respons e l -2 - 1 a
system function H( s). It is clear that the pole-zero diagrams of H( s ) and
—*
(T
'
' 72
function R( s ) is the superposition of
E( s). Consider the system function
H(s) =
H ( s z0)
(s Po )( s Pi ) -
„ —- (6.172) -74
(a )
-74
( b)
-74
(C )
En( s Z,) (6.173) -•
. .
FIG 6.15 (a) System function. (6) Response to unit step excitation, (c) Response to
and the excitatio n £ ( s) = s Pi - excitation e{t ) = 3 cos 2/. \
l - a Kl '
if
I
The Laplace transform 163
162 Network analysis and synthesis
of a When we interpret the Eq. 6.180 from a 17«
course, is true. We therefore conclude that the system function must - -O 20
complex plane viewpoint, we see that each one —
differentiator must have a zero at the origin
have a pole at the origin.
; that of an integrato r
of the terms ( pt
— represents a vector drawn
from a zero Zj to the pole in question, p{.
Pix

—
Similarly, the terms ( pt pk ), where i k , -S
represent vectors from the other poles to the
6.7 EVALU ATION OF RESIDU ES
pole pt. In other words, the residue Kt of any
the behavio r
Poles and zeros give a powerful graphical description of pole pi is equal to the ratio of the product of
of a system. We have seen that the poles are the complex frequenc
ies of the vectors from the zeros to pi to the prod- }PI*
o
answer
the associated time responses. What role do the zeros play
? To uct of the vectors from the other poles to />,.
FIG. 6.16. Poles and zeros
this question, consider the partial fraction expansion To illustrate this idea, let us consider the pole- of F( s).
N v
zero plot of
F(S) = Z
— s-
-s i (6.176)
F( s ) =
4Q(5 - Zp)(s ZX)
(6.181) -
i (
-
(5 P0 )( S ~ PJ)(5 - p )
^
where we shall assume that F( s ) has only simple poles and
no poles at
given in Fig. 6.16. The partial-fraction expansion of F( s ) is
s = co. I
The inverse transform is
(6.182)
( / 0 = 2 KS *
(6.177) s - Po 5 - Pl S - Pl*
!
where the asterisk' denotes complex conjugate. Let us find the residues K0
complex
It is clear that the time response f ( t ) not only depends on the and Ki by means of the graphical method described. First we evaluate Kt
frequencies st but also on the constant multipliers Kt . These . We will
constant s Kt
by drawing vectors from the poles and zeros to plt as shown in Fig. 6.17a.
are called residues when they are associated with first-order poles The residue Kx is then
show that the zeros as well as the poles play an important
part in the
determination of the residues K{. CD
(6.183)
Earlier we discussed a number of different methods for
obtaining !
graphical
the residues by partial-fraction expansion. Now we consider a where symbols in boldface represent vectors We know that the residue .
method whereby the residues are obtained directly from a pole zero - of the conjugate pole /),* is simply the conjugate of Kx in Eq. 6.183. Next, ;
diagram. Suppose we are given
17« 7«
F( s ) = A(5 ~ Zp)(5 ~ Z,) • • ' (5 - „) Z
(6.178)
i

(S - Po)(5 - Pl) • • • (5 - Pm ) / A
R
c> M .
n and all the poles are simple. Let us expand F( s ) as
'
where m > N
Ko Ki + Km (6.179)
a cr \
F( s) = •• •
+ 5- D B
S - Po 5 - Pi Pm N’ L .
Our task is to determine the residues K( We know that . i

_ MPi ~ Zo )( Pi ~ 2l ) ' ' ‘ ( Pi - Zn ) | i

Ki = ( s - Pi ) F( s) L= > j ( Pt — Po ) •‘ • ( Pi — Pi -i )( jPi — Pi+ ) l ' ' ’ ( Pi ~~ Pm )
(6.180)
(a )

. . Determining residues by vector method.

FIG 6.17
( b)

I
i
 The Laplace transform 165
164 Network analysis and synthesis
from the From Fig. 6.19b we find the value of the residue K2 to be
\ jo> to evaluate Ka , we draw vectors
in Fig. _
C= 3 poles and zeros to p0, as indicated 3x 2
6 Mb. We see that * = - ^ /2 /
2
— 135° x J 2 / + 135°
' fi XpRL (6.184)
K0 =
MN Therefore, the partial-fraction expansion of F( s) is
, 3.
With the use of a ruler and a protractor
3
- a
— 2 1 - we determine the lengths and the angles
the vectors so that the residues can
of
be deter -
Hs ) = s
+ l - jl
'
s+ l + jl s +2
(6.187)

x -
n mined quickly and easily , Consider the fol - 6.8 THE INITIAL AND FINAL VALUE THEOREMS
of
lowing example. The pole-zero plot
In this section we will discuss two very useful theorems of Laplace
3s
F(s) = transforms. The first is the initial value theorem. It relates the initial
2)(s + - jl)(s + 1 + 71)
'

. -
FIG. 6.18 Pole zero diagram of
(s + 1 value of f ( t ) at / = 0 + to the limiting value of s F( s ) as s approaches
(6.185) infinity, that is,
F( s).
lim f { t ) = lim s F(s) (6.188)
is shown in Fig. 6.18. The partial-fraction
expansion of F(s) is i $ -* oo

The only restriction is that fit ) must be continuous or contain, at most, a

F(s) = +
*
+
K2 4 (6.186)
step discontinuity at t = 0. In terms of the transform, F( s ) = £[/(/)];
s+2
s l jl S + 1 + ;l
'

+ - :
i this restriction implies that F(s) must be a proper fraction, i.e., the degree
the poles and zeros to the pole of the denominator polynomial of F(s) must be greater than the degree of
First let us evaluate Kr . The phasors from the numerator of Fis ). Now consider the proof of the initial value theorem,
—
at 1 + j\ are shown in Fig. 6.19c. We
see that Kx is
which we give in two parts.
(a) The function fit ) is continuous at / = 0, that is, /(0 ) = /(0 + ). —
Kr = 3 X
/451 x 2/ 90° 2 From the relationship
72
|;w C[/'(0] = rf ' it )e-sidt = s F i s ) - f i0-) (6.189)
\ jw J 0-
C= 3
C =3 we obtain lim L[/'(0) = lim s F(s) - /(0-) = 0 (6.190)
> -
a + oo CO
\n 1

^ 135» / Therefore lim s F(s) = /(0 ) = /(0 + ) — (6.191)

-2 X 45 - 135'

-T’
180° «-* oo

(h) The function fit ) has a step discontinuity at t = 0. Let us represent

fit ) in terms of a continuous part f f t ) and a step discontinuity D «(/ ,
)

DE A n - -n as shown in Fig. 6.20. We can then write fit ) as

fit ) = + Duit ) (6.192)

(a) (b)

.
where D = /(0 + ) /(0 — — ). The derivative of /( /) is

/'(0 = /'i(0 + Ddit ) (6.193)

. .
FIG 6.19 Evaluation of residues of F
( s)

J The Laplace transform 167

Network analysis and synthesis
166
fi(0
Du(t ) so that limsF(s) = lim 1 |
-. CC J J-. 00 \
- = 00 )
^ +1 J
- — (6.204)

f( D Next we consider the final value theorem, which states that

D lim fit ) = lim s F(s) (6.205)
< -* oo s ~* 0

provided the poles of the denominator of F(s) have negative or zero real
0 t
+ 10 t parts, i.e., the poles of F(J) must not be in the right half of the complex-
0 frequency plane. The proof is quite simple. First
. function into a continuous function
FIG 6.20. Decomposition of a discontinuous
plus a step function.
Since f f t ) is continuous at t = 0, we know
from part (a) that
\- j it ' )e-at dt = s Fis ) - fiO- ) (6.206)

lim s Fj(s) = /i(0 ) = /(0 )

« -* co
of Eq 6.193, we have
— —. (6.194) Taking the limit as s * 0 in the Eq. 6.206, we have

f>
— 0
( dt = lim s Fis ) /(0 — — ) ( 6.207)
Taking the Laplace transform of both sides
-
s Fis ) fi0 ) s F f i s ) /i(0 ) + D
s F( s ) = 5 Ffs ) + D
- - —— (6.195)
(6.196)
Jo
Evaluating the integral, we obtain
fi 00) - /(0-)
s- »0

s F(s) - /(0-)
which simplifies to
Now, if we take the limit of Eq. 6.196 as s *
D, we have
co and let /(0+ ) /(0 )

lim s Fis ) = lim s F f s ) + /(0 +) /(0

) (6.197)
--
- -
— —= Consequently, / ( 00) = slim
~* 0
= lim
« -* 0

s Fis )
(6.208)

( 6.209)
S ~* OO S ~* co Example 6.20. Given the function
By Eq. 6.194, we then obtain fit ) = 3u( t ) + 2e~l (6.210)
(6.198) which has the transform
lim sF(s) = / (0 + )
3 “* 00
_ 3 + 2 5J 3
F( )x
Example 6.18. Given the function ' = -, + 7TT = sis + 1) (6.211)
2is + 1)
Fis ) = 5* + 2s + 5
is greater than the degree
(6.199) let us find the final value /( 00 ). Since the poles of Fis ) are at s
we see the final value theorem applies. We find that
—
= 0 and s = 1,

let us find /(0 + ) . Since the degree of the denominator

applies. Thus lim s Fis ) = 3 (6.212)
of the numerator of Fis), the initial value theorem 3 — O

which is the final value offit ) as seen from Eq. 6.210.

limsFCs) = lim :
-
S * Q0
(6.201) Example 6.21. Given = 2*‘
fit ) (6.213)
= 2e~ cos 21 r-; j
Since C
" i[F(i)]
‘ we see that lim fit ) = co (6.214)
I
we see that /(0 + ) = 2.
Example 6.19. Now let us consider a case where
not apply. Given the function
the initial value theorem does But from ,
* F( ) = — 2s
(6.215) 1

:
fit ) = <5(0 + 3e~
we see that (0 + ) = 3 . The transform of fit
/ ) is
(6.202)
‘ - we have
s 0
—
lim s Fis ) - 0

We see that the final value theorem does not apply in this case, because the pole
(6.216)

*V> -' + 7TT

3 (6.203) s = 1 is in the right half of the complex -frequency plane. !

.
 The Laplace transform 169
168 Network analysis and synthesis
TABLE 6.1 TABLE 6.1 (cont .)
s
Laplace Transforms 18. cos cuf s2 + CO 2
Fis )
no a
19 . sinh at s2 - a2
F( s ) = dt
1. no 20 . cosh at
s

2. fli /xW + a2 f2( t ) a1 Fj( s ) + a2 F2( S ) s2 -a 2

CO

3.
d sF ( s) — / (0 — ) 21 . e at sin cot (J + <f ) + <o2
dt
(* + «)
dn 22. c-*' cos a) l
4 - dT« no p =i
(J + a)2 + o>2
1 e-cUtn 1
- F( s )
5. f^nodr s
23 .

t
n\ (i + a)n+1
s
1
—
JJ sin cot
*
24. (s2
6. /(r) dr da 2 co + CO2 )2
1
dn
25 . -an1- J„( at ) ; n = 0 , l , 2, 3 , . . .
'

-» ' 5= (s2 + «2)^[(s2 + a2)^ - s] " “

7. ( "/( )
(Bessel function of first kind ,
8 . fit —• — a) u(t a) e-a 3 F ( s )
F( s - a)
wth order)
9. «» /(/) 26. (TTT ) X ~
s M
~

1
10. S( t ) T(k + 1)
27 . tk ik need hot be an integer) slc+ 1
dn Sn
"
12.
•
dT O
H(
^ 0
1
5
Problems

1 6.1 Find the Laplace transforms of

13 . t 72 (a)
ib )
-
f (0 sin («/ + P )
fit ) = <r< ‘+«> cos { a>t + p )
14.
tn
7! sn+1
1
ic )
(d) / (0 =
-
fit ) (/3 + l )e-“
K is a real constant > 1.
1 fit = + ‘
) (/ 2 ae - )e-M
15 . e-*‘ S +a (/ ) /(o = t m
1 ( ) fit ) = u[ t ( t 2 - 4)]
__L_ ie * - e-K *6.2
16 .
P -«
~
‘ ‘ )
is + a)(s + /?) Find the Laplace transforms of
a) («) /(/) = cos (/ - 7T-/4) «(/ - W4)
17. sin ft) / s2 + o>2 ( )
* /(/) = cos (1 - -77-/4) «(/)

174 Network analysis and synthesis

6.16 It is given that
as2 + bs + c
Fis ) = s2 + 2s2 + s + l
Knd
- *°
6.17 Using
nd

and /( «> ) for

(a)
js + i )(s + 2)
Fis ) = is + 3 )(J + 4)
= 1.
. ,.
PP y

chapter 7

Fis ) = is +
s Transform methods in
ib ) l )(s - 1 )

ic )
s2 + 3s + 2
Fis ) = J3 + 3 j2 + 3s + T network analysis
s{ s + 4)(s + 8 )
id ) Fis ) = is + l )(s + 6)

7.1 THE TRANSFORMED CIRCUIT

In Chapter 5 we discussed the voltage-current relationships of network

elements in the time domain. These basic relationships may also be
represented in the complex-frequency domain. Ideal energy sources , for
example, which were given in time domain as v( t ) and i( t ) , may now be
represented by their transforms V( s ) = C[y(f )] and / ( s ) = C[/( f )]. The
resistor, defined by the v- i relationship
v( t ) = Ri( t ) (7.1 )
is defined in the frequency domain by the transform of Eq . 7.1 , or
V( s ) = R I( s ) (7.2)
r For an inductor, the defining v - i relationships are
7
•

di
v( t ) = L -
dt
(7.3)
i( t ) = ff
L Jo-
v( r ) dr + i (O - )
Transforming both equations, we obtain
F(s) = sLI ( s ) - Li(O - ) ::
(7.4)
I( s ) = 4 V( s) +
sL s
175

:
 Network analysis and synthesis Transform methods in network analysis 177
176
I<s ) ii( t ) hit )
O-
-o O

I< s)

sL
• M •

(T)
-
Li (O )
V( s ) A
^ V( s)
vi ( t ) Ll Li V 2( t )

-o o .
FIG 7.3
. .
FIG 7.1 Inductor.
in The process of solving network differential equations with the use of
The transformed circuit representation for an inductor is depicted transform methods has been given in Chapter 6. To analyze the circuit
.
Fig. 7.1 For a capacitor, the defining equations are on a transform basis, the only additional step required is to represent all
v( t ) = —f
i
i(r) dr + v( Q — )
the network elements in terms of complex impedances or admittances with
associated initial energy sources.

The frequency domain

,w ~ c7,
C Jc
dv

counterparts of these equations are then

_ (7.5) Consider the example of the transformer in Fig. 7.3. If we write the
defining equations of the transformer directly in the time domain, we have

dt dt
(7.7)
nS) 4 sC -
« i5 -)
’ s
7(s) = sC F(s) - C »(0-)
») + = (7.6) MO =
Transforming this set of equations, we obtain
M T1 + L2 T
dt dt
“

as depicted in Fig. 7.2.

From this analysis we see that in the complex-frequency representation
= , -,
F,(s) sL /,(i) L 1,(0-) + sM J2( s ) - M i2(0-)
(7.8)
the network elements can be represented as impedances and admittances in - -
V2( s ) =. sM /,(s) Mt\( 0 ) + SL 2 I2( S ) - L2 / 2(0-)
series or parallel with energy sources. For example, from Eq. 7.4, we see This set of transform equations could also have been obtained by repre-
is sL, senting the circuit in Fig. 7.3 by its transformed equivalent given in Fig. 7.4 .
that the complex-frequency impedance representation of an inductor
and its associated admittance is 1 jsL Similarly, the impedance of a . In general, the use of transformed equivalent circuits is considered an
capacitor is 1/sC and its admittance is sC This fact is very useful in circuit . easier way to solve the problem.
analysis Working from a transformed circuit diagram, we can write,
.
mesh and node equations on an impedance or admittance basis directly.
£lil(0 ) - £212(0 ) -
-O
-°+
^= s
I( ) m . \ *M .
I sL1 sLz

+
*<>->
V( s )
I ° 0 Cv ( 0 -)
V( s) 1 VM
ATi2(0 ) -
v*w

Jo -5
. .
FIG 7.2 Capacitor.
.
FIG 7.4
•
'

:
s
1
178 Network analysis and synthesis Transform methods in network analysis 179

+
| 1
2? ^— -
0
Li
— —vW
1
3 3 2
+

+
2
V + +
QuiW R 5.

-i- . 2

..
FIG 7 S FIG. 7.8
.
Example 7.1 In Fig. 7.5, the switch is thrown from position
1 to 2 at / = 0. Since the switch is closed at / = 0, we can regard the 5-v battery as an equiva-

-
Assuming there is no coupling between L and L 2 , let us write the ,
mesh equations lent transformed source 5/s. The circuit is now redrawn in Fig. 7.8 as a trans-
from the transformed equivalent circuit in Fig. 7.6 . The mesh equations are formed circuit. The mesh equation for the circuit in Fig. 7.8 is

- - *0 = + h) Il(s) ~ 7cUs )
^0 5
V1( s ) + L1 iLl( 0 )
(7.9)
- -
s s (
= 3 +s + 7(s) (7.10)
Solving for 7(s), we have
0 -) = ~ 7 ( s} + (i + sLi + R ) h( s )
ch 7(s) = (s
2s +3 1 1
+ +s + 2
~ (7.11)
+ l )(s + 2) s 1

+
— sL\

ili£ (0 ) , -+
MQ ) -
sLi

Li iz2(0 —
+
)
Therefore,
Example 7.3 .
/(/) = c-J[/(s)] = r* + e~ 2t (7.12)
Consider the network in Fig. 7.9. At t = 0, the switch is opened.
,
Let us find the node voltages t> (r) and v2( t ) for the circuit. It is given that
L =i h C = If
QVi(s)
±
R
G 1 mho
— F=1v
7c
—
1
—
i (0 )
.
FIG 7.6 +

,
.
Example 7.2 In Fig 7.7, the switch is thrown from position 1 to 2 at
.
the initial conditions are iL{ 0 )
time
2 —. =
c =£ uc (0 )-
t 0. Just before the switch
= is thrown
amp, VQ(0 = 2
) -
v . Let us find the current /
'
( / ) after the switching action

FIG. 7.9
Before we substitute element values, let us write the node equations for the
transformed circuit in Fig. 7.10. These are
Node Vp .
— -— + Cvd0 -) =|/sC + —1 j F,(s) - — V ( s)
»V,(0 )
j
\ 1
2

Node V2 : (7.13)
id0-) 1
.
FIG 7.7 s (n: + c) m
 Transform methods in network analysis 181
180 Network analysis and synthesis

Vi V2
Network N
n voltage sources I( s ) Zi( s )
©— m current sources

© Cuc (0 - TsC
)
(0 )

% G
.
FIG 7.11
Thevenin’s theorem
From the standpoint of determining the current I( s) through an element
FIG. 7.10 of impedance Z1(s), shown in Fig. 7.11, the rest of the network N can be
replaced by an equivalent impedance Ze( s) in series with an equivalent
If we assume that prior to the switch opening the circuit had been in steady
— —
state, then we have vrfO ) = 1 v, iL(0 ) = 1 amp. Substituting numerical
values into the set of node equations, we have
voltage source Ve( s ), as depicted in Fig. 7.12. The equivalent impedance
Ze( s ) is the impedance “looking into” N from the terminals of Zx( s ) when
all voltage sources in N are short circuited and all current sources are
1 -1
JFI
- = ( +- J (5) - IK*( )
5
(7.14)
open circuited. The equivalent voltage source Fe(s) is the voltage which
appears between the terminals 1 and 2 in Fig. 7.11, when the element Zx( s )
is removed or open circuited. The only requirement for Thevenin’s
1 2
(H y2( s )
theorem is that the elements in Zx must not be magnetically coupled to any
element in N.
Simplifying these equations, we obtain The proof follows. The network in Fig. 7.11 contains n voltage and m
-
s 1 = (s* + 2)V1( s) - 2V2{s) current sources. We are to find the current I( s) through an element that
(7.15) is not magnetically coupled to the rest of the circuit, and whose impedance
1 = -2 Fj(i) + (s + 2 )V2( s ) is Zjis). According to the compensation theorem,1 we can replace Zx( s)
Solving these equations simultaneously, we have by a voltage source V( s), as shown in Fig. 7.13. Then by the superposition
s+1 s +1 principle, we can think of the current I(s) as the sum of two separate parts
Vi( s ) = s2 + 2s + 2 ( s + l )2 + 1 ~
I( s) = IM + I2( s) (7.18)
(7.16)
y (s) = 5 +2 = ( r.+ 1)2 + 1 Let the current Ix( s) be the current due to the n voltage and m current
s2 + 2* + 2 ^
so that the inverse transforms are
=0< * ?!
Vl( i ) = e * cos t ,
~

7.2 THEVENIN’S AND NORTON’S THEOREMS

v2( t ) = e-
\cos t + sin t ) (7.17)
I\ ZeU )
Ve s )

\ Zi( s )
I( s )
In network analysis, the objective of a problem is often to determine a
single branch current through a given element or a single voltage across 2
an element. In problems of this kind, it is generally not practicable to
write a complete set of mesh or node equations and to solve a system of . .
v
FIG 7.12 Th6venin’s equivalent circuit.
equations for this one current or voltage. It is then convenient to use two
very important theorems on equivalent circuits, known as Thevenin' s 1 See H. H. Skilling, Electrical Engineering Circuits, 2nd Ed. John Wiley and Sons,

and Norton' s theorems. New York, 1965 .

182 Network analysis and synthesis Transform methods in network analysis 183
sources alone ; i.e., we short circuit the source V( s ), as shown in Fig . 7.14a.
Therefore J2( s ) is equal to the short- circuit current 7SC . Let I 2( s ) be the
current due to the voltage source V ( s ) alone, with the .rest of the voltage
7.146).
sources short circuited and current sources open circuited (Fig.
With the m current and n voltage
sources removed in Fig. 7.146, we see +
+1 that the network N is passive so that
Network
N
n voltage sources
m current sources
1( e ) ) (/ ) v<s ) I2( s) is related to the source V(s) by the
* [ - relation
Network
N
9s>
.
FIG 7.13 = - Zin(s) (7.19 )

of the
where Zin( s ) is the input impedance of the circuifafthe terminals
source V( s). We can now write f( s ) in Eq. 7.18 as
.
FIG 7.14 b
I( S ) = Isc( s ) - Zin(s) (7.20)
so that we can rewrite Eq. 7.20 as
case
Since Eq. 7.20 must be satisfied in all cases, consider the particular
when we open circuit the branch containing Z s ). Then I( s ) =
0 and V( s ) ZlD( s )l( s ) - Voc(s) = - V(s) (7.22)
is the open circuit voltage Voc( s ). From Eq 7.20 we have
- .^ In order to obtain the current 7(s) through Zt(s), the rest of the network
N can be replaced by an equivalent source Ve( s ) = V 0c( s ) in series,
(7.21) with an equivalent impedance Ze( s) = Zm(r), as shown in Fig. 7.12.
-
Zin( S)
Example 7.4. Let us determine by Thevenin’s theorem the current I s ) flowing
vi
+
through the capacitor in the network shown in Fig. 7.15. First, let us obtain ^
V2 +

©
Li(0 - )

+
+
.
I( s ) '
Network
N 9i L

.
FIG 7.15
A
Zt{ s) by opening all current sources and short circuiting all voltage sources.
Then, we have in the network in Fig. 7.16, where

h
> .
Z (s) = R + sL (7.23)
h Next we find V ,( s ) by removing the capacitor so that the open-circuit voltage
.
FIG 7.14c • I between the terminals 1 and 2 is Ve( s ), as shown in Fig. 7.17. We readily

i
 184 Network analysis and synthesis Transform methods in network analysis 185
ol
Ls

R Ze( s )

o2

FIG 7.16.
FIG. 7.19
determine from Fig. 7.17 that

By Thevenin’s theorem we then have

Ve( s ) = / ( s )R + Li( fi — ) -
VrXO — ) (7.24) and the input impedance to the left of N is
JII(1/JQ
Ze{ s ) = SL2 + sLfi T 1 fsC (7.27)

I( s ) R + 17(0 -) - rc(0 -)/s

m = Z ,( sVe) +( s Z( s)
) (7.25) RVe( s )
R + sL + 1 IsC We know that Vo( s ) = Ze( s ) + R (7.28)

RV ( s )V / sC )
( R 4- sL sL, + 1/sC) + LJC
Therefore V0( s ) = (7.29)
Li( 0 -+
)
Ls +]
>C (0 ~ > Finally v0( t )
^
= C-HKoW] (7.30)
IMC R
+ 41 Norton’s theorem
Ve( s )

FIG. 7.17
— o2
When it is required to find the voltage across an element whose admit-
tance is yi(s), the rest of the network can be represented as an equivalent
admittance Ye( s ) in parallel with an equivalent current source Ie( s ), as
shown in Fig. 7.20. The admittance Ye( s ) is the reciprocal of the Thevenin
Example 7.5 For the network in Fig. 7.18, let us determine the voltage v0{ t )
.
across the resistor by Thevenin’s theorem. The switch closes at / = 0, and we
impedance. The current Ie{ s ) is that current which flows through a short
assume that all initial conditions are zero at t = 0. circuit across Tf (s). From Fig. 7.20,

S Li Li = ye( s) V1( s )
+ n(s)
(7.31) .
I ( s)

The element whose admittance is Yx must not be magnetically coupled to

any element in the rest of the network .
Qv( t )
* % Vo( t )
Z Consider the network in Fig. 7.21. Let us find the voltage across the
capacitor by Norton’s theorem. First, the short-circuit current source
4(s) is found by placing a short circuit across the terminals 1 and 2 of the
FIG 7.18 . Ei

First let us redraw the circuit in terms of its transformed representation

, I 1
inspection the
which is given in Fig. 7.19. We can almost determine by that
' h(s ) Ye( s ) Yi( s )
Thevenin equivalent voltage source of the network to the left of N in Fig. 7.19 is

K( s) =
V ( s )( l / sC )
LjS + 1 IsC
(7.26)
T
FIG. 7.20
X

I
186 Network analysis and synthesis Transform methods in network analysis 187
sL sL 1 0.1s 1
?i -''TnffiT' ?i
__ +
h(* )

mC R
T~ «c
i
I( s )
0 R ' ’ h( s ) V( s ) 2

4> 2
.
2
t
FIG. 7.21 FIG 7.22 FIG 7.24.
capacitor, as shown in Fig. 7.22. From Fig. 7.22 Ie( s) is
Us ) is then Us )
Us ) =
.
/ (s) =
I ( s )R
sL + R
(7.32)
R2 [ Ye( s )

1
+ G2 ]
(7.37)

The admittance Ye( s) is the reciprocal of the Thevenin impedance, or (s + 5 )%s + 6)

(7.38)

.
Y ( s) =
sL + R
1 ( 7.33) By inspection , we see that Us) can be written as
( s + 6) - (s + 5) 1 1
Then the voltage across the capacitor can be given as Us ) = (* + 5 )2( s + 6) = (s " (7.39)
+ 5)2 (s + (j
6) + 5)
I( s ) R
F(s) = Y s )IJM ( s ) ( sL + R )sC + l
(7.34) Repeating this procedure, we then obtain
£ + Yc
Example 7.6. In the network in Fig. 7.23, the switch closes at / = 0. It is given 1 1 1
that v( t ) = 0.1e 5i and all initial currents and voltages are zero. Let us find the
' U.s ) = (r + 5)2 s -(- 5 +s+6 (7.40)
current /2(0 by Norton’s theorem.
The transformed circuit is given in Fig. 7.24. To find the Norton equivalent Taking the inverse transform of I2( s ) , we finally obtain
current source, we short circuit points 1 and 2 in the network shown. Then
Ie( s ) is the current flowing in the short circuit, or 4(/) = (/e-5{ - <r5t + e ~6t ) u( t ) (7.41)
V( s ) 0.1 1
/e( j) (7.35)
Rt + sL L( s + RlL )( s + 5) (j + 5 )( s + 10) 7.3 THE SYSTEM FUNCTION
The equivalent admittance of the circuit as viewed from points 1 and 2 is
As we discussed earlier, a linear system is one in which the excitation
1 i LC + sRxC + 1
? 0.5J2 + 5s + 10 (7.36) / e( t ) is related to the response r( t ) by a linear differential equation. When
Ye( s) = sC + iq + sL i?! + sL s + 10 the Laplace transform is used in describing the system, the relation between
the excitation E( s ) and the response R( s) is an algebraic one. In particular,
i = 01 *1 = 1 1
k( t )
I when we discuss initially inert systems, the excitation and response are
related by the system function H( s ) as given the relation

R2 = 2
R( s ) = E( s )H( s ) (7.42)
C =|
We will discuss how a system function is obtained for a given network,
and how this function can be used in determining the system response.
O - h- As mentioned in Chapter 1, the system function may assume many forms
.
FIG 7.23 and may have special names such as driving-point admittance, transfer
 188 Network analysis and synthesis Transform methods in network analysis 189

^( v
3! 4 =
t
-J
sL
T *
V0( s ) +
Vg ( s ) Z2( s ) V0( s )

FIG. 7.25
.
FIG 7.27

-
i m p e d a n c e , v o l t a g e or c u r r e n t r a t i o t r a n s f e r f u n c t i o n . This is because the current
form of the system function depends on whether the excitation is a voltage
or current source, and whether the response is a specified current or I( s ) =
V f s)
(7.45)
voltage. We now discuss some specific forms of system functions. Zx(s) + Z f s)
Impedance Since V0 f ) = Z f s ) I f ) (7.46)
When the excitation is a current source and the response is a voltage,
V0( s ) Z 2( s )
then the system function is an impedance. When both excitation and then ( 7.47)
response are measured between the same pair of terminals, we have a Kf ) Z f s ) + Z2(s)
driving-point impedance. An example of a driving- point impedance is )
Current-ratio transfer function
given in Fig. 7.25, where When the excitation is a current source and the response is
another
current in the network, then H f ) is called a current ratio transfer
)
H f ) = Kf = R +
( l /sC)sL
( 7.43) -
As an example, let us find the ratio IJIg for the network given
function.
hf ) s L + 1/ s C in Fig. 7.28.
Admittance Referring to the depicted network, we know that
When the excitation is a voltage source and the response is a current ,
H f ) is an admittance. In Fig. 7.26, the transfer admittance I2/ V g is
obtained from the network as
hf )
— h i s ) + i o f ), hf ) — = I 0( s )( R + sL) (7.48, 7.49)

1
Eliminating the variable f , we find
Hf ) = (7.44)
F„(s) R ,+ sL+ ] / s C\ ' hf )=hf ) i ( + 1+s C
R sL
) (7.50)
Voltage-ratio transfer function /
When the excitation is a voltage source and the response is also a voltage,
i so that the current-ratio transfer function is
then H f ) is a voltage-ratio transfer function. In Fig. 7.27, the voltage-
, Iof ) 1/s C
_,
ratio transfer function V0 f ) j V g f ) is obtained as follows. We first find the
L
hf ) R+ s L+ l / s C
(7.51)

+1
— —— sL
nnnr' 1 ( wv
*C Ri

Ri
h( s )
n m^
sL

' h( s )
!;
CKfs )
I
M s) 7
*W ()J sC
R

FIG. 7.26 FIG. 7.28

i
-

190 Network analysis and synthesis Transform methods in network analysis 191
s 2. For the square pulse in Fig. 7.30, i j f ) can be written as

T 4(0 = «(0 ~ «(< - a) (7.55)

^> (t) C: L G v( t ) Its transform is I a f ) = - f - e-°° )

s
(7.56)
The response F(s) is therefore given as
FIG. 7.29
Vf ) = 1 - e~a s
2 (7.57)
C[s + s( G/ C) + 1/LC]
From the preceding examples, we have seen that the system function is a
function of the elements of the network alone, and is obtained from the Note that in obtaining the inverse transform C-1[ F( j)], the factor e~t t S
network by a straightforward application of Kirchhoff ’s laws. Now let us must be regarded as only a delay factor in the ti me domain. Suppose we
obtain the response transform R f ), given the excitation and the system rewrite F(s) in Eq. 7.57 as
function. Consider the network in Fig. 7.29, where the excitation is the
current source i f t ) and the response is the voltage v( t ). We assume that
the network is initially inert when the switch is closed at t = 0. Let us find
Vf ) = ^ s
(1 - e~a s ) (7.58)

the response V f ) for the excitations : Then if we denote the inverse transform by v f t ),
1. /,(/) = (sin co0t ) u( t ).
2. i f t ) is the square pulse in Fig. 7.30.
3. i f t ) has the waveform in Fig. 7.31.
|i
i!

we obtain the time response

rfo - -psa]
t (7.59)

First, we obtain the system function as ‘

vf ) = v f t ) - v f t - a ) u( t - a) (7.60)
I s
Hf ) = (7.52) Observe that v f t ) in Eq. 7.59 is the response of the system to a unit step
sC + l /sL+ G C[s 2 + s(G / C) + 1/LC] i
excitation.
1 ift)
— (sin co0t ) u( t ). The transform of i g f ) is
i:
3. The waveform in Fig. 7.31 can be represented as

hf ) “
s TVS
+ o CO
(7.53) t (
i g f ) = u f ) + - u( t ) - t ~ a ) u ( t - a) (7.61)
! a a
so that V(s) = I g f )H ( s ) = s 2 —+^ 02 s
(7.54)
<u C[s2

hm
+ s{ G j C ) + 1/LC] and its transform is hf ) = - l
s \
( +-
as
- a s /) — (7.62)

h i t )| V f ) is then F(s) = (\ l + a s -— —) 2
1
a s C [s + s( G / C ) + 1 / LC ]
/
(7.63)
2
If we denote by v t ) the response of the system to a ramp excitation, we
l
i:
see that ^
t
0! t m .
C 1[ F(.s)]
— j - - v 2( t )
vft) |
a — — —- vtf
a
a) u f a) (7.64)

.
FIG 7.30 FIG. 7.31 where v f f ) is the step response in Eq . 7.59.

A
 192 Network analysis and synthesis Transform methods in network analysis 193

Let us now discuss some further ramifications of the equation for the It is apparent that the free response is 0 Ae , and the forced response is
0.4 cos t + 0.2 sin t.
— ~2 f

response R( s ) = H( s ) E( s ). Consider the partial-fraction expansion of

R( s ) : As a final topic in our discussion, let us consider the basis of operation
for the R-C differentiator and integrator shown in Figs 7.33a and 7.33b. .
m = 2i S Si + 2i S h.S j (7.65)
—
where s, represent poles of H( s ), and s } represent poles of £(J). Taking
— We use the Fourier transform in our analysis here, so that the system
function is given as H( jco ) , where co is the ordinary radian frequency
variable. Consider first the system function of the differentiator in Fig.
the inverse transform of R( s ), we obtain 7.33a .
(7.66) Vo( j<o ) _ R jcoRC
r(f) = 2
i
+ 2i Bfi'* V ( jco ) R + 1IjcoC„ jcoRC + 1
(7.71)

The terms Ate it are associated with the system H( s ) and are called free
3
Let us impose the condition that
response terms. The terms are due to the excitation and are known
as forced response terms. The frequencies sf are the natural frequencies of coRC «1 (7.72)
the system and j, are the forced frequencies. It is seen from our discussion We have, approximately,
of system stability in Chapter 5 that the natural frequencies of a passive
network have real parts which are zero or negative. In other words, if we
YAM ~ jcoRC (7.73)
Vfjco )
denote st as s4 = + jcoiy then er,- < 0.
Example 7.7. For the initially inert network in Fig. 7.32, the excitation is
Then the response V0( jco ) can be expressed as
vg ( t ) =|cos t «(/). Let us find the response v0( t ) and determine the free and
forced response parts of v0( t ). The system function is
Vfjco ) a RCljco Vfjco)] (7.74)

1/ sC 2
(7.67) Taking the inverse transform of V0( jco ), we obtain
~
Vff ) “ R + 1IsC s + 2
V 0( 0 — RC at
J - Vg( t ) (7.75)
„
Since K (s) is (7.68)

Note that the derivation of Eq. 7.75 depends upon the assumption that the
the response is then
s 0.4 0.4r + 0.2 -
R C time constant is much less than unity. This is a necessary condition.
Fo0O = Vfs ) H( s) = (s7+ T)(* + 2 ) ~ ~
s + 2 + s2 + 1
(7.69) -
Next, for the R C integrator in Fig. 7.33b, the voltage-ratio transfer
function is
We next obtain the inverse transform Vo( jco ) 1/ jcoC 1
(7.76)
v0( t) = —0 Ae ~2* + 0.4 cos t + 0.2 sin t (7.70) Vg( jco ) 1IjcoC + R jcoCR + 1
R = 2U C
-A/WV +
I
+
vg ( t ) C = if
= : vo( t ) i Vg ( t ) R vo( t ) vo( t )

(a) ( b)
:
FIG . 7.32 -
FIG , 7.33. ( a ) R C differentiator. (b) R C integrator. -

Transform methods in network analysis 195

194 Network analysis and synthesis
If we assume that coRC » 1, then
KOco ) ^ jcoRC —
— KG*> ) 1 l
(7.77)
C m
Under these conditions, the inverse transform is +
t
m
— — Jo „
bit ) R
«o(0
1
RC
v ( f ) dr (7.78)
~R
<- /
i RC

*C
.
so that the R C circuit in Fig. 7.33b is approximately an integrating circuit
- °FIG. 7.34 FIG . 7.3S. Current impulse
response of R C network in
Fig. 7.34.
-
7.4 THE STEP AND IMPULSE RESPONSES

In this section we will show that the impulse response h( t ) and the system Since the step response is the integral of the impulse response, we can
and
function H( s) constitute a transform pair, so that we can obtain step use the integral property of Laplace transforms to obtain the step response
impulse responses directly from the system function . as
,
We know, first of all, that the transform of a unit impulse <5(/) is unity
i.e., C[<5(/)] = L Suppose the system excitation were a unit impulse, then «(0 = r1 ® [ ] (7.84)
the response R( s) would be
where a(f ) denotes the step response. Similarly, we obtain the unit ramp
R( s ) = E( s) H( s ) = H ( s ) (7.79)
response from the equation
( s)
We thus see that the impulse response h( t ) and the system function H (7.85)
constitute a transform pair, that is,
C[fi(0] = H( s ) where y( t ) denotes the ramp response. From this discussion, it is clear
(7.80)
that a knowledge of the system function provides sufficient information to
C-W)] = h( t )
obtain all the transient response data that are needed to characterize
Since the system function is usually easy to obtain, it is apparent that
we the system.
can find the impulse response of a system by taking the inverse transform
Example 7.9. Let us find the current step response of the R-L circuit in Fig.
of H( s ). 7.36. Since I( s) is the response and VB( s) is the excitation, the system function
Example 7.8. Let us find the impulse response of the current i( t ) in the R
-C is
circuit in Fig. 7.34. The system function is : H {S )
I( s ) 1
(7.86)
J( s ) 1 1 V ( s)„ R + sL
* (7.81) Therefore H( s )/ s is
,
V ( s) R + 1 IsC R( s + 1/ RC ) ff ( s ) 1 1(' 1 \ (7.87)
Simplifying H( s ) further, we have s s( R + sL) R \s s + R/ LJ
1 IRC
“ i( -1
s + 1 IRC ) (7.82) The step response a(/) is now obtained as

The impulse response is then

Kt ) = £"

W = [
__ ] )J <5(0
l
u{ t ) (7.83)
«(0 =
1

^ 0 - e -WW ) u( t ) (7.88)

which is shown in Fig. 7.35 . ^ RC

To check this result, let us consider the impulse response of the R L circuit, -

V
L
 Transform methods in network analysis 197
196 Network analysis and synthesis
which we found in Chapter 5. 7.5 THE CONVOLUTION INTEGRAL
h( t ) = 2e
1
- ( RIL )l u( t) (7.89)
In this section we will explore some further ramifications of the use of
response, or the impulse response h( t ) to determine the system response r( t ). Our
The step response is the integral of the impulse discussion is based upon the important convolution theorem of Laplace
(or Fourier) transforms. Given two functions /j(0 and f 2it ), which are
-JAM*
^ co (7.90)
«(< ) = - e-< K/iW) « a zero for t < 0, the convolution theorem states that if the transform of
response of an initially inert fx( t ) is Fx( s ) , and if the transform of f 2( t ) is F2( s), the transform of the
It is readily seen that if we know the impulse
of the system due to any other convolution of fxit ) and fft ) is the product of the individual transforms,
linear system, we can determine the response -
excitation. In other words, the impulse resp
onse alone is sufficient to charac
. Fi( s ) F2( S ), that is,
and response
terize the system from the standpoint of excitation
o -o [ J fi( t - / ( ) drj = Fi(
C T) 2 T s) F2(S) (7.97)

Inert uo( t )
Vg ( s )
+ Vi( t ) network where the integral jVi(* - ) / ( ) dr T 2 T

is the convolution integral or folding integral, and is denoted operationally

FIG 7.36 . .
FIG 7.37
as
about the system
Example 7.10 In Fig 7.37, the only information we possess
. . v {( t ) = <3(0,
(7.98)
inert linear system ; (2) when
in the black box is : (1) it is an initially
then (7.91) Proof. Let us prove that £[/ */2] = FXF2. We begin by writing
v0( t ) = ( e~2 he~ ) u( t ) ‘- ‘
3 j

what the excitation vt(t ) must be in order

With this information, let us determine
to produce a response v0( t ) = te u( t ). First
~n , we determine the system function t[/i(0* /«(0] = j/ (* - )/*00 dT\ dt
i T (7.99)

to be
V&) 1 1 2^ + 5 , (7.92) From the definition of the shifted step function
W ~ V ( s) ~ s + 2 + s + 3 (s + 2)( s +3)
<
~u
We next find the transform of te u( t ) . :
u( t — r) =1 r<t
(7.100)
1 (7.93) =0 r>t
we have the identity
The unknown excitation is then found from the equation
Vo( s) (* + 2 )( s + 3) (7.94)
Jjft - )/ ( ) dr =
T 2 T frit - r) u( t - r)/ (r) dr 2 (7.101)
M H ( s) 2( s + 2.5)(J + 2 f
Then Eq. 7.99 can be written as
, we have
Simplifying and expanding V s ) into partial fractions
^<s +
Vi( s ) = 2( s + 2.5 )( s + 2)
3)
s
1
+2
0.5
s + 2.5
(7.95) t[/i(0*/i(0]
-JV J7 - “ 0 0 u( t - T)/2(T) dr dt
i
(7.102)

The system excitation is then V

If we let x = t — r so that
— __ —
g st g s(*+r
)
(7.103)
vft ) = (c " 2
‘ - 0.5e-2-5') u( t ) (7.96)

Transform methods in network analysis 199

198 Network analysis and synthesis
This is readily seen from the relationships
then Eq. 7.102 becomes
£[/i(0V»(0] = r [COfi ( x ) u( x ) f 2( r )e-"e-’ xdrdx W 0* M 0 ] = Ffs ) Ffs ) (7.112)

[
= Jo
—
Jo Jo

7(*) u( x)e
i

Ffs ) F 2(s)
- x
dx f7a(r)c-" dr
Jo
(7.104)
and Wt )* fm = Ffs ) Ffs)
To give the convolution integral a more intuitive meaning, let us examine
the convolution or folding process from a graphical standpoint. Suppose
we take the functions
fiij ) = u( r )
(7.113)

The separation of the double integral in Eq. 7.104 into a product

of two (7.114)
integrals is based upon a property of integrals known as the separability Mr ) = r u( r )
property? as shown in Figs. 7.38a and 7.39a. In Fig. 7.38, the various . steps for
( j) = e~ 2t u( t ) obtaining the integral
.
Example 7.11 Let us evaluate the convolution of the functions fx
and f2( t ) = t u( t ) , and then compare the result with the inverse transform of ffi
( t r ) f 2( f ) dr -
Fx( s) F2( S ), where
FM = W)1 = 7
1 are depicted. Part (b) of the figure shows fx( r ) = «( r). The function — —
^ (7.105) flit - T) = u( t r) -(7.115)
Ms ) = 7
1

product
, Ml ~ T ) MT ) = «(<
—
in part (c) merely advances fx{ r ) by a variable amount t. Next, the
r )r u( r ) -(7.116)
the dummy
The convolution of fx( t ) and f2( t ) is obtained by first substituting
variable / - v for t in /i( / ), so that is shown in part ( d ). We see that the convolution integral is the area under
the curve, as indicated by the cross-hatched area in part ( d ). Since the
fft - r) = e~w ~7 u( t - T) (7.106)
)
convolution integral has a variable upper limit, we must obtain the area
Then MO* MO is
—
under the curve offft r ) f2{r ) for all t . With t considered a variable in
j/ lit ~ r ) f2( r ) dr =
£ — Te -W r) ( JT = e ~ 2t j Tfi 2 r dr (7.107) Fig. 7.38rf, the area under the curve is
fi* f 2 =
t2
= u( t )m - (7.117)
Integrating by parts, we obtain
_ as plotted in part (e) of the figure.
/i(0* /a(0 = c 2< ) U{ t ) (7.108)
In Fig. 7.39, we see that by folding /2(T) about a point t , we obtain the
From Eq. 7.105, we same result as in Fig. 7.38. The result in Eq. 7.117 can be checked by
Next let us evaluate the inverse transform of Fx( s) F2( s ). taking the inverse transform of Ffs) F 2( s ), which is seen to be
have y
piis ) F2( S ) = 7 J * l
“ - + J+ 2 (7.109) £ MFX ( S ) F 2( S ) ] =
{
= -2 u( t ) (7.118)
^
-i [ Fx( s ) F2( S ) ]
2)

(7.110) i Let us next examine the role of the convolution integral in system
so that Z analysis. From the familiar equation
An important property of the convolution integral is expressed by the R( s) = E( s ) H( s ) (7.119)
equation we obtain the time response as
jut -
r)/2(r) dr = /iW/aCf ~ r ) dr (7.111)
J r( l) = tr' iEis ) H ( s ) ] =
JV Kt - ) r ) dr (7.120)
1 See, for example, P. Franklin, A Treatise on Advanced Calculus, John Wiley and
where e(r) is the excitation and A(r) is the impulse response of the system.
Sons, New York, 1940 .
 Transform methods In network analysis 201
200 Network analysis and synthesis
fi ( T ) Using the convolution integral , we obtain the response /(f ) as
f1( r)

Ko -U , — r) h( r ) dr =2 Jdr ( 7.124)
o
to)
T 0
= 2e ‘ e
~
j * ~r dr = 2(e “

‘ ~ KO

fl ( - T) 2
A/WW
1

o o
( b)
\
Mt - r
(b)
C) v t
( ) 1< o - -o
fl ( t - r)
1
)
eft )
Ideal
amplifier
r( t )

t
0 t T 0
(0
T
FIG. 7.40 .
FIG 7.41
< c)
Example 7.12b. The ideal amplifier in Fig. 7.41 has a system function H( s ) = K,
h( t - T) h(T ) f\ ( T ) hit - T > where AT is a constant. The impulse response of the ideal amplifier is then
KO = K <5(0 (7.125)

0 t f Let us show by means of the convolution integral that the response r( t ) is related
<d ) (d ) to the excitation e( t ) by the equation
K0 = Ke( t ) (7.126)
h* h h* h
Using the convolution integral , we have

0
(e)
t t
K0 = JV ) hit - r ) dr = xj‘ e( r ) 6( t - T) dr = Ke{ t ) (7.127)

FIG. 7.38 FIG. 7.39 Since / is a variable in the expression for r(0 we see that the ideal amplifier in.
the time -
the time domain is an impulse scanning device which scans the input e( t ) from
Using Eq . 7.120, we obtain the response of a system directly in / = 0 to t = oo . Thus, the response of an ideal amplifier is a replica of the
domain . The only information we need about the system is its impulse ^ input K0 multiplied by the gain K of the amplifier.
response . *;
Example 7.12a. Let us find the response /(f ) of the R-L network in Fig
. 7.40 due
7.6 THE DUHAMEL SUPERPOSITION INTEGRAL
to the excitation (7.121 )
= 2e 0 K0 ‘K In the Section 7.5, we discussed the role that the impulse response plays
The impulse response for the current is
:
in determining the response of a system to an arbitrary excitation. In
1 (7.122) this section we will study the Duhamel superposition integral , which also
K0 = e~WL)l 1 K0
describes an input-output relationship for a system. The superposition
Therefore, for the R- L circuit under discussion integral requires the step response a(f ) to characterize the system behavior.
(7.123) We plan to derive the superposition integral in two different ways.
K0 = e~2t K0

;
.

Transform methods in network analysis 203

202 Network analysis and synthesis
Taking the inverse transform of H( s )/ s, we obtain the step response
The simplest is examined first . We begin with the excitation-response
relationship R( s ) = E( s ) H( s ) (7.128) KO = ~ e~4lKCu( t ) (7.135)

Multiplying and dividing by s gives From the excitation in Eq . 7.132, we see that K0 ) = 0 and —
K(S) =
^
5
- S £(S) (7.129)
The response is then
v\( t ) = Ax 6(0 + A2 K0 (7.136)

Taking inverse transforms of both sides gives ,

i( Q = jj ^ >' r ) *( t - r ) dr
C-W)] = s £(s) ] = j At
J
•
<5( T) a( t ~ r ) dr + A2 a( t - r) dr

-" f f l ^
(7.130)
* [ s E( s ) ]
= Ax *( t ) + e-( t-r )IRC gT (7.137)
which then yields
— r
^\
r(0 = a(0*[e'(0 + K0 ) <5(0]
e-
R ‘^Jo-
= R e-‘/*° u(0 + ~ QT! RC

— e(0 ) a (0 + — J— — e’( r ) a( t r) dr (7.131)

= j^
tlRC + C( 1 - K0
where e' ( t ) is the derivative of e(0, K0 ) is the value of e( t ) at t = 0
and a( / ) is the step response of the system . Equation 7.131 is usually
referred to as the Duhamel superposition integral .
As we see from the Example 7.13, KO ) = 0, which is the case in many
transient problems. Another method of deriving the Duhamel integral
—
Example 7.13. Let us find the current /(/ ) in the R C circuit in Fig. 7.42 when - avoids the problem of discontinuities at the origin by assuming the lower

^
the voltage source is (7.132 )
limit of integration to be / = 0 + . Consider the excitation e ( t ) shown by
t)
A
^=^ +
as shown in Fig. 7.43 . The system function of the R C circuit is -
the dotted curve in Fig. 7.44. Let us approximate e( t ) by a series of step
functions, as indicated in the figure. We can write the staircase approxi-
s
mation of e( t ) as '
/(0 (7.133)
H ( s) =
VB( s ) R( s + 1 IRC ) K0 = K0 + ) K0 + AEx u( t - AT) (7.138)
Therefore the transform of the step response is + AE2 u( t 2AT) + -!- AEN u( t «AT)
: " " "

H ( s)
s R( s
I
+ 1 IRC )
(7.134) x
.
e( t)
-r "

Vg ( t )
}A £ 2

+ /
vg ( t ) Slope = A2 e (0 + )
R Ai t
1
i

J'
i FIG 7.42 .
0

.
FIG 7.43
t 0

.
24r 3Ar

FIG 7.44. Staircase approximation to a signal.

4Ar t

::
 204 Network analysis and synthesis Transform methods in network analysis 205

where AEk is the height of the step increment at t = kAr. Since we assume
2

the system to be linear and time invariant, we know that if the response
to a
—
unit step is a( t ), the response to a step Kt u( t r) is K{ <x.( t r . There-
) - — +
fore we can write the response to the step approximation in Fig

r( t ) = e(0 + ) a(< ) + AEX a( t — AT)

. 7.44 as vg( 01
A-T, R. VO( t )

+ AEt a(/
If AT is small, r( t ) can be given as
— 2Ar) + + AEn a(f - nAt )
(7.139)

PROB 7.2 .
7.3 In the circuit shown, all initial currents and voltages are zero. Find

—^
r(t) = e(0 + ) «(<) + a( t - AT) AT /(/) for / > 0 using Thevenin’s theorem.
AT (7.140)
+ AT
1
a( / - 2AT) AT + " "

^
+ AT a( I - « AT) AT v( t )

which, in the limit, becomes 40

4
r( t ) = e(0+ ) a(0 + Arlim-*0 i=J0+

= e(0 + ) a( t) + |
Jo+
» -
+ oo

e' ( r ) a( t - T) dr
^^
~ a(f — IAT) AT
'

(7.141)
WO Q) if :
lh

Problems PROB. 7.3

7.1 In the circuit shown v( t ) = 2u( t ) and //,(0 -) =2 amps, Find and
sketch »2(0- 7.4 In the circuit , shown in (a), the excitation is the voltage source e( t )
described in (b). Determine the response i( t ) assuming zero initial conditions.
1000
e( t )
U "'“l
5h 1000 /
• 1

e( t )

-^
0 1 'it

PROB 7.1 . 1
(a) (b)
at 1
7.2 The switch is thrown from position 1 to 2 at~ t = 0 after having been
for a long time . The source voltage is v ( t ) = Vxe „
at sin §t . PROB 7.4 .
(a) Find the transform of the output voltage vQ( t ).
(b) Find the initial and final values of » (/). „
(c) Sketch one possible set of locations for the critical frequencies in the plane
s 7.5 Determine the expression for v0( t ) when /(/) = 6( t ), assuming zero initial
and write the form of the response vft ). ( Do not take the inverse transform.) conditions. Use transform methods.

Amplitude, phase, and delay 213

I M(<»)
1
Amplitude
0707
"
response

-co -coc 0[ «c
(a)
:
chapter 8 \, <t>( u )
i

Amplitude, phase, and delay Phase response

— co ~ uc
°>c CO

X
X

8.1 AMPLITUDE AND PHASE RESPONSE ( b)

In this section we will study the relationship between the poles and
. .
FIG 8.1 Amplitude and phase response of low pass filter
- .
zeros of a system function and its steady-state sinusoidal response. In
other words, we will investigate the effect of pole and zero positions upon response <f>( cn ) is linear, then minimum pulse distortion will result. We see
the behavior of H( s) along the jco axis. The steady-state response of a from the phase response <f>( co) in Fig. 8.1 that the phase is approxi
-
system function is given by the equation
H ( jto ) = M (ft>y*( 0 ) (8.1)
)
—
mately linear over the range coc co < + coc. If all the significant
harmonic terms are less than coc, then the system will produce
minimum phase distortion With this .
where M(co) is the amplitude or magnitude response function, and is an example, we see the importance of an
even function in to. represents the phase response, and is an odd amplitude-phase description of a system. o-
R
AA/WV -o
function of co. In the remaining part of this chapter,
The amplitude and phase response of a system provides valuable we will concentrate on methods to
information in the analysis and design of transmission circuits. Consider obtain amplitude and phase response Vi ( s ) ic =r v* <*>
the amplitude and phase characteristics of a low pass filter shown in - curves, both analytically and graph-
Figs. 8.1a and 8.16. The cutoff frequency of the filter is indicated on the ically.
amplitude response curves as coc. It is generally taken to be the “half - / To obtain amplitude and phase FIG. 8.2
power ” frequency at which the system function \ H( j<oc )\ is equal to 0.707 curves, we let s = jco in the system
of the maximum amplitude \ H( j<om& x )\. In terms of decibels, the half- function, and express H( jco ) in polar form. For example, for the amplitude
power point is that frequency at which 20 log \ H( jcoc )\ is down 3 db from and phase response of the voltage ratio F2
20 log \ H( jcom* x )\. The system described by the amplitude and phase in Fig. 8.2, the system function is
/ Fx of the R-C network shown
characteristics in Fig. 8.1 shows that the system will not “ pass” frequencies
that are greater than coc. Suppose we consider a pulse train whose H( s ) ^ w me (8.2)
amplitude spectrum contains significant harmonics above co0. We know K( s) s + 1IRC
that the system will pass the harmonics below coc, but will block all Letting s = jco, we see that H( jaS) is
.
harmonics above coc Therefore the output pulse train will be distorted
when compared to the original pulse train, because many higher harmonic H ( jco ) = 1/ RC
terms will be missing. It will be shown in Chapter 13 that if the phase (8.3)
jco + 1 fRC
212
 Amplitude, phase, and delay 215
214 Network analysis and synthesis
ATM
1
0.707

0
0 o>
<j>( w )
- 45"
-90'
FIG . 8.3. Amplitude and phase response of R-C network .

In polar form H( jco ) becomes

H ( jco ) = 2 IRC2 IA g
1 j tan-1 <„ RC
—
( 8.4)
_ FIG . 8.4. Evaluation of amplitude and phase from pole-zero diagram.

*
( co + 1IR C ) In general, we can express the amplitude response M ( co ) in terms of the
The amplitude and phase curves are plotted in Fig. 8.3. At the point following equation .
co = 0, the amplitude is unity and the phase is zero degrees. As co increases,
the amplitude and phase decrease monotonically. -When co = 1 jRC , the
PJ vector magnitudes from the zeros to the point on the jco axis
amplitude is 0.707 and the phase is -45°. This poiht is the half-power
point of the amplitude response. Finally as co oo, M( co ) approaches
M ( co ) = —XT
zero and <f>( co) approaches 90°. —
Now let us turn to a method to obtain the amplitude and phase response
3 =0
vector magnitudes from the poles to the point on the jco axis

from the pole-zero diagram of a system function. Suppose we have the Similarly, the phase response is given as
system function
A 0 (s - z0)(s - zt) (8.5) < j ( co ) = 20 angles of the vectors from the zeros to the jco axis
H(s) = (s ~ Po )( s ~ Pi )( s ~ P2 ) t =
H( jco ) can be written as
A0( jco - z0 )( jco zj)
H ( jco ) =
( jco p0 )( jco -
pj)( jco - p2 )
(8.6)
-
— y'
—2 3= 0
angles of the vectors from the poles to the jco axis

It is important to note that these relationships for amplitude and phase

Each one of the factors jco - zt or jco - p } corresponds to a vector are point-by-point relationships only. In other words, we must draw
,
from the zero zt or pole p directed to any point jco on the imaginary axis. vectors from the poles and zeros to every point on the jco axis for which we
Therefore, if we express the factors in polar form, wish to determine amplitude and phase. Consider the following example.
jco Zi = N e*',- jco - p = M ei ' , 6
, , (8.7)
4s 4s
then H{ jco ) can be given as F(s) = s2 (8.9)
+ 2s +2 (s +1+ jl )( s + 1 - jl )
H ( jco ) =
AQNQ
M0M1M2
g3( v o+ vi-so-Si-eO
^ (8.8)
Let us find the amplitude and phase for F( j 2 ). From the poles and zeros
of F( s ), we draw vectors to the point co = 2, as shown in Fig. 8.5. From
as shown in Fig. 8.4, where we note that is negative.

216 Network analysis and synthesis Amplitude, phase , and delay 217
jw of the amplitude and phase response. At oo = 0, we see that the vector
magnitude from the zero at the origin to co = 0, is of course, zero.
>2
Consequently, M ( j0 ) = 0. From Eq. 8.9 for F( s ) , F( jO) is
A/2,
i
lim F( jco ) =
400) (8.10)
( 45'
jl
<o -> 0
cu > 0
(1 + )1 ) ( 1 — jl )

From this equation , we see that the zero at the origin still contributes a
/10
A
90° phase shift even though the vector magnitude is zero. From Fig.

-1 V
8.6a we see that the net phase at co = 0 is </> (0) = 90° 45° + 45° = 90°.
Next, at a very high frequency coh , where o>h » 1, all the vectors are
—
approximately equal to cohem° , as seen in Fig. 8.66. Then

, 71.8°

- jl coh coh
and j( coh ) ~ 90 - 90
° ° - °
90 = -90°
Extending this analysis for the frequencies listed in Table 8.1, we obtain
values for amplitude and phase as given in the table. From this table we
FIG . 8.5 . Evaluation of amplitude and phase from pole zero diagram - . can sketch the Amplitude and phase curves shown in Fig. 8.7.
the pole-zero diagram , it is clear that Next let us examine the effect of poles and zeros on the jco axis upon
frequency response. Consider the function
M ( j 2) =4
V
2
2 X VlO,
= 1.71 F( s ) =
s2
s2
+ 1.03
+ 1.23
—
(s + jl .015)(s jl .015)
( s + ;1.109)(s -;1.109)
( 8.11)

and j>( j 2 ) = 90° - 45° - 71.8° = -26.8° 2.5

With the values M ( j 2 ) and j ( j 2 ) and the amplitude and phase at three or
four additional points, we have enough information for a rough estimate
/
70

jl jl
/ 45'

90" .
< 90'
1.0 <T

45'
— jl
— jo)

( a) (b) Frequency, u
FIG . 8.6 . Determining amplitude and phase at zero and very high frequencies. FIG. 8.7. Amplitude and phase response for F( s ) in Eq 8.9. .
.

i
 218 Network analysis and synthesis Amplitude, phase, and delay 219
TABLE 8.1
Frequency, co Amplitude Phase, degrees 10

1.0 1.8 25.8 -D

1.5 2.0 -5.3 £
3.0 1.3 -50.0 i
a. 5
I
5.0 0.8 -66.0
10.0 0.4 -78.5
0
$
s
whose pole-zero diagram is shown in Fig. 8.8. At co = 1.015, the vector 8
180
from zero to that frequency is of zero magnitude. Therefore at a zero
on the jco axis, the amplitude response is zero. At co = 1.109 the vector £
from the pole to that frequency is of zero magnitude. The amplitude o
response is therefore infinite at a pole as seen from Eq . 8.11. Next,
1.015 1.109
consider the phase response. When FIG. 8.9. Amplitude and phase for F( s ) in Fig. 8.8 .
jco co < 1.015, it is apparent from the pole -
: j1.109 zero plot that the phase is zero. When
co > 1.015 and co < 1.109, the vector in the right-half plane. Consider the pole-zero diagrams in Figs. 8.12a
' 71.015 and 8.12b. Both, pole-zero configurations have the same poles ; the only
from the zero at « = 1.015 is now
_
pointing upward, while the vectors from
the other poles and zeros are oriented
difference is that the zeros in ( a ) are in the left-half plane at s = 1 ± j1,
while the zeros in (b) are the mirror images of the zeros in ( a), and are
—
->1.015 in the same direction as for co < 1.015. located at s = + 1 ± /1. Observe that the amplitude responses of the
We see that at a zero on the jco axis, the two configurations are the same because the lengths of the vectors corre-
-71.109 spond for both situations. We see that the absolute magnitude of the
phase response has a step discontinuity
FIG. 8.8 of + 180° for increasing frequency.
Similarly, at a pole on the jco axis, the
phase response is discontinuous by 180°. These observations are —
illustrated by the amplitude and phase plot for F( s ) in Eq. 8.11 for the -o
o
•

1.
frequency range 0.9 co 1.3, shown in Fig. 8.9. f
D

With a simple extension of these ideas, we see that if we have a zero at <
I
—
z = a ± jco{ , where a is very small as compared to cop then we will have
a dip in the amplitude characteristic and a rapid change of phase near 0 M 0
"7 u
co = coi as seen in Fig. 8.10. Similarly, if there is a pole at p = a ±
}

with a very small, then the amplitude will be peaked and the phase will
— i
%
decrease rapidly near co = cop as seen in Fig. 8.11. A contrasting situation £ </)

occurs when we have poles and zeros far away from the jco axis, i.e., cr is I
large when compared to the frequency range of interest. Then we see that o W; O)
0 co
COj
these poles and zeros contribute little to the shaping of the amplitude and
phase response curves. Their only effect is to scale up or down the overall
amplitude response . . .
FIG 8.10 Effect of zero very near . .
FIG 8.11 Effect of pole very near
From stability considerations we know that there must be no poles in
.
the right half of the s plane However, transfer functions may have zeros
the jco axis. the jco axis .

220 Network analysis and synthesis Amplitude , phase, and delay 221
jco jco
11 left-half plane. Consequently, the vector jco
72 drawn from a pole to any point co ± on
X 72 /2
the jco axis is identical in magnitude with K=1
ji j1 o the vector drawn from its mirror image
° to cov It is apparent that the amplitude
X 71 o
a a
-2 - 1 - 2 1 response must be constant for all frequen-
cies. The phase response, however, is 2 1 1 2 a - -
o 71- -71 o
anything but constant, as seen from the * 7 l o -
x -72 -72 amplitude and phase response curves
given in Fig . 8.15 for the pole-zero config- FIG. 8.14. All-pass function.
( a) ( b)
uration in Fig. 8.14.
A system function whose poles are only in the left-half plane and whose
.
FIG 8.12. (a) Minimum phase function , ( b ) Nonminimum phase function.
zeros are mirror images of the poles about the jco axis is called an all-pass
i function . The networks which have all -pass response characteristics are
phase of (b) is greater than the phase of ( a ) for all frequencies. This is i often used to correct for phase distortion in a transmission system . I;
because the zeros in the right-half plane contribute more phase shift
(on an absolute magnitude basis) than their counterparts in the left-half 8.2 BODE PLOTS
plane. From this reasoning, we have the following definitions. A system
function with zeros in the left-half plane, or on the jco axis only , is called a In this section we will turn our attention to semilogarithmic plots of
minimum phase function . If the function has one or more zeros in the amplitude and phase versus frequency . These plots are commonly known
right-half plane, it is a nonminimum phase function . In Fig. 8.13, we see as Bode plots. Consider the system function
the phase responses of the minimum and nonminimum phase functions
in Figs. 8.12a and 8.12 . * (8.12)
Let us next consider the pole-zero diagram in Fig. 8.14. Observe that D( s )
the zeros in the right-half plane are mirror images of the poles in the

S
I
i
1.0
—
I 1
2 4 6 8 10
co
+ 180

0
L 8
t -180
£

-3600 4 6 8 10 12
co
. .
FIG 8.13 Comparison of minimum and nonminimum phase functions. . .
FIG 8.15 Amplitude and phase of all-pass function in Fig. 8.14.
 222 Network analysis and synthesis Amplitude, phase, and delay 223
The constant K 2 is either negative if |/sT| < 1, or positive if | | > 1. The
1*1 >1 r
phase response is either zero or 180° depending on whether K is positive or
*
*2
1 negative. The Bode plots showing the magnitude and phase of a constant
£ are given in Fig. 8.16.
J? log
(ft) The factor s. The loss (gain) in decibels associated with a pole
j (zero) at the origin is ± 20 log co. Thus the plot of magnitude in decibels
I *I < I versus frequency in semilog coordinates is a straight line with slope of
Magnitude
M ± 20 db/decade or ± 6 db/octave. From the Bode plots in Fig. 8.17, we
see that the zero loss point (in decibels) is at w = 1, and the phase is
constant for all co .
( c ) The factor s ± a. For convenience, let us set a = 1. Then the
0
K positive magnitude is
1 ± 20 log | jot ± 1| = ± 20 log (co2 ± l )1 (8.16) ^
1-f Phase as shown in Fig. 8.18a. The phase is
a-& _
Arg ( jco ± l )*1 = ± tan 1 co (8.17)
- ir
K negative
logu as shown in Fig. 8.18ft .
( b) A straight-line approximation of the actual magnitude versus frequency
curve can be obtained from examining the asymptotic behavior of the
.
FIG 8.16 . Magnitude and phase of constant. ;
factor jco + 1. For co « 1, the low-frequency asymptote is
We know that the amplitude response is 20 log \ jco ± 1| |m <u ^ 20 log 1 = 0 db (8.18)
_ \ N ( ja )\ For co » 1, the high-frequency asymptote is
M ( co ) = \ H ( jco )\ ( 8.13)
I -DO)| 20 log|jco + 1| 20 log co db (8.19)
If we express the amplitude in terms of decibels, we have t
+ 20
20 log M ( co ) = 20 log \ N( jco )\ - 20 log \ D( jco )\ (8.14)
In factored form both N( s ) and D( s ) are made up of four kinds of factors :
rf +
1
°
(a) a constant, K
I1 - ° 10
E
(ft) a root at the origin , s < - 20
(c) a simple real root, s ± a
(d) a complex pair of roots, s2 ± 2as ± a2 ± /S2
' + i
ft
To understand the nature of log-amplitude plots, we need only examine the c

amplitude response of the four kinds of factors just cited. If these factors IE
are in the numerator, their magnitudes in decibels carry positive signs. \ '

2- f .%
"
If these factors belong to the denominator, their magnitudes in decibels I I
carry negative signs. Let us begin with case (a).
(a) The factor K. For the constant K , the db loss (or gain) is 0.1 0.2 0.3 0.5 0.7 1.0 2.0 3.0 5.0 7.0 10.0
Frequency, co
20 log K = K2 (8.15) . .
FIG 8.17 Magnitude and phase of pole or zero at s = 0.
I
(

224 Network analysis and synthesis Amplitude , phase , and delay 225
TABLE 8.2
Actual Straight-Line
Magnitude, Approximation, Error,
Frequency db db db

w =J 2 octaves below ± 0.3 0 ± 0.3

co =\ octave below ±1 0 ±1
co =1 break frequency ±3 0 ±3
co =2 octave above ±7 ±6 ±1
co =4 2 octaves above ± 12.3 ± 12 ± 0.3

which , as we saw in (ft), has a slope of 20 db/ decade or 6 db/ octave. The
-
low and high-frequency asymptotes meet at co = 1 , which we designate
as the break frequency or cutoff frequency. The straight-line approximation
is shown by the dashed lines in Fig. 8.18a. Table 8.2 shows the comparison
between the actual magnitude versus the straight-line approximation. We
see that the maximum error is at the break frequency co = 1, or in un-
normalized form : co = a .
For quick estimates of magnitude response, the straight-line approxi-
mation is an invaluable visual aid. An important example of the use of
these straight-line approximations is in the design of linear control
systems.
( d) Complex conjugate roots. For complex conjugate roots, it is con-
venient to adopt standard symbols so that we can use the universal curves
that result therefrom . We describe the conjugate pole (zero) pair in terms
of a magnitude co0 and an angle 6 measured from the negative real axis,
as shown in Fig. 8.19. Explicitly, the parameters that describe the pole
(zero) positions are co0 , which we call the undampedfrequency of oscillation ,
and £ = cos 6 , known as the damping
jco
factor . If the pole (zero) pair is
given in terms of its real and imaginary
parts,
/± 2 = - a ±;/9 (8.20) juo V1 - f 2

a and /9 are related to £ and co0 by the

following : ~{ O 0 <0
a = co0 cos 6 = co0 £ (8.21) -jco W5
/9 = co0 sin 6 = co0 V 1 £
2

Returning to the definition of the

— .
FIG . 8.19 Pole location in terms of
0

. .
FIG 8.18 Magnitude and phase of simple real pole or zero. damping factor, £ = cos 6 , we see that £ and co„.
226 Network analysis and synthesis
the closer the angle 0 is to w/ 2, the smaller is the damping factor. When
the angle 0 is nearly zero degrees, the damping factor is nearly unity.
To examine the Bode plots for the conjugate pole (zerd) pair, let us set
co0 = 1 for convenience. The magnitude is then
± 20 log |1 - w 2 + J 2 fa\ = ± 20 log [(1 - (8.22)
2 £w (8.23)
and the phase is <f>( co ) = tan
'

1 —o (
2

If we examine the low- and high-frequency asymptotes of the magnitude ,

we see that the low-frequency asymptote is 0 decibels ; the high-frequency
»
asymptote (for w 1) is ± 40 log co , which is a straight line of 40
db/decade or 12 db/octave slope. The damping factor £ plays a significant
part in the closeness of the straight-line approximation , however. In
Fig. 8.20 the asymptotic approximation for a pair of conjugate poles
(co0 = 1) is indicated by the dashed line. Curves showing the magnitude
for £ = 0.1, £ = 0.6, and £ = 1.0 are given by the solid lines. We see
that only for £ 0.6 is the straight-line approximation a close one.
Universal curves for magnitude and phase are plotted in Figs. 8.21 and
8.22 for the frequency normalized function
1 (8.24)
G( s ) =
(*K)* + 2 £(s/ a>0) + 1
We see that the phase response, as viewed from a semilog scale, is an odd
— —
function about co/ co0 = 1. The phase at m = tu0 is 90° or 7T/ 2 radians.

*
'

. -
FIG. 8.20 Magnitude versus frequency for second order pole.

.iii ..

228 Network analysis and synthesis Amplitude , phase, and delay 229
For a conjugate pair of zeros, we need only reverse the signs on the scales
of the magnitude and phase curves.
Example 8.1. Using Bode plot asymptotes, let us construct the magnitude
versus frequency curve for
0.1s
G( s ) = (8.25)

M( s2
16 x 104
s
+ 103 +1

We see there are two first -order break frequencies at co = 0 and co = 50. In
)
addition, there is a second-order break frequency at co = 400. With a quick
-
calculation we find that £ = 0.2 for the second order factor. The asymptotes
are shown in Fig. 8.23. The magnitude and phase plots are given in Fig. 8.24
through a microfilm plot computer program .
i
i + 6 db/octave
?
•
g

3
2
<

|
1 i 50 400 log co —-»

3 s
I3 - 12 db/ octave
•a
2

£ -6 db/ octave
rj
CO

i . .
FIG 8.23 Asymptotes for G( s ) in Eq 8.25. .
/

i
t. 8.3 SINGLE-TUNED CIRCUITS

We will now study a class of circuits whose system functions can be

!l described by a pair of conjugate poles. These circuits are called single -
tuned circuits because they only need two reactive elements an inductor
and a capacitor. The undamped frequency of oscillation of the circuit is
—
.
then co0 = ( LC )~ l/i An example of a single-tuned circuit is the R L C --
-
circuit in Fig. 8.25, whose voltage ratio transfer function is
r > 1/sC 1/LC
saajSap '(S)£)3JV
«(*) =!K%
,(s) R + sL + 1/sC 8
s + (.R /L)s + 1 / LC
(8.26)

&
 230 Network analysis and synthesis
Amplitude, phase, and delay 231 i.
1

7
“
M2
K = a2 + f } 2

'Mi
-a
-JP

.
FIG 8.26
:
The poles of H ( s ) are

Pi ,* = 2L 2\LC I2 /
-a ± jP (8.27)

where we assume that ( R2/ L2 ) < (4/LC). In terms of a and P in Eq. 8.27,
H( s ) is
+ P2 «2
H( s) =
(s + a + jp)( s + a
From the pole-zero diagram of H( s ) shown in Fig 8.26, we will determine
—. jP )
(8.28)

.
the amplitude response \ H( jco )\ Let us denote the vectors from the poles
.
FIG. 8.24a Magnitude of G( s ) in Eq . 8.25 . to the jco axis as |MJ and |M2| as seen in Fig. 8.26. We can then write
FIG. 8.24b . Phase of G( s ) in Eq. 8.25.
K
\ H (J o )\ =
(
, (8.29)
|M ||M2|
where K =a +P2 2 and
A sL
-o ,
IMJ = [«2 + (ft> + mA
|M2| = [«2 + (a, - pyV' i
+ (8.30)
Vi ( s )
In characterizing the amplitude response, the point a> = eoniax, at which
j
IH( jco )\ is maximum, is highly significant from both the analysis and design
.
FIG 8.25. Single tuned circuit. aspects. Since \ H( J <o )\ is always positive, the point at which \ H( jco )\ 2 is
- maximum corresponds exactly to the point at which \ H( jco )\ is maximum.
Since \ H( jio )\ 2 can be written as

M
232 Network analysis and synthesis Amplitude, phase , and delay 233
8 2 2

= [a2 + (co + p f ] [ aP2 + (co - 0 )2 ]

2 (a + ) 7
“
\H( jco )\
(8.31)
(a2 + p2 ) 2
JP I
co + 2co (a - p2 ) + (a2 + p2)
4 2 2 2
“max
we can find comax by taking the derivative of |H( jco )\ 2 with respect to co 2 and P/
setting the result equal to zero. Thus we have

d \ HQ<o )\2 + P2 )2 [ 2co2 + 2(a2 - p2 )]

(a2
(8.32) —
H a ->
dco2 [co4 + 2co2(a2 - p2 ) + («8 + P 2 f f
d \ H( jco )\2 _ (8.33)
From the equation
dco2
-7 /3 x -
-“ max
we determine COmax = P 2 - OL2 (8.34)
FIG. 8.27. Peaking circle.
Expressed in terms of the natural frequency of oscillation co0 and the
p, comax in Fig. 8.27.
^
damping factor £, co ax is
V
(C00 1 - C f = 02(l - 2 ) )2 2
with sides a,
that
By the Pythagorean theorem, we find
COmax = ^ ~
^ ( (8.35)
0

Since comax must always be real, the condition for comax to exist, i.e., the
CO
COmax = P 2 ~ OC 2 (8.39)
The circle described in Fig. 8.27 is called the peaking circle. When a = p,
condition for \ H( jco )\ to possess a maximum, is given by the equation
the peaking circle intersects the jco axis at co = 0, as seen in Fig. 8.28a.
(8.36) When « > p , the circle does not intersect the jco axis at all (Fig. 8.286) ;

so that
^
1 < 0.707. Since £ = cos 6 , comax
2 2 l
does not exist for 6 < 45°.
therefore, comax cannot exist. When the imaginary part of the pole is
much greater than the real part, i.e., when p » a, then the circle intersects
When 6 = 45°, we have the limiting case for which ci>max exists. In this the jco axis at approximately co = co0, the natural frequency of oscillation
case, £ = 0.707 and the real and imaginary parts of the poles have the same of the circuit (Fig. 8.28c).
magnitude, i.e., a = /?, or A figure of merit often used in describing the “peaking” of a tuned
£cu0 = cojl £2 -(8.37) circuit is the circuit Q , which is defined in pole-zero notation as
We see from Eq. 8.34 that, when a = p, then ow = 0. This is the lowest
frequency at which comax may be located. For £ > 0.707, or 2£
—
2 cos 6
(8.40)

£«0 > "oV1 - £2 (8.38) From this definition, we see that poles near the jco axis ( £ small) represent
high- 2 systems, as given in Fig. 8.28c, and poles far removed from the jco
Wmax is imaginary ; it therefore does not exist. To summarize, the key axis represent low- Q circuits (Fig. 8.28a). Although the Q of the circuit
point in this analysis is that the imaginary part of the pole must be greater given by the pole-zero plot of Fig. 8.286 is theoretically defined, it has no
or equal to the real part of the pole in order for wmax to exist. Interpreted
graphically, if we draw a circle in the s plane with the center at a and the
radius equal to p , then the circle must intersect the jco axis in order for comax
— practical significance because the circuit does not possess a maximum
point in its amplitude.
By means of the peaking circle, we can also determine the half-power
to exist, as seen in Fig. 8.27. Moreover, the point at which the circle inter - point, which is the frequency coc at which the amplitude response is
sects the positive jco axis is comax. This is readily seen from the triangle
. .
\ H( jcoc )\ = 0.707 \ H( jcoms x )\
. i
 Amplitude , phase , and delay 235
234 Network analysis and synthesis
jo> For a given pole pair the parameters 0, a, and Kare prespecified.
jw ju Therefore, we have derived \ H jco )\ in terms of a single variable parameter,
(
the angle ip. When the angle ip = TT/2 rad, then sin ip = 1, co, = comax , and,
r jp jp f “ max
|f 7(jCOinax)| — 20 a
(8.46)

-a c*> max
— a When ip = 7T/ 4 rad, then sin ip = 0.707 and
\ H( jco,)\ = 0.707|H(;comax)|
- - jpx -« max so that Wj = coc. Let us consider now a geometric construction to obtain
a>c Let us first draw the peaking circle as shown in Fig. 8.30. We will
.
( a)
denote by A the point at which the peaking circle intersects the positive
real axis. 'Now we draw a second circle with its center at A , and its radius
. .
FIG 8.28 Examples of peaking circles. (a) a = jS, comsx = 0. (b) a > ft , cumax equal to AB, the distance from A to either one of the poles, as seen in
undetermined , (c) ft » o, comsx ft. Fig. 8.30 The point where this second circle intersects the jco axis is u>c.
.
The reason is that, at this point, the inscribed angle is ipx = TT/ 4 because it
We will now describe a method to obtain coc by geometrical construction. is equal to one-half the intercepted arc, which, by construction , is TT/2.
Consider the triangle in Fig. 8.29, whose vertices are the poles
and a point ooi on the j<o axis. The area of the triangle is
—
When (Umax 0, the half-power point coc is also called the half-power
bandwidth of the tuned circuit. In Fig. 8.31a the half-power point is given
when cumax = 0. For a high- 2 circuit, where comax ^ w0 > the amplitude is
Area (Ap
^/ co,) = 0a
In terms of the vectors |MX| and |M2|from the poles to (oit the area can also
(8.41)
_
highly peaked at to = comax, as shown in Fig. 8.316. In this case, if
\ H( jO )\ < 0.707 |//( /ft>max) ] , there are two half power points <oCi and «Cj -
be expressed as / jw
|MX |M2
| | sin ip
Area ( Ap1p1* w,-) = (8.42)
7c
%“
2
fix;
where ip is the angle at as seen in the figure. From Eqs. 8.41 and 8.42, Peaking
" circle
we see that the product |Mi||M 2| is equal /
. AB
ju to
Pi 20a
IMJ |M2| = sin (8.43)
'

Mi
jp
ip -a Otr a

20 Ui Since the amplitude response is

— a <T
|MJ |M2|
K (8.44)
~ JU max

where K is a constant, then ->wc

P\*
-ip -

K sin ip
= 20a (8.45)
.
FIG 8.29
|H(M) I
.
FIG 8.30 . Geometric construction to obtain half-power point.

1
236 Network analysis and synthesis 1 Amplitude , phase , and delay 237
ju

1 _ 0i
H

0- 707 \ Hmax\
Pix
00 “
0 PiX 1
—— uo
Ui
Uo

1.0
0.707

I Hoi 1
0 0 — a 0

0
"c u 0
“Clmax “
“( b )
C2 (O
—-
>

. .
FIG 8.31 ( a) Low
(a)
-6 circuit response , ( b ) High-fi circuit response. pi*
(a)
x0o
P1* X -

( b)
,02
k
about the point <wmax, as seen in Fig. 8.316. By the construction process (c )
just described , we obtain the upper half-power point coCj. It can be shown1 FIG. 8.32. Several steps in obtaining phase response for high- Q circuit , (a) w , = 0.
that the point rnmax is the geometric mean of coc and coCi , that is, ~
(6) co, COQ. (c) Ut = a.
^
= tCmsix (8.47) Finally, as an example to illustrate our discussion of single-tuned
circuits, let us find the amplitude response for the system function
As a result, the lower half-power point is
34
H( s ) = (8.50)
s2
wCi — WC
Pmax
2
(8.48) + s + 34
6
Now we determine the maximum and half-power points comax and coc,
The bandwidth of the system for a high - g circuit of this type is described by and also the amplitudes |tf (/«max)| and \ H( jcoc )\. In factored form, H( s) is
7 (8.49) ; 34
BW =
In design applications these high- g circuits are used as narrow bandpass
!
H( s) =
(s + 3 + jS )
(8.51)
j5 )( s + 3
and the poles of H( s) are shown in Fig. 8.34. We next draw the peaking
—
filters.
Finally, there are certain aspects of the phase response of high- g circuits
/ 1

—
circle with the center at s = 3 and the radius equal to 5. At the point
where the circle intersects the ja> axis, we see that comax = 4. To check this
that are readily apparent. In Fig/8.32 we see several steps in the process
of obtaining the phase response; The phase shift at co = 0 is 0, as seen
from Fig. 8.32a. At cu = oo, the phase shift is n rad, as shown in part ( c )
—
result, the equation w a!C = 02 a 2 gives
^ COmax
—
= (52 - 32)* = 4 (8.52)
of the figure. Finally, in the neighborhood of co = eo0 comax, the phase ~ 0 coo CJ

—
shift resulting from the pole in the lower half plane is approximately
—-
02 = 7J /2 (Fig. 8.326). The change in phase in this region is controlled
in large by the pole px . It is readily seen that the phase response in the
region of cu0 has the greatest negative slope, as seen from a typical phase
» - 1
£ - 90°
-
response of a high g circuit shown in Fig. 8.33. ;
-180’
1
See for example, F. E. Terman, Electronic and Radio Engineering , McGraw Hill
. - . .
Book Company, New York, 1953 <; FIG 8.33 Phase response of high Q circuit. -
 238 Network analysis and synthesis Amplitude, phase, and delay 239
ju The amplitude \ H { j(omd.x )\ is then the special case when the R , L , and C elements in the primary circuit are
C equal in value to their counterparts in the secondary. Since the primary
6.78 = wc 34
mm = and secondary inductances are equal, the mutual inductance is
By
4 = <vmax
(3 +
—
—
j9)(3
"
jl ) I

= 30 = 1.133 (8.53)
M = KL (8.58)
In this analysis we assume the coefficient of coupling K to be a variable
5 The point A at which the peaking circle parameter. Let us determine the amplitude response for the voltage-ratio
intersects the positive real axis is located transfer function F2(s,)/ K1(s). From the mesh equations
LA
-3 0 12 X ) at s' = 2.0. With the center at A , we
draw a circle of radius AB (equal to K( S ) = (R + SL + L(S) - sM J 2( s )
V
5 2 in this case). At the point C where ( 8.59)
this new circle intersects the jco axis, we 0 = -sM It( s ) +
have coc. By measurement, we find
we readily determine
coc 6.78 (8.54) F2(S) RM/ L2 S
3
H( s ) = (8.60)
.
FIG. 8.34 Peaking circle construc- Let us check this result. Referring to [ s + ( R / L)s + 1/LC]2 - s“ K 2
Fj(s) 2

tion example. Fig. 8.34, we know that the line segment

Using tuned -circuit notation , we set
AB is of length 5%/ 2 ; it follows that AC
is also 5\/ 2 units long. The line segment AO is of length 2£co0 =£
AO = 5 - 3 = 2 units (8.55) (8.61)
Then coc is given as
coc - V( AC )2 - ( AO )2 = V46 = 6.782 (8.56)
<u0 = LC— 1

H ( s ) can then be written as

Finally, we obtain \ H( jcoc )\ as
s3RM / L1
34 H( s ) =
1^06.782) 1 = V(34 -
46) 2
+ (6 V46) 2 = 0.802 (8.57)
(
(1 - X 2) s2 +1
+^ K
2 u0
s +1 )(
(Up

+K '
2
+ 1-
'
2CoJo
K
s+
o 02
1-i
)
)
which is precisely 0.707 \ H ( jcomakX )\ . (8.62)
8.4 -
DOUBLE TUNED CIRCUITS
/ .
If we set A=
RM _ 2ga >0 JC
(8.63)
L (l - K 2 )
2
1 - K2
In Section 8.3, we studied the frequency response for a pair of conjugate then we can write
poles. Now we turn our attention to the amplitude response of two pairs As3
of conjugate poles in a high- Q situation. The circuit we analyze here is the H( s ) = (8.64)
(s - Sj)(s - s1*)(s - s2)(s - s2*)
-
double-tuned or stagger tuned circuit given in Fig. 8.35. We will consider
where {Si , Si*}
1+ K
±
+ K (1 + Kf 3“ (8.65)
{s2, s2*} = -1- 1
— K
11
(1 - Kf 3"
Let us restrict our analysis to a high- Q circuit so that £2 « 1. Furthermore,
. j let us assume that the circuit is loosely coupled so that X « 1. Under
FIG. 8.35 Double tuned circuit.
-

240 Network analysis and synthesis Amplitude, phase, and delay 241
these assumptions, we can approximate the pole locations by discarding clarity purposes. Note that we have a triple zero at the origin. In terms
the terms involving I 2 under the radicals in Eq. 8.65. Then the poles of the vectors in Fig. 8.36, the amplitude response is
(si, Si *} can be given approximately as
A IMpl 3
\ H( jco )\ = |M,||M | |M | |M | (8.68)
{si, sx*} ~ -£ > ft 0 ± -j ) (8.66) ,
Since the circuit is high- g in the vicinity of co
2 3 4

Similarly, { j2, ^2*} can be given as

= w0, we have
|M3| ~ |M | 4 2 |M0| ~ 2co 0

{s2, s2*} -£&> o ± +Y ) (8.67) so that in the neighborhood of co0

-
The pole zero diagram of H( s ) is given in Fig. 8.36. The real part of the Aa>0
( 8.70)
poles — is greatly enlarged in comparison to the imaginary parts for 4 \ M ,\ |M 2|
It is evident that the amplitude response
of \ H( jco )\ in the neighborhood of «0
depends only upon the pair of vectors
IMJI and |M2|. The double-tuned prob-
lem has thus been reduced to a single-
tuned problem in the neighborhood of
w 0. Consequently, we can use all the
results on the peaking circle that were
derived in Section 8.3. Let us draw a
peaking circle with the center at
Mo s = - t,(oo + /« „ (8.71)
. .
FI G 8.37 Peaking circle for double-
M3 and with a radius equal to co0 Kj 2 , as tuned circuit .
shown in Fig. ' 8.37. The inscribed
^-
3
- <r
-fw0
II I
angle y> then determines the location of the maxima and half- power points
of the response. Without going into the derivation , the amplitude response
can be expressed as a function of y> according to the equation
_
I H( jco )\ ~ 4co
Aco0 sin y> __ sin ip
(8.72)
M<
0 Kaco0 ) 2(1 - K 2 )
Referring to the peaking circle in Fig. 8.37, let us consider the following
situations :
1. w 0 £ > OJ 0 K / 2 : In this case, the peaking circle never intersects the jco
axis ; rp is always less than TT/2 (Fig. 8.38a), and the amplitude response
never attains the theoretical maximum
H max
1
—
2(1 - K 2 )
(8.73)

as seen by the curve labeled ( a ) in Fig. 8.39. In this case, K < 2 £, and the
.
FIG. 8.36 Poles and zeros of a double- tuned circuit. circuit is said to be undercoupled.

.
 242 Network analysis and synthesis Amplitude, phase, and delay 243
jm ju> jm
H max

w 2 max Hmax
7T \ «
£
ft
V2

C >0 COO I cop I

*
-fcoo
I 0 co
X-
r« i “ Cj
“0 0> c
2

FIG . 8.40. Haif -power points of overcoupled circuit.

(a) (W
maximum at mt and a> 2. In the case of overcoupling and critical coupling,
.
FIG . 8.38. (a) Undercoupling (b) Critical coupling, (c) Overcoupling.
we can determine the half-power points by using the geometrical con -
struction method given in Section 8.3. Observe that there are two
2. co0 l = cogK/ 2 : Here the peaking circle intersects the jm axis at a
half-power points mCi and coCj, as shown in the overcoupled curve in
single point co = m0 (Fig. 8.38b ). At m0 , the amplitude is equal to Hmax
Fig. 8.40. The bandwidth of the circuit is then
in Eq. 8.73. In this case £ = K / 2 , and we have critical coupling .
3. G> 0 £ < m0 K/ 2 : The peaking circle intersects thejm axis at two points. BW = coC - toCi (8.75)
2
mt and co2, as seen in Fig. 8.38c. The intersecting points are given by the
equation -
Example 8.2. The voltage ratio transfer function of a double-tuned circuit is
given as
0>1,8 max
— O)0 ± O0 [ ( fl -' T (8.74)
/ i
: H ( s)
As3
= ( s + 2 +;100)(s + 2 -;100)(s + 2 + y106)(r + 2 - 106)
/ '
(8.76)

Consequently, the amplitude response attains the theoretical maximum

//max at two points, as shown by curve (c) in Fig. 8.39. In this situation, From H( s ) , let us determine the following: (a) the maximum points ml max and
co 2 max ; (6) the 3 db bandwidth BW ; (c) the damping factor ? ; ( d ) the coefficient
the circuit is said to be overcoupled.
i of coupling K ; ( e ) the gain constant A ; and ( f ) the maximum of the amplitude
Note that in Fig. 8.39 the undercoupled and critically coupled curves
response //max.
have their maximum points at co0. The overcoupled curve, however, is Solution , ( a ) The natural frequency of oscillation co0 is taken to be approxi -
mately halfway between the two poles, that is, co0 = 103 radians. In the neigh -
Umax
•
\H \ borhood of co0, we draw the poles s = 2 + /100 and s = 2 + y106, as —
shown in Fig. 8.41. From the peaking circle centered at the point s = 2 + jco0 ,
— —
shown in Fig. 8.41, we obtain

[( b )
.M max “ ®0 = V32 - 22 = 2.236 radians (8.77)

so that co 2 max = co0 + 2.236 = 105.236 radians (8.78)

Wj max —
= co0 2.236 = 100.764 radians
'( a ) (6) Next we draw a circle centered at s = 1 + /<*> with radius 3 V 2. Where „
this circle intersects the jm axis, we have mCt so that
0 co
“ l max "0 "2 max
mCi - m0 = V(3 V2) - 1 = 4.123 radians (8.79)
2
.
FIG . 8.39. ( a ) Undercoupled case (b) Critically coupled case, (c) Overcoupled case. .

i* 6,

244 Network analysis and synthesis Amplitude , phase, and delay 245

8.5 ON POLES AND ZEROS AND TIME DELAY

What is time delay ? How do we relate it to frequency response ? We

will attempt to answer these questions in this section. First consider the
= 105.236 transfer function of pure delay
H ( s ) = e~ sT (8.86)
For a system described by Eq . 8.86, any excitation e( t ) produces an
identical response signal e( t —
T ) , which is delayed by time T with respect
to the excitation . This is shown by the Laplace transform relationship,
R( s ) = Z [ e( t - T ) ] = Z [ e( t ) ] (8.87)
“ 1 max = 100.764 Let us examine the amplitude and phase response of the pure delay.
From the equation
H ( jco ) = e~ i<oT (8.88)
we obtain the amplitude response

FIG . 8.41 . Peaking circle for Example 8.2 .

\ H( jco )\ = 1 (8.89)

The 3 db bandwidth is then

and the phase response <fi( w) = — coT (8.90)
We see that the delay T is equal to minus the derivative of the phase
BW
— 2( mCz — eo0) = 8.246 radians (8.80)
response, that is,
d <f>( w)
(c) The damping factor £ is obtained from the real part of the
poles £ mn = 2’ T=- (8.91)
from which we obtain dm
2 The magnitude, ’ phase and delay characteristics of H ( jm ) = er 3aT are
£ = 103
~ = 0.0194 (8.81) . given in Fig. 8.42a, b, and c.
If we define delay as in Eq . 8.91, we can readily deduce that for the
( d ) The coefficient of coupling K is obtained from the
radius of the peaking
6 response to be nearly identical to the excitation, the system amplitude
circle, which is
response should be constant, and its phase response should be linear over
0K _ the frequency range of interest. If the phase is not linear, we have what is
^ 2 (8.82)
known as delay distortion. To visualize delay distortion more clearly,
We thus have

( e ) The gain constant A is equal to

K
—
= m = 0.0582
6
0
(8.83)
A( m ) 4>( m ) Afw)

^
__ 2 £ m0 K
~
2(2)(0.0582)
= 0.2328 (8.84)
[0 w
Nj
0 0 m
1 - K2 1 - (0.0582)2
(/) Finally, the maximum amplitude //
max is
(a ) (c )
1 FIG . 8.42. Amplitude, phase, and delay of ideal delay function , ( a ) Amplitude.
//max = 2(1 - K 2) = 0.5009 (8.85) ( b) Phase, (c) Delay .

L
 246 Network analysis and synthesis Amplitude, phase , and delay 247
we recall from Fourier analysis that any signal is made up of different joi
frequency components. An ideal transmission system should delay each -
010 + O 0
frequency component equally. If the frequency components are delayed
by different amounts, the reconstruction of the output signal from its
Fourier components would produce a signal of different shape as the input.
For pulse applications, delay distortion is an essential design considera-
tion. D«lay
Let us next examine how we relate delay, or envelope delay (as it is "o k
bandwidth
sometimes called) to the poles and zeros of a transfer function. For any
transfer function

m=
n (* - zi ) - Co - 1

—
O>0
(8.92)
IT (* - Pi )
1
a -ao o
3= 1

with zeros at z( = — rt- ±

< ja>i and poles at p } = — cr* ± ja s, the phase for
> .
FIG. 8.43 Graphic construction to obtain delay bandwidth.
real frequencies is
<f>( cd ) = i2
=1
tan- . » ± 2!
cr
*
_2
1
tan—!
CO ± ft *
0)
)
(8.93)
I
2. The frequency at which the delay is half the maximum, or l /2c0, is
a> x = «0 ± (8.97)
^
Envelope delay is
,
d <f ( co ) Of '
3. The “effective delay bandwidth” is then co0 a0 < <o < a>0 + a0 or
simply 2cr0. The upper and lower half-bandwidth points can be obtained
—
- 2= 0 2 graphically by drawing a circle with center at co = oo0 and radius cr0.
+ (ft ± ft ,)2 i2 o/ + (ft ± ft *)2
doj
«
)
+ ) ) )
(8.94)
The intersections of the circle with the j(x> axis are the half- bandwidth
-
i i f 1

We see that the shapes of the delay versus frequency characteristic are points, as shown in Fig. 8.43.
the same for all poles and zeros. The zeros contribute “ negative” delay ; 4. The product of the maximum delay and delay bandwidth is always
the poles, positive delay. However, linear physical systems do not have 2. Thus, if we wish to obtain large delay by placing zeros or poles near
transfer functions with zeros alone. The inductor H( s) = Ls is the only the jco axis, the effective delay bandwidth is then very narrow.
exception. Its phase is = TT / 2 ; thus the delay is zero. .
5 The delay of an all-pass function is twice the delay due to the poles
Now let us consider the delay due to one singularity, for example, a alone.
—
pole atp0 = <r0 + jeo0. The delay due to the one pole is The delay versus frequency curve for the pole is shown in Fig. 8.44.
0o
-
We see that the delay bandwidth concept is only useful for a rough
A,0(ft>) = (8.95) approximation, since the delay versus frequency characteristic only falls
0O
2
+ (ft) - ft 0)2 )
as 1/co. To calculate the delay versus frequency characteristic for a
The following points are pertinent : transfer function with a number of poles and zeros, it is convenient to
obtain the delay curves for the individual singularities and then to add the
1. The maximum delay due to this pole is separate delays. It is not as desirable to obtain the total phase response
and then differentiate numerically.
1 Finally, it should be pointed out that envelope delay only has meaning
Am = - (8.96)
.
00 when the phase response goes through the origin If it does not, there is
and occurs at co = ft>0. The delay versus frequency curve is symmetric a frequency-shift component in addition to the delay that is hard to
about co = ft)„. account for analytically.

Network analysis and synthesis Amplitude, phase, and delay 249

248
( b ) When -
= 10 + ylO3, s2 = -10 -ylO3, and y( j0) = lO mhos, find
2 '

A (wJ
the values of R , L, and C and determine the numerical value of s3.

\L
Y(s> ~C
R

wo
— 4<ro wo - 3<7Q WO ~ 2o-o wo - <ro WQ wo + co wo + 2cro WO + 3<TO WQ + 4 <ro w
PROB. 8.2

8.3 Find the amplitude and phase response for the following functions and
sketch.

Problems
. .
FIG 8.44 Plot of delay versus frequency. ® w -r h
8.1 Find the poles and zeros of the impedances of the following networks
(c) F( s ) = S-2 + W02
•
s
( d ) F( s ) = s—
2 —
+ f»o2
and plot on a scaled s plane.
Note that K and to0 are positive quantities.
4h 8.4 Given the function
A{ oi ) + jBjoi )
Z ( s)
— ^ GO) C(co) + jD( <o)
-
(a) determine the amplitude and phase of G(Joi ) in terms of A , B, C, D. Show that
the amplitude function is even and the phase function is odd.
AAA 8.5 By means pf the vector method, sketch the amplitude and phase response
2a
Z( s )
ifd for

( a ) F( s )
j
= s( s + 10)
+ 0.5 (C) F(S) =1 1
_
( b)

AV
2Q — 4h
(6) F( s ) = s2 + 2s + 2
s ( d ) F( S ) =
7^ 2TT 2
s
i

- - 2s +
Z( s ) 5
(e) F(r) = ( g ) F( s ) = 7 + 2)( s + 1)
-

(c)
s2 + 4
^
s* + 2s + 5
i
PROB 8.1 . (/) F( s ) =
(s + 2)(s2 + 9)
(/i) F( s ) =
( s + 2 )( s + 1)
8.2 The circuit shown in the figure is a shunt peaking circuit often used in 8.6 Plot on semilog paper the Bode plots of magnitude and phase for
video amplifiers.
( a ) Show that the admittance F(s) is of the form

Y( s )
_ K{ s - st)(s - s2)
(ft) F( s ) =
100(1 + 0.5J)
s( s + 2)
- ^3) F( s ) =
50(1 + 0.025s)(l + 0.1s)
Cb) (1 + 0.05J)(1 + 0.01s)
Express slt s2, and s3 in terms of R, L, and C.

( '
 252 Network analysis and synthesis
Use the peaking circle to determine the maximum and half-power points and
the circuit Q . Find the gain constant A and the coefficient of coupling K .
8.17 For the double- tuned circuit shown, determine the maximum and half-
power points and the circuit Q . Find the gain constant A and the coefficient of
coupling K .

chapter 9
Network analysis II
M = 5 x 10-6 h
PROB 8.17 .
8.18 Determine the delay at eo = 0, 1 , and 2 for
9.1 NETWORK FUNCTIONS
(«) 1
F( S ) =
7+ 2 In electric network theory, the word port has a special meaning. A
(6) -3 ;
port may be regarded as a pair of terminals in which the current into
F( s ) = -
s +3 one terminal equals the current out of the other. For the one-port network
3s shown in Fig. 9.1 , 1 = 1' . A one-port network is completely specified
(c) F( s ) = when the voltage-current relationship at the terminals of the port is given.

(d )
( s + l )(r + 2)

F( s ) = 2 +
s 1
s + 2s + 5 - —
For example, if V 10 v and 1 2 amp, then we know that the input
or driving point impedance of the one-port is
—
O) F( s ) = ( s + 2)(r2 + 2s 2)
+ +
1) Zjn
—— 1
= 5Q (9.1)

-
Whether the one port is actually a single 5-Q resistor, two 2.5-Q resistors
in series, or two 10-Q resistors in parallel, is of little importance because
the primary concern is the current-voltage relationship at the port.
Consider the example in which I = 2s + 3 and V = 1 ; then the input
I admittance of the one-port is

1
Ym =
which corresponds to a 2-f capacitor in parallel with a
its simplest case (Fig. 9.2).
^ = 2s + 3
- -Q resistor in
j
(9.2)

o - y,n
One-port r±r 2 f in
V network
o-
r
.
FIG 9.1 .
FIG 9.2
: 253
-l

254 Network analysis and synthesis Network analysis II 255

1 h h 2
parameters. Note that zn relates the current and voltage in the 1-1' port
o- - o only ; whereas z22 gives the current-voltage relationship for the 2-2' port.
Vi Two-port v2 Such parameters are called open-circuit driving-point impedances. On
network
o- ~2' the other hand , the parameters z12 and z21 relate the voltage in one port to
the current in the other. These are known as (open-circuit) transfer
FIG. 9.3 impedances.
As an example, let us find the open-circuit parameters for the T circuit
Two-port parameters in Fig. 9.4. We obtain the z parameters by inspection
-
A general two port network, shown in Fig. 9.3, has two pairs of voltage-
current relationships. The variables are Vx , V 2 , /1 /2. Two of these are
dependent variables ; the other two are independent variables. The number
(
zn =
h
— 12=0
= + z»
of possible combinations generated by four variables taken two at a time
is six. Thus there are six possible sets of equations describing a two-port
network. We will discuss the four most useful descriptions here.
*22 =
Vt
h — l /i-= 0
+ Zc
(9.5)
The z parameters
A particular set of equations that describe a two port network are the -
*13 = ^
h l / i =o
= Z6
-
z parameter equations
= Z„
V1 =
1171 + Z1272 Z h / 2 =0
(9.3)
V 2 = hJl + Z2272 Observe that z12 = z 21. When the open-circuit transfer impedances of a
two-port network are equal, the network is reciprocal. It will be shown
In these equations the variables Vx and V are dependent, and Iv I2 are later that most passive time-invariant networks are reciprocal.1

_
2
independent. The individual z parameters are defined by Most two-port networks, whether passive or active, can be characterized
by a set of open-circuited parameters. Usually, the network is sufficiently
= s complicated so that we cannot obtain the z parameters by inspection, as
h i /2=0 h l 7i =0 we did for the T circuit in Fig. 9.4. The question is now, “How do we
(9.4)
iS I obtain the z parameters for any circuit in general ?” The procedure is as
71 1 / =0
*22
72 follows. We write a set of node equations with the voltages at the ports
2 l /!=0
Vx and F2, and other node voltages within the two-port V 3, V4 , . . , Vk .
It is observed that all the z parameters have the dimensions of impedance. as the dependent variables. The independent variables are the currents Ix
Moreover, the individual parameters are specified only when the current in and /2, which we will take to be current sources. We then proceed to
one of the ports is zero. This corresponds to one of ports being open write a set of node equations.
circuited, from which the z parameters also derive the name open circuit - h «1k K
I
h - «11^1 + «12^2 + «13^3 H

T CEm l
—
Iz ^21 ^1 t ^22 ^ 2 + ’ ‘
" "

+ ...
* • • •
+ «2kVk
+ n3kYjc (9.6)
Vi Zb v2
0 = «31
^+ 1
• ••

T f
+ nkkVk
FIG 9.4 . 1 One important exception is the gyrator discussed later in this chapter.

I.

M
 256 Network analysis and synthesis Network analysis II 257
where nu represents the admittance between the ith and y th nodes, that is, '
The determinant for this set of equations is
AY = YAYB + YAYC + YBYC (9.11 )
na = Ga + sCu + —— (9.7)
sLu In terms of A T, the open-circuit parameters for the Pi circuit are
If the circuit is made up of R- L-C elements only, then it is clear that Yo
ni } = njt . As a result, the (/th cofactor of the determinant of the node zu = AY
221 "
Ay
equations, Ay , must be equal to the //th cofactor, Aj ( , that is, Atj = AH . ( 9.12)
This result leads directly to the reciprocity condition z21 = z12 , as we
212 “
Yc Z 22 =
YA + YC
shall see. AT AY
Returning to the set of node equations in Eq . 9.6, let us solve for Vk Now let us perform a delta- wye transformation for the circuits in Figs .
and V2 . We obtain
9.4 and 9.5. In other words, let us find relationships between the im-
An A 21
Vi = Afii + ~
- h mittances of the two circuits so that they both have the same z parameters.
A We readily obtain
(9.8)
Yc
v* = ~ A + ~~ i »
A A
~ z12 — Zj = Ay
In relating this last set of equations to the defining equations for the z
parameters, it is clear that
z22— Zb + Zc — YA + YQ Ay
(9.13)

zn — A
Z
Z12
— ~
A
(9.9)
zn — Za + Z _ YB +
yc
6
Ay

A ,1 YB
Z 21 = -Af zzii ——
.
A
We then find
Ay
; (9.14)
Since for a passive network A21 = A12, it follows that z21 = z12, the YA
network is then reciprocal . Zc = Tt
AY
As an example, let us find the z parameters of the Pi circuit in Fig. 9.5.
The y parameters
First, the node equations are
Suppose we were to write a set of mesh equations for the two port in
Fig. 9.3. Then the voltages V1 and V 2 would become independent sources,
h = ( YA + YJV , - YCV2 and the currents lx and /2 would be just two of the dependent mesh
(9.10)
h = ~ YCV± + ( YB + YC )V 2 currents . Consider the general set of mesh equations
Vx = mnh + ml 2 I2 -I- • • • + mlkIk
'
V 2 = ma /, + m22 I2 + • • • + m2kIk
r 0 = m3 Jk + + m3kIk (9.15)

0 = mkili + • ' ••
+ kkIk
where m{ i represents the sum of the impedances in the /th mesh and mtj is
FIG 9.5 . the common impedance between mesh / and mesh j . We note here again

258 Network analysis and synthesis Network analysis II 259

that for an R- L-C network , mtj = mH for all / and /. Thus reciprocity The short -circuit parameters are then
holds .
Solving the set of mesh equations for lx and /2 , we obtain the following 2 s2 + 4s + 1
equations. Vn = Va = 2(2s + 1)

—^ ^
h=
A
y ,+ A
Vi 2s2 + 2s + 1
(9.20)
(9.16) ya = y12 = -
2(2s + 1)
I2 v2
A \+ A
~~ V
When yn = y22 or zn = z 22 , the network is symmetrical 2
The equations of 9.16 define the short - circuit admittance parameters as Returning to Eq . 9.17, which defines the y parameters, we see that the y
parameters are expressed explicitly as
A = VnVi + VI 2V 2 (9.17)
A = 2/21 VI + y22V2
where yiS = AJA for all / and j .
Let us find the y parameters for the bridged -T circuit given in Fig . 9.6 .
ya =V ^I 1 IK 2=O

The mesh equations for the circuit are 2/12 =A

V3 Iv^o ( 9.21 )

V\ IK2 —0
VS = I1 + (9.18)
2/ 22
Vi IK,=O
0 = s
- ~ Ii + -s Ii + 2\l -s + l /) l 3 The reason that the y parameters are also called short - circuit admittance
In straightforward fashion we obtain parameters is now apparent . In obtaining yn and y21 , the 2-2' port must
be short circuited, and when we find y 22 and yl 2 , the 1-1' port must be
A
_ 2(2s_+_l ) short circuited, as shown in Figs . 9.1a and 9.1b .
s2 As a second example, let us obtain the y parameters of the Pi circuit in
2
Fig. 9.5 . To obtain yn and y 2V we short circuit terminals 2-2'. We then
An — A22 = 2s +1
-f 4s
2 (9.19) have
s
2 s2 + 2s +1 2/n = YA + Yc
AJ 2 A21 = (9.22)
s2 2/ 21 — YC
~

2
vW r:
i 2 .
yn X2 i >22. y12

+
1 I +
yi. 1 V* (a ) (b)
FIG 9.7.
V T 1A
symmetrical network is easily recognized because by interchanging the 1-1'
FIG. 9.6
-
and 2 2' port designations, the network remains unchanged .

V
 260 Network analysis and synthesis Network analysis II 261
We next short-circuit terminals 1-1' to obtain h h
o -
2/ 22 ~ Ys + YQ NIC

The h parameters
VL 2 — ~ Yc
(9.23) ZL V2

A set of parameters that are extremely useful in describing transistor FIG . 9.8. Negative impedance converter with load impedance .
circuits are the h parameters given by the equations
Yi
I2
—
=*
hnf + hl 2V2
211\ H h22V 2
“
(9.24)
Observe that for the Pi circuit, h2l = h12. This is the reciprocity
condition for the h parameters and can be derived from their relationships
to either the z or y parameters.
—
The individual parameters are defined by the relationships Next let us consider the h parameters of an ideal device called the
negative impedance converter ( NIC), which converts a positive load
impedance into a negative impedance at its input port.3 Consider the
* 11 = Vi * =*
12
V NIC with a load impedance ZL shown in Fig. 9.8. Its input impedance is
h \v 2=o 2 /1=0 (9.25)
= - ZL
^
Z< n (9.29)
* 21 = * = V,
22
A \ vt ~o / 1 =0
We see that hu and h2 X are short-circuit type parameters, and h12 and h22
which can be rewritten as
YI _ V2
(9.30)
are open-circuit type parameters. The parameter hn can be interpreted h h
as the input impedance at port 1 with port 2 short circuited . It is easily The following voltage-current relationships hold for the NIC .
seen that hn is merely the reciprocal of yn.
Vi = kV2
*
ii = —
?/ n
The parameter h 22 is an open-circuit admittance parameter and is related
(9.26) h = kh
If we interpret Eq . 9.31 using h parameters, we arrive at the following
(9.31)

to z22 by conditions.
1 11 = 22 = 0
22 = -
* (9.27) * *
—* —
Z 2'i
1 (9.32)
Both the remaining h parameters are transfer functions ; h2l is a short
circuit current ratio, and h12 is an open-circuit voltage ratio. Their
- 12 , k * 21

relationships to the z and y parameters is discussed later in this chapter.

For the Pi circuit in Fig. 9.5, the h parameters are
We see that since h12 —
h21, the NIC is nonreciprocal.
In matrix notation, the h matrix of the NIC is
'
0 k
* 11 =
1 * *
11 12
= A (9.33)
YA + YC 0
Yc * *
21 22
Ik J
*12 = YA + Y0 The NIC is a convenient device in the modeling of active circuits It .
(9.28) is not, however, a device that exists only in the imagination. Practical
Yc
*21 = - YA + YC 3
For a lucid discussion of the properties of the NIC, see L. P Huelsman, Circuits , .
-
Matrices , and Linear Vector Spaces , McGraw Hill Company, New York, 1963,
* 22 — YB + YA + YC YAYQ
Chapter 4.

Network analysis II 263

262 Network analysis and synthesis
Let us find, as an example, the h
-o
realizations of NIC’s have been achieved using transistors. Some of
/1
o-
these are described in an article by Larky.
4 ABCD parameter for the ideal trans - n:l
former in Fig. 9.9, whose defining V2
Vi
The A B C D parameters equations are
Let us take as the dependent variables the voltage and current at the
port 1, and define the following equation.
Vx = nV2
~ V{
\ VA B1VV 2
~
(9.34)
1
/1 = -n ( —
(9.38)
/ 2) FIG. 9.9. Ideal transformer .
If we express Eq. 9.38 in matrix form, we have
This matrix equation defines the A , B, C , D parameters, whose matrix is ~
[n 0 ] P V
"
2
known as the transmission matrix because it relates the voltage and current (9.39)
at the input port to their corresponding quantities at the output. The 0 - -h
"

h
reason the current I2 carries a negative sign is that most transmission nj
engineers like to regard their output current as coming out of the output so that the transmission matrix of the ideal transformer is
port instead of going into the port, as per standard usage.
In explicit form, the ABCD parameters can be expressed as
'
X f] pi 0

A
-* V2 1 / 2 =0
B=

D
_
= h
— —^ I
2 (9.35)
<
C D
"
0

Note, incidentally, that the ideal transformer does not possess an im

1
nJ
(9.40)

-
Y2 1 /2=0 12 lFis=0 pedance or admittance matrix because the self- and mutual inductances
From these relations we see that A represents an open-circuit voltage are infinite.6
transfer function ; B is a short-circuit transfer impedance ; C is an open- For the ideal transformer terminated in a load impedance shown in
circuit transfer admittance; and D is a short - circuit current ratio. Note Fig. 9.10a, the following set of equations apply.
that all four parameters are transfer functions so that the term trans-
one . Let us describe the short-circuit Vx = nV2
mission matrix is a very appropriate (9.41)
transfer functions B and D in terms of y parameters , and the open -circuit V2
transfer functions A and C in terms of z parameters . Using straight - 1
nZi
forward algebraic operations, we obtain
Taking the ratio of Vx to Iu we find the input impedance at port 1 to be
A= ^ B= - 1
Z21

1 _ V 21
yn
(9.36) Z! - —h ~ n2ZL (9.42)

C =—
Z 21
D=
2/21 Thus we see that an ideal transformer is an impedance transformer. If
For the ABCD parameters, the reciprocity condition is expressed by the the load element were an inductor L (Fig. 9.106), at port 1 we would see
equation
.
an equivalent inductor of value n2 L Similarly, a capacitor C at the load
would appear as a capacitor of value C/n at port 1 ( Fig. 9.10c).
2
"
A BT
det
c D. =
AD BC = 1 -
(9.37) As a second example indicating the use of the transmission matrix in
- , 4 For a detailed discussion concerning ideal transformers, see M. E. Van Valkenburg
,
4
-
A. I. Larky, “ Negative Impedance Converters,” Trans IRE on Circuit Theory . Introduction to Modern Network Synthesis, John Wiley and Sons, New York, 1960.
- .
CT 4, No 3 September
( 1957 ), 124-131.
 Network analysis II 265
264 Network analysis and synthesis
h Therefore we have h12 = Yx
_* 12
(9.46)
Yg 171=0 *22
v2 Similarly, since h2l is defined as a short-circuit type parameter, we can
\ZL derive A21 in terms of y parameters as
u 2i
«
——^21
= Vn (9.47)

We can express all the h parameters as functions of the z parameters or y

parameters alone. An easy way to accomplish this task is by finding out
L
what the relationships are between the z and y parameters themselves.
Certainly, by their very nature, the z and y parameters are not simply
reciprocals of each other (as the novice might guess), since one set of
parameters is defined for open-circuit conditions and the other for short -
circuit.
. .
FIG 9.10 Ideal transformer as an impedance transformer . The z and y relationships can be obtained very easily by using matrix
network analysis, consider the ABCD parameters of the Pi circuit in notation. If we define the z matrix as
Fig. 9.5.
A = Ys + Yc
Yc
[Z] =
*11 *12
*21 *22
_ (9.48)

and the y matrix as

B= ±
Yc (9.43) [Y ] =
Vn 2/i 2
(9.49)
c _ YAYB + YBYC + YAYC lj/ 21 2/ 22.
Yc
= s fr
In simplified notation we can write the two sets of equations as
l)
+
Yc [ V ] = [Z } [ I ] (9.50)
If we check for reciprocity from Eq. 9.43, we see that and [7] = [ Y ) [ V ] (9.51)

- BC _ YA + YC YB + Yc - YAYB + YBYC + YAYC ) Replacing [/] in Eq. 9.50 by [ Y ] [ V ] , we obtain

( )( ) (
AD 2
Yc [V ] = [Z ] [ Y ] [V ] (9.52)
(9.44) so that the product [Z][ Y] must yield the unit matrix [ U ] . The matrices
V [ 7] and [Z] must therefore be inverses of each other, that is,
9.2 RELATIONSHIPS BETWEEN TWO PORT PARAMETERS - [Z]-1 = [ Z] and [ 7]-1 = [Z] (9.53)
K '

-
The relationships between two port parameters are quite easily obtained 1
From the relationship, we can find the relations between the individual
because of the simple algebraic nature of the two port equations. For - z and y parameters.
example, we have seen that hn = l / yu and h22 = l /z22. To derive A12 in
-
terms of open circuit parameters, consider the z parameter equations A,
VM = 2An,
when port 1 is open circuited : Ix 0. —
Ki = z12/2 2/12 = 2/21
_
= 5 «
(9.54)

(9.45) A, A,
V 2 = z22/2

!
: .

266 Network analysis and synthesis Network analysis II 267

TABLE 9.1 without load and source impedances. These transfer functions can be
Matrix Conversion Table .
described by means of z or y parameters alone For example, let us derive

[*]
A = *11*22 *12*21
*
[2/1
— w [ T]
-
the expressions for the open circuit voltage ratio V 2/ V1 by using z param-
eters first and y parameters next. Consider the z parameter equations
for the two-port when port 2 is open circuited.
2/22 2/12 A AT
~ V2 = z21/j
*11 *12 Ay A„ >h- C C
[*]
*21 *22
_ 2/21 vn
A,
7*21 1
>

1 D Vi = *uA
If wc take the ratio of V 2 to V1, we obtain
(9.57)

Ay h22 c c
*22 *12 1
h* D _ AT Yi _ hk
\y]
A, A, Vn 2/12
£ - An B B Yi- *n
(9.58)

*21 fii A21 A

* 1 A By letting /2 of the second y parameter equation go to zero, we derive
2/21 2/22
Ay A, An An B B -
the open circuit voltage ratio as
Ay fl2 1
An Ai2
B A ji Yi= _ VLI (9.59)
*22 *22 2/ii 2/n D D Vi 2/ 22
[A ]
*21 l
^ 21
7*21 A22
_
1
D
C In Similar manner we can derive the short circuit current ratio of a
two-port as
-
*22 *22 2/n 2*ii D
Ay 1 _ h = y*i (9.60)

[T ]
fn
*21
1
*21
*22
- 2yJl/
21
_ 2/21
2/n
^
A22
t
A 2I - hi
A 21
A B

and
h
/.2 _
= *
2/n
21
(9.61)
C D h
*22
*21 *21 2/21 2/21 -
h2\ A2I
-
The open- and short circuit transfer functions are not those we usually
deal with in practice, since there are frequently source and load im-
where Az = zuz22 — _
z12z21 ; and
pedances to account for. The second category of two-port transfer
zn J/22 2/22
2/u
= Ay functions are those including source or load impedances. These transfer
Ay functions are functions of the two-port parameters z, A, or y and the
*
12 *
_
= 2f i 2
21
_
= 2 /?i
(9.55)
source and /or load impedance. For example, let us derive the transfer
Ay Ay admittance I2/ V1 of a two-port network that is terminated in a resistor
—
where A „ = yny22 yuyn. Using these identities we can derive the h of R ohms, as given in Fig. 9.11. For this two-port network, the following
or ABCD parameters in terms of either the z or y parameters. Table 9.1
provides a conversion table to facilitate the process. Note that in the
equations apply.

V 2 = I2 R
— —^
It 2/21 1 + 2/22 2 ^ (9.62)
table
Ar = AD BC — (9.56)
- h h

i;
o
9.3 TRANSFER FUNCTIONS USING TWO PORT PARAMETERS - Vi Two-port
network
In this section we will examine how to determine driving-point and
o
transfer functions of a two-port by use of two-port parameters. These
functions fall into two broad categories. The first applies to two-ports FIG. 9.11
 268 Network analysis and synthesis Network analysis II 269
r ~
1 terminated in a load impedance ZL. If we write the mesh equation for
*11 — { *22 M-45 +
the I2 mesh , we have

— — Z2 lA (Z22 + Z2,) /2 (9.64)

Vl /l
+1212 ^2 V2
Since F2 = — ZL,
I2 we readily obtain
_ _
22lA 22lZ£
Z2i (9.65)
A Z22 + ZL
l' Z
It also is clear that the current- ratio transfer function for the terminated
_
!
. . -
FIG 9.12 Two port equivalent .
two-port network is
A _ z2i
(9.66)
A Z22 + Z^,
By eliminating the variable V 2 , we obtain
In similar fashion, we obtain the voltage-ratio transfer function for the
A yJR circuit represented in Fig. 9.13 as
Ai
VI 2/22 + 1JR
(9.63)
F2
=
_ 2/21 (9.67)
Note that Y 2 l and y21 are not the same. Y 21 is the transfer admittance K A + y22
of the two-port network terminated in a resistor R , and y 21 is the transfer
Next, suppose we are required to find the transfer function F2/ Fe for
admittance when port 2 is short circuited. We must be careful to make
the two-port network terminated at both ends, as shown in Fig. 9.14.
this distinction in other cases of a similar nature.
We first write the two mesh equations
In order to solve for transfer functions of two-ports terminated at
either port by an impedance ZL, it is convenient to use the equivalent
circuit of the two-port network given in terms of its z parameters (Fig.
Vg
0
—= + ZJLI) A + zuh
(2? j
z2iA + (z22 + R2 ) I2
(9.68)
9.12) or y parameters ( Fig. 9.13). The equivalent voltage sources z12/2
and z21/ j in Fig. 9.12 are called controlled sources because they depend Next, we solve for I 2 to give
upon a current or voltage somewhere in the network.6 Similarly, the
current sources y12V 2 and y 2 XV1 are controlled sources. For the circuit Vgzil
in Fig. 9.12, let us find the transfer impedance Z21 = V 2lh , with port 2 { Rl- + Zll )( ^ 2 + Z22) — Z12Z2X
(9.69)

r ]2
From the equation V 2
solution.
= — R2 I2 , we may now arrive at the following

_
TE
Vl
721 Vl I 722 Y2
+

V2
Vz
Vg
RJi
Vg ( Ri + 21I)(/? 2 + z22)
^
Z21 2

—z z 21 12
(9.70)

JT
-
r
1
I_ 1
i
? *1
211

/1
+
221/1
222

*2
+

v2
. .
FIG 9.13 Two-port equivalent.
+.
vg ^ 12 /2

* For a lucid treatment of controlled sources, see E. J Angelo, Electronic Circuits, . .

FIG 9.14
-
2nd Ed., McGraw Hill Book Company, New York, 1964.

270 Network analysis and synthesis Network analysis II 271

Vl
+ —- AV\A
zii 212

212
AW
-
222 212 — ^O1 *
-
('221 212) /! . —
— 1
V2
4.
r
m
/in

— +1
h
n

Vl A12 V2 Q (j )/»2l/l 1
v2
Zm
FIG. 9.15 Two port equivalent circuit with one controlled-voltage source.
. -
Note that the equivalent circuits of the two-ports in Figs. 9.12 and
9.13 are not unique. Two other examples are given in Figs. 9.15 and 9.16. . .
FIG 9.17 Hybrid equivalent circuit.
Observe that the controlled sources are nonzero in these equivalent
circuits only if the circuit is nonreciprocal. We can easily check to see that Eq . 9.75 is dimensionally correct since hn
Finally, let us consider the hybrid equivalent circuit shown in Fig. 9.17. has the dimensions of impedance, h22 is an admittance, and h12 , h2l are
Observe the voltage-controlled source h2lV 2 at port 1 and the current- dimensionless since they represent voltage and current ratios, respectively.
controlled source at port 2. Let us find the input impedance Zin.
The pertinent equations are
(9.71)
9.4 INTERCONNECTION OF TWO PORTS -
Vi = hnIx + h12V 2
In this section we will consider various interconnections of two-ports.
and = ZLJ2 =
V2 — — { huh + h22V2 ) ZL (9.72) We will see that when a pair of two-ports are cascaded , the overall trans
mission matrix is equal to the product of the individual transmission
-
Solving Eq. 9.72 for V 2 , we find matrices of the two-ports. When two two-ports are connected in series,

V2 = — h 2iZLh
1 + h22ZL
(9.73)
their z matrices add ; when they are connected in parallel, their y matrices
add. First let us consider the case in which we connect a pair of two-
ports Na and Nb in cascade or in tandem, as shown in Fig. 9.18. We see
Substituting V 2 in Eq. 9.73 into Eq. 9.71, we have that
~
1
_ — AaJ \L Ab _
V > V
* " (9.76)
Vi =
h h
Un - 1 +12 h2iZi
,

22 ZL
)A ( 9.74)

The transmission matrix equation for Na is

so that Zin
A
hx 2h2lZL
1 + h22Zz
(9.75)
fVl
'
A . Bal rv2a - (9.77)
712
£-
LA ca AJL -A*.
vA/W 0

/20 /1b /2
£ -o
Vl ! 7ll + 712 ( p? V2
t *+
722 + 712 Vl Na V 2a Vlb Nb V2
S
*
.
FIG. 9.16 Two-port equivalent circuit with one controlled current source. - . .
FIG 9.18 Cascade connection of two ports .

*
 272 Network analysis and synthesis Network analysis II 273
so that the transmission matrix of the gyrator is
h r h lx
*a 2c
h'
-o 0 al
"

1
Vi v2 Vi - 26 v2 We see that for the gyrator
AD
La
- BC = -1
0

(9.82)
o- Therefore the gyrator is a nonreciprocal device, although it is passive.8
L Gyrator J The overall transmission matrix of the configuration in Fig. 9.19 is
. .
FIG 9.19 Gyrator in tandem with T circuit . obtained by the product of the individual transmission matrices
+ z„g;
~
A B 0 a
2a + zjsb + zbzc
Correspondingly, for Nb we have Z6 Z6

Bbirv2 - 1 26 + Zc
- Aa _ _ _ _
fj6 Cb Db 72 _— (9.78) C D
La
1
0
L *6 *6 (9.83)
Substituting the second matrix into the first, we obtain a a(zt + *c)
Z6 Z6
'

q p , Bai VAb Bbirv2 -

Ji J LCa Aj LCb Db l - I2 \ _ (9.79) Za T~ Zft
. azb
ZaZ6 + Z6Zc +
azb
We see that the transmission matrix of the overall two-port network is If we check the configuration in Fig. 9.19 for reciprocity, we see that for
simply the product of the transmission matrices of the individual two - transmission matrix in Eq. 9.83
ports.
As an example, let us calculate the overall transmission matrix of a
AD BC = -1 - (9.84)
gyrator 7 in tandem with a T network shown in Fig. 9.19. The ideal We thus see that any reciprocal network connected in tandem with a
gyrator is an impedance inversion device whose input impedance Zin is gyrator yields a configuration that is nonreciprocal.
related to its load impedance ZL by Next consider the situation in which a pair of two- ports Na and
Nb are connected in parallel, as shown in Fig. 9.20. Let us find the
Zin = a* YL
l2a
a2 (9.80) 1:1
+
Via
h
The constant a in Eq. 9.80 is defined as the gyration resistance. If we +
Ideal
regard the gyrator as a two-port, its defining equations are transformer V2

Vx = «(-/2) sr V26
1 (9.81)
h = -a v2
.
FIG 9.20. Parallel connection of two ports.
7
B. D. H. Tellegen, “The Gyrator, a New Electric Network Element,” Phillips
-
Research Repts., 3 (April 1948), 81 101, see also Huelsman, op cit , pp. 140-148. . . 8
Most gyrators are microwave devices that depend upon the Hall effect in ferrites.
.

Network analysis II 275

274 Network analysis and synthesis
ha
.2
1
ID I f + +
Via
Z3
2’
Vx
hb
1 Zjj T GE
Vi6
Ze 5-
T . .
FIG 9.22 Series connection of two ports .
1' 2'
FIG. 9.21
so that the y parameters of the overall two-port network can be expressed
in terms of the y parameters of the individual two-ports as
y parameters for the overall two-port network. The matrix equations
for the individual two ports are - Vll Vxz Vila + 2/116 Vl 2a + 2/126
(9.89)
V ViXa 2/12« Ka Ha-
-2/ai Vila + 2/216 2/22a + 2/226
- -
(9.85)
lf we connect two ports in series, as shown in Fig. 9.22, we can express
-
-ha- -Vila 2/ 220-1 l- Vja-
-
the z parameters of the overall two port network in terms of the z param -
and
V 2/llb Vx 2b \^ 16
(9.86) eters of the individual two-ports as
L/26-
- 2/216 • 2/226 -1 LF - 26 Z11 Z12 Zlla + Z116 Z12a + 2126 (9.90)
From Fig. 9.20, we see that the following equations must hold. LZ2 IA 4 z22a h
- Z21 Z 22 -l " % 2 lb “
^226-
Vx = Vla = 16 V 2 = V2a = V 2b ^ (9.87) We may summarize by the following three points.
= ha + hb h ha + hb
be careful
— 1. When two- ports are connected in parallel, find the y parameters
In connecting two-ports in series or in parallel, we must first , and, from the y parameters, derive the other two-port parameters.
that the individual character of a two-port network is not altered when
connected in series or parallel with another two port. For example when, - -
2. When two ports are connected in series, it is usually easiest to find
the z parameters.
we connect the two-ports in Fig. 9.21 in parallel, the impedances Z7 and 3. When two-ports are connected in tandem, the transmission matrix
Z8 will be short circuited. Therefore, to insure that a two-port network is generally easier to obtain.
does not interfere with the internal affairs of the other, ideal transformers As a final example, let us find the y parameters of the bridged -T circuit
are used to provide the necessary isolation. In matrix notation, the sum in Fig. 9.6. We see that the bridged-T circuit would be decomposed into
of the individual [/<] matrices of the two-ports in parallel must equal the a parallel connection of two-ports, as shown in Fig. 9.23. Our task is to
[/] matrix of the overall two-port network. Thus we have first find the y parameters of the two-ports Na and Nb. The y parameters
of Nb are obtained by inspection and are
V V+ V
U 2-1 Uj -/*»- (9.88)
2/m = 2/216 = —i (9.91)
2/ii6 = 2/226 = 1
VlXa 2/l 2a
+
2/ll6 2/ rnljV Na is a T circuit so that the z parameters can be obtained by inspection.
Ly2ia 2/22aJ -2/216 2/226- / -^ - 2

•
:
 1
Network analysis II 277
276 Network analysis and synthesis
there are no parasitic resistances associated with the L and C Since, at .
T 1
1 _ high frequencies, parasitic losses do play an important role in governing
system performance, we must account for this incidental dissipation
-
lo o2 somehow. An effective way of accounting for parasitic resistances is to
, , KJ “load down” the pure inductances and capacitances with incidental
- dissipation by associating a resistance r, in series with every inductor Lit
l'o r
I

I
I
A
^
2

Nb
h-
I

I
o 2'

:
and, for every capacitor Cit we associate a resistor whose admittance is gu
as depicted in Fig. 9.24. Suppose We call the system function of the
network without parasitic dissipation ( Fig. 9.24a) H( s ) , and the system
function of the ‘loaded ” network Hfs ) (Fig. 9.24 b ). Let us consider the
FIG 9.23. the relationship between H( s ) and H s ) when the dissipation is uniform ,
These are
i.e., in a manner such that ^
— — z22a
s +1 — = Ct— = a (9.95)
zlla
s (9.92)
where the constant a is real and positive.
^
z12o
We then find the y parameters from the equations
— — z21a 1 When a network has uniform dissipation or is uniformly loaded, the
sum of impedances in any mesh of the unloaded network

- y -_ gA ~_ s2(ss ++ _1l ) mu = Ri + sLf + —

1
„
Vna y
22«
22g , (9.96)
?a sCi
(9.93) becomes, after loading,
s2
2/l 2a ~ 2/ 21« = 1
2s +1 m'ij = Rt + sL; + rt +
sCi + g{
Since both Na and Nb are symmetrical two ports, we know that yn = yi2 - (9.97)
for the overall bridged-T circuit. The y parameters for the bridged-T
circuit are now obtained as
= Ri + L;(s + a) + — s
1
Ci + a)
(

Vll = Vila + Vnb

_ s(s + 1) 1
+2
_ 2s 2
+ 4s + 1 o
— A/V
Ri
0 o o
Ri
AA/V
2s + 1

2/12 = 2/120 +
s 2
2/126
1
2(2s + 1)

2
2s + 2s + 1
(9.94) o
— —'
Li
' OCIP
Ci
o -
o ^WT'
Li

Ci
ri
-AA/V ° —
2s +1 2 2(2s + 1) -o
9.5 INCIDENTAL DISSIPATION o
L-Wv 1
Si
— -o
H( s )
As we have seen, the system function H( s ) of an R L C network consists -- H s ) = H( s + a )
of a ratio of polynomials whose coefficients are functions of the resistances, (a)
^ (b)
inductances, and capacitances of the network. We have considered , up
to this point, that the inductors and capacitors are dissipationless ; i.e., . .
FIG 9.24 (a) Original network . (6) Loaded network .

Network analysis II 279

278 Network analysis and synthesis
8Q
if HWV-i

2n 2h in if in

»
2n < 6fl : =r * f
>10
FIG 9.25 . 2h

Similarly, on node basis, if the admittance between any two nodes of the . .
FIG 9.26 Loaded network a = J.
original (unloaded) network is
1 In series with the inductor L = 2 h, we have a resistor
nu = Gt + sC + -sLi~ { (9.98)
r3 = aL = 1 Q (9.103)
then the same node admittance after loading is Now we determine the y parameters for the loaded network to be
1 , 1 s 1 s 1
n' u = Gf + sCi + g{ + sLf
+ r,-
— Gt + Cfs + a) +
1
Lt + a)
( s
(9.99)
-= £i ++ + i) + Ks + J)
( 2(5 + i)
Since any system function can be obtained through mesh or node equa - = 1 + 2s - +1
1
+1+4 8
y' 22
tions, it is readily seen that the original system function H( s ) becomes
H( s + a) after the network has been uniformly loaded , that is Hfs ) = )2 2
= 1 + 4+( £ +)
(s
H( s + a).
Consider the following example. Let us first find the y parameters for *+i
_ +
the unloaded network in Fig. 9.25. By inspection we have ! 2/ 12 — ( 1)
4
+ = 2
) (9.104)

2/ii = -2 + -3 + -4 = i2 + 1 2 — We see that the y parameters of the loaded network could have been
obtained from the y parameters of the unloaded network by the relation-
— s2 ship Hfs) = H( s + a).
1

2/ 22 = 1 + 2s
1
=1+ +2 (9.100)
4s r We will make use of the uniform loading concept to prove an important
theorem concerning network realizability in Chapter 10.
s
2/12 = 2/21 = - 4
9.6 ANALYSIS O F LADDER NETWORKS
Then, for a loading constant a =| , the loaded network is shown in Fig.
9.26. In parallel with the capacitor Cx = £ f, we have an admittance In this section we will consider a simple method of obtaining the
network functions of a ladder network in a single operation.9 This
= aCx = £ mho
gi (9.101) method depends only upon relationships that exist between the branch
In parallel with the capacitor C2 = J f, the associated admittance is * F. F. Kuo and G . H. Leichner, “An Iterative Method for Determining Ladder
Network Functions,” Proc. IRE, 47, No. 10 (Oct. 1959), 1782-1783.
gt = a C2 =| mho (9.102)
 !
280 Network analysis and synthesis Network analysis II 281
the 2-2' port of the ladder and work towards the 1 1' port. Each suc- -
1 Z> Zj + 2 1 ceeding equation takes into account one new immittance. We see,
-H
i rL~\ ^' Vj + 3 further, that with the exception of the first two equations, each subsequent
equation is obtained by multiplying the equation just preceding it by the
i *J + 3 immittance that is next down the line, and then adding to this product
T T the equation twice preceding it . For example, we see that I2{ s ) is obtained
by multiplying the preceding equation V2( s ) by the admittance y4(s).
. .
FIG 9.27 Ladder network . The next immittance is Z3(s). We obtain Va( s) by multiplying the previous
equation J2(s) by Z3(s) to obtain I 2 Z 3 ; then we add to this product the
currents and node voltages of the ladder. Consider the network shown equation twice preceding it, V 2( s ), to obtain Va = I 2 Z 3 + V 2 The process .
in Fig. 9.27, where all the series branches are given as impedances and is then easily mechanized according to the following rules : (1) alternate
all the parallel branches are given as admittances. If the v denote node writing node voltage and branch current equations ; (2) the next equation
voltages and i denote branch currents, then the following relationships is obtained by multiplying the present equation by the next immittance
apply. (as we work from one port to the other), and adding to this product the
vi+i = h+2Zj+2 + vj+3 results of the previous equation.
(9.105)
ij vi+i j+i + ij+i
— ^ '
Using this set of equations, we obtain the input impedance Zin( s ) by
dividing the equation for Vy( s) by the equation for Iy( s) We obtain the
voltage-ratio transfer function V / V s ) by dividing the first equation
.
These equations form the basis of the method we discuss here.
To illustrate this method , consider the network in Fig. 9.28, for which
the following node voltage and branch current relationships apply.
^ ^
V 2( s ) by the last equation Vy( s ). We obtain other network functions such
as transfer immittances and current ratios in similar manner. Note that
every equation contains V2( s ) as a factor. In taking ratios of these equa -
V2( s ) = V 2( s )
tions, the F2(s) term is canceled . Therefore, our analysis can be simplified
m = r4( j) V2( s ) if we let V 2( s) = 1.
Va( s ) = I2( s) Z 3( s ) + V2( s ) = [1 + Z 3( s) Y4( S ) ] V 2( S ) If the first equation of the set were a current variable It( s) instead of the
II ( S ) = Y 2( S ) Va{ s ) + / (*) , (9.106) f voltage V 2( s), the subsequent equations would contain the current variable
= { F2(S)[1 + Z3(s) YJs ) ] + Y4( S ) } V2(S) -
Ii( s ) as a factor, which we could also normalize to It( s) = 1. An example
in which the first equation is a current rather than a voltage equation
V1( s) = I 1( s ) Z 1( s ) + Va( s ) : .
may be seen by determining the y parameters of a two-port network.
,
= { Z [ Y2( I + z3 y4) + y4] + (1 + z3 y4)} v2( s ) Before we embark upon some numerical examples, it is important to
note that we must represent the series branches as impedances and the
Upon examining the set of equations given in Eq. 9.106, we see that shunt branches as admittances. Suppose the series branch consisted of a
each equation depends upon the two previous equations only. The first resistor R = 1 £2 in parallel with a capacitor C =\ f. Then the impedance
equation, V 2 = V 2 is, of course, unnecessary. But, as we shall see later, of the branch is
it is helpful as a starting point. In writing these equations, we begin at
2 Z(S) = « =± (9.107)
rHJL 1/sC +R s+2
and must be considered as a single entity in writing the equations for the
Vi Y2 Y V2 ladder. Similarly, if a shunt branch consists of a resistor R2 = 2 Q and
*
an inductor Lx = 1 h, then the admittance of the branch is
T T 2' 1
y(s) =2 (9.108)
FIG. 9.28
+s
,

Network analysis II 283

282 Network analysis and synthesis
The key point in this discussion is that we must use the total impedance
or admittance of a branch in writing the equations for the ladder.
Example 9.1. Let us find the voltage ratio VJVj, the current ratio I2 IIi , the
input impedance Zx = VJIj , and the transfer impedance Z21 = F2/A f r the
°
network in Fig. 9.29. First, we must represent the series branches as impedances
and the shunt branches as admittances, as shown in Fig. 9.30. The branch
current and node voltage equations for the network are
FIG. 9.31
h Va h h Va h illustrated in Fig. 9.32. The pertinent equations are
o-
V + 3s +
if 3h
4 /2 = i Fa = 2
Vi ih 2f V2 f 2s
^= Y2 /„ = £s(2) + 1 = s + 1
1
Yi = 2( s + 1) ( s + 1) + 2 = -
5
(9.111)
.
FIG 9.29 FIG . 9.30

v2( s) = l
h = + ( s + 1) = 6s + 1 + —
5/ 2
s
We now obtain our short-circuit functions as
f 2( s ) = 2s
2/11
.= h 12
yj = T 5 + 5 + 5 ’
2 1 _! _
2/21 " i
2
(9.112)
^
" “
5
Va( s) = 3s( 2s ) + 1 = 6i2 + 1
2 (9.109) 2

^
2
h( s) = -s (6J2 + 1) + 2s = 14J + -s +
1
2(s + l ) -Wr-
8
+ 6 + 1 = 6sz + 57
^ Vi
2» 2 + l
is 'h

The various network functions are then obtained. I

Vx 6 si+ 57 s2 + 8 FIG. 9.32

(«)
^ in ~ ~
h
"
14s3 + 2s
A number of other contributions on ladder networks have appeared
v2 S2 in the recent literature. To name but a few, there are the works of
Cb) V3
~
6s4 + 57s2 Bashkow,10 O’Meara,11 Bubnicki,12 Walker,13 and Dutta Roy.14
+8
12 _
(9.110)
2s2 10
T. R. Bashkow, “A Note on Ladder Network Analysis,” IRE Trans, on Circuit
(c)
A
”
14s2 + 2 11 T.
-
Theory, CT 8, (June 1961), 168.
R. O’Meara, “Generating Arrays for Ladder Network Transfer Functions,”
IEEE Trans, on Circuit Theory , CT 10, (June, 1963), 285.
-
id ) — V2
r
~
Iy 14s2 + 2
s 11 Z. Bubnicki,
“Input Impedance and Transfer Function of a Ladder Network,”
IEEE Trans, on Circuit Theory, CT-10, (June, 1963), 286.
Example 9.2. Find the short-circuit admittance functions yn and y21 for the
13 F.
-
Walker, “The Topological Analysis of Non Recurrent Ladder Networks,”
Proc. IEEE, 52, No. 7, (July, 1964), 860.
network in Fig. 9.31. To obtain the short circuit functions, we must short - 11 S. C. Dutta
Roy, “ Formulas for the Terminal Impedances and Transfer Functions
circuit the 2-2' port. Then we represent the series branches as impedances and
the shunt branches as admittances so that the resulting network is given as is
of General Multimesh Ladder Networks,” Proc. IEEE, 52, No. 6, (June, 1964), 738 .

!
i
 '
284 Network analysis and synthesis I Network analysis II 285
Because of the recursive nature of the equations involved, the digital 9.3 Find the z parameters for the lattice and bridge circuits in the figures
computer is an ideal tool for the analysis of ladder networks. In Chapter shown. (The results should be identical .)
15 a detailed description is given of a digital computer program based I» {JD -o 2
upon the algorithm described in this section for evaluating ladder network
functions. Zb Zb

Problems
9.1 Find the z, y , h , and T parameters of the networks shown in the figures.
Some of the parameters may not be defined for particular circuit configurations.

Z zi 1 0
l 'o. I .
- s [ or
y Zi Lattice network
(a )

(a) ( b) (c)
PROB 9.1 .
9.2 Find the z , y , h, and T parameters for the networks shown in the figures.
2 fi lfi
1 o- V\MA AAAAr -02
lfi 3 fi

i'o
Bridge network
( b)

PROB 9.3 .
9.4 For the lattice and bridge circuits in Prob. 9.3 find the y parameters in
terms of the admittances Ya = l /Za and Yb = l / Zt .
9.5 For the circuit shown, find the voltage-ratio transfer function VJV 1 and
the input impedance Z1 = FJ /IJ in terms of the z parameters of the two-port
network N and the load resistor RL.
1 h
lfi 10 +
lfi < 2f i Vi N V2

o- r
(C)
PROB 9.2 . PROB 9.5 .
!

:
1
Elements of realizability theory 291
h( t ) Ht - T )

10 chapter T
h: 0 t 0 t

Elements of realizability . .
FIG 10.1 (a) Nonrealizable impulse
( a) (b)
response. (b) Realizable impulse response.

theory is not causal. In certain cases, the impulse response could be made
realizable (causal) by delaying it appropriately. For example, the impulse
response in Fig. 10.1a is not realizable. If we delay the response by T
seconds, we find that the delayed response h( t T ) is realizable (Fig.
10.16).
—
10.1 CAUSALITY AND STABILITY In the frequency domain, causality is implied when the Paley Wiener -
criterion1 is satisfied for the amplitude function \ H( jco )\ . The Paley Wiener -
In the preceding chapters we have been primarily concerned with the criterion states that a necessary and sufficient condition for an amplitude
problem of determining the response, given the excitation and the network ; function \ H( jco )\ to be realizable (causal) is that
this problem lies in the domain of network analysis. In the next five
chapters we will be dealing with the problem of synthesizing a network
f “ [ log \ H ( ja> )\ 1 dco < oo (10.4)
j-
I + CO 2
given the excitation E( s) and the response R( s ). The starting point for
00

any synthesis problem is the system function The following conditions must be satisfied before the Paley Wiener -
criterion is valid : /?(/) must possess a Fourier transform H( ja>) ; the
square magnitude function \ H( jco )\2 must be integrable, that is,
H( s ) = —
E( s )
(10.1)

Our task is to synthesize a network from a given system function.

The first step in a synthesis procedure is to determine whether H( s)
J \ H ( jco )\ 2 dco <
The physical implication of the Paley-Wiener criterion is that the amplitude
co (10.5)

can be realized as a physical passive network. There are two important \ H( ja>)\ of a realizable network must not be zero over a finite band of
—
considerations causality and stability. By causality we mean that a frequencies. Another way of looking at the Paley Wiener criterion is that -
voltage cannot appear between any pair of terminals in the network the amplitude function cannot fall off to zero faster than exponential
before a current is impressed, or vice versa . In other words, the impulse .
order For example, the ideal low-pass filter in Fig. 10.2 is not realizable
response of the network must be zero for / < 0, that is, I because beyond a>c the amplitude is zero. The Gaussian shaped curve
,2
h( t ) =0 t <0 (10.2) \ H ( jco )\ = e~a ( 10.6)

As an example, the impulse response shown in Fig. 10.3 is not realizable because
| log \ H( j d)\ | = <o2
( (10.7)
h[ t ) = <r 0 < u( t ) (10.3a)
1 R . E . A . C. Paley and N . Wiener, ‘‘Fourier Transforms in the Complex Domain,”
is causal, whereas h( t ) = (10.36)
Am . Math . Soc . Colloq . Pub ., 19 ( 1934), 16-17.
290
 292 Network analysis and synthesis Elements of realizability theory 293
we then can approximate h( t —
td ) by a causal response hL( t ) which is
discussion of the Paley-Wiener
zero for t < 0 . ( For a more detailed
criterion, the reader is referred to an excellent treatment by Wallman.2)
If a network is stable, then for a bounded excitation e( t ) the response
Ao ( /) is also bounded . In other words, if

«c w KOI < Ci o t < 00

“ 0 cic CO 0
then KOI < c2 o / < co
FIG . 10.2. Ideal filter char - FIG . 10.3. Gaussian filter character-
acteristic istic. where Cj and C2 are real, positive, finite quantities. If a linear system is
stable, then from the convolution integral we obtain
so that the integral
:
/ 1 + co8
is not finite. On the other hand , the amplitude function
dco ( 10.8)
KOI < ciJ"Wr l ) dr < C* (10.11)

Equation 10.11 requires that the impulse response be absolutely integrable,

1 ( 10.9) or
I H ( jco )\ =
Vi + w
does represent a realizable network . In fact, the voltage-ratio transfer
8

J |h (r)| dr < oo (10.12)

function of the R- C network in Fig. 10.4 has an amplitude characteristic One important requirement for h{ t ) to be absolutely integrable is that the
given by \ H( jco )\ in Eq . 10.9 . impulse response approach zero as t approaches infinity, that is,
For the ideal filter in Fig. 10.2, the inverse transform h( t ) has the form
lim h( t ) -+ 0
Ao sin cocl < -> CO
h( t ) = (10.10)
irt Generally , it can be said that with the exception of isolated impulses, the
where A0 is a constant . From the sin x / x curve in Fig. 3.14, we see that impulse response must be bounded for all t , that is,
h( t ) is nonzero for t less than zero. In fact , in order to make h( t ) causal, |A( /)| < C all t ( 10.13)
it must be delayed by an infinite amount . In practice, however, if we
delay h{ l ) by a large but finite amount td such that for / < 0 the magnitude where C is a real, positive, finite number.
—
of h( t td ) is less than a very small quantity e , that is, Observe that our definition of stability precludes such terms as sin co0t
from the impulse response because sin co0t is not absolutely integrable .
I h( t - td )| < e t <0 These undamped sinusoidal terms are associated with simple poles on
10
the jco axis. Since pure L- C networks have system functions with simple
O vw \A poles on the jco axis, and since we do not wish to call these networks
+ unstable, we say that a system is marginally stable if its impulse response
is bounded according to Eq . 10.13 , but does not approach zero as t
Vi ( s ) If 4: Vo( s ) approaches infinity .
In the frequency domain , the stability criterion requires that the system

o -o * G. E. Valley Jr. and H. Wallman, Vacuum Tube Amplifiers, McGraw Hill Book -
FIG . 10.4 Company, New York, 1948, Appendix A, pp. 721-727.

294 Network analysis and synthesis '

Elements of realizability theory 295
function possess poles in the left-half plane or on the jco axis only. More- polynomials. A polynomial P( s ) is said to be Hurwitz if the following
over, the poles on the jco axis must be simple.3 As a result of the require- conditions are satisfied :
ment of simple poles on the jco axis, if H( s ) is given as
1 . P( s ) is real when s is real .
ansn + an^sn
m = bmsm
1
+ + a0
1- ats (10.14) 2. The roots of P( s ) have real parts which are zero or negative.
+ *> - m i + • •
+ blS + b0 As a result of these conditions, if P( s ) is a Hurwitz polynomial given by
then the order of the numerator n cannot exceed the order of the de-
nominator m by more than unity, that is, n m <. 1 . If n exceeded m — P( s ) = ans" + an _lsn~1 + 1- + a0 ( 10.15)
by more than unity, this would imply that at s = jco = oo, and there then all the coefficients at must be real ; if .v ,- = a * + jp is a root of P( s ) ,
would be a multiple pole . To summarize, in order for a network to be then ctf must be negative . The polynomial
stable, the following three conditions on its system function H{ s ) must
be satisfied : P( s ) = ( s + 1)(J + 1 + yV2 )( + 1j
— j l2 )
\ ( 10.16)
1 . H{ s ) cannot have poles in the right-half plane . is Hurwitz because all of its roots have negative real parts . On the other
2 . H( s) cannot have multiple poles in the jco axis. hand ,
3 . The degree of the numerator of H( s ) cannot exceed the degree of G(j) = { s - l )(j + 2 ){ s + 3) ( 10.17)
the denominator by more than unity.
is not Hurwitz because of the root s = 1 , which has a positive real part .
Finally, it should be pointed out that a rational function H( s) with poles Hurwitz polynomials have the following properties :
in the left-half plane only has an inverse transform h( t ) , which is zero for
t < 0.4 In this respect, stability implies causality. Since system functions 1 . All the coefficients at are nonnegative . This is readily seen by
of passive linear networks with lumped elements are rational functions examining the three types of roots that a Hurwitz polynomial might
with poles in the left-half plane or jco axis only, causality ceases to be a have . These are
problem when we deal with system functions of this type. We are only s = ~Yi Yi real and positive
concerned with the problem of causality when we have to design a filter s = ± jco , co, real
for a given amplitude characteristic such as the ideal filter in Fig . 10.2. s = - oc{ ± jPi a,- real and positive
We know we could never hope to realize exactly a filter of this type
because the impulse response would not be causal . To this extent the The polynomial P( s) which contains these roots can be written as
Paley-Wiener criterion is helpful in defining the limits of our capability.
P( s ) = ( s + YiW + C0 * ) [( s + a,.)2 + p?] • • • (10.18)
10.2 HURWITZ POLYNOMIALS Since P( s ) is the product of terms with only positive coefficients, it follows
In Section 10.1 we saw that in order for a system function to be stable, that the coefficients of P( s ) must be positive . A corollary is that between
its poles must be restricted to the left-half plane or the jco axis . Moreover , the highest order term in s and the lowest order term, none of the coeffi-
the poles on the jco axis must be simple. The denominator polynomial of cients may be zero unless the polynomial is even or odd. In other words,
the system function belongs to a class of polynomials known as Hurwitz an-1, an-2 » • • , a2 , ax must not be zero if the polynomial is neither even
nor odd . This is readily seen because the absence of a term a( implies
3
In Chapter 6 it was shown that multiple poles on the jco axis gave rise to terms as
cancellation brought about by a root s —
with a positive real part.
A 0t sin co 0 r .
2. Both the odd and even parts of a Hurwitz polynomial P( s ) have
roots on the jco axis only. If we denote the odd part of P( s ) as n( s ) and
.
4 G Raisbeck, “A Definition of
Passive Linear Networks in Terms of Time and
the even part as m( s) , so that
Energy,” /. Appl. Phys., 25 (Dec., 1954), 1510-1514. The proof follows straight -
forwardly from the properties of the Laplace transform. P( s ) = n( s ) + m( s ) ( 10.19)

..
 296 Network analysis and synthesis Elements of realizability theory 297
then m( s) and n( s ) both have roots on tne jco axis only. The reader is We see that the process of obtaining the continued fraction expansion
referred to a proof of this property by Guillemin.5 of f ( s ) simply involves division and inversion. At each step we obtain a
3. As a result of property 2, if P( s ) is either even or odd , all its roots are quotient term qts and a remainder term, Ri+1( s)/ Ri( s ). We then invert the
on the jco axis. remainder term and divide Ri+i( s ) into Rfs ) to obtain a new quotient .
4. The continued fraction expansion of the ratio of the odd to even There is a theorem in the theory of continued fractions which states that
parts or the even to odd parts of a Hurwitz polynomial yields all positive the continued fraction expansion of the even to odd or odd to even parts
quotient terms. Suppose we denote the ratios as y>( s ) = n( s )/ m( s) or of a polynomial must be finite in length .7 Another theorem states that, if
>p{ s ) = m( s )/ n( s ), then the continued fraction expansion of ip( s) can be the continued fraction expansion of the odd to even or even to odd parts
written as of a polynomial yields positive quotient terms, then the polynomial must
1 be Hurwitz to within a multiplicative factor fV( s ).8 That is, if we write
V>( s ) = qxs + (10.20)
1
qtf + m = W(s) Ffs) (10.25)
q3s 4 then F( s) is Hurwitz, if fV( s ) and Fjs ) are Hurwitz. For example, let us
test whether the polynomial
1 F( s ) = s4 + s3 + 5s2 + 3s + 4 (10.26)
—
+ qns
is Hurwitz. The even and odd parts of F( s) are
.
where the quotients qx , qz , . . , qn must be positive if the polynomial m( s) = s4 + 5s3 + 4
P( s ) = n( s ) + m( s ) is Hurwitz.6 To obtain the continued fraction expan- (10.27)
sion, we must perform a series of long divisions. Suppose y>( s) is n( s ) = s3 + 3s
We now perform a continued fraction expansion of y>( s) = m( s )ln( s ) by
(10.21)
n( s ) dividing n( s ) by m( s ) , and then inverting and dividing again, as given by
the operation
where m( s) is of one higher degree than n( s ) Then if we divide n( s ) into . s3 + 3s ) s4 + 5s2 + 4 ( s
m( s ), we obtain a single quotient and a remainder
s4 4- 3s2
(s)
y>( s ) = qxs + Rin( s ) (10.22) 2s3 4- 4 ) s3 4- 3s ( si2
s3 4- 2s
The degree of the term Rx( s ) is one lower than the degree of n( s ) Therefore . s ) 2s2 4- 4 ( 2J
if we invert the remainder term and divide, we have
2J2
4 ) s ( s/ 4
RW . « ,( <>
(10.23)
s
Inverting and dividing again, we obtain so that the continued fraction expansion of yt( s ) is
Ri ( s ) R (s) 1
R (s)
= q3s + R*( s ) (10.24) (10.28)
* * n( s ) 1
6
E. Guillemin , The Mathematics of Circuit Analysis, John Wiley and Sons, New York , 2
1949 . An excellent treatment of Hurwitz polynomials is given here. 2s 4- T7
“

s/4
8
A proof can be undertaken in connection with L C driving-point functions ; see -
M . E. Van Valkenburg, Introduction to Modern Network Synthesis, John Wiley and 7
See Van Valkenburg, loc . cit .
Sons, New York , 1960. 8 fV( s ) is a common factor in m( s ) and n( s ).

298 Network analysis and synthesis Elements of realizability theory 299

Since all the quotient terms of the continued fraction expansion are The continued fraction expansion is
positive, F( s ) is Hurwitz.
s3 + 3s )s4 + 2s3 + 2 ( s
Example 10.1 . Let us test whether the polynomial s4 + 3s2

G( s) = s3 4- 2s2 4- 3s 4- 6 (10.29)
—s 3
4- 2)
^s — 2s ( — s
3
4- 3f

5s) -s2 + 2 { — s/ 5
is Hurwitz. The continued fraction expansion of n( s )/ m( s ) is obtained from the -s*
division 2) 5s (fs
2 s3 4- 6) 4- 3s (s/2
s® 5s
s3 4- 3s
We see that F( s) is not Hurwitz because of the negative quotients.
0
Example 10.4. Consider the case where F( s ) is an odd or even function . It is
We see that the division has been terminated abruptly by a common factor
impossible to perform a continued fraction expansion on the function as it
stands. However, we can test the ratio of F( s ) to its derivative, F'( s ) .s If the
s3 4- 3s. The polynomial can then be written as
ratio F( s )/ F'( s ) gives a continued fraction expansion with all positive coefficients,
then F( s ) is Hurwitz . For example, if F( s ) is given as
G(s) = (s3 4- 3s) (\ + (10.30) F( s ) = s7 4- 3s5 4- 2s3 4- s s ( 10.35)
then F' ( s ) is F'( s ) = 7s6 4- 15s4 4- 6s2 4- 1 ( 10.36)
We know that the term 1 4- 2/s is Hurwitz . Since the multiplicative factor
s3 4- 3s is also Hurwitz, then G(s) is Hurwitz. The term s3 4- 3s is the mul - Without going into the details, it can be shown that the continued fraction
tiplicative factor fV ( s ) , which we referred to earlier. expansion of F(s)/F'(s) does not yield all positive quotients. Therefore F( s ) is
not Hurwitz.
Example 10.2. Next consider a case where JF(s) is non-Hurwitz.
10.3 POSITIVE REAL FUNCTIONS
F( s ) = s7 4- 2s« 4- 2s6 4- s4 4- 4s3 4- 8s2 4- 8s 4- 4 ( 10.31 )
In this section we will study the properties of a class of functions
The continued fraction expansion of F( s) is now obtained .
known as positive real functions. These functions are important because
they represent physically realizable passive driving-point immittances. A
n( s) _s 1
function F( s) is positive real (p.r.) if the following conditions are satisfied :
m( s) 2 1
f* + (10.32) 1. F( s ) is real for real s; that is, F{ o) is real.

We thus see that fV( s ) = j4 4- 4, which can be factored into

—
( si K4) 2. The real part of F( s ) is greater than or equal to zero when the real
part of s is greater than or equal to zero, that is,
Re [F(.y)] > 0 for Re J 0
IV( s) = (i2 4- 2s 4- 2 Xs3 - 2s + 2 ) ( 10.33) Let us consider a complex plane interpretation of a p.r. function.
Consider the s plane and the F( s ) plane in Fig. 10.5. If F( s ) is p. r., then a
It is clear that F( s ) is not Hurwitz . point o0 on the positive real axis of the s plane would correspond to, or
Example 10.3. Let us consider a more obvious non-Hurwitz polynomial
map onto , a point F( o0 ) which must be on the positive real axis of the
F( s ) plane. In addition, a point s{ in the right half of the s plane would
F( s) = s4 + s3 + 2s3 + 3s 4- 2 (10.34) * See Guillemin, loc. cit .
300 Network analysis and synthesis Elements of realizability theory 301
;' lmF in Fig. 10.6, the average power dissipated by the network is
jia
s plane F( s ) plane
Average power = i Re [ Zin( jw ) ] |/|2 > 0 (10.39)
We then conclude that , for any passive network
Si F( si )
Re [Z n( jco)] > 0 , (10.40)
<r0 F( <r 0 ) ReF We can now prove that for Re s = a > 0, Re Zin(<r + jco ) > 0. Consider
the network in Fig. 10.6, whose driving -
/
point impedance is Zin(.s). Let us load
the network with incidental dissipation
.
FIG 10.5. Mapping of s plane onto F( s ) plane. such that if the driving-point impedance Passive
network
of the uniformly loaded network is
map onto a point F( s,) in the right half of the F( s ) plane. In other words,
for a positive real function, the right half of the s plane maps onto the
Zx( s), then o-

right half of the F( s ) plane. The real axis of the s plane maps onto the Zt{ s) = ZJs + «) (10.41) .
FIG 10.6
real axis of the F( s ) plane.
A further restriction we will impose is that F( s) be rational. Consider where a, the dissipation constant, is real and positive. Since Z,(s) is the
the following examples of p.r. functions : impedance of a passive network,
1. F( s ) = Ls (where L is a real, positive number) is p. r. by definition. Re ZJiJm) > 0 (10.42)
If F( s) is an impedance function, then L is an inductance.
2. F( s ) = R (where R is real and positive) is p.r. by definition. If F( s ) so that Re Zin(« + jto ) > 0 (10.43)
is an impedance function, R is a resistance. Since a is an arbitrary real positive quantity, it can be taken to be a.
3. F( s ) = Kjs ( K real and positive) is p.r. because, when s is real, F( s ) Thus the theorem is proved.
is real. In addition, when the real part of 5 is greater than zero, Re ($) = Next let us consider some useful properties of p.r. functions. The
a > 0. proofs of these properties are given in Appendix D.
Then
M! a +
) =
&
2
to2
>0
Therefore, F( s ) is p.r. If F( s) is an impedance function, then the corre-
(10.37) 1. If F( s ) is p.r., then 1/ F( s ) is . also p. r. This property implies that if a
driving-point impedance is p. r., then its reciprocal, the driving-point
admittance, is also p. r.
sponding element is a capacitor of 1/ K farads. 2. The sum of p.r. functions is p.r. From an impedance standpoint,
We thus see that the basic passive impedances are p.r. functions. we see that if two impedances are connected in series, the sum of the
Similarly, it is clear that the admittances impedances is p. r. An analogous situation holds for two admittances in
£ parallel. Note that the difference of two p.r. functions is not necessarily
Y(s) = K
Y(s) = Ks
(10.38)
—
p.r. ; for example, F( s ) = s \ / s is not p.r.
3. The poles and zeros of a p. r. function cannot have positive real
parts, i.e., they cannot be in the right half of the s plane.
K
Y( s ) = - .
4. Only simple poles with real positive residues can exist on the jco axis.
s 5. The poles and zeros of a p.r. function are real or occur in conjugate
are positive real if K is real and positive. We now show that all driving- pairs. We know that the poles and zeros of a network function are
point immittances of passive networks must be p.r. The proof depends functions of the elements in the network. Since the elements themselves
upon the following assertion : for a sinusoidal input, the average power are real, there cannot be complex poles or zeros without conjugates
dissipated by a passive network is nonnegative. For the passive network because this would imply imaginary elements.

.. £
302 Network analysis and synthesis Elements of realizability theory 303
6. The highest powers of the numerator and denominator polynomials If AT, is found to be positive, then F( s) satisfies the second of the three
may differ at most by unity. This condition prohibits multiple poles and conditions.
zeros at s = oo. In order to test for the third condition for positive realness, we must
7. The lowest powers of the denominator and numerator polynomials first find the real part of F( jto ) from the original function F( s ) To do this, .
may differ by at most unity. This condition prevents the possibility of let us consider a function F( s ) given as a quotient of two polynomials
multiple poles or zeros at s = 0.
8. The necessary and sufficient conditions for a rational function with F( s ) (10.45)
real coefficients F( s) to be p.r. are Q(s)
(a) F( s) must have no poles in the right-half plane. We can separate the even parts from the odd parts of P( s) and Q( s ) so
(b) F( s ) may have only simple poles on the jto axis with real and positive that F( s ) is
residues.
F ( s ) = Mi + Ai
(» ) (s)
( c ) Re F( jco ) > 0 for all to . M2(s) + N 2( s )
(10.46)

Let us compare this new definition with the original one which requires where Mjs ) is an even function and Njs ) is an odd function. F( s ) is now
the two conditions. decomposed into its even and odd parts by multiplying both P( s ) and
1. F( s ) is real when s is real.
2. Re F( s ) > 0, when Re s > 0.
Q( s) by M %
—N 2 SO that
F( s ) = + Ai M2 - N 2
In order to test condition 2 of the original definition, we must test M2 + N2 M2 - N2
(10.47)
every single point in the right-half plane. In the alternate definition, Mr M 2 - N j N 2 , M 2 N 1 - Mx N 2
condition (c) merely requires that we test the behavior of F( s) along the M22 - N 22 M 2 - N 22
jco axis. It is apparent that testing a function for the three conditions
given by the alternate definition represents a considerable saving of We see that the products M 2 and Nx N2 are even functions, while
effort, except in simple cases as F( s ) = l / s. Mx 2
N and M 2 Nx are odd functions . Therefore, the even part of F( s) is
Let us examine the implications of each criterion of the second definition.
= MxM . 22 - NNx2
Condition (a) requires that we test the denominator of F( s ) for roots in M - N2
Ev [F(s) j (10.48)
the right-half plane, i.e., we must determine whether the denominator of
F( s ) is Hurwitz. This is readily accomplished through a continued fraction and the odd part of F( s ) is
expansion of the odd to even or even to odd parts of the denominator. ! M2 NX - Mt N 2
Odd [F(s) j =
— —
The second requirement condition (b) is tested by making a partial
fraction expansion of F( s ) and checking whether the residues of the poles
M 22 - N 2
(10.49)

on the ja> axis are positive and real. Thus, if F( s ) has a pair of poles at If we let s = jco, we see that the even part of any polynomial is real,
while the odd part of the polynomial is imaginary, so that if F( jco ) is
s = ± jcolt a partial fraction expansion gives terms of the form shown.
written as
F( jco ) = Re [ F( jco ) ] + jlm [ F( jco ) ] (10.50)
s - jcox
1'
5 +M it is clear that Re (F(yco)] = Ev [F(*)J |„ (10.51)
The residues of complex conjugate poles are themselves conjugates. If
—
the residues are real as they must be in order for F( s ) to be p. r. then
Kx = Kx* so that
—
and j Im [ F( ja> )] = Odd [
Therefore, to test for the third condition for
]
-
^) U positive
realness, we
(10.52)

Kx | Kx* 2 Kxs _ (10.44)

determine the real part of F( jco ) by finding the even part of F( s ) and then
letting s = jco. We then check to see whether Re F( jco ) > 0 for all co.
—
s jmx s + jto1 s + co 2
2
 304 Network analysis and synthesis
Elements of realizability theory 305
A ( io )
when b = 0, then a = 0. To summarize, the conditions that must be
A ( oi ) fulfilled in order for F( s ) to be positive real are
1. a, b, c 0.
2. b > a.
s -\- 2
Single
O)
Double W
We see that F 1( s ) = s2 (10.59)
root root + 3s + 2
.
FIG 10.7 .
FIG 10.8 is p.r., while the functions

FA) = s2 + 2
S 1
(10.60)
The denominator of Re F( jco ) is always a positive quantity because
+
M 2( ja> )* - N { jcof =
+ Nt( co )* 0 (10.53)
FA) = s2 +2s 1
2 s 4 (10.61)
That is there is an extra j or imaginary term in N jco ), which, when
,
—
squared , gives 1, so that the denominator of Re F( jco ) is the sum of
two squared numbers and is always positive. Therefore, our task resolves
^ + +
are not p.r. As a second example, let us determine the conditions for the
into the problem of determining whether biquadratic function
s2
A( co ) =
—
M 2( jco ) N -y(Jco ) N 2( jco ) > 0 (10.54)
If we call the preceding function A( co ), we see that A{ co ) must not have
F(s) = 2 + flis + o
s + V + b0
a (10.62)

positive, real roots of the type shown in Fig. 10.7; i .e. , A( co ) must never to be p.r. We will assume that the coefficients aly a0 , blt b0 are all real,
have single, real roots of to. However, A( co ) may have double roots positive constants. Let us test whether F( s ) is p.r. by testing each require-
(Fig. 10.8), because A( co ) need not become negative in this case. ment of the second definition.
As an example, consider the requirements for '
S ,
First, if the coefficients of the denominator b and b0 are positive, the
denominator must be Hurwitz. Second , if bx is positive, we have no poles
F( s ) =
s +a (10.55) on the jco axis. Therefore we can ignore the second condition.
s 2
+ bs + c The third condition can be checked by first finding the even part of
to be p.r. First , we know that, in order for the poles and zeros to be in F( s ) , which is •
the left-half plane or on the jco axis, the coefficients a, b , c must be greater (s 2 + a 0)(s2 + b0) - a, 2
or equal to zero. Second , if b = 0, then F( s) will possess poles on the
Ev [F(s)] =
(s2 + b0)2 b * s* - ^
jco axis. We can then write F( s ) as
F& = -
rr + irr (10.56)
s4
(s2 + bo)2 -
—
+ [(fl0 + b0 ) at bjs2 + Upfep (10.63)
S + s +c
C 2

We will show later that the coefficient a must also be zero when b = 0. The real part of F( jco ) is then
Letlis proceed with the third requirement, namely, Re F( jco ) > 0. From co4 - [(a 0 + b0) - + a0 b0
the equation
Mi( jco ) Mljco) N jto ) N 2( jco ) > 0 -
^
(10.57)
Re [ F ( ja> ) ] =
—
( co2 + b0)2 +b
^
to 2
(10.64)

we have ) 2
—
fl( ft + c ) + bco > 0
2
(10.58a) We see that the denominator of Re [ F( jco)] is truly always positive
so it remains for us to determine whether the numerator of Re [F( yft>)]
which simplifies to A( co ) = ( b
— a)co2 + ac > 0
It is evident that in order to prevent A( co ) from having positive real roots
(10.58b) ever goes negative. Factoring the numerator, we obtain
_ 4a b,
of co , b must be greater than or equal to a, that is, b a. As a result, co2.
( fl + feo) - « A
° ± 17 _
[( fl # + bo) 0 lblf (10.65)
* = 0

_ __

306 Network analysis and synthesis Elements of realizability theory 307

There are two situations in which Re [F(jco)] does not have a simple real The examples just given are, of course, special cases. But they do
root. illustrate the procedure by which functions are tested for the p. r. property.
Let us consider a number of other helpful points by which a function
1. When the quantity under the radical sign of Eq. 10.65 is zero (double, might be tested quickly. First, if F( s ) has poles on the jco axis, a partial
real root) or negative (complex roots). In other words, fraction expansion will show if the residues of these poles are positive
K«o + W - «A]2 4 aobo 0 “ (10.66) and real. For example,
3s2 5
F( s ) = 2 +
or
If +
— aA
l( a0 + b0) aA]2 < 4a0b0
(a0 b0) - 0
(10.67)
(10.68)
s(s + 1)
(10.82)

has a pair of poles at s = ±/l . The partial fraction expansion of F( s ),

then + — aA Va
(a0 b0) 2 b
0 0
(10.69) 25
F{ s ) = s2 -f— (10.83)
> Va — Vb
1 s
or ajby ( 0)
2 (10.70)
0
shows that the residue of the poles at s = ± j is negative. Therefore F( s )
If + - A<
(«o bo) 0 0 (10.71) is not p.r.
aA — ao + < V A b0) 2 a (10.72) Since impedances and admittances of passive time-invariant networks
then (
are p. r. functions, we can make use of our knowledge of impedances
but
so again
(a0 + b0)
—
< < — +
albl
^
a A > Va - Vb (
0

0
a

0 )2
(a0 b0) (10.73)
(10.74)
connected in series or parallel in our testing for the p. r. property. For
example, if ZA) and Z 2( s ) are passive impedances, then Z1 connected in
parallel with Z2 igives an overall impedance
2. The second situation in which Re [ F( jco ) ] does not have a simple
real root is when co\ % in Eq. 10.65 is negative so that the roots are imagi- Z(s) = A(s) Z2(s) (10.84)
nary. This situation occurs when Us) + Z2(s)
[(flo + b0 ) ~ U ,? ~ 4a0b0 > 0 (10.75) Since the connecting of the two impedances in parallel has not affected
the passivity of the network, we know that Z(5) must also be p.r. We see
and
From Eq . 10.75 we have
(a0 + b0) — aA < 0 (10.76) that if F-y( s ) and F2( s ) are p.r. functions, then

F( s ) = FA) F 2( s ) (10.85)
a A ~ ( ao + io) > 2\/a„b0 > (a + b„) — aA
0 (10.77) FA) + FA)
Thus aA > (Va0 “ Vb )2 (10.78) must also be p.r. Consequently, the functions
0
Ks
We thus see that Eq. 10.70 is a necessary and sufficient condition for a F( s ) = (10.86)
biquadratic function to be positive real. If we have s +a
aA = (Va0 - Vbo )2 (10.79) and F( s ) = —
J -I- a
(10.87)

then we will have double zeros for Re [ F(/<w)]. where a and K are real and positive quantities, must be p. r. We then
Consider the following example: observe that functions of the type

We see that
aA = 2 x 5
F( s ) =

(
s2 + 2s
s2 + 5s

Va - Vb, )
0
+ 25
+ 16
2
= (V25 - Vl 6 )2
(10.80)

( 10.81 )
F(s)
- 5i2
s+a
s
a, /S

l
>0
(10.88)
s+a s +a
so that F( s ) is p.r. must be p.r. also.
 308 Network analysis and synthesis Elements of realizability theory 309
Finally, let us determine whether Let us examine these cases one by one. Criterion 1 is satisfied because
the poles of Z2( s) are also poles of Z(s). Criterion 2 is satisfied by this
Ks
F( s ) = <x , K 0 (10.89) same argument. A simple partial fraction expansion does not affect the
s2 + a residues of the other poles. When s = ja>, Re [Z(y'w) = D/ j(o] = 0.
is p. r. If we write F( s ) as Therefore we have
1 (10.90) Re Z2( joj ) = Re Z( jw ) > 0
F (s) = - (10.94)
s/ K + a / Ks
From the foregoing discussion, it is seen that if “Z(i) has a pole at s = 0,
we see that the terms s/ K and a / Ks are p.r. Therefore, the sum of
the
.r., a partial fraction expansion can be made such that one of the terms is of
two terms must be p. r. Since the reciprocal of a p.r. function is also p
the form Kjs and the other terms combined still remain p r. .
we conclude that F( s ) is p.r. A similar argument shows that if Z( s ) has a pole at s = oo (that is,

10.4 ELEMENTAR Y SYNTHESIS PROCEDURE S

—
n m = 1), we can divide the numerator by the denominator to give a
quotient Ls and a remainder term R( s ), again denoted as Z,(s) and Z2(s).
' '
The basic philosophy behind the synthesis of driving-point functions , Z( s) = Ls + R( s ) = Z1( s) + Z 2( s ). (10.95)
is to break up a p.r. function Z( s ) into a sum of simpler p. r functions . .
Here Z2(s) is also p.r If Z( s ) has a pair of conjugate imaginary poles on
.
Z,(s), Z 2( s), . . , Z „( s), and then to synthesize these individual Zt( s ) as the imaginary axis, for example, poles at s = ± jo>i , then Z( s ) can be
elements of the overall network whose driving point impedance is Z(
s). - |
expanded into partial fractions so that
Z( s ) = Zx( s) + Z 2( s ) + • + Zn( s ) (10.91)
2 Ks
Z( s ) = + Z2(s) (10.96)
First, consider the “ breaking-up” process of the function Z(s) into the s2 + ft,,2
sum of functions Zt( s). One important restriction is that all Zt( s ) must be
p.r. Certainly, if all Zt( s) were given, to us, we could synthesize a network jlcoK
) °
V — ft,2 + ft,,2 "
Here Re (10.97)
whose driving-point impedance is Z( s) by simply connecting all the Z{( s)
in series. However, if we were to start with Z( s) alone, how would we so that Z 2( s ) is p.r.
\S + ft>i /« -y<D

decompose Z(s) to give us the individual Zt( s ) ? Suppose Z [ s ) is given in V " Finally, if Re. [ Z( jw ) ] is minimum at some point and if the value of
general as
ansn + a „_,s" - + P( s)
1 •••
(10.92)+ n,s + a0 _ Re Z( ja> f ) = Kt 'as shown in Fig. 10.9, we can remove a constant K < Kt
from Re [Z(ya,)] so that the remainder is still p.r. This is because Re [Z(/co) j
= + b ,s + b0
Z(s)
bmsm + b^ s + ” Q( s )
1
will still be greater than or equal to zero for all values of co.
Consider the case where Z( s ) has a pole at s = 0 (that is, b0 = 0). Let us Suppose we have a p.r. function Z(s), which is a driving-point im-
divide P{ s) by Q{ s) to give a quotient Djs and a remainder R( s ), which pedance function . Let Z( s ) be decomposed , as before, so that
we can denote as Zx( s) and Z2( s ) . ,
Z( s) = Z (s) + Z ( s ) , (10.98)
Z(s) = -s + R( s ) D >0 ReZ( ju ) \
= Z,(s) + Z2(s) (10.93)
Are Z, and Z 2 p.r. ? From previous discussions, we know that Z, = D / s
is p.r. Is Z2( s ) p.r ? Consider the p.r. criteria given previously.
.
1. Z 2( s ) must have no poles in the right-half plane.
2. Poles of Z 2( s) on the imaginary axis must be simple, and their
w,-
residues must be real and positive. ^
3. Re [Z»(;« )] 0 for all ft,.
, .
FIG 10.9

Elements of realizability theory 311

310 Network analysis and synthesis
the remainder as Y2(J), as shown in Fig. 10.12. The remainder function Y2(J) is
] Z\( s ) [ p.r. because we have removed only the minimum real part of Y2(Jto ) Y2( s) .
is obtained as
Z( s ) Z2( s )
Y2( S ) = (10.103)

FIG . 10.10
It is readily seen that Y2( s ) is made up of a £-fl resistor in series with a f -farad
,( from Z( s ) to capacitor. Thus the final network is that shown in Fig. 10.13.
where both Z, and Z2 are p.r. Now let us “ remove” Z s)
. in Fig.
give us a remainder Z 2{ s) This removal process is illustrated
10.10 and shows that removal corresponds to synthesis of Z (
j). ,
Example 10.5. Consider the following p.r. function 3«
2a
s2 + 2s + 6 2n
Z( s ) = (10.99)
TOO Y 2( S ) Y <s )
4s + 3) ^
We see that Z( s ) has a pole at s = 0. A partial fraction expansion of Z( s ) yields .
FIG 10.12 .
FIG 10.13
2 s
Z( S ) =
J +7T1 :
Example 10.7 . Consider the p.r. impedance
= Z,(s) + Z2( s ) ( 10.100)
6s3 + 34 + 3s + 1
Z( s ) = (10.104)
,
If we remove Z (s) from Z( s ), we obtain Z2( s ), which can be shown
by a resistor 6s3 + 3s
in parallel with an inductor, as illustrated in Fig. 10.11 .
The real part of the function is a constant, equal to unity. Removing a constant
of 1 £2 , we obtain (Fig. 10.14)
34 + 1
,
Z (.s) = Z( s ) 1 =
64 + 3s -(10.105)

,
The reciprocal ofZ (s) is an admittance

Z( s ) Z2 64 + 3s
Yi( s) = 34 + 1
(10.106)

.
FIG 10.11 which has a pole at s = °o. This pole is removed by finding the partial fraction
expansion of T,(s) ;
Example 10.6 TiW = 2s + 34 + 1 (10.107)
Is + 2 (10.101)
Y( s ) = 2s + 4
where Y( s) is a p.r. function.
Ml
( ) ]. The real
Let us synthesize the network by first removing min [Re Y Jai °—ViAAA/V
part of Y( joj ) can be easily obtained as
8 + 14 a,2
Z( s ) Zi( s )
'

Re [ Y( ja> ) ] = 16 + 4a,2 (10.102)

, ,
We see that the minimum of Re [ Y( j(u ) ] occurs at a = 0 and
is equal to .
FIG 10.14
min [Re Y ja> = J
( )] . Let us then remove T =\ mho from Y ,
(s) and denote

i
 312 Network analysis and synthesis Elements of realizability theory 313
lfl 10.2 Determine whether the following functions are p.r For the functions .
o WWW with the denominator already factored, perform a partial fraction expansion
first.
Z ( jl y\
2 f
=tz to
s2 1
= s3 ++ 4s
m
2 s2 2s + 4
FIG. 10.15 to F( s ) = (s + +l )(r2 + 2)
and then by removing the term with the pole at J = a> to give a capacitor
of
( Fig. 10.15). Y ( s ) is now obtained as (s + 2 )( s + 4)
2 farads in parallel with Y 2( s ) below 2 to F( s ) =
(J + 1)(* + 3)
s
Y2( S ) = Fx(s) - 2s = 3s2 + 1 (10.108)
s2 4
(d ) F( s ) = .s3 + 3s2 ++ 3s + 1
The reciprocal of Y2( s ) is
1
Z2( s ) = 3s + - (10.109)
5s2 + s
to F( s ) = s2 + 1
which is, clearly, an inductor of 3 h in series with a capacitor of 1 farad . The
final network is shown in Fig. 10.16. 10.3 Suppose FjCv) and F2( s ) are both p.r. Discuss the conditions such that

° WWW
lfl 3h
—
F( s ) = Fj(s) F2( S ) is also p.r.
10.4 Show th'at the product of two p.r. functions need not be p.r. Also show
that the ratio of one p.r. function to another may not be p.r. (Give one example
of each.)
Z( s ) 2f if
=
:
10.5 Given Z(s) =
s2
s2
+ Xs
+ 5s + 4 '

(«) What are the restrictions on Xfor Z( s ) to be a p.r. function ?

FIG 10.16 . to Find X for Re [Z( jcu)] to have a second-order zero at co = 0.
These examples are, of course, special cases of the driving point syn - - ( c ) Choose a numerical value for X and synthesize Z( s ).
thesis problem. However, they do illustrate the basic techniques involved. 10.6 Prove that if Z2( s ) and Z2( s ) are both p.r.,
In the next chapter, we will discuss the problem of synthesizing a network
with two kinds of elements, either L C , R C , or R L networks. The syn-
- - - Z( s ) = Zt
(s)Z2(s)
thesis techniques involved, however, will be the same. Zi(s) + Z2(s)
must also be positive real.

= s2 ++ s ++ 1
Problems 2s2 s 2
10.7 Z( s) is p.r. Determine min [Re Z(/cu)] and synthesize
10.1 Test the following polynomials for the Hurwitz property. Z( s ) by first removing min [Re Z( jco )].
(«) s3 + s2 + 2s + 2 10.8 Perform a continued fraction expansion on the ratio
s* + s2 + s + 1
*
( )
to s7 + s5 + s2 + s s3 + 2s2 + 3s + 1
Y( s) =
(d ) s3 + 4s2 + 5s + 2 s 3
+ s2 + 2s + 1
to s® + 2s3 + s What does the continued expansion imply if Y( s ) is the driving point admittance -
(f) s7 + 2r* + 2s5 + s4 + 4s3 + 8s2 4- 8r 4- 4 of a passive network ? Draw the network from the continued fraction .

314 Network analysis and synthesis

s3 + 4s
to s2 4- 2
s +1
to s(s 4- 2) chapter II
2s 4- 4
to 2s 4- 3 i
Synthesis of one-port networks
s 4- 3s 4- 1
2

(d ) s2 4- 1
with two kinds of elements

In this chapter we will study methods for synthesizing one-port net-

works with two kinds of elements. Since we have three elements to choose
from, the networks to be synthesized are either R C , R-L, or L C networks. - -
We will proceed according to the following plan. First we will discuss the
properties of a particular type of one-port network , and then we will
synthesize it. Let us first examine some properties of L C driving-point -
functions.

II . I -
PROPERTIES OF L C IMMITTANCE FUNCTIONS

Consider the impedance Z( s ) of a passive one-port network. Let us

represent Z( s) as
:

Z ( s ) = Mt +
(s) NJs ) (HD
. M2(S) 4- Nt( s )
where Mx , M 2 are even parts of the numerator and denominator, and
Nu N 2 are odd parts. The average power dissipated by the one-port is
Average power =\ Re [Z(y& )]|f|2 j (11.2)
where / is the input current. For a pure reactive network, it is known that
the power dissipated is zero. We therefore conclude that the real part of
V - Z( jw ) is zero ; that is
-
Re Z ( jco ) - Ev Z ( joj ) = 0 (11.3)
(s) M2(s) - N s ) N 2( S)
where Ev Z(s) = Mt
M\2 S)
315
-
N2 ( S ) ^ (11- 4)
 316 Network analysis and synthesis
In order for Ev Z( jco ) = 0, that is,
Synthesis of one port networks 317 -
terms, i.e., no two adjacent terms of either polynomial may differ
Mjjco ) M 2( joi)
either of the following cases must hold :
— Nt( jco ) N 2( jco ) = 0 (11.5) than two powers. For example, for Z( s) given in Eq. 11.9, if b
then Z{ s) will have poles when
by more
3 = 0,

b3ss + bts = 0 (11.10)

(«) Mx = 0 = N 2
(11.6) so that the poles will be at s = 0 and at
(b) M 2 = O = N3
In case ( a ), Z( s ) is eH 2k -l ) vH k = 0, 1, 2, 3 ( 11.11)
(11.7) It is clearly seen that none of the poles sk are even on the jco axis, thus
M2
violating one of the basic properties of an L-C immittance function.
and in case (b ) Z(5) =
^
We see from this development the following two properties of L C
functions :
N2
(11.8)

-
From the properties given in Eq. 11.11, we can write a general L C
impedance or admittance as
K( s2 + co?)( s2 + m2 ) • • (s2 + o> * ) • • •
Z( s ) =
-
(11.12)
s(s2 + ft)22)(s2 + co? ) • • • (s2 + co?)
1. ZLC( s) or YLC( s) is the ratio of even to odd or odd to even poly- Expanding Z( s ) into partial fractions, we obtain
nomials.
2. Since both Mt( s ) and Njs ) are Hurwitz, they have only imaginary
roots, and it follows that the poles and zeros of ZLC( s ) or YLC( s ) are on
the imaginary axis.
Z(s) = ^+S
2
S
2
2 K 2S
+ ? CO s2
2 Kts
+ M4
- + • • • + Kxs (11.13)
where the K( are the residues of the poles. Since these poles are all on
Consider the example of an L-C immittance function given by the jco axis, the residues must be real and positive in order for Z ( s ) to be
positive real. Letting s = jco, we see that Z( jco ) has zero real part, and
-f
sa + b,_ s
q4s4 q 2s2 + fl0
Z(s) = 5
(11.9) can thus be written as a pure reactance jX( co ) Thus we have .
b5s + b3
Let us examine the constraints on the coefficients at and i,-. We know,
first of all, that in order for the impedance to be positive real, the coeffi-
ZQco ) = j (
V —— co
2 K 2CO
—
co? ft) 2 + • •
'
+ KnCo j
cients must be real and positive. We also know that an impedance
function cannot have multiple poles or zeros on the jco axis. Since co is
= jX ( co ) (11.14)
defined to be on the jco axis, the highest powers of the numerator and the Differentiating X( co ) with respect to co , we have
denominator polynomials can differ by, at most, unity For example, if . dX ( co ) _ K + Kx + 2/C ( > + co?) • • • 8 ft
2

—
0
the highest order of the numerator is 2n , then the highest order of the
( co? ft) 2)2 +
(11.15)
—
2
dco w
denominator can either be 2« 1 so that there is a simple pole at s = oo,
or the order can be 2n + 1 so that there is a simple zero at s oo.
Similarly, the lowest orders of numerator and denominator can differ by —. Since all the residues Kt are positive, it is seen that for an L C function,
dX ( co )
-
at most unity, or else there would be multiple poles or zeros of Z(s) at 0 (11.16)
s = 0. dco
Another property of the numerator and denominator polynomials is A similar development shows that the derivative of Im [ Y( jco)] = B( co )
that if the highest power of the polynomial is 2n , for example, the next is also positive, that is,
—
highest order term must be 2n 2, and the succeeding powers must
differ by two orders all the way through. There cannot be any missing
dB( co )
0
dco
(11.17)

L
318 Network analysis and synthesis Synthesis of one port networks - 319
Consider the following example. Z( s ) is given as X ( to )
Ks( s2 + co?)
Z(s) = 2 (11.18)

Letting s = jco , we obtain X( co ) as

(s + co?)( s2 + co?)
—o - ’
+1 +2 +T 03

Kco( — 2
+ co?)
—
ft)
Z( jco ) = j X ( co ) = + j
(
— ft)
2
+ ? co )(
Let us draw a curve of X( co ) versus co. Beginning with the zero at co = 0,
ft)
2
+ co? )
(11.19)
.
FIG 11.2

let us examine the sequence of critical frequencies encountered as co 2. The poles and zeros are simple and lie on the jco axis.
increases. Since the slope of the X( co ) curve is always positive, the next 3. The poles and zeros interlace on the jco axis.
critical frequency we encounter is when X ( co ) becomes infinitely large or 4. The highest powers of numerator and denominator must differ by
the pole is at co2. As we pass co2 , X ( co ) changes sign and goes from + unity ; the lowest powers also differ by unity.
to In general, whenever we pass through any critical frequency, 5. There must be either a zero or a pole at the origin and infinity.
there is always a change of sign, as seen from the way j X( co ) is written in
the last equation. After we pass through co2 , with the slope of X( co ) -
The following functions are not L C for the reasons listed at the left.
always positive, it is easy to see that the next critical frequency is the zero
Ks( s2 + 4)
at co3. Thus, if an impedance function is an L C immittance, the poles - 3. Z(s) = (s 2 l )(s2 3)
and zeros of the function must alternate . The particular X( co ) under + +
discussion takes the form shown in Fig. 11.1 / Since the highest powers s5 + 4s3 + 5s
of the numerator and the denominator always differ by unity, and the
2. Z(s) = 3s4 + 6s2 (11.20)
lowest powers also differ by one, we observe that at s = 0 and at s = co,
there is always a critical frequency, whether a zero or a pole. 1. K ( s2 + l )(s2 + 9 )
For the example just discussed , there is a zero at s = 0 and a zero at
Z(s) = (s2 + 2)(s2 + 10)
s = oo. The critical frequencies at s = 0 and s = co are called external On the other band , the function Z(s) in Eq. 11.21, whose pole-zero
critical frequencies, whereas the remaining finite critical frequencies are
referred to as internal. Thus, in the previous example, ft>2, co3, and ft>4
diagram is shown in Fig. 11.2, is an L C immittance. -
are internal critical frequencies.
Z( s ) =
2( sa + l)(s 2
+ 9)
-
Finally, let us summarize the properties of L C impedance or admittance s(s2 + 4)
(11.21)
functions.
1. ZLC( s) or YLC( S ) is the ratio of odd to even or even to odd poly - 11.2 SYNTHESIS OF L-C DRIVING POINT IMMITTANCES -
nomials.
-
We saw in Section 11.1 that an L C immittance is a positive real function
with poles and zeros on the jco axis only. The partial fraction expansion
X ( o> )
-
of an L C function is expressed in general terms as

o 03
F( s ) = -° + 2
s s + co2
+ + K „s ,
(11.22)
The synthesis is accomplished directly from the partial fraction expansion
by associating the individual terms in the expansion with network elements.
If F( s ) is an impedance Z ( s ), then the term K0 js represents a capacitor of
. .
FIG II I l / K0 farads; the term Kxs is an inductance of Kx henrys, and the term

!
 Synthesis of one- port networks 321
320 Network analysis and synthesis
,
1

^
2 f
2]?, h

-^ RRHh
jhf

p* i
Z? hh
Y( s )
ifa h

.
FIG 11.5
T
Z( s ) 3

The partial fraction expansion of Y( s ) is

FIG 11.3.

IKiS / is* + co,2) is a parallel tank circuit that consists

i , farads in parallel with an inductance of 2 Q co
of a capacitor of
/ 2 Thus a partial
,. .
y(s)
- s +?
tl ^ + 1 +
|s
(11.26)
1/2/: /
-
,

fraction expansion of a general L C impedance would -

yield the network from which we synthesize the network shown The L C in Fig. 11.6.
shown in Fig. 11.3. For example, consider the following L C function. - networks synthesized by partial fraction expansions are sometimes called
Foster-type networks.1 The impedance form is sometimes called a Foster
Z( s ) =
2(s8 + l )(sa + 9) (11.23) series network and the admittance form is a Foster parallel network.
s(s2 + 4) -
A useful property of L C immittances is that the numerator and the
denominator always differ in degree by unity. Therefore, there is always
A partial fraction expansion of Z(s) gives
,
Z(s) = 2s + I +
_ a zero or a pole at s = oo. Suppose we consider the case of an L C
impedance Z(s), whose numerator is of degree 2n and denominator
-

We then obtain the synthesized network in Fig. 11.4.

(11.24)
—
is of degree 2n 1, giving Z ( s ) a pole at s = oo. We can remove this
pole by removing an impedance Lrs so that the remainder function Z2(s)
-
is still L C :
The partial fraction expansion method is based upon the elementary
synthesis procedure of removing poles on the jco axis. The advantage
with L C functions is that all the poles of the function lie on the ju axis
-
Z 2( s) = Z( s) Lxs (11.27)
The degree of the denominator of Z 2( s ) is 2n 1, but the numerator is of
— —
so that we can remove all the poles simultaneously. Suppose F{ s ) in
Eq. 11.22 is an admittance Y{ s ). Then the partial fraction expansion of
—
degree 2n 2, because the numerator and denominator must differ in
degree by 1. Therefore, we see that Z2(s) has a zero at s = oo. If we
y( j) gives us a circuit consisting of parallel branches shown in Fig. 11.5. invert Z 2( s ) to give y3(s) = 1/ Z2(s), y2(s) will have a pole at s = oo ,
For example, which we can again remove to give a capacitor C 2s and a remainder
8
(s2 4) Y 3( S ), which is
y(s) = s( s + 2) 8 + (11.25) Y 3( S ) = Y 2( S ) C2 s. -
(11.28)
( s* + l)(s + 3)

£f 2h h
o
2h if
— KSISISU — Y( sl
if

if
T ’ifT
Z( s ) .
FIG 11.6

FIG 11.4 . 1R . M. Foster, ‘‘A Reactance Theorem,” Bell System Tech. J., No. 3 (1924), 259-267.

322 Network analysis and synthesis -

Synthesis of one port networks 323

c
'mnp
2h
—
o oytyip

-
71 La
Ci
z Zi_
2h
if
§h
Remainder

.
FIG 11.10
.
FIG 11.7 FIG 11.8.
as shown in Fig. 11.10. The admittance y4(s) = 1/ Z4(s) has a pole at
We readily see that oo, which we can invert and
Ya( s) has a zero at s =
is zero. Each time
s = oo, which we remove to give a capacitor of J farad and a remainder
remove. This process continues until the remainder
we remove a pole, we remove an inductor or a capacitor depending upon
Yb( s ) = 3/2s, which represents an inductor of f h. Removing this in -
ductor gives us zero remainder. Our synthesis is therefore complete and
whether the function is an impedance or an admittance. Note that the the final network is shown in Fig. 11.11.
final structure of the network synthesized is a ladder whose series arms Since we always remove a pole at s = oo by inverting the remainder
are inductors and whose shunt arms are capacitors, as shown in Fig. 11.7. and dividing, we conclude that we can synthesize an L-C ladder network
Consider the following example. by a continued fraction expansion. The quotients represent the poles at
2s5 + 12s3 + 16s (11.29) s = oo, which we remove, and we invert the remainder successively until
Z(s) =
s4 + 4s8 + 3 the remainder is zero. For the previous example, the continued fraction
We see that Z(s) has a pole at s = co, which we can remove by first expansion is
dividing the denominator into the numerator to give a quotient 2s and a
remainder Z 2( s ), as shown in Fig. 11.8 Then we have . s4 + 452 + 3)2s5 + Us3 + 16J(2J <-> Z
2s* + 8s3 + 6 s
(11.30) 4 ss + 105 4 + 4i2 + 3(ir <-> y >
s* + fi2
v
Observe that Z 2( s) has a zero at J = oo. Inverting Z2( s ) , we again remove
the pole at infinity. Then we realize a capacitor of\ farad and a remainder
I*2 + 3)4i3 + 10s(
4J3 + 8$
> Z

y3( j), as may be seen in Fig. 11.9. y

)- J2 + 3(fj
2J|
as8 3
s= 3 + —
„/ x 1 (11.31)
F3(s) = YJs )
4 4s + 10s
3)2s(Js <~> Z
Removing the pole at s = oo of Z3(s) = 1/ y3(s), gives a series inductor 2s
of f h and .

Z4(s)
— —^
Z3(s)
8
s — 3 2
^
2s
2
(11.32)
:
We see that the quotients of the continued fraction expansion give the
elements of the ladder network. Because the continued fraction expansion
2h
o
'UUIP o
—nnfiT'— —'TTuiT'— — —
2h
fh §h
if 4 = Ya( s ) Ifi if

.
FIG 11.9 FIG. II II .
 Synthesis of one-port networks 325
324 Network analysis and synthesis
statement, it can be said that both the Foster and the Cauer forms give
always inverts each remainder and divides, the successive quotients alter- the minimum number of elements for a specified L-C driving-point
nate between Z and Y and then Z again, as shown in the preceding function. These realizations are sometimes known as canonical forms.
expansion. If the initial function is an impedance, the first quotient must j
necessarily be an impedance. When the first function is an admittance,
the first quotient is an admittance.
-
11.3 PROPERTIES OF R C DRIVING POINT IMPEDANCES -
Since the lowest degrees of numerator and denominator of an L C - -
The properties of R C driving-point impedances can be derived from
admittance must differ by unity, it follows that there must be a zero or a -
known properties of L C functions by a process of mapping the jco axis
pole at s = 0. If we follow the same procedure we have just outlined ,
and remove successively poles at s = 0, we will have an alternate real-
—
onto the a axis.3 We will not resort to this formalism here. Instead,
we will assume that all driving-point functions that can be realized with
ization in a ladder structure. To do this by continued fractions, we two kinds of elements can be realized in a Foster form. Based upon this
a r r a n ge both numerator and denominator in ascending order and divide assumption we can derive all the pertinent properties of R C or R L - -
the lowest power of the denominator into the lowest power of the numer - driving-point functions. Let us consider first the properties of R C -
ator ; then we invert the remainder and divide again . For example, in driving-point impedance functions.
the case of the impedance we have Referring to the series Foster form for an L-C impedance given in
(s2 + l )(sa + 3) (11.33)
Fig. 11.3, we can obtain a Foster realization of an R C impedance by -
Z(s) = simply replacing all the inductances by resistances so that a general R-C
s(s2 + 2)
impedance could be represented as in Fig. 11.13. The R C impedance, -
The continued fraction expansion to give the alternate realization is as seen from Fig. 11.13, is
2s+ s3)3 + 4i2 + \ s 3l 2s Z K , K2
3 + -f .s 2

fs2 + s* )2s + S3(4/5J <-> Y

Z(s) =
^+
.s
K „+
+ ff 2
+ •••
S + <T, S
(11.34)

where C0 = 1 jK0 , Rm = Km , Cj = 1/ Klr Rt = Kjo and so on. In

2s + js3 ^
) J2 + si( 25 j2s <-> Z
order for Eq. 11.34 to represent an R C driving point -
impedance , the -
I*3! constants Kt and oi must be positive and real. From this development ,
I*2 two major properties of R-C impedances are obtained , and are listed in
s4)Ji3( l / 5 j <-*. Y the following.
h3 Cl C2

The final synthesized network is shown in Fig. 11.12. "f he ladder networks
Hh AA/^-
realized are called Cauer ladder networks because W. Cauer 2 discovered
the continued fraction method for synthesis of a passive network.
Note that for both the Foster and the $
c 0 VW
Hi
Wv
E2
—
Cauer-form realizations, the number of
If elements is one greater than the number of
If
5h
internal critical frequencies, which we defined
previously as being all the poles and zeros r
FIG . 11.13
of the function, excluding those at 5 = 0 and
FIG . 11.12 j
= oo. Without going into the proof of the 5
M. E. Van Valkenburg, Introduction to Modern Network Synthesis, John Wiley and
Sons, New York, 1960, pp. 140-145.
‘ Wilhelm Cauer, “The Realization of Impedances with Prescribed Frequency
Dependence,” Arch. Electrotech., 15 (1926), 355-388. :i

326 Network analysis and synthesis Synthesis of one-port networks 327

-
1. The poles of an R C driving-point impedance are on the negative
—
real ( a) axis. It can be shown from a parallel Foster form that the
poles of an R-C admittance function are also on the axis. We can thus
Next, let us see whether the poles and zeros of an R-C impedance
function alternate. We have already established that the critical frequency
nearest the origin must be a pole and the critical frequency nearest a = oo
-
conclude that the zeros of an R C impedance are also on the a axis.
2. The residues of the poles, Kit are real and positive. We shall see later
— must be a zero. Therefore, if Z(r) is given as

that this property does not apply to R C admittances. - Z(s) = (s +

(s <r3)(s+ <r4) (11.40)
Since the poles and zeros of R-C impedances are on the a axis, let
us examine the slope of Z(or) along the a axis. To find the slope, — — -
+ Oi )(s + <T3)
Then, if Z( s ) is R C , the singularity nearest the origin must be a pole
dZ ( o)/ do, we first let s = a in Z(s), and then we take the derivative of which we will assume to be at s = - or,; the singularity furthest from
Z ( o) with respect to a. Thus we have the origin must be a zero, which we will take to be s = ot . Let us plot
—
Z ( o) =
^
a
2
+ Ox
K
+ *«, + a *1 + a o +
+
. 2

2
• • • (11.35) Z(cr) =
( o + o2 )( o
( a + ffjXff
+ <r4)
+ o3 )
(11.41)
_
and
dZ ( o )
do
K0
ff
2
(o
dZ ( o )
+ ffj)
2
~K2
( a + o2 f
+ (11.36) —
versus a, beginning at a = 0 and extending to a =
Z(0) is equal to a positive constant — oo. At o — 0,

Z(0) = £s s
®
It is clear that
do
<0 (11.37) (11.42)
°i°2
Let us now look at the behavior of Z(s) at the two points where the
real axis and the imaginary axis intersect, namely, at <r = to = 0 and at
Since the slope of Z(o) is always positive as a increases, Z(cr) must
—
increase until the pole s = oq is reached (Fig. 11.14). At o = olt
— —
a = co = oo. This is readily done by examining the general R C network - Z( o) changes sign, and is negative until the next critical frequency is
in Fig. 11.13 at these two frequencies. At a = 0, (d c), if the capacitor C0 - reached . We see that this next critical frequency must be the zero,
is in the circuit, it is an open circuit and there is a pole of Z ( s ) at a = 0. —
s = o2 . Since Z ( o ) increases for increasing a, the third critical
— —
——
If C0 is not in the circuit, then Z(0) is simply the sum of all the resistances frequency must be the pole s = o3. Because Z( o) changes sign at CT3,
in the circuit.
Z(0) = Rx + R2 + • • • + R „ (11.38)
the final critical frequency must be the zero, s
——
cr4. Beyond o = at ,
the curve becomes asymptotic to Z( co) = 1. From this analysis we see
that the poles and zeros of an R-C impedance must alternate so that for
I
because all of the capacitors are open circuits at a = 0.
At a = co, all the capacitors are short circuits. Thus, if is in. the Z ( <r )
.
circuit, Z( oo) = Rx If Rm is missing, then Z( oo) = 0. To summarize
these last two statements, we have
oo, C0 present
i
Z(0) = m .i.. 2( 0 )
I Rt , „
C missing ~
Z( )
^ !
i

0, Rx missing
Z( oo) =
RM present
If we examine the two cases for Z(0) and Z( co), we see that
i
Z(0) > Z( co) (11.39)

.... .
 328 Network analysis and synthesis -
Synthesis of one port networks 329

the case being considered -

Fig. 11.15. We see that an R C impedance, ZRC( s ), also can be realized
as an R-L admittance YBL( s ). All the properties of R- L admittances are
oo > n4 > a3 > a2 > oj 0. * (11.43) -
the same as the properties of R C impedances. It is therefore important

In addition, we see that l (11.44)

to specify whether a function is to be realized as an R C impedance or an -
<*1
R-L admittance.
°3
which shows that Z(0) > Z( co).
To summarize, the three properties we need to recognize an R C - -
11.4 SYNTHESIS OF R C IMPEDANCES OR R L ADMITTANCES -
impedance are :
We postulated in Section 11.3 that the Foster form realization exists
1. Poles and zeros lie on the negative real axis, and they alternate. -
for an R C impedance or an R- L admittance. Since Foster networks are
2. The singularity nearest to (or at) the origin must be a pole whereas synthesized by partial fraction expansions, the synthesis is accomplished
the singularity nearest to (or at) a = oo must be a zero.
3. The residues of the poles must be real and positive.
— with ease. An important point to remember is that we must remove the
minimum real part of Z( jco ) in the partial fraction expansion It can be .
shown 4 that min [Re Z ( ja> ) ] = Z( oo), so that we have to remove Z( oo)
An example of an R C impedance is : - as a resistor in the partial fraction expansion. In cases where the numer -
l)(s + 4)(s + 8)
Z(s) = +
(s ator is of lower degree than the denominator, Z( oo) = 0. When the
(11.45)
s( s + 2)(s + 6) numerator and the denominator are of the same degree, then Z( oo) can
be obtained by dividing the denominator into the numerator. The
The following impedances are not R-C.
quotient is then Z( oo). Consider the following example.
l )(s + 8)
Z(s) = +
(s
(s + 2)(s + 4) F(s) =
3(i ±s( s2_)(£3)± 4) (11.48)
(s 2)(s + 4) +
Z(s) = + (11.46)
(s + 1) The partial fraction expansion of the remainder function is obtained as
( s + l)(s + 2)
Z(s) = 8 1
s(s + 3) F(s) = - + +3 (11.49)
-
s s +3
Let us reexamine the partial fraction expansion of a general R C
impedance. where F( oo) = 3. If F( s ) is an impedance Z(s), it must be an R C im- -
pedance and it is realized in the series Foster form in Fig. 11.16. On the
^ ^ -
*
F(S) = + K„ + - + • • (H 47)
S s + ai '
other hand, if F( s ) represents an admittance, we realize Y( s) as an R L -
.
network in the parallel Foster form (Fig. 11.17).
Instead of letting F( s ) represent an impedance, consider the case where
F( s) is an admittance Y( s ). If we associate the individual terms in the
expansion to network elements, we then obtain the network shown in
:
o
—— — w—
1(
if
AV
30
/
in
if
Z( s )

!
^ h
.
FIG 11.16

.
FIG 11.15 * Van Valkenburg, loc cit .

1
Network analysis and synthesis
330 Synthesis of one port networks - 331

3 ft 11.5 PROPERTIES OF R L IMPEDANCES AND

-
R C ADMITTANCES
-
Y( s )
Sh ini lh The immittance that represents a series Foster R L impedance or a -
-
parallel Foster R C admittance is given as

.
FIG 11.17
F( s ) = KKS + K 0 + s Kts< + (11.50)
+ Ji
An alternate method of synthesis is based on the following fact. If The significant difference between an R C impedance and an R L im- - -
—
we remove min Re [Z( jeo ) ] Z( oo) from Z{ s ), we create a zero at s = oo
for the remainder Z s). If we invert Z s ), we then have a pole at s = co,
pedance is that the partial fraction expansion term for the R-C “tank”
-
^ ^
which we can remove to give Z 2( s ). Since min Re [ Y 2( ja> ) ] = Y 2( oo), if
circuit is KJ ( s + a{ ) ; whereas, for the R L impedance, the corresponding
term must be multiplied by an s in order to give an R L tank circuit -
we remove Fz( oo), we would have a zero at s = co again, which we consisting of a resistor in parallel with an inductor.
again invert and remove. The process of extracting Z( co) or Y( co ) and -
The properties of R L impedance or R C admittance functions can be -
the removal of a pole of the reciprocal of the remainder involve dividing derived in much the same manner as the properties of R C impedance -
the numerator by the denominator. Consequently, we see that the whole functions. Without going into the derivation of the properties, the more
synthesis process can be resolved by a continued fraction expansion. significant ones are given in the following :
The quotients represent the elements of a ladder network .
For example, the continued fraction expansion of F( s ) in Eq. 11.48 is 1. Poles and zeros of an R- L impedance or R-C admittance are located
on the negative real axis, and they alternate.
s2 + 3J)3S2 + 18r + 24(3 2. The singularity nearest to (or at) the origin is a zero. The singularity
3s2 + 9s nearest to (or at) s = oo must be a pole.
3. The residues of the poles must be real and negative .

s2 + Is
>
9s + 24 2 + 3s($s
Because of the' third property, a partial fraction expansion of an R L -
i>s )9 s + 24(27 impedance function would yield terms as
9s
(11.51)
24) (s/72
^
is
s + Ci
This does not present any trouble, however, because the term above does
. ..
not represent an R- L impedance at all. To obtain the Foster form of an
If F( s ) is an impedance Z( s ) , the resulting network is shown in Fig. R L impedance, we will resort to the following artifice. Let us first expand
-
11.18. IfF( j) is an admittance 7(i), we have the R L network of Fig. 11.19. - '

Z( s )/ s into partial fractions. If Z( s ) is an R- L impedance, we will state

°
—
Z( s )
Wv
3 ft

if = 4=
w
27 ft^
-
72f "p
Y( s )
In
Jh
^ h
without proof here that the partial fraction expansion of Z( s )/ s yields
positive residues.5 Thus, we have
Z( s ) K0
s
= s
_ + K„ +
s
Ki
+ at
+ ... (11.52)

.
FIG 11.18 .
FIG 11.19
5
Actually, ZXL( s )ls has the properties of an R-C impedance; see Van Valkenburg
loc . cit .
 Synthesis of one- port networks 333
332 Network analysis and synthesis
W
N/ If
o | Af
WW * f
In:
in — Ah — vMiU 212 /
yfs) 2 fi ¥n

Z( s )
4 fi
fn :
^
Y ( sl FIG. 11.23

. we can then remove the pole at s = 0. Since the value Z(0) is obtained
FIG. 11.20 FIG 11.21
by dividing the lowest power of the denominator into the lowest power
where K0 , Ku .. . -
, K„ 0. If we multiply both sides by s, we obtain term of the numerator, the synthesis could be carried out by a continued
fraction expansion by arranging the numerator and denominator poly-
Z(s) in the desired form for synthesis. Consider the following function :
nomials in ascending order and then dividing. For example, the following
F( s ) =
2(s + l )(s + 3) (11.53) -
function is either an R C impedance or an R C admittance. -
( s + 2)(s + 6)
F( s ) =
2(s + l)(s + 3) _ 6 + 8s + 2s 2
(11.57)
-
F( s ) represents an R L impedance or an R C admittance because it - (s + 2)(s + 6)
12 + 8s +s 8

satisfies the first two criteria cited. The partial fraction expansion of The continued fraction expansion of F( s ) is
F( s ) is
XA
F( S ) = 2
2 $+6
-7T (11.54) 12 + 8s + s2)6 + 8s + 2s2(£
6 + 4s +| s2
so we see that the residues are negative. The partial fraction expansion of 4s + fs2)12 + 8s + s2(3/s
F( s )[ s, on the other hand, is
_
F( s) 2(s + l )(s + 3) 1 (11.55)
12 +
| s
Is + s2)4s +| (
2 -
s|
s s(s + 2)(s + 6) s s +2 s +6 4s + fs*
If we multiply both sides by s, we obtain AS2|) -S + s2(49/5s
is is 1 (11.56) Is
F( s ) = ~ +
s+2 s+6 s8)iV(A
If F( s ) represents an impedance Z(s), it is synthesized in series Foster As8
-
form, giving the R L network in Fig. 11.20. If F( s ) is an admittance Y ( s ) ,
then the resulting network is the R-C network shown in Fig. 11.21. If F( s ) is an impedance function, the resulting network is the R- L network
1
'

To synthesize an R- L impedance in ladder form, we make use of the shown in Fig. 11.22. If, on the other hand, F( s) is an R-C admittance
fact that min Re [ Z( j<o ) ] = Z(0). If we remove Z(0) from Z(s), the y(s), the network is synthesized as in Fig. 11.23 .
remainder function Zt(s) will have a zero at s = 0. After inverting Zx(s),

o
— Wv
in
ih
Wv
?n
Wv—
An
11.6 SYNTHESIS OF CERTAIN R - L- C FUNCTIONS

-- -
Under certain conditions, R L C driving point functions may be syn-
thesized with the use of either partial fractions or continued fractions.
Z( s ) , For example, the function
58 2s 2
Z( s ) = 2 + + (11.58)
FIG 11.22. s +s+1

334 Network analysis and synthesis -

Synthesis of one port networks 335
-
is neither L C , R-C, nor R- L. Nevertheless, the function can be synthesized
by continued fractions as shown .

s2 + s + l) + 2s + 2(1 Z
s2
Y( s ) fn
fa 60
1
s2 + s + 1 ^ |h
s + l )s2 + s + 1(5 Y
s2 + s
.
FIG 11.25
1 > ++ 11( + 1 ^- z
J
J

The network that is realized from the expanded function is given in Fig.
11.25.
The network derived from this expansion is given in Fig. 11.24.
If we try to expand T(s) by continued fractions, we see that negative
In another case, the poles and zeros of the following admittance are all
quotients result. However, we can expand Z( s) = 1/ Y( s ) by continued
on the negative real axis, but they do not alternate.
fractions, although the expansion is not as simple or straightforward as

°— vW
Hi lfl
Y( s ) =
(s
(s
+ 2)(s + 3)
+ l)(s + 4)
(11.59)
in the case of an R-C function, because we sometimes have to reverse the
order of division to make the quotients all positive. The continued
fraction expansion of Z(s) is
ii f lh ' The partial fraction expansion for Y( s ) is
6 + 5s + s2)4 + 5s + J2(f
FIG. 11.24 Y ( s) = 1 + —s— +
71
s+ s
±
Z
+4
(11.60) 4+
|
+ 2
^
5 + JJ 2)6 + 5.s + i2(18/ 5
^
Since one of the residues is negative, we cannot use this expansion for 6 + fJ
synthesis. An alternate method would be to expand Y( s)/ s and then l g S + S 2 )$ s2 +
multiply the whole expansion by s . is2 + Hs
Y( s ) f B As )?s + S 2( ¥
(11.61)
s 5 s +1 s +4
s2 ) As( 6l\ 5s
When we multiply by s , we obtain, * As
y(s) = -2 — —— s
~
+1
Note that Y( s ) also has a negative term. If we divide the denominator of
1
s +4
(11.62)
§
i:
As we see, the division process giving the quotient of 1/3 involves a
reversal of the order of the polynomials involved. The resulting ladder
network is given in Fig. 11.26.
this negative term into the numerator, we can rid ourselves of any terms
with negative signs.
°— In Wv wvo
4
if
V5 f
- K2 - 7TT) 9S

Z( s )
Ah An

-i | h
+s 1+s +4
+
(11.63)

I
FIG. 11.26
 336 Network analysis and synthesis Synthesis of one- port networks 337
In the beginning of this section, it was stated that only under special
-
conditions can an R- L C driving-point function be .synthesized with the fh 2h
use of a ladder form or the Foster forms , These conditions are not lh
given here because they are rather involved . Instead , when a positive Z( s ) if
real function is given, and it is found that the function is not synthesizable 1f
by using two kinds of elements only, it is suggested that a continued ^p
fraction expansion or a partial fraction expansion be tried first. PROB. 11.4

Problems
11.5 The input impedance for the network shown is
-
11.1 (a) Which of the following functions are L C driving point impedances ?
Why ? 2s2 + 2
sO2 + 4)02 + 16) 02 + 1)0* + 8) Z\a =
s3 + 2s2 + 2s + 2
-
Zi(s) = 02 + 9 )( s + 25) ’ Z2{ s) = s( s2 + 4)
If Z0 is an L-C network : (a) Find the expression for Z0 . (b) Synthesize Z0 in a
( b ) Synthesize the realizable impedances in a Foster and a Cauer form . Foster series form.
11.2 Indicate the general form of the two Foster and the two Cauer networks
that could be used to synthesize the following L C impedance. -
Z( s)
* 1)0* + 9)02 + 25)
=0 +
T Zo
s( s2 + 4)(i2 + 16)
^ in

There is no need to calculate the element values of the four networks.

11.3 Synthesize the L-C driving point impedance-
T.
PROB 11.5

6s4 + 42s2 48
Z( s) = s5 + 18s3 +48s 11.6 Indicate which of the following functions are either R C , R-L, or L-C-
+
impedance functions.
in the form shown in the figure, i .e., determine the element values of the network
s3 2s
in henrys and farads. («) Z(s) = s4 + 4+s2 + 3
Li
s2 + 6s + 8
(b) Z(s) = 2
Ci j
s + 4s + 3
Z( s ) Ci
Ci
=fC
3
(c) Z(s)
s2 + 4s + 3
= s2 + 6s + 8

.
PROB 11.3
id ) Z(s) =
s2 + 5s + 6
s2 + s
-
11.4 There exists an L C network with the same driving-point impedance as
the network shown in the figure. This alternate network should contain only s4 + 5s2 + 6
.
two elements Find this network.
(e) Z(s) = s3 + s

338 Network analysis and synthesis Synthesis of one- port networks 339
11.7 An impedance function has the pole-zero pattern shown in the figure.
—
If Z( 2) = 3, synthesize the impedance in a Foster form and a Cauer form.
-
11 12 Find the networks for the following functions in one Foster and one
Cauer form.
0 + 1)0 + 3)
TOO =
jo> 0 + 2)(s + 4)
2Q + 0.5)Q + 4)
Z( s ) =
s( s + 2)
11.13 Synthesize the following functions in Cauer form.
s3 + 2s2 + s + 1
Z( s ) = s3 + s2 + s
s3 + s2 + 2s + 1
Z( s ) = 44 + J8 + 3J* + J + 1
PROB 11.7 .
4s3 + 3s2 + 4s + 2
Z( s) =
11.8 From the following functions, pick out the ones which are R C admit-
tances and synthesize in one Foster and one Cauer form.
- 2s2 + s
0 + 0 + into 2) 4)
the form shown in the figure.
+ 1)0 + 3)
2(s + 1)Q + 3)
4Q 11.14 Synthesize Z( s ) =0 + 1 )0 + 5)
TO) = 0 + 2)0 + 4) TO) =
sis + 2 )
_ + 4)0 + 8)
TO) sO
Q + 1)Q + 4)
TO) = sis + 2)
0 + 1)0 + 6)
11.9 Find the networks for the following functions Both Foster and ladder.
forms are required.
ZJsl
Zis ) = 0 + 0 2)
1) + 4)
(«)
sis +
Zis ) 30 + 1)0* + 4) PROB 11.14 .
ib) s+3 11.15 Of the three pole-zero diagrams shown, pick the diagram that repre-
11.10 For the network shown, find Y when sents an R- L impedance function and synthesize in a series Foster form.
1
Vi jos
2 + T
V0
sjs2 + 3)
2s + s3 + 6s + 1
3
-4 - 3 -2 - 1 a

-
Synthesize Y as an L C admittance.
J 01

+
i T +

Toj Y i v2
T joi
PROB 11.10.
11.11 Synthesize by continued fractions the function
s3 + 2s2 + 3s + 1
TO) = s3 + s2 + 2s + 1
. nr^T—fo
.
PROB 11.15
—
it
 340 Network analysis and synthesis
-
11.16 Synthesize a driving point impedance with the pole zero pattern shown
in the figure in any form you choose. { Hint: Use uniform loading concepts.)
-
1 ju

V2 *
v 1
12 chapter
i' Elements of transfer
1

4 function synthesis
PROB 11.16 .
11.17 Following are four successive approximations of tanh s.
.. 3s J2 + 15r
<a) ?T3 (6)
6s2 + 15 12.1 PROPERTIES OF TRANSFER FUNCTIONS
10s3 + 105s s5 + 105s3 + 945s
^ + s4 45s2 + 105 15s4 + 420s2 + 945
A transfer function is a function which relates the current or voltage
at one port to the current or voltage at another port. In Chapter 9 we
Synthesize networks for the functions above whose input impedances approxi
mate tanh s.
- discussed various descriptions of two-port networks in terms of the open-
circuit parameters zif and the short-circuit parameters yit . Recall that for
the two-port network given in Fig. 12.1, the open circuit transfer im- -
pedances Zj2 and z21 were defined as

* .-*1 h l / =0
l
(12.1)

h 1 / =0
2

-
In terms of the open circuit transfer impedances, the voltage-ratio transfer
function is given as
Yl hk - (12.2)
Vi
-
In terms of the short circuit parameters, the voltage ratio is shown to be

(12.3)
Vx Vn
When the network is terminated at port two by a resistor R, as shown in
Fig. 12.2, the transfer impedance of the overall network is

Z21
_ — *^
_ Yl 21
(12.4)
h * + 22 R
341

.
.
.

342 Network analysis and synthesis Elements of transfer function synthesis 343
& I*
t * ft Two-port
+ and is an even function in a>. The phase response is
Vi Two-port V2 Vi network
R <£ v2
network
2 Arg T( jco ) = arctan arctan (12.10)
.
FIG 12.1 .
FIG 12.2
If arg T( j0 ) = 0, we see that the phase response is an odd function in to.
The transfer admittance of the overall structure in Fig. 12.2 is

F21 Jz
_ _
2/ 21G (12.5)
Now let us discuss some specific properties of the open-circuit and
short-circuit parameters.
Vi 2/22 + G 1. The poles of z21( j) are also the poles of zn(y) and z22(s). However,
.
where G = IjR When both ports are terminated in resistors, as shown in not all the poles of zu( j) and z22(s) are the poles of z2l( j). Recall that in
Fig. 12.3, the voltage ratio transfer function F2/ F„ is
- Chapter 9 we defined the z parameters in terms of a set of node equations
F2 *21- ^ as
—
2
= ( 11 + ( 12.6)
* 1» )
^ *21
l * )(*22
12 + ^- 2 An
Other transfer functions such as current ratio transfer functions can *!l(S) = A *»(*) = AA
-
also be described in terms of the open and short circuit parameters. In -
Chapter 10 we discussed the various properties of driving point imped
ances such as zu and z22. This chapter deals with the properties of the
- - *Z12 — —.
*z21 A
transfer immittances z21 and y2l for a If there is no canpellation between each numerator and denominator ofzu,
passive reciprocal network. First, let z22, and Zj2, then the poles are the roots of the determinant A, and all three
_
I + us discuss certain properties which functions have the same poles. Consider the two port network described -
fti Two-port S apply- to all transfer functions of passive by the black box in Fig. 12.4a. Let z'u, z'22, and z'i2 be the z parameters
+ network R S Vz linear networks with lumped elements. of the network. Let us examine the case when we attach the impedances
*
We denote a transfer function as T( s ). Zx and Z2 to ports one and two, as shown in Fig. 12.46. The z parameters
1. T ( s ) is real for real s. This property -
for the two port network in Fig. 12.46 are
FIG. 12.3 is satisfied when T( s) is a rational ,
function with real coefficients. *11 = z n + Zj
-
2. T( s ) has no poles in the right half plane and no multiple poles on *22 = z 22 4 Z2
,
"

the ja) axis. If T(s) is given as T( s ) = P( s )/ Q( s ), the degree of P( s ) cannot

exceed the degree of Q( s ) by more than unity. In addition , Q( s ) must be *12 = z 12
a Hurwitz polynomial. It is clear that the poles of zn include the poles of the poles of z 22
3. Suppose P( s ) and Q{ s ) are given in terms of even and odd parts, include the poles of Z2. However, the poles of z12 include neither the
that is, poles of Z, nor Z2. Consequently, we see that all the poles of z12 are also
T(s)
_
p(s) Mi( ) + NM
* (12.7) poles of zn and z22. The reverse is not necessarily true.
Q( s ) M2(s) + iV2(s)
where Mf ( s) is even and Nt( s ) is odd. Then T( JOJ ) is h h
{ joy ) + Ni( jco )
O- -O
rHn£-
T ( jco ) = Mx (12.8) Vi
z'll , z'22
v2 Vi * '11, z'22
v2
Ms jco ) + Na( jeo)
( z'l2 Z'l2

The amplitude response of T( ja> ) is

'A
^ ^]
i

^
( jc0 ) + N (a) ( b)
\T( ja) )\ = ° (12.9)
.
W + N *(Ja>) FIG 12.4

i ..
 344 Network analysis and synthesis Elements of transfer function synthesis 345

h 3f h h
h h
-o o- -o +
+ V2
VI Yi
.
y'n y'22
y'12
Y2 v2 Vi ih '
ih v2

.
FIG 12.5 .
FIG 12.6
h h
i
2. The poles of y12( s) are also the poles of yn( s ) and y 22( s ). However, + +
not all of the poles of yn( s ) and y 22( s ) are the poles of y12(s) . This property Vi V2
is readily seen when we examine the two-port network in Fig. 12.5 .
The y parameters are
Vn = it' ll + Yi
FIG. 12.8
y22 = y’ 22 +

Vn = y\B 12.2 ZEROS OF TRANSMISSION

Clearly, the poles of ylt( s ) do not include the poles of either Yx and Y 2.
A zero of transmission is a zero of a transfer function. At a zero of
Consider the network in Fig. 12.6. The y parameters are transmission, there is zero output for an input of the same frequency.
2 For the network in Fig. 12.7, the capacitor is an open circuit at s = 0,
Jfuto = - + 3s so there is a zero of transmission at s = 0. For the networks in Figs.
12.8 and 12.9, the zero of transmission occurs at s = dcj/ y/ LC . For
2/22(5) = -s + 3s
4
'

the network in Fig. 12.10, the zero of transmission occurs at s = 1/ RC

In general, all the transfer functions of a given network have the same
— .
zeros of transmission, except in certain special cases. For example if
2/12(5) = -3s Z12(J) has a zero of transmission at s = sx , than y12( s ), V 2( s ) jV1( s ) etc.,
’
l

Observe that yu( s ) and y 22( s ) have poles at s = 0 and s = co, whereas
y12( s ) only has a pole at s = 00.
3. Suppose 2/11(5)» 2/ 22(5), and 2/12(5) all have poles at s = sv Let us
denote by kn the residue of the pole at sx of the function yn( s ) The .
—
residue of the pole s sx of y 22( s ) will be denoted as k 22 , and the residue
of the same pole of 2/12(5) will be denoted as k12. Without going into the
-
proof,1 a general property of L C , R-C , or R L two-port networks is that -
k\\k 22 kx 2 > 0 (12.11)
1
:
This equation is known as the residue condition. For example, for the
-
L C network in Fig. 12.6, the residue condition applied to the pole at
s = 00 gives 3 X 3 - 3a = 0; whereas for the pole at s 0, we have
—
2 x 4 02 = 8 > 0. Thus we see that the residue condition is fulfilled
—
for both poles.
. .
1 For a general discussion, see M E Van Valkenburg, Introduction to Modern
.
FIG 12.10
Network Synthesis, John Wiley and Sons, New York, 1960, pp. 305-313.
;

346 Network analysis and synthesis Elements of transfer function synthesis 347

rd } | l ~~ 23 (~ ~i
| hj
Zr> ~
L C -
ladder

Vi Y2 y4 Ye
.
FIG 12.13

T T T. .
.
FIG 12.11 '
transmission at s = 00 due to the elements Lx and C3. It would also have a
zero of transmission at s = 0 due to Clt a zero at s = 1 / R2C2 due to —
-
will also have a zero at s = sx . This fact is clearly seen when we examine
the relationships between the transfer functions. For example, we have
the parallel R C branch, and a zero of transmission at s = jl{ L2C A
-
due to the L C tank circuit . It is seen that none of the transmission
zeros are in the right-half plane. We also see that a transfer function may
^
2/ 21
possess multiple zeros on the jeo axis.
-
«21 =
2/112/22
Z 2l
— 2/122/21
(12.12) In Section 12.4 it may be seen that lattice and bridge circuits can easily
be nonminimum phase. It can also be demonstrated that when two ladder
networks are connected in parallel, the resulting structure may have
and 2/21 = - zllZM
212Z21
(12.13)
right-half-plane zeros.2
-
In addition, the voltage- and current ratio transfer functions can be
expressed in terms of the z and y parameters as 12.3 SYNTHESIS OF Y21 AND Z21 WITH A l -« TERMINATION
F2 Z2 I I2 _ Vii (12.14) In this section we consider the synthesis of an L C ladder network -
Vi *n ’ h 2/n with a l -fl resistive termination to meet a specified transfer impedance Z21
In Chapter 8 we saw that transfer functions that have zeros of trans-
^-
or transfer admittance I i In terms of the open- and short circuit -
mission only on the jw axis or in the left-half plane are called minimum -

—
parameters of the L C circuit, Z21(s) can be expressed as
phase functions. If the function has one or more zeros in the right-half
plane, then the function is nonmimimum phase It will be shown now that . Zn z 22^+ 1 (12.15)
any transfer function of a passive reciprocal ladder network must be
minimum phase. Consider the ladder network in Fig. 12.11. The zeros of
transmission of the ladder occur at the poles of the series branch im- and 72I(J) is Y21 = (12.16)
pedances or at the zeros of the shunt branch impedances. Since these
y22 +1
branch impedances are themselves positive real, the poles and zeros of &' as depicted in Figs. 12.13 and 12.14.
these impedances cannot be in the right-half plane. Consequently, the Before we proceed with the actual details of the synthesis, it is necessary
transfer functions of ladder networks must be minimum phase. For the to discuss two important points. The first deals with the ratio of the odd
network in Fig. 12.12, a transfer function would have two zeros of to even or even to odd parts of a Hurwitz polynomial Q( s ). Suppose

«1
Cl
If
R2
rn C4
Li
$
y L C -
ladder

J- 02
C3
^ J?3

. .
FIG I 2 M

.
FIG 12.12 * Van Valkenburg, op . cit., Chapter 11.

mm
 348 Network analysis and synthesis Elements of transfer function synthesis 349
Q( s) is given as
From this we obtain
Q{ s ) = M ( s ) + N( s ) (12.17) m (12.23)
where M { s ) is the even part of Q( s ), and N( s ) is the odd part . We know M( s )
VVL =
that the continued fraction expansion of M ( s )IN( s) or N( s )/ M ( s ) should m
yield all positive quotients. These quotients can, in turn, be associated On the other hand, if P( s ) is odd, we divide by M( s ) so that
with reactances. Therefore it is clear that the ratio of the even to odd or
the odd to even parts of a Hurwitz polynomial is an L C driving point - - yi, =1 P( s )/ M ( s )
function. (12.24)
The second point to be discussed is the fact that the open-circuit transfer + [ N ( s )lM ( s ) ]
-
impedance z21 or the short circuit transfer admittance y2l of an L C circuit - P( s )
is an odd function. To show this, we must remember that in an L C - and 2/ 21
M( s )
circuit with steady-state input, the currents are 90° out of phase with the
(12.25)
voltages. Thus the phase shifts between the input currents and output
voltages or input voltages and output currents must be 90° out of phase, 2/ 22
m
M(s)
or
We assume that P( s ), M ( s ), and N( s ) do not possess common roots. For
Arctan = ± -2 rad (12.18) i our purposes, we will consider only the synthesis of 721 or Z21 with zeros
fiO '
w)
of transmission either at s = 0 or s = oo In a ladder network, a zero of .
ftQ transmission at s = 0 corresponds to a single capacitor in a series branch
Arctan iaO =
'
and ± -2 rad (12.19) or a single inductor in a shunt branch. On the other hand, a zero of
VM transmission at s = oo corresponds to an inductor in a series branch or
-
so that Re z21( jco ) = 0 = Re y21( j(o ) for an L C network. In order for the a capacitor in a shunt branch. In terms of the transfer impedance
real parts to be equal to zero, the functions z21 and y 2l of an L C two-port
network must be odd.
-
2 (5) =
_^ + Un-lS” 1 + ‘ ‘ ‘ + QiS + a0) (12.26)
i (s” "

Suppose, now, that the transfer admittance YZ 1 is given as the quotient 6(s) 5“ + + - .. + blS + b0
the presence of n zeros of Z21( ) at s = 0 implies that the coefficients a„_1,
of two polynomials
j

an~ 2 , .. . , aua0 are all zero. The number of zeros of Z 21( s ) at s = 00

^
P( s )
Z2i = 6(s) M ( s ) + N ( s)
(12.20)
is given by the difference between the highest powers of the denominator
where P( s ) is either even or odd. Now, how do we determine the short - —
and the numerator, m n. We know that n can exceed m by at most
unity, while m can be greater than n by more than one. For example,
_
circuit parameters yzi and y22 from the Eq. 12.20 to get it into the form

(12.21) !
—
if m n = 2, and n = 3 with an 1; . . . , au a0 = 0, we know that the
transfer function has three zeros of transmission at s = 0 and two zeros
of transmission at s = 00.
1 + Vu We can now proceed with the matter of synthesis. Consider the
The answer is quite simple. We divide both the numerator P( s ) and the following example.
denominator Q{s) by M{ s ) or N( s), the even or the odd part of Q( s). Z,i(5) =
2
(12.27)
Since y21 must be odd, if P( s ) is even, we divide by N( s ) so that s 3 2
+ 3s + 4s + 2
p( s )im We see that all three zeros of transmission are at s = 00 . Since the
(12.22) numerator P( s ) is a constant, it must be even, so we divide by the odd
1 + [ M( s )l N ( s ) ]

Ls
- JLi

f
350 Network analysis and synthesis Elements of transfer function synthesis 351

part of the denominator s3 + 4s. We then obtain

2 If
- T
*21 = s3
+ 4s '© I ": 2h in ; v2
2
3s + 2
*22 = s3 4- 4s
(12.28)

FIG 12.17 .
i
We see that both z21 and z22 have the same poles. Our task is thus simpli- 1 numerator and denominator by the even part of the denominator so that
fied to the point where we must synthesize z 22 so that the resulting network
s3 s3 4- 4s
has the transmission zeros of z 21. This requires that we first examine
the
the required zeros of *
21 =
3s 4- 22 *22 = 3s2
(12.30)
possible structures of the networks which have +2
transmission and see if we can synthesize 22z in one of those forms . For
The network that gives three zeros of transmission at s = 0 is a high-pass
the example that we are considering, a network which gives us three zeros structure which is realized by a continued fraction expansion of z22.
of transmission at s = 00 is shown in Fig. 12.15. We can synthesize z22 The final realization is shown in Fig. 12.17.
to give us this structure by the following continued fraction expansion Finally, consider the transfer admittance
of 1/ z22 .
s2
>s 4-
3J2 4- 2 * 4
3
Y
1
Y21( S ) =
s 4- 3s + 4s + 2
3
(12.31)
2

-120s)3s24- 2( iV ^- Z which has two zeros of transmission at s = 0, and one zero at s = 00.

-
3s2 Since the numerator is even, we divide by s3 + 4s so that
2)W(f » Y 3s2 -2
2/22 = s3 +4 4s (12.32)

The question remains as to how we synthesize y22 to give a zero of trans -

Since z22 is synthesized from the 1-Q termination toward the input end, mission at s = 00 and two zeros at s = 0. First, remember that a parallel
the final network takes the form shown in Fig. 12.16. Examining the inductor gives us a zero of transmission at s = 0. We can remove this
network more closely, we see that it takes the form of a low pass filter. - parallel inductor by removing the pole at s = 0 of y22 to give
Thus the specification of all zeros at J = 00 is equivalent to the specification
of a low-pass filter.
As a second example, consider the transfer impedance
Vl =^ 22 - —=
1
2s
5s/2
s 4- 4
2
(12.33)

If we invert yx , we see that we have a series L C combination, which gives -

Z2i(s)
— s 4- 3s 4- 4s 4- 2
3 2
s3 (12.29) us another transmission zero at s = 0, as represented by the f- farad
capacitor, and we have the zero of transmission at s = 00 also when we
-
Since the numerator of Z21(s) is an odd function, we have to divide both remove the inductor of f h. The final realization is shown in Fig. 12.18.

III =Mf in;

T
v2
+
Vi
fh 1'
2h
12

.
i .
FIG 12.18
in

.
FIG 12.15 FIG 12.16

Jl&L
 352 Network analysis and synthesis Elements of transfer function synthesis 353

h Zn \ Network lfi Vi «

(a) . . -
FIG 12.21 Constant resistance networks in tandem.

-
+
Vi
^
1
Network in
be decomposed into a product of simpler voltage ratios, which could be
realized as constant-resistance networks, and then connected in tandem .
For example, suppose our objective is to realize

n _ K( s - - z,)( s - z )
Z0)(s 2
(12.35)
( s ~ Po )( s - Pi )( s - pt )
(b)
. K
FIG 12.19
in terms of constant-resistance networks. We can first synthesize the
individual voltage ratios
An important point to note in this synthesis procedure is that we must
place the last element in series with a voltage source or in parallel with a Vi K0( s - z0) _
current source in order for the element to have any effect upon the transfer K s - Po
function. If the last element is denoted as Z,„ then the proper connection Vt _ Kt( s - Q
of Z„ should be as shown in Fig. 12.19. (12.36)
Vi s - px
V„ _ K ( - z )
12.4 SYNTHESIS OF CONSTANT RESISTANCE NETWORKS - 2 S 2

Vi s ~ Pi
In this section we will consider the synthesis of constant resistance two- -
port networks. They derive their name from the fact that the impedance -
as constant resistance networks and then connect the three networks in
looking in at either port is a constant- tandem to realize VjVa.
resistance R when the other port is termi - Although there are many different types of constant-resistance net-
in the same resistance R , as depicted works, we will restrict ourselves to networks of the bridge- and lattice-type
S_
Two- port nated
network R
in Fig. 12.20. Constant-resistance net- structures as shown in Figs. 12.22a and 12 ,22b. These networks are
works are particularly useful in transfer balanced structures ; i.e., the input and output ports do not possess com-
function synthesis, because when two mon terminals. Upon a close examination, we see that the bridge and
FIG. 12.20. Constant-resistance lattice circuits in Figs. 12.22 are identical circuits. The bridge circuit is
network . constant-resistance networks with the
same R are connected in tandem, as shown merely the unfolded version of the lattice. Consider the open-circuit
.
in Fig 12.21 , neither network loads down the other. As a result, if the parameters of the bridge circuit in Fig. 12.23. First we determine the
-
voltage ratio transfer function Na is V 2 / Vl and that of Nb is F3/ F2, the
of impedance zu as
voltage-ratio transfer function of the total network is Zg + Zb
*u = 2
(12.37)
F_FF (12.34)
F Vx F Next we determine the transfer impedance z21, which can be expressed as
Equation 12.34 implies that, if a voltage-ratio transfer function is to be
-
realized in terms of constant resistance networks, the voltage ratio could h
(12.38)

354 Network analysis and synthesis Elements of transfer function synthesis 355

- za
so that Z21 = zb 2
(12.41)

From the lattice equivalent of the bridge circuit we note that z22 = zn.
Now let us consider the lattice circuit that is terminated in a resistance
Vr
R, as shown in Fig. 12.226. What are the conditions on the open-circuit
parameters such that the lattice is a constant-resistance network ? In
other words, what are the conditions upon zu and z21 such that the input
impedance of the lattice terminated in the resistor R is also equal to R 1
In Chapter 9 we found that the input impedance could be expressed as

+
h h Zn — — Zn
2*22
zn 2
1
““
^
(12.42)

Zb Zb Since z22 = zn for a symmetrical network, we have

Yi R ~ zai 2
Zu = (12.43)
zu + R
In order for Zn = R, the following condition must hold .

. .
( b)
FIG 12,22. ( a ) Bridge circuit (6) Constant-resistance lattice.
zu2 z212 = R2
For the lattice network, we then have
— (12.44)

and is obtained as follows. We first obtain the current 7' as H(Za + Z„)2 - (Za - Zb )2 ] = R2 (12.45)

which simplifies to give Z„Z* = R2 (12.46)

r= (12.39)
za + zb 2 Therefore, in order for a lattice to be a constant-resistance network,
Eq. 12.46 must hold.
Next we find that F2 - F2 - = z - za r
( » ) Next, let us examine the voltage ratio F2/ F„ of a constant-resistance
lattice whose source and load impedances are equal to R (Fig. 12.24).
= (Z6 - Z ) ^h
2
0 (12.40) From Chapter 9 we can write
F 2 z iR
h =
Vg (zn + K 22 + R ) - z21z12
)(Z
(12.47)

Vi

.
FIG. 12.23 Analysis of bridge circuit. .
FIG. 12.24 Double-terminated lattice.

i
 356 Network analysis and synthesis Elements of transfer function synthesis 357
V* z2l R which, as we recall, is an all-pass transfer function By associating Eq 12.54 . .
which simplifies to = ( ZU + with Eq. 12.50, we have
Vg
^)
2
~ Z 212
_ Z2 lR ( 12.48)
= 5, R = 1 (12.55)
1
2 Rzn + 2R 2 Since ZbZa = 1, we then obtain (12.56)

In terms of the element values of the lattice, we have -

We see that Zb is a 1- h inductor and Za is a 1 farad capacitor. The final network
is shown in Fig. 12.25.
V» _ KZ> - Za ) R (12.49) if
Vg R( Zb + Z0) + 2 R 2
From the constant-resistance condition in Eq . 12.46 , we obtain l
+ 1 V2
V2 _ \[ Zb - ( R 2lZb ) ] R
2
y$
Vg R [ Zb + ( R2/ Zb ) ] + 2 R
\{ Zb - R2 )
FIG 12.25.
( Z 2 + R2 ) + 2 RZb Example 12.2 . Let us synthesize the all-pass function
_ \{ Z -R)2 2
V2
Vg ~
1 5 - 1 s2
2 s + \ s2
- 2s + 2 (12.57)
( Zb + Rf + 2s + 2
-
whose pole zero diagram is shown in Fig. 12.26 Since the portion .
i( Z „ - R ) . (12.50) _1s-
Z„ + R
V2 1
Vg = 2 7+ 1
In Eq . 12.50, the constant multiplier £ comes about from the fact that has already been synthesized , let us concentrate on synthesizing the function
the source resistance R acts as a voltage divider. If we let
V2
~
s2 - 2s + 2 (12.58)
(12.51) Va 2
s + 2s +2
K First, we separate the numerator and denominator function into odd and
we can express Z „ in Eq . 12.50 in terms of G as even parts. Thus we have
V2 ( s2 + 2 ) 2s _ (12.59) -
Va (s2 + 2) + 2s
z6 = [11-+GG((ss) ) ]
* ( 12.52) V
ju

In terms of Z0 , the voltage ratio can be given as X n

V2 1 R Za _ — (12.53)
Vg 2 R + Za
In the following examples, we will usually let R be normalized to unity.
Example 12.1 . The voltage ratio is given as
V2 _1s —1
~ (12.54)
x -A
Vg 2 7 +T .
FIG 12.26

Am
358 Network analysis and synthesis Elements of transfer function synthesis 359
If we divide both numerator and denominator by the odd part 2s, we obtain
V2 [( s* 4 2 )l2s] 1 - -
(12.60)
vw ^-r A/v
Va [(i2 + 2)/2i] + 1 R ^R 1
We then see that
25 VL _R zb R V2

=2 +5
s 1
(12.61) T
. .
FIG 12.29 Constant-resistance bridged -T circuit .
which consists of a £-h inductor in series with a -
1 farad capacitor. The im -
pedance Za is then
2s Under the constant-resistance assumption, the voltage-ratio transfer
Za ~ 2 (12.62) •: function can be given as
/
-
s +2
and is recognized as a J farad capacitor in parallel with a 1 h inductor. The -
V* R _ (12.64)
voltage ratio V2 IVa is thus realized as shown in Fig. 12.27. The structure that Vx R + Za Zb + R
Example 12.3 . Let us synthesize the voltage ratio
V2 s2 +1
Vx = s2 + 2s + 1
(12.65)

- -
as a constant resistance bridged T network terminated in a 1 fl resistor. First
let us write VJV 1 as
-
Vx 1
(12.66)
3
Vx 1 + [2s/(s + 1)]
.
FIG 12.27
so that Z , = 52 25+ 1 (12.67)
realizes the transfer function VJVg in Eq. 12.57 is formed by connecting the
networks in Figs. 12.25 and 12.27 in tandem, as shown in Fig. 12.28. Finally, and (12.68)
25
it should be pointed out that constant-resistance lattices can be used to realize
-
other than all pass networks. We recognize Za as a parallel L C tank circuit and - Zb as a series L-C tank
circuit. The final network is shown in Fig. 12.30.

\,
:

Next, let us consider the constant-resistance bridged-T network in Fig.

12.29. If the resistances in the network are all equal to R ohms, the
network has constant-resistance if
ZaZ> = R2 ( 12.63) FIG. 12.30

...
'
 synthesis 361
Elements of transfer function

Network analysis and synthesis

5 +4 (12.76)
360
and
ZW = 2T
Example 12.4. Let us synthesize the voltage ratio
is shown in Fig. 12.31.
V2 ( s + 2 )( s + 4) (12.69) The final synthesized network
Vy (5 + 3)(35 + 4)
in tandem. Problems
in terms of two constant -resistance -
bridged T circuits connected
into two separate voltage 12.1 Give an example of a network where: (a) a transfer function
of a transfer function
has
on
, we break up the voltage ratio in Eq. 12.69 multiple zeros on the ju> axis ; (b) the residue of a pole
At first
ratios the jta axis is negative. shown in the
Vg +2 (12.70)
condition holds for the networks
S

Vy J +3 12.2 Show that the residue

figure.
V2 5 +4 (12.71) lh
~ 35 + 4 in
and Va
1 2f
o2 iWWW—| —
, 02

we have
For the voltage ratio VJVlt i* in

5 +2 _ Zbl (12.72) in
5 + 3 Zby + 1

Zfll
_ 1 (12.73)
(a)
PROB 12.2 .
( b)

so that Zbl = + 2 and

5 5 + 2

we have
For the voltage ratio V2/ Va , find by inspection the zeros of
transmission
12.3 For the network shown
5 +4 = 1 (12.74)
and plot on a complex plane
.
35 + 4 1 + Za2
5h
25 (12.75)
Z "
from which we find °2 5 +4
If 2n

l f ^p :2 f in
lh
H
- O

PROB 12.3 .

driving point impedances

the figure show that the
12.4 For the networks in R2.
Zin are equal to R when ZaZb =
.
FIG 12.31

:
Elements of transfer function synthesis 363
362 Network analysis and synthesis
(6) Synthesize Za and Zb if
R s2
V2 _ + 35 + 2

Vi zm = R nr
Zb R < V.2
I
(«)
Vl

Vi
s3 + 452 + 55 + 2
12.8 Synthesize the following voltage ratios in one of the forms of the
networks in Prob. 12.4
V2 5 + 2
5 +3
V, _
(a )
T Vi
2(5» + 3)
Is2 + Is + 6

(c)
V2 _ 3(5 + 0.5)
Vy 45 + 1.5
Zb
+
12.9 Synthesize Na with termination resistors R2 = 4 f 2, Ry = 1 Q to give
12s2
R
Vg^= 15s2 + 75 + 2
R > V2

Zg
T
Ng

(b)
PROB. 12.4 PROB. 12.9

.
12.5 For the networks in Prob 12.4, find the voltage ratio transfer functions - .
12.10 For the network in Prob 12.9, realize network Na to give
VJV , .i
12.6 For the network shown in the figure (a) show that V? l 1
: Vg = 2 2s + 3
V2 1
V0
=2 Y
+ ( а ) Synthesize Na as a constant-resistance lattice. ( R = 1 A ) .
(b) Synthesize Y when -
(б) Synthesize Na as a constant resistance ladder as in Prob 12.4. (R = 1 A.) .
V2 0.5( s2 + 2)
V0 s2 + 25 + 2
- -
(c) Synthesize Na as a constant resistance bridged T circuit. ( R = 1 A.)
12.11 Synthesize the following functions into the form shown in the figure
1
W Z2 y = 5® + 352 + 35 + 2
in
5
V2 ( )
+
Y
* Z2 y = s
5 + 352 + 35 + 2
Vo s2
(c) Y21 = 5s + 352 + 35 + 2
PROB 12.6 . (4 )
5s
'
* 21 = 5s + 352 + 35 + 2
12.7 ( a ) For the constant-resistance bridged-T circuit, show that if ZaZb = 1,
i then (e) 52
Y21 =
n= 1 (5 + 2)*
Vl l + Zg
 364 Network analysis and synthesis
r
l
I
L -C
network H2
l 'o -
PROB. 12.11

-
12.12 Synthesize as a constant resistance lattice terminated in a 1 H resistor. -
13 chapter
(a) Vt s+1
s2 + s + 1
s* - 1 Topics in filter design
Vx 1
/2 s2 - 3s + 2
(6)

( c)
V1 s2 + 3s + 2
Vt( s ) J* 20s2 + 5s 20 - -
m
Fi(s) s3 + 20s2 + 5s + 20 ( 3.1 THE FILTER DESIGN PROBLEM
12.13 Synthesize the functions in Prob. 12.8 as constant resistance bridged T - -
circuits.
In the preceding chapters we examined different methods for synthesizing
a driving point or transfer function H( s ). Most problems have as their
initial specification an amplitude or phase characteristic, or an impulse
response characteristic instead of the system function H( s ) Our problem .
is to obtain a realizable system function from the given amplitude or
phase characteristic. For example, a typical design problem might be to
synthesize a network to meet a given low-pass filter characteristic The .
specifications might consist of the cutoff frequency coc, the maximum
allowed deviation from a prescribed amplitude within the pass band, and
the rate of fall off in the stop band. We must then construct the system
function from the amplitude specification. After we obtain H( s ), we
proceed with the actual synthesis as described in the Chapter 12 Another .
problem might consist of designing a low-pass filter with a linear phase
characteristic within the pass band. Here, both amplitude and phase are
specified. We must construct H( s ) to meet both specifications. Problems
of this nature fall within the domain of approximation theory. In this
chapter we will consider selected topics in approximation theory and then
present examples of filter design where both the approximation and the
synthesis problems must be solved. *

13.2 THE APPROXIMATION PROBLEM IN

NETWORK THEORY
m
The essence of the problem is the approximation of a given func -
tion f ( x ) by another function/, ; ax, . . . , a„) in an interval xk < x < x2.
I The parameters <Xj,\ ^
in the approximating function are fixed
365

&
y
Network analysis and synthesis
1 Topics in filter design 367
366
by the particular error criterion chosen. When
fa( x; 0L l t , a„), the following error criteria are most
we let e f ( x )
common : — — $ h*( t ) such that the squared error

e = jj°[/ (0 - h*( t )] dt
i
2

1. Least squares. The value of /(a a„) is minimized where

^ is minimum.
rxt
= Jxi M 2
w(z) dx * A generally effective procedure in time-domain approximation( t utilizes
.
orthonormal functions <f>k( t ) 1 The approximating function h* ) takes
error in certain sub- 01 the form
and w( x) is a weighting function which stresses the
intervals.
2. Maximally flat. The first n 1 derivatives
at x = x0.
—
of e are made to vanish

interval xl < x < x2

so that the error
h* ( t ) =
k=l
<f>k( t ) (13.3)

3. Chebyshev. The value of p is minimized in the

J [, 0 — 2 Oj dt
1

* <At
e = /( a (
where p = |«|max.
at a set of n points
4. Interpolation. The value of e is made to vanish is minimized when
in the interval xk < x < x2 .
After an error criterion is chosen, we must determine the particular d k = 1 , 2, . . . , « (13.4)
*=
a
‘
.

form of the approximating function This depends .

upon whether we
choose to approximate in the time or frequency
domain . Suppose f {x) a sum
domain and the approxi- as we saw in Chapter 3. If the orthonormal set is made up of
represents a magnitude function in the2frequency of exponentials es* , then the approximate impulse response
mating function is to be rational in cu ; then ‘
2

fa(* > a l> • • • > an)

«i + «33 + a-gX + • • • (13.1) h*(0 = 2 wSkl (13.4)
a2 + a4a: + <x6 xz + • • k=i

.. 2 .. In addition, the values of <x* must

be restricted to insure
where x = to (13.5)
thatyj,(a;; ax a „) > 0. In the time domain , /(x) might represent an has a transform ~~~
impulse response of a system to be synthesized . In
the case of an R C - k ~1 S Sfc

Realizability is insured if in the orthonormal set {aAeSt‘} R e j k <

, 0;
transfer function, we have
( s)
fa( x « , • • « )
> ! • . + «3 + •
n = (13.2) ..
k = 1, 2, . , «. Synthesis
obtained in Eq. 13.5.
then proceeds from the system function
H( jca )
H *

\\ \
-
where x = t . Since an R C transfer function
, are
must have its poles on the
restricted to negative real -
Frequency domain approximation
negative real axis, the values of a*, k even In frequency-domain approximation
numbers. the principal problem is to find fa
in the choice of a
The keystone of any approximation problem lies rational function H( s ) whose magnitude
suitable error criterion subject to realizability restrictions . The problem -1 w o>
a ’s are assigned before applying the \ H( ja> )\ approximates the ideal low-pass +1
can be simplified when some of the
characteristic in Fig. 13.1 according to
. ,3X ideal low-pass filter char-
cited , except the Chebyshev , can then
error criteria. All the error criteria the unknowns a predetermined error criterion. In the FIG
be reduced to a set of linear algebraic equations for acteristic.
next few sections we examine several
<*« •

-
Time domain approximation
The principal problem of time domain approximation
consists of
- 1 W. H. Kautz, Transient Synthesis in the Time Domain,” IRE Trans , on Circuit
“
( ) an approximating function Theory, CT 1, No. 3, September 1954, 29-39.
-
approximating an tmpulse response h t by
I
K
 368 Network analysis and synthesis
different ways to approximate the ideal low-pass : the maximally flat or
Butterworth approximation , the equal-ripple or Chebyshev approximation,
and the optimal or Legendre approximation. Another major problem is
that of obtaining a transfer function Hx( s ), whose phase is approximately
linear or whose delay is approximately flat over a given range of frequencies.
Here again , there are two different methods : the maximally flat or the
equal-ripple methods. Our discussion will center around the maximally
flat method . The joint problem of approximating both magnitude and
phase over a given frequency range is possible, but will not be discussed
here .

13.3 THE MAXIMALLY FLAT LOW PASS FILTER

APPROXIMATION
-
In Chapter 10 we saw that the ideal low-pass filter in Fig. 13.1 is not
realizable because its associated impulse response is not zero for t < 0.
However, if we use a rational function approximation to this low-pass
filter characteristic, the Paley-Wiener criterion will be automatically
satisfied . We will therefore restrict ourselves to rational function approxi-
mations.
In low-pass filter design , if we assume that all the zeros of the system
function are at infinity , the magnitude function takes the general form

M(co) = * 0
( 13.6) . .
FIG 13.2 Amplitude response of Butterworth low-pass filters .
+ /(«2)]* [1
where K0 is the d-c gain constant and /(co2) is the polynomial to be selected that the higher n is, the better the approximation . The amplitude approxi -
to give the desired amplitude response . For example, if mation of the type in Eq. 13.8 is called a Butterworth or maximally flat
response . The reason for the term “maximally flat” is that when we
/(co2) = CO2" ( 13.7) expand M(w) in a power series about tu = 0, we have
then the amplitude function can be written as ,n
M ( o> ) = *o( l ~ Wn 4+ i" ” - A 1 ® + AWn + ) ( 13.10)

M(co) =
K0
(1 + co2")*
( 13.8)
co = 0. For co »
—
We see that the first 2n 1 derivatives of M( cS) are equal to zero at
1 , the amplitude response of a Butterworth function
We see that M ( 0) = K0 , and that M(co) is monotonically decreasing with can be written as ( with K0 normalized to be unity)
co. In addition , the 0.707 or 3-decibel point is at co = 1 for all n , that is, 1
M( co) —
co"
co »1 ( 13.11)
M( l ) = -

^ all n

The cutoff frequency is thus seen to be co = 1 . The parameter n con-

trols the closeness of approximation in both the pass band and the stop
band . Curves of M { oo ) for different n are shown in Fig. 13.2. Observe
(13.9)
We observe that asymptotically , M( w ) falls off as co " for a Butterworth
response. In terms of decibels, the asymptotic slope is obtained as
20 log M ( a> ) = — 20n log co ( 13.12)

'o

P-
370 Network analysis and synthesis
Topics in filter design 37|
Consequently, the amplitude response falls asymptotically at a rate of
6n db/octave or 20« db/decade. Jco
One question remains. How do we obtain a transfer function H( s)
from only the amplitude characteristics M(cu) ? The procedure is as
follows. We first note that the amplitude response M( co ) and the complex 1

system function H( jco ) are related by 1.0 \

/ \
,
/ 60”, \
M\u> ) = H ( jco ) H ( - jeo )
If we define a new function h( s2 ) such that
( 13.13)
—L -60”' o J—
\ /
\ /
Ks2 ) = B( s ) H ( — s) ( 13.14)
\
N
1.0 /

we see that M 2( cd) = h( - a>2) ( 13.15)

—
From h( co2) all we need do is to substitute s2 = co 2 to give h( s2 ) . —
—
Then we factor h( s2 ) into the product H( s ) H( s ) . Since the poles and . .
FIG 13.3 Poles of H ( s ) H {
— s ) for an n = 3 Butterworth filter.
zeros of H( s) are the mirror images of the poles and zeros of H( s ) , we
simply choose the Hurwitz factors of h( s2 ) as H( s ). An example will serve
— For a Butterworth response, the poles of H ( s ) H( - s) are the roots of
to clarify this discussion . Consider the third- order (« = 3) Butterworth
response given by
— _ eMk
( 1 ) V" = - 1
-\ ) z ( 13.21)
1 k = 0, 1 , 2, . . . , 2n — 1
M\w ) = (13.16) K; The poles sk are then given by
1 +. oo*
1 Sk — e« -D/ 2n>
( 2fc
n even
(13.17) (13.22)
1 - ( - co2)3 eHk / n )a n odd
1 or simply by sk e « ( 2*+»-i )/ 2«]ir
— k = 0, 1 , 2, . . . , 2 n ( 13.23)
We see that h{ s2 ) is h( s 2 ) = ( 13.18)
1 - (s2)3 Expressing sk as sk = ak + jcok , the real and imaginary parts are given
Factoring h( s2 ) , we obtain by

h(s2) = 1 1 ak = cos
2k + n 1
7T
=
. ( 2k - 1 W
sin 1-—
1 + +2s 2s2 + s3 1 - 2s + 2s2 - s3 2n \ n 12
( 13.24)
= H( s) H ( - s ) 2k + n 1 — ( )f
^
( 13.19) cok = sin TT = COS
We then have 2n
H( s) = |It is seen from Eqs. 13.22 and 13.23 that all the poles of H( s) H( ~ s ) are
s3 + 2s2 + 2s + 1 p located on the unit circle in the s plane , and are
symmetrical about both
the a and the jco axes. To satisfy realizability conditions, we associate the
1
(s + l )(s + i
V
+ 3/ 2)(s +
; i - // 3/2) ( 13.20)
t£ plane with H( s ) . —
poles in the right-half plane with H( s ), and the poles in the left- half

The poles of H ( s ) and H( - s ) are shown in Fig . 13.3. Observe that the As an example, consider the construction of an H( s ) that gives an n 4
=
—
poles of H( s) are mirror images of the poles of H( s ) , as given by the Butterworth response. From Eq . 13.23, it is seen that the poles are
given by
theorem on Hurwitz polynomials in the Chapter 2.
/, • S /t — eiim+ z m* )
( 13.25)
372 Network analysis and synthesis Topics in filter design 373
H( s ) is then given as
13.4 OTHER LOW- PASS FILTER APPROXIMATIONS
H( s ) = 1
(13.26)
(s + e^' Xs + e' M' Xs + emuX* + eil"AH ) In Section 13.3, we examined the maximally flat approximation to a
low-pass filter characteristic. We will consider other low pass filter -
If we express sh in complex form and expand, we obtain approximants in this section.
The Chebyshev or equal - ripple approximation
m = (s 2
+ 0.76536s +
1
l )(s2 + 1.84776s + 1)
(13.27) We have seen that the maximally flat approximation to the ideal low -
pass filter is best at co = 0, whereas, as we approach the cutoff frequency
To simplify the use of Butterworth functions, H ( s ) is given in Tables (o = l , the approximation becomes progressively poorer . We now
13.1 and 13.2 for n = 1 to n = 8, in factored form as in Eq. 13.27, or consider an approximation which “ ripples” about unity in the pass band
multiplied out as and falls off rapidly beyond the cutoff co = 1 . The approximation is
1 equally good at co = 0 and co = 1 , and , as a result is called an equal -
m = ansn + a „_ is“ 1
+ • • •
+ axs + 1
(13.28) ripple approximation . The equal - ripple property is brought about by the
use of Chebyshev cosine polynomials defined as
TABLE 13.1 Cn( co ) = cos ( n cos-1 co ) M<1 (13.29)
= cosh ( cosh -1 co )
M>l
n
Butterworth Polynomials ( Factored Formf n
For n = 0 we see that C0( co ) = 1 (13.30)
1 s+1 and for n = 1, we have Ci( co ) = co (13.31)
2 s2 + V 2s + 1 Higher order Chebyshev polynomials are obtained through the recursive
3 (J* + S + l )(s + 1)
4 (s2 + 0.76536s + lXs2 + 1.84776s + 1)
formula
Cn( co ) = 2co Cn X (o ) C„ 2(<u) _ - _
(13.32)
5 (s + l )(s2 + 0.6180s + l )(s2 + 1.6180s + 1) Thus for n = 2, we obtain C2(co) as
6 (s2 + 0.5176s + lXs2 + V 2s + l )(s2 + 1.9318s + 1)
7 (s + l )(s2 + 0.4450s + l )(s2 + 1.2456s + l )(s2 + 1.8022s + 1) C2(a>) = 2co( co ) — 1 (13.33)
8 (s2 + 0.3986s + l )(s2 + 1.1110s + l )(s2 + 1.6630s + l )(s2 + 1.9622s + 1) = 2co2 - 1
In Table 13.3, Chebyshev polynomials of orders up to n = 10 are given.
TABLE 13.2 TABLE 13.3
n Chebyshev polynomials C„(co) = cos { n cos-1 <o )
Butterworth Polynomials *
n «2 «3 at «5 0 1
1 CO
1 1 2 2 co2 1 -
———
2 V2 1 3 4 co3 3co
3 2 2 1 4 8 co4 8 co2 + 1
4 2.613 3.414 2.613 1 5 16 co5 20co3 + 5 eo
5 3.236 5.236 5.236 3.236 1 6 -
32 co6 48 co4 + 18 co 2 - 1
6 3.864 7.464 9.141 7.464 3.864 1 7 64 co7 - 112 co5 + 56 co3 - 7co
7 4.494 10.103 14.606 14.606 10.103 4.494 1 8 128 co8 - 256 co6 + 160 co4 - 32 co2 + 1
8 5.126 13.138 21.848 25.691 21.848 13.138 5.126 9 256 co9 - 576co7 + 432 co5 - 120co3 + 9 co
10 512 co10 - 1280co8 + 1120 cos - 400 co4 + 50co2 -1
* = 1.

374 Network analysis and synthesis Topics in filter design 375

(1 + e 2 ) -*

. .
FIG 13.4 C3(co) and C4(co) Chebyshev polynomials.

The pertinent properties of Chebyshev polynomials used in the low pass -

filter approximation are :

.
1 The zeros of the polynomials are located in the interval |co| < 1 , o l CO

as seen by the plots of C3(co) and C4(co) in Fig. 13.4. FIG 13.5 . . Chebyshev approximation to low-pass filter.
2. Within the interval ( co| < 1, the absolute value of Q,(co) never
exceeds unity ; that is, |C„(co)| < 1 for |w| < 1.
3. Beyond the interval |co| < 1, jC„(co)| increases rapidly for increasing maximum and minimum in the pass band is given as
values of |<w|.

Now, how do we apply the Chebyshev polynomials to the low-pass

Ripple =1 — (1 +1<?)'A (13.35)
filter approximation ? Consider the function e2 Cn2( co ), where c is real and
small compared to 1. It is clear that e 2 C„2(co) will vary between 0 and e 2 1
(13.36)
At co = 1, \ H( jco )\ is \ H(J 1 )\ = (1
in the interval |co| < 1. Now we add 1 to this function making it 1 + + e2)*
e2 Cn2( co ). This new function varies between 1 and 1 + e2, a quantity
because C„2( l) = 1.
slightly greater than unity, for \co\ <, 1. Inverting this function, we obtain 1, as co increases, we reach a point
In the stop band , that is, for |co|
the function which we will associate with \ H( jco )\ 2 ; thus
cok, where e Cn\co ) 1 so that
2
»
1 1
! #(»! =
:
\ H ( jco )\ 2 = 1 e2Cn
(13.34) CO > (
Ok (13.37)
+ \co ) eCn(«)
The loss in decibels is given as
Within the interval \ co\ < 1, \ H( jco )\ 2 oscillates about unity such that the
maximum value is 1 and the minimum is 1/(1 + e2). Outside this interval,
Cn%co ) becomes very large so that, as co increases, a point will be reached
Loss = — 20 log 10 \ H( jco )\
(13.38)
»
where c2 Cn 2( w) 1 and \ H( jco )\ 2 approaches zero very rapidly with ^ 20 log c + 20 log C„(w)
further increase in rt>. Thus, we see that \ H( jco )\ 2 in Eq . 13.34 is indeed a
suitable approximant for the ideal low-pass filter characteristic. But for large co, C„(<o) can be approximated by its leading term 2n con ,
1

Figure 13.5 shows a Chebyshev approximation to the ideal low-pass so that

Loss = 20 log e + 20 log 2n-1wn I
filter. We see that within the pass band 0 co < 1, \ H( jco )\ ripples
-
between the value 1 and (1 + e 2)-* *. The ripple height or distance between = 20 log + 6( n — 1) + 20« log co
(13.39)
 376 Network analysis and synthesis I Topics in filter design 377
We see that the Chebyshev response also falls off at the rate of 20« £ normalized pole locations s' k are given by
db/decade after an initial drop of 20 log e + 6(« l ).decibels. However,
in most applications, e is a very small number so that the 20 log e term is
— s', = -f y
actually negative. It is necessary, therefore, cosh pk
to compensate for this decrease in loss in 5
ju the stop band by choosing a sufficiently (13.42)
jv -
- cosh Pk large « . cosh cosh /?
^jl From the preceding discussion, we see 1 V *
that a Chebyshev approximation depends = , + j<°\
<*'
upon two variables, e and «, which can be The normalized pole locations can be derived as
determined from the specifications directly.
The maximum permissible ripple puts a
( )
JT sinh Pk bound on

band fixes «
e Once e 'determined , any
desired value of attenuation in the stop

The derivation of the system function

H( s) from a Chebyshev amplitude approxi-
.
.
^ a' k = tanh fik sin

a>' k = cos ^
Comparing the normalized Chebyshev pole locations with the Butterworth
! (13.43)

mation \ H( j(o )\ is somewhat involved and

- jl will not be given here.2 Instead, we will pole locations iri Eq . 13.24, we see that the imaginary parts are the same,
simply give the results of such a derivation. while the real part a' k of the Chebyshev pole location is equal to the real
First we introduce a design parameter. part of the Butterworth poles times the factor tanh /3k . For example, with
. .
FIG 13.6 Locus of poles of n = 3 and tanh fik = 0.444, the Butterworth poles are
Chebshev filter. 1 . ._
Pk = ~n sinh i l (13.40)
= - i + JO
where « is the degree of the Chebyshev polynomial and e is the factor
controlling ripple width. The poles, sk = ak + jook , of the equal-ripple
J2,3 =
—
0.5 ± y 0.866 '

approximant H( s) are located on an ellipse in the s plane, given by so that the normalized Chebyshev poles are given by
W(, = -1(0.444) + jO
sinh /3* ^ cosh2 /3 2 =1 (13.41)
*
I p
= -0.444 + jO
The major semiaxis of the ellipse is on the jco axis and has a value w =
± cosh /3*. The minor semiaxis has a value a = ± sinh /?*, and the foci s2>3 = -0.5(0.444) ± y‘0.866 s
are at co = ± 1 (Fig. 13.6). The half-power point of the equal-ripple = - 0.222 ± y0.866 ‘

amplitude response occurs at the point where the ellipse intersects the jco
axis, i.e., at co = cosh fik . Recall that for the Butterworth response, the Finally, to obtain the denormalized Chebyshev poles, we simply multiply ;i
half-power point occurs at co = 1. Let us normalize the Chebyshev poles s' k by cosh pk , that is,
sk such that the half-power point also falls at co = 1 instead of at co = 3 sk = { o' k + P\) cosh pk (13.44)
cosh (ik ; i.e., let us choose a normalizing factor, cosh fik , such that the
There is an easier geometrical method to obtain the Chebyshev poles,
2
Interested parties are referred to M. E. Van Valkenburg, Introduction to Modern given only the semiaxis information and the degree «. First we draw two
Network Synthesis, John Wiley and Sons, New York, 1960, Chapter 13. circles, the smaller of radius sinh j8k and the larger of radius cosh /3 , as
*

378 Network analysis and synthesis Topics in filter design 379

ju find /3*. From Eq. 13.40 we have

/
/ 1
NT
-
cosh Pk

\
:
Pk = ~n sinh-1 -
1 1
(13.49)
/ \
f \ \ \ 1
/ / \ = £ sinh ”
1.965 = 0.476
i \Q> M \
In order to find the normalized Chebyshev poles from the Butterworth
* sinh Pk s b | -' I
I
S poles, we must first determine tanh flk . Here we have
\ /1
\
\
\\ J y
i
I
/
I tanh ft. = tanh 0.476 = 0.443 (13.50)
\
_
/
1

/ From Table 13.1, the n = 3 Butterworth poles are

\! /
^K^
Chebyshev
pole locus
Butte orth
pole locus ,=
j
- 1.0, .r2 3 = -0.5 ± y 0.866
Multiplying the real parts of these poles by 0.443, we obtain the normalized
'
(13.51)

FIG 13.7 n . . = 3 Chebyshev filter poles. Chebyshev poles.

shown in Fig. 13.7. Next, we draw radial lines according to the angles of = -0.443, $'
23 = -0.222 ±;0.866
the Butterworth poles (Eq. 13.22) as shown. Finally, we draw vertical Finally, the denormalized Chebyshev poles are obtained by multiplying
dashed lines from the intersections of the smaller circle and the radial lines, the normalized ones by cosh /3 = 1.1155 so that the denormalized poles
*
and horizontal dashed lines from the intersections of the large circle and i are $x = -0.494 and $2 3 = -0.249 ±;0.972.
>
the radial lines. The Chebyshev poles are located at the intersection of R. H( s ) is then
•1
the vertical and horizontal dashed lines, as shown in Fig. 13.7.
Consider the following example. We would like to obtain a system
1
‘
m = (s
+ 0.494)(s + 0.249
0.502
- j0.972)(s + 0.249 + j0.972)
function H( s ) that exhibits a Chebyshev characteristic with not more than (13.52)
-
1 decibel ripple in the pass band and is down at least 20 decibels at co = 2. 0.502
When we design for 1-decibel ripple, we know that at co = 1, \ H( jl )\ is s3 + 0.992s2 + 1.255s + 0.502
down 1 decibel so that
In Fig. 13.8, the amplitude responses of the Chebyshev and an n = 3 i|
1 Butterworth filter are shown.
20 log = 20 log (1 e2)* -1 (13.45)
+ K Monotonic filters with optimum cutoff
1 K In comparing Butterworth filters with Chebyshev filters, the following
We then obtain
(1 + C )*
2 = 0.891 (13.46)
B 030 be said. The Butterworth response is a maximally flat, monotonic
K response, whereas the Chebyshev response is equal ripple in the pass band.
and e = 0.509 (13.47) W In the stop band , the Chebyshev response falls off more rapidly than the
H Butterworth (except when e is very, very small). In this respect, the Cheby-
Our next task is to find n from the 20 decibels at co = 2 specification.
From Eq. 13.39 the loss can be given as approximately
K shev filter is a better filter than the Butterworth. However, as we shall see
K *n Section 13.5, the transient response of the Chebyshev filter is very poor.
20
^ 20 log 0.509 + 6(n - 1) + 20« log 2 (13.48) 1 Ifthewe require sharp cutoff characteristics for a given degree n3, however,
'

Solving for «, we obtain « = 2.65. Since « must be an integer, we let « = 3.

P Butterworth filter is quite unsatisfactory. In 1958, Papoulis proposed

With the specification of « and e, the pole locations are completely speci-
fied. Our next task is to determine these pole locations. First we must —
R
> -

'
3
A. Papoulis, “Optimum Filters with Monotonic Response,” Proc. IRE, 46 , No. 3,
March 1958, pp. 606-609.
 380 Network analysis and synthesis Topics in filter design 381

Radian frequency, to
The polynomial L„(<y2) has the following properties :
0.1 0.2 («) 4,00) = 0
0 (b) 4,( 1 ) = 1
-1 (c) dLn( w2) < 0
1
d (o
rfL„( tus
00 da ) IG>=1
=M ( M maximum)
-4
Properties a , b and c are the same as for the Butterworth generating poly-
.a nomial /(w2) = a)2 . Property c insures that the response M ( oS) is mono-
6 tonic and property d requires that the slope of Ln( a>2 ) at to = 1 be the
4 "

steepest to insure sharpest cutoff.

s? Papoulis originally derived the generating equation for the polynomials
2
-8 Ln (for n odd) to be
-9
- 10
.*
where n = 2k +
Ln( <o2 ) =
1 and the
C?oa < dx

are the Legendre polynomials of the

(13.55)

Pt( x )
- li first kind4
± P0( x ) = 1
- 12
P,( x ) = *
- 13
- 14
P2(z) = K 3*2 - 1)
( 13.56)
FIG . 13.8. Amplitude response of n = 3 Chebyshev filter with 1.0-decibel ripple in pass P3( x ) = l ( 5 x* - 3 x )
band and Butterworth response (n = 3). i
and the constants at are given by
a class of filters called Optimum or “L” filters, which have the following
properties : n„
__ -_ £ _ 2 _ ak
( 13.57)
1
° 3 + 5 y / 2( k + 1) 2k 1
1 . The amplitude response is monotonic. Later Papoulis5 and, independently, Fukada,6 showed that the even-
2. The fall-off rate at co cutoff is the greatest possible, if monotonicity is ordered Ln polynomials can be given by
assumed .
3. The zeros of the system function of the L filter are all at infinity.
[2 *<(*)] dx
t
4fc+ 2(w2) (13.58)
Recall that the magnitude response of a low-pass filter with all zeros at
infinity can be expressed as n = 2k +2
M( co ) = [1 *(0.
(13.53) 4 E . Jahnke and F . Emde, Tables of Functions , Dover Publications, New York , 1945.
+ / *>2)f 5
A . Papoulis, “On Monotonic Response Filters,” Proc . IRE, 47, February 1959,
332-333.
Let us denote the polynomial generating the L filter by 6
M . Fukada, “Optimum Filters of Even Orders with Monotonic Response,”
/(o>2) = L>2) ( 13.54) Trans. IRE, CT-6, No. 3, September 1959, 277-281.

382 Network analysis and synthesis Topics in filter design 383 1

where the constants at are given by : Frequency, rad/sec

Case 1 ( k even) :
0 1 I
*L (13.59)
5 2k + 1 V(k + l )( fc + 2) |
<*i = ‘
— ak-1 0
Case 2 ( k odd) : i
r
ak 1
£1 = £3 = ...
( 13.60)
3 7 2k +1 V(fc + l )(fe + 2)
«0 = a 2 = •"
= atk = 0
Fukada tabulated the Ln(«2) polynomials up to n = 7 together with
dLn( co2 )ld(o evaluated at co = 1 to give an indication of the steepness of
the cutoff . This is shown in Table 13.4 .
To obtain the system function H ( s ) for the L filter, we must factor the
equation for h( s2) and choose the Hurwitz factors as H ( s ) .
1
/,(s2) = H(s) H( - s) =1 ( 13.61)
+ Ln( -s2)
For example, for n = 3, the magnitude response squared is
1
M 2(co) =1 ¥
+ L3(CO2) FIG . 13.9.
( 13.62) Amplitude response of Optimum versus Butterworth filters.
1 1 After we factor h( s2 ) , we obtain
1 + co2 - 3 co4 + 3co6 0.577
H( s ) = ( 13.64)
Substituting — w2 = s2 , we obtain s3 2
+ 1.31s + 1.359s + 0.577
1 where the numerator factor, 0.577, is chosen to let the d-c gain be unity.
/i (s2) = H ( s ) H ( - s ) = 6
( 13.63)
1 - s - 3s4 - 3s
2
The poles of H( s ) are sx = -0.62 ; s2 3 = -0.345 ± /0.901 . The ampli -
tude response of third - order Optimum (L) and Butterworth filters are
TABLE 13.4 compared in Fig . 13.9 Note that the amplitude response of the Optimum
L„( co2) Polynomials filter is not maximally flat, although still monotonic. However, the cutoff
dLn{ 1 ) characteristic of the Optimum filter is sharper than the cutoff of the
n L >) 2
duo Butterworth filter.
Linear phase filters
2 CO4 4
8
Suppose a system function is given by
3 3 co6 - 3 co4 + co2
4 6 «>8 - 6o>6 + 3 co4 12 H ( s ) = Ke~ sT ( 13.65)
5 20a,10 - 40a,8 + 28 a,6 - 8 co4 + a,2 18 where K is a positive real constant . Then the frequency response of the
6 50a,12 - 120a,10 + 105 a,8 - 40a,6 + 6a,4 24
system can be expressed as
7 175 a,14 - 525a,12 + 615a,10 - 355a,8 + 105 a,6 - 15 a,4 + a,2 32
H ( jw ) = Ke~ iaT ( 13.66)
!
»
 384 Network analysis and synthesis Topics in filter design 385
so that the amplitude response M( co ) is a constant K , and the phase re - If sinh s and cosh s are expanded in power series, we have
sponse
<f>{ m ) = — coT (13.67) , , , 5* s4 . s6
+ ••
is linear in <w. The response of such a system to an excitation denoted by
the transform pair { e( t ), E( s)} is
sinh s = s + I +
sj . -
s
+ +
2!
_
4!
7
• • •
(13.73)

R( s) = K E( s )e~sT (13.68) ;
3! 5! 7!

so that the inverse transform r( /) can be written as From these series expansions, we then obtain a continued fraction expan-
sion of coth s as
r( t ) = C-W)] (13.69) l l
coth s =s+3
-
“

= Ke [ t T )u( t - T) 1
-
+ (13.74)
We see that the response r( t ) is simply the excitation delayed by a time T , ' -1 + 7 1
and multiplied by a constant . Thus no signal distortion results from
transmission through a system described by H( s) in Eq. 13.65. We note
-s + -
further that the delay T can be obtained by differentiating the phase
response <f>( co ), by co ; that is,
Delay
__ _ M S2
dm
T (13.70)
If the continued fraction is terminated in n terms, then H( s ) can be
written as
m = Bn*(<s> ) . (13.75)

Consequently, in a system with linear phase, the delay of the system is

obtained by differentiating the phase response <f>( co ). where Bn( s) are Bessel polynomials defined by the formulas
A system with linear phase and constant amplitude is obviously desirable
from a pulse transmission viewpoint. However, the system function H( s ) *o = l
in Eq . 13.65 is only realizable in terms of a lossless transmission line called * =s+1 i (13.76)

Bn = (2n - !)*„_, +
a delay line . If we require that the transmission network be made up of
lumped elements, then we must approximate H{ s ) = Ke~ sT by a rational
function in s. The approximation method we shall describe here is From these formulas, we obtain
due to Thomson .7 We can write H( s ) as
B2 = J2 + 3s + 3 (13.77)
H( s) = - 4°
esT B3 = s3 + 6 s2 + 15J + 15
(13.71)
Higher order Bessel polynomials are given in Table 13.5, and the roots of
sinh sT + cosh sT Bessel polynomials are given in Table 13.6. Note that the roots are all in
the left-half plane. A more extensive table of roots of Bessel polynomi-
where K0 is chosen such that 77(0) = 1. Let the delay T be normalized to
als is given by Orchard .8
unity and let us divide both numerator and denominator of H( s ) by sinh s The amplitude and phase response of a system function employing an
to obtain
unnormalized third-order Bessel polynomial
H(s) = Ko/
sinh s
coth s + 1
- (13.72)
H( s ) =
15 (13.78)
3 2
s + 6s + 15s + 15
7 W
. .
E Thomson, “ Network with Maximally Flat Delay,” Wireless Engrg., 29, Oct.
—
1952, 256 263.
8
H. J. Orchard, “The Roots of Maximally Flat Delay Polynomials” IEEE Trans
-
on Circuit Theory, CT 12 No. 3, September 1965, 452 454
,

- .

Topics in filter design 387

386 Network analysis and synthesis
TABLE 13.5 Frequency, rad/sec
0.1 0.2 0.4 0.6 0.8 1 3 4 5 6 7
Coefficients of Bessel Polynomials •
0
n b0 bl b2 b3 b4 bb b9 b7
\

0
1
1
1 1
-4 \
Vfc Bessel

2 3 3 1 - 6 \
\
3 15 15 6 1 -8 \
4 105
945
105
945
45
420
10
105
1
15 1 S
“ 10 Butterworth
—
-r-
5 12 \
6 10,395 10,395 4,725 1,260 210 21 1 \
1
7 135,135 135,135 62, 370 17, 325 3,150 378 28 1 14
\
-16 \
v
are given by the solid lines in Figs. 13.10 and 13.11. These are compared
- 18 \ \
of an unnormalized third order Butterworth - - 20 \ \
with the amplitude and phase
function given by the dotted lines. Note that the phase response of the -22 \
A
-
constant delay function is more linear than the phase of the Butterworth -24 .

\
function. Also , the amplitude cutoff of the constant delay curve is more - - 26
gradual than that of the Butterworth . FIG . 13.10. Amplitude response of n = 3 Bessel and Butterworth filters.
TABLE 13.6
Frequency, rad /sec
Roots of Bessel Polynomials 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
0
0
n Roots of Bessel Polynomials -15 PN v
1 -1.5
1.0 + y O ' - 30
2 --2.32219
± y 0.866025
+ JO
' -45
- 60
Bessel n =3
3
- 1.83891 j\.75438
± -75 \
-2.89621 ± y 0.86723 \
r\ \ j
'

|-90
4
- 2.10379 ± /2.65742
1-105
-3.64674 + /0 8 - 120
5 -3.35196 ± /1.74266 I -135 X\
- 2.32467 ± /3.57102 Butterworth n = 3
- 4.24836 ± / 0.86751 -150 X,
6 - 3.73571 ± / 2.62627 -165
- 2.51593 ± /4.49267
-180
' -4.97179 + /0
- 4.75829 ± /1.73929 -195
7
- 4.07014 ± /3.51717 -210
. -2.68568 ± /5.42069 . .
FIG 13.11 Phase responses of low-pass filters.
 388 Network analysis and synthesis
Topics in filter design 389
Most of the foregoing figures of merit are related to frequency response,
13.5 TRANSIENT RESPONSE OF LOW PASS FILTERS - particularly bandwidth and phase linearity. Some of the quantities, such
In this section we will compare the transient response of the filters as rise time and delay time are intimately related to each other but have
discussed in Section 13.4. In particular, we will compare the step response rather tenuous ties with overshoot. Let us examine qualitatively the
of the filters according to the following figures of merit : relationships between the transient response criteria just cited and
frequency response.
1. Rise time tR. The rise time of the step response is defined here as Rise time and bandwidth have an inverse relationship in a filter. The
the time required for the step response to rise from 10 % to 90 % of its
wider the bandwidth, the smaller the rise time ; the narrower the band-
final value as depicted in Fig. 13.12 . width, the longer the rise time. Physically, the inverse relationship could
2. Ringing . Ringing is an oscillatory transient occurring in the response
be explained by noting that the limited performance of the filter at high
of a filter as a result of a sudden change in input (such as a step). A
frequencies slows down the abrupt rise in voltage of the step and prolongs
quantitative measure of the ringing in a step response is given by its the rise time. Thus we have
settling time.
3. Settling time. The settling time is that time ts beyond which the TB x BW = Constant (13.79)
step response does not differ from the final value by more than ± 2 %, Rise time is a particularly important criterion in pulse transmission. In
as depicted in Fig. 13.12.
an article on data transmission,9 it was shown that in transmitting a pulse
4. Delay time , tD. Delay time is the time which the step response
of width Tx through a system with adjustable bandwidths, the following
requires to reach 50 % of its final value as shown in Fig. 13.12. results were obtained :
5. Overshoot. The overshoot of the step response is defined as the
difference between the peak value and the final value of the step response Bandwidth Rise Time
(see Fig. 13.12) expressed as a percentage of the final value. ( fc = l / T’i ) (milliseconds)

/ Overshoot fc 0.5

i
2/c 0.25
3/c 0.16
1 4/c 0.12
1.0
5/c 0.10
0.9

The table shows a definite inverse relationship between rise time and
bandwidth.
tE 0.5 < — tD — A definition of time delay is given by Elmore as the first moment or
: centroid of the impulse response
L "
TD = f (0 dt J *
(13.80)

0.1
provided the step response has little or no overshoot. Elmore’s definitio
of rise time is given as the second moment
0 h<
I
—
0 2 4 6 8 10 12 14 16 18 20 TR (13.81)
Time sec
.
FIG 13.12. Figures of merit for step response. ts = rise time ; t , = setting time ; 9
—
R. T. James, “Data Transmission The Art of Moving Information,” IEEE
tD = delay time . Spectrum, January 1965, 65-83. i

390 Network analysis and synthesis Topics in filter design 391

O These definitions are useful because we

t
V( s )
m
R
can obtain rise time and delay time directly
from the coefficients of the system function
H( s ). Without going into the proof, which
1.0
0.9
is in Elmore and Sands,10 if H( s ) is given as
1 + ats + a 2s2 + • • • + a „ sn
I= 0.708
5
H( s) = 1 + bxs + bzsz + • • • + bmsm 0.6
FIG . 13.13. R- C network . S
(13.82) 3 -5
-
°
the time delay
and the rise time is
TD .is TR — ~ ai ( 13.83) |0.4
I 0.3
0.2
TR = { 2Tr [b * - a * + 2(a, - bj) }* (13.84) o. i
For the R-C network in Fig. 13.13, H ( s) is 0

=M
R
H(S) (13.85) o 4 6 8 10 12 14 16 18 20
I( )s 1 + sRC Time t , sec
so that TD = RC _ FIG . 13.15. Step response of normalized Butterworth low-pass filters .

-
TR = JlirRC
It should be emphasized that Elmore’s definitions are restricted to step
(13.86)
The step responses of the n = 3, n = 7, and n = 10 Butterworth filters
are shown in Fig. 13.15. Note that as n increases, the overshoot increases.
responses without overshoot because of the moment definition . The more This is because the higher order Butterworth filters have flatter magnitude
general definition of rise time is the 10-90 % one cited earlier, which has characteristics (i .e., there is more gain at frequencies just below the cutoff ).
no formal mathematical definition .
Ringing is due to sharp cutoff in the filter magnitude response, and is
Overshoot is generally caused by “excess” gain at high frequencies. By accentuated by a rising gain characteristic preceding the discontinuity.
excess gain we normally mean a magnitude characteristic with a peak The step response of an n = 3 Bessel (linear phase) filter is compared
such as the shunt peaked response shown by the dashed curve in Fig.
13.14. A magnitude characteristic with no overshoot is the magnitude
-
to the response of an n = 3 Chebyshev filter with 1-decibel ripple in
Fig. 13.16. We cannot compare their rise times since the bandwidths of
-
characteristic of an R C interstage shown by the solid curve in Fig. 13.14. the two filters have not been adjusted to be equal. However, we can
compare their ringing and settling times. The Chebyshev filter has a
y \ sharper cutoff, and therefore has more ringing and longer settling time
\
-
Q
\ than the Bessel filter. Note also the negligible overshoot of the Bessel
filter that is characteristic of the entire class of Bessel filters.
1 Simple
\

^ — \
- Shunt peaked response The decision as to which filter is best depends upon the particular
situation. In certain applications, such as for transmission of music,
R-C response \
\ phase is not important. In these cases, the sharpness of cutoff may be the
dominant factor so that the Chebyshev or the Optimum filter is better
than the others. Suppose we were dealing with a pulse transmission
Log frequency
. .
FIG 13.14 Comparison of shunt-peaked and simple R- C magnitudes.
system with the requirement that the output sequence have approximately
the same shape as the input sequence, except for a time delay of T =
10 W. C. Elmore and M
. Sands, Electronics , National Nuclear Energy Series, Div .
V, 1, McGraw-Hill Book Company, New York , 1949, pp. 137-138.
—
Tz 7\, as shown in Fig. 13.17a. It is clear that a filter with a long rise
time is not suitable, because the pulses would “smear” over each other
 392 Network analysis and synthesis
Topics in filter design 393
1.2
1.0
/
0.9
1.0
/
0.8
0.7 T
I 0.8
Step response of n = 3 Bessel filter
g 0.6 Step response of n 3 Chebyshev =
|, 0
filter with 1 db ripple -
|0 , Step response

< 0.4 t
0.3
/
I 0.4
0.2 l
0.1
/ 0.2
Second derivative
0
0
0

.
4

.
6
10
Time, sec
8
12 14 16
FIG 13.16 Comparison of filter transient responses.
18 20
-1.2
0 2 4 6 8
V . /

10 12 14 16 18 20
Time t , sec
.
FIG 13.18

Ti
Input pulse train
System
with short
rise and
settling
times
M
T2
Output pulse train
A note that after the first peak, the ringing of the step response has an
approximate sinusoidal waveshape. Let us now consider the second
derivative of the step response shown by the dashed curve in Fig. 13.18.
fa) Beyond the first peak of the step response, the second derivative is also
(approximately) sinusoidal, and is negative when the step response is
System
greater than unity, and positive when the step response is less than unity.
with long If we add the second derivative to the step response, we reduce the over-
rise and shoot and ringing.11
settling
Ti times T2 Suppose a(f ) is the step response and H ( s ) is the system function of the
Input pulse train Output pulse train filter. The corrected step response can be written as
( b)

.
FIG 13.17. Smearing of pulses in systems with long rise and setting times. <*i (0 = «(0 + K 0 (13.87)
2
t/ f
as seen in Fig. 13.176. The same can be said for long settling times.
where K is a real, positive constant. Taking the Laplace transform of
Since a pulse transmission system must have linear phase to insure un-
distorted harmonic reconstruction at the receiver, the best filter for the
Eq. 13.87, we have
system is a linear phase filter with small rise and settling times. cwoj - f Wpr] -
(13.88)
13.6 A METHOD TO REDUCE OVERSHOOT IN FILTERS

We present here a method to reduce the overshoot and ringing of a

filter step response. The step response of a tenth-order Butterworth filter
= ( Ks’ + I) — 5

u F. F. Kuo, “A Method to Reduce Overshoot in Filters,” Trans. IRE on Circuit

is shown in Fig. 13.18. It is seen that the overshoot is about 18 %. We
-
Theory, CT 9, No. 4, December 1962, 413-414.

394 Network analysis and synthesis Topics in filter design 395

1.2 will have negligible effect. Therefore the zeros should be added some -
where near the band edge. Figure 13.19 shows the effects of adding
V zeros at w = ± 1 (i.e., right at the band edge), and at ft> = 1.5. We see
1.0
/' that the further away the zeros are placed from the band edge, the less
0.8 effect they will have. The addition of the zeros will decrease the 3-decibel
7
bandwidth of the filter, however, as seen in Fig. 13.20. Therefore, a
I compromise must be reached between reduction in bandwidth and
to.6
Tenth order Butterworth
E reduction in overshoot .
0.4
Zeros at to ± 1.0
Zeros at to = i 1
= , An effective way to overcome this difficulty is to scale up the bandwidth
by a factor of, for example, 1.8. Then the zeros are placed at eo = ± 1.8,
0.2 l
///
which will reduce the 3-decibel bandwidth to approximately its original
figure cu = 1.0, as shown in Fig. 13.20. The overshoot, however, will be
reduced as though the zeros were at the band edge.
0
0 2 4 6 8 10 12 14 16 18 20
Time t , sec
13.7 A MAXIMALLY FLAT DELAY AND CONTROLLABLE
.
FIG 13.19 MAGNITUDE APPROXIMATION
From Eq. 13.88 we see that by adding a pair of zeros on the ju> axis at 1
In this section we will examine an interesting result, which is due to
VK
s = ±y/ , the overshoot and ringing are reduced. Budak.12 The result deals with linear phase approximation with con-
For low-pass filters, the factor 1 fsj K must in general be greater than trollable magnitude. In Section 13.4 we discussed approximation of a
the bandwidth of the system. For normalized Butterworth filters the flat delay using Bessel polynomials. The resulting rational approximant
bandwidth is to = 1 so that K < 1. The factor K also controls the amount was an all- pole function (all transmission zeros at J = oo), whose de-
of overshoot reduction . If K is too small, adding the zeros on the jm axis nominator was a Bessel polynomial. There was no control of the magni-
-0 tude using the all-pole approximant. In Budak’s method the magnitude
is controllable, while the phase is as linear as the standard Bessel approxi-
-4 NJ mation .
\\
\
Budak’s approximation is obtained by introducing the parameter k to
-6
-8 T 1r \ split e~s into two parts such that

- to \ e~ks
T e~ s =
s -12 + ( -!> »
*
0 < k <1 (13.89)

i -14
Tenth order Butterworth
Zeros at tv = i 1.0 ,1
li
and then approximate independently e~ ks and e— ( fc — i ) $ with all pole Bessel -
•
<3 -16 Zeros at to = ± 1.5 1
Zeros at to = ± 1.8: with
1 polynomial approximations. Thus the resulting approximation for e~ s
-18 original bandwidth also
equal to 1.8 T will have Bessel polynomials for both numerator and denominator. The
- 20 poles of the e~{*—!>» approximant will be the zeros in the final approximant,
-22 1
while the poles of the e~ ks approximant remain as poles in the final
- 24 l
/ \ approximant. For realizability, the degree of the e-dfc-l >s approximant
- 26 l V should be less than the degree of the e~ki approximant.
-280.1 \
0.2 0.3 0.4 0.6 0.8 1.0 1.5 2.0 3.0
Radian frequency, w ** A. Budak , “A Maximally Flat Phase and Controllable Magnitude Approximation,”
FIG. 13.20 -
Trans , of IEEE on Circuit Theory , CT 12, No. 2, June 1965, 279.
 396 Network analysis and synthesis

M (u) A0

1.0 1.0
^ *=
0.6
0.8 ^ 0.8

0.6 - * = 0.8^CJi 0.6 * = 1.0

(Bessel) ^
^
* = 1.0
0.4 — ( Bessel) -
\ 0.4
\ /
0.2 0.2 k = 0.7- 4 >
1 2 3 4 5 to 1 2 3 4 5
M (b)
.
FIG 13.21

As an example, consider the approximation with three zeros and four

poles.

e~ks 105
( ksY
(13.90)
+ 10(ks) + 3
45( /cs)2 + 105(*s) + 105
15
(13.91)
P - l )s]3 + 6[(fc - l)s]2 + 15 p - l )s] + 15
We then perform the operations as indicated by Eq . 13.89 to obtain

7{P - l )s]3 + 6[( k - l )s]2 + 15[( k - l )s] + 15}

e (13.92)
( ksY + 10( /cs)3 + 45(ks)2 + 105(fcs) + 105

In Fig. 13.21a the magnitude characteristic of Eq. 13.92 is plotted with 13.8 SYNTHESIS OF LOW - PASS FILTERS
k as a parameter. The phase characteristic is given in terms of deviation ft Given the system function of the low-pass filter as derived by the
of phase from linearity A <f> = w - #co) and is shown in Fig. 13.216.
methods described in Section 13.4, we can proceed with the synthesis of
The improvement in phase linearity over the all-pole Bessel approximation
the filter network. If we consider the class of filters terminated in a 1-Q
k = 1 is shown by these curves. Note as the bandwidth is increased
load, and if we let the system function be a transfer impedance,
{ k decreasing), phase linearity is improved. Figure 13.22 shows the step
response of Eq . 13.92 also with k as a parameter. Since the effect of
decreasing k increases bandwidth, the corresponding effect in the time
domain is to decrease rise time.
z iW TT 1+z
.-
5-
22
(13.93)

Budak also observes that as k decreases from unity, the poles and zeros or a transfer admittance np> = -?- (13.94)
1 + Vn
migrate to keep the phase linear. The zeros move inward from infinity
along radial lines, while the poles move outward along radial lines. we can synthesize the low-pass filter according to methods given in
Chapter 12. For example, consider the n = 3 Optimum (L) filter function

. ..

L
398 Network analysis and synthesis Topics in filter design 399
given as a transfer impedance L2 Le
0.577 -HTQTp
Z2l(s) = 3
s + 1.31s2 + 1.359s + 0.577
(13.95)
T
We see that the zeros of transmission are all at infinity. Since the numer-
* <D Yf .Cs IQ V 2
'

ator of Z21 is even, we divide both numerator and denominator by the

odd part of the denominator s3 + 1.359s. Thus we have
. .
FIG 13.24 Canonical form for filters described in Tables 13.7, 13.8, and 13.9.
1
0.577
*21 — s3 + 1.359s
(13.96) In Tables 13.7, 13.8, 13.9 are listed element values (up to n = 7) for
1.31s2 + 0.577 single-terminated Butterworth , Chebyshev (1-decibel ripple), and Bessel
*22 = s3 + 1.359s filters, respectively.13 These apply to the canonical realization for a
The structure of the low-pass filter with three zeros of transmission at transfer impedance Z21(s) shown in Fig. 13.24. If a F2 l(s) realization is
infinity is given in Chapter 12. We must synthesize z22 to give the n
reactance structure. This we accomplish through the following continued
- desired, we simply replace all shunt capacitors by series inductors and
vice versa. The element values all carry over.
fraction expansion of l /z 22 : In Chapter 14, we will consider some examples of synthesis of double
terminated filters. To stimulate the curiosity of the reader, note that the
1.31s2 + 0.577 ) s3 + 1.359s ( 0.763s voltage-ratio transfer function V 2 jV0 of the network in Fig. 13.25 is
s3 + 0.440s precisely the n = 3 Butterworth function.
Recall that in’ Chapter 12, when we cascaded two constant-resistance
0.919s ) 1.31s2 + 0.577 ( 1.415s networks, the overall system function H0( s ) was the product of the in-
1.31s2
dividual system functions H s ) H 2( s ). We can apply this property to
0.577 ) 0.919s ( 1.593s ^
networks which are not constant-resistance if we place an isolation ampli-
0.919s fier between the networks, as shown in Fig. 13.26. Since pentodes provide
the necessary isolation , our task is simplified to the design of the individual
The optimum filter is shown in Fig. 13.23. For the n = 3 Butterworth filter structures H s ) , '//2(s), . . . , Hn( s ), which we call interstage networks.
given by the transfer impedance ^
Some common interstage structures are shown in Fig. 13.27. In Fig.

1 fer impedance of the shunt-peaked network is

-
13.27a a structure known as the shunt peaked network is shown. The trans-
Z2i(s) — s + 2s + 2s + 1
3 2
(13.97)

we have Z2 I(S) = s3
1
+ 2s ’
2s2 + 1
*22(S) = s3 2s
+
(13.98)
Z2i(s) — 1
„
CS2 +
s + R/ L
SR / L + (1/ LC)
(13.99)

We then synthesize z 22(s) by a continued fraction expansion to give the 2

filter shown in Fig. 13.23. l
l
+ 1 :1 V2
1.415 h L= j

_ t
Vo]

'a
0.763 fd L
Ti.593 fd of 5 V2 .
FIG 13.25. n = 3 double-terminated Butterworth filter.

( a ) Optimum filter ( b ) Butterworth filter

.
13 More extensive tables are given in L Weinberg’s excellent book, Network Analysis

and Synthesis, Chapter 13, McGraw-Hill Book Company, 1962.

.
FIG 13.23

l - ,- .
 400 Network analysis and synthesis Topics in filter design 401
TABLE 13.7
Normalized Element Values for a Single Terminated
Butterworth Filter
Hi H2 H3
n CI L3 c3 Z4 - c5 L6 C7
Pentode Pentode
Interstaige Interstfage Interstaige
1 1.000 netwoirk netwoirk netwoirk
2 0.707 1.414 FIG . 13.26. Pentodes used as isolation amplifiers.
3 0.500 1.333 1.500
4 0.383 1.082 1.577 1.531
5 0.309
We see that Z21(J) has a real zero and a pair of poles which may be com-
0.894 1.382 1.694 1.545
6 0.259 0.758 1.202 1.553 1.759 1.553 plex depending upon the values of R , L, and C. In Fig. 13.276, a simple
7 0.222 0.656 1.055 1.397 1.659 1.799 1.558 -
R C interstage is shown, whose transfer impedance is

Z21(» = C s + 1IRC (13.100)

TABLE 13.8
Observe that all the filter transfer functions considered up to this point
Normalized Element Values for a Single Terminated
Chebyshev Filter with l -db Ripple
are made up of pairs of conjugate poles and simple poles on the a axis.
It is clear that if we cascade shunt-peaked stages and R-C stages, we can
—
n Ci L2 C3 Li C5 Ls C7 adjust the R, L, and C elements to give the desired response characteristic.
The only problem is to cancel the finite zero of the shunt-peaked stage.
1 0.509 For example, if we wish to design an amplifier with an n = 3 low-pass
2 0.911 0.996 Butterworth characteristic, we first break up the system function into
3 1.012 1.333 1.509 complex pole pairs and real pole terms, as given by
4 1.050 1.413 1.909 1.282
5 1.067 1.444 1.994 1.591 1.665 1
6 1.077 1.460 2.027 1.651 2.049 1.346 ZM (5) = (s2
7 1.083 1.496 2.044 1.674 2.119 1.649 1.712
+ + l)(s + 1)
s
(13.101)
5 + 1 1 1
s2 + s + ls + ls + l
TABLE 13.9 We then associate the individual factors with shunt-peaked or simple
Normalized Element Values for a Single Terminated -
R C stages and solve for the element values. The n = 3 Butterworth
Bessel Filter amplifier is given in Fig. 13.28.
:
n Ci L2 C3 -
Z4 C5 LS C7 o -
1 1.000 % L
c c R 5
2
3
0.333
0.167
1.000 ;
0.480 0.833
4 0.100 0.290 R
0.463 0.710
5 0.067 0.195 0.310 0.422 0.623 o
6 0.048 0.140 0.225 0.301 0.382 0.560 (a) (b) :
7 0.036 0.106 0.170 0.229 0.283 0.349 0.511
FIG . 13.27. (a) Shunt-peaked interstage. (f >) R C interstage .-

..

402 Network analysis and synthesis Topics in filter design 403

m
so that the actual value of the capacitance is

r#Vr# l
c=£
co
"
0
(13.106)

-
Q> 4= 4 1 1 Vi
Since resistances, ideally, are independent of frequency, they are unaf
fected by frequency normalization.
l; Consider, next, impedance denormalization. Suppose the actual
impedance level should be R0 ohms instead of 1 Q. Then a denormalized
impedance Z is related to a normalized impedance Z„ by
FIG . 13.28. Butterworth amplifier.
Z = R0Z „ (13.107)
13.9 MAGNITUDE AND FREQUENCY NORMALIZATION where R0 is taken to be dimensionless here. Thus, for a normalized resistor
Rn , the denormalized (actual) resistance is
In Section 13.8, we discussed the synthesis of low-pass filters with a
cutoff frequency of 1 rad /sec and a load impedance of 1 Q Filters designed. R = R0 Rn (13.108)
with these restrictions are considered to be normalized in both cutoff For an inductance, the corresponding relationship is
.
frequency and impedance level We will now discuss methods whereby
sL = R0( sL„) (13.109)
the normalized filters can be converted into filters which meet arbitrary
cutoff frequency and impedance level specifications. Let us denote by a so that the actual inductance value is
subscript n the normalized frequency variable s„ and the normalized
element values Ln , Rn , and C„. The normalized frequency variable s„ is L = R0 Ln (13.110)
related to the actual frequency s by the relation Similarly, for a capacitor we have
s
Sn =
M0
(13.102) J_ = * 0
(13.111)
sC sCn
where co0 , the normalizing constant, is dimensionless and is often taken to so that the actual capacitance is
be the actual cutoff frequency.
Since the impedance of an element remains invariant under frequency c = C" (13.112)
R0
normalization, we obtain the actual element values from the normalized
values by setting the impedances in the two cases equal to each other. For For combined frequency and magnitude denormalization, we simply
example, for an inductor, we have combine the two sets of equations to give
snLn = sL = u>g?nL (13.103) R -- R0 Rn
From this equation we then obtain the denormalized value of inductance
as RgCOo H / (13.113)

L= —
co0
(13.104) L=
^ «0
Similarly, from the impedance 1/ snCn of a frequency normalized capacitor Let us consider an actual example in design. In Section 13.8, we
C„, we obtain the denormalized value of capacitance through the equation synthesized a transfer impedance Z2i with an n = 3 Butterworth amplitude
1 1 characteristic with a cutoff frequency of 1 rad /sec and a load impedance of
(13.105) 1 Q. Let us redesign this filter for a cutoff frequency of 104 rad /sec to work
s„C„ sC

'

JL
 404 Network analysis and synthesis Topics in filter design 405
simplest transformation equation, that of low-pass to high-pass, which is
66.7 mh
s = — (13.115)
Q > 500 a
> r‘ where sn represents the normalized low pass frequency variable, 5 is the -
regular frequency variable, and co0 is the cutoff frequency of the high- pass
filter. In terms of real and imaginary parts, we have
. .
FIG 13.29 Denormalized low-pass filter . «o
a + j(o =
<*n + j >n
into a load of 500 Q. From the original network in Fig. 13.23, we take the
° (13.116)
element values and denormalize with the normalizing factors, <o0 = 104
and R0 = 500.
Then the denormalized element values are Since we are interested principally in how the jco„ axis maps into the jco
R = 500RL = 500 axis, we let <r„ = 0 so that
to
_ (13.117)
C ,= 500(104) = 0.1 /if On
(

which is the equation that transforms normalized low-pass filters to

(13.114) denormalized high-pass filters. From Eq. 13.117 we see that the point
3 (500)
L _
10,000
= 0.0067 h (
un— 1 corresponds to the point co
transformation maps the segment |
—
± ft>0. It is also clear that the
con < on to the segments defined by
| 1
I co0 < M °°seeas shown in Fig. 13.30 .
C2 = 500(104) = 0.3 [it Now let us
>
how the frequency transformations change the network
The final design is shown in Fig. 13.29.
elements. For convenience, let us denote the normalized low pass network -
elements with a subscript n , the high-pass elements with a subscript h , the
-
band pass elements with a subscript b, and the band -elimination elements
13.10 FREQUENCY TRANSFORMATIONS with a subscript‘e. For the low-pass to high-pass case, let us first consider
the changes for the capacitor Cn. The transformation is given by the
Up to this point, we have discussed only the design of low-pass filters,
while neglecting the equally important designs of high-pass, band -pass, join jo>
and band-elimination filters. We will remedy this situation here, not by
introducing new design procedures but through a technique known as a
frequency transformation, whereby, beginning from a normalized low-pass s„ plane
filter, we can generate any other form of filter. Using frequency trans-
+7 7
“o s plane

formations, the elements of the normalized low-pass filter are changed

-
into elements of a high-pass, band pass, or band-elimination filter. 0 On 0 a
Analytically, a frequency transformation simply changes one L C -
-
driving-point function into another L C driving point function . Therefore, - - j
-;«o '

the transformation equations must be L C functions themselves. Also, -

-
since we proceed from normalized low pass filters, the transformation
equations include built-in frequency denormalization factors so that the
resulting networks need only be scaled for impedance level. Consider the .
FIG 13.30 Low. -pass to high-pass transformation.
•
. i
?

/
406 Network analysis and synthesis
equation Topics in filter design 407
1
L/lS
Ch = 1.5 x 10 - 9
f
C „sn (13.118)
co0Cn
For the inductor L„, we have
= -i%- h
Lih =
- 1
-
3 h
10 RL = 50012
LA L S A -L (13.119) l
s CAs
We observe that a capacitor changes into an inductor and an
inductor . .
FIG 13.32 Transformation of low -pass filter in Fig. 13.23 into high-pass filter.
-
changes into a capacitor in a low pass to high-pass transformation (
13.31). The element values of the high - pass filter are given
Fig.
in terms of the high- pass-filter circuit shown in Fig. 13.32. Its element values are :
-
normalized low pass filter elements as
1 RL = 500 Q
Lh ~ ~ (13.120)
(OQ C. 500 3
Llh = 108( ) = IQ "
h
and Ch ( 7
1
oaLn
- (13.121) *1 (13.122)
Consider the following example. From the normalized third (500)106(£)
= 1.5 x 10 9
"
f
Butterworth filter given in Fig. 13.23, let us design a corresponding-
order
high- 500
pass filter with its cutoff frequency co0 106 rad sec and the
= /
level of 500 Q. From the low-pass filter, we can draw by
impedance U * = 106( ) = 0.333 x 10-3 h
inspection the f
Next, let us examine the low-pass to band-pass transformation (also
Low-pass High-pass Band-elimination Band-pass
-
an L C function) :
0
( o
Sn = BW
~
(13.123)
\ w0 si
Ln where, if coC 2 and, a>ci denote the upper and lower cutoff frequencies of the
band -pass filter, BW is the bandwidth
c* =
^r BW
and u>0 is the geometric mean of a>C 2 and cocl
= (OQ
2 - a>m ( 13.124)

= V 0 J2 0

—
ft>0 ( ( (
Q1 (13.125)
cn Lei
—1 Ce2
(
^ _
The low-pass to band -pass transformation maps the segment |cu„| 1 to
<
the segments coC 2 |co| > <oCl , shown in Fig. 13.33. The normalized
low-pass elements are then modified according to the following equations:

LA £
BW
L,
+ aBWs
fia
(13.126)
.
FIG . 13.31 Element changes resulting from
frequency transformations. = UiS + ~
CMs
 408 Network analysis and synthesis
Topics in filter design 409

Ml
l /w given above . The element values of the band-pass filter are given in the
"C2 following equations :
Wo
+j "Cl 6 x 104
s„ plane s plane Lx = (4 x 104)2Q) = 0.75 x 10~4 h

0 <r« 0 a Q= i = —1 x 1(T4 f
6 x 104 12
- j &

--" = 6 x 104 = - x 10-4 h

ci L2
W0 9
~"C2 (13.129)
6 X 104
. .
FIG 13.33 Low-pass to band-pass transformation.
C2 =
(4 x 104)2(-4) = 32 — x 10 4 f “

We note that the inductor 6 X 104

Ln is transformed into a series-tuned tank, L3 = = 0.25 X 10-4 h
shown in Fig. 13.31, whose elements are given as (4 X 104)2(f )
A
I C3 = = 0.25 x 10-4 f
Lbl = ~ ; 6 x 104
BW
Cbl
The capacitor Cn is transformed into a parallel tuned
— BW
co02 Ln
(13.127) Finally, the band -elimination filter is obtained through the transfor-
mation
BW
whose elements are
- tank (Fig. 13.31) (13.130)

r Bw
+
?)
where BW and to0 are defined in a manner similar to that for the band-pass
_ (13.128) filter. The transformation maps the segment of the jcon axis in Fig. 13.35a
CM
Let us transform the third-order Butterworth low-pass
= -BW ^ onto the segments shown on the jco axis in Fig. 13.356. For the low-pass
filter in Fig. 13.23 if"
into a band -pass filter with a 1 fi impedance level,
- whose bandwidth is Mt
BW = 6 x 104 rad /sec, and its band-pass is
“centered ” at &>0 = 4 x 104
rad /sec. We draw the band -pass filter shown "«
in Fig. 13.34 by the rules j wo
L2 C2 "Cl
nmn—if 0
<r„
0

/( ? ) £*! C3 Cl i V2 - j
--"Wo Cl

—"c 2

. .
FIG 13.34 Band-pass filter transformed from low-pass filter in Fig. 13.23.
(a) ( b)
. .
FIG I 3.3S Low-pass to band-elimination transformation.

J
410 Network analysis and synthesis Topics In filter design 411
to band -elimination transformation we, therefore, have the following 13.2 Find the poles of system functions with n = 3, n = 4, and n = 5
element changes: Butterworth characteristics. (Do not use the tables.)
1 13.3 Show that the half- power point of a Chebyshev low-pass amplitude
L„s„ =
( s/ LnBW ) + ( a>02ILnBWs ) response is at co = cosh pk for e « 1.
13.4 Determine the system function for the following filter specifications:
1 ( а ) Ripple of £ db in band |w| 1 ;
4 (б) at co = 3, amplitude is down 30 db.
CeiS + ( l / Lels ) (13.131)
1 _ s. go
| ( ) (a)
13.5 Compare the slopes at co = 1 of the following polynomials (for n = 3):
/(co2) = co2n
Cns„ \
CnBW 0 s /
co
(6) /(co 2) =|C„(2co2 - 1) +\ = Cn\co )
(c) /(co2) = L„(co2).
13.6 Determine the polynomials L4( co 2) and Z 5(co2). .
Observe that the normalized low-pass inductor goes into a parallel - 13.7 Expand cosh s and sinh s into power series and find the first four terms
tuned circuit and the capacitor Cn goes into a series tuned circuit, as - of the continued fraction expansion of cosh s/sinh s Truncate the expansion at .
.
shown in Fig 13.31. In Table 13.10, we have a composite summary of the n = 4 and show that H ( s ) = KjB s ).
^
various transformations. 13.8 Synthesize the n = 3 linear phase filter as a transfer impedance termin -
-
ated in a 1 D load.
TABLE 13.10 -
13.9 Synthesize the low pass filter, which, when terminated in a 1 H resistor, -
Table of Various Frequency Transformations will have a transfer admittance whose poles are shown in the figure.
Transformation
jo)
-
Low Pass to Equation

o
High-pass s n = "S ~
V

Band-pass ,*n = 4BWWJo1 + M) (j. *-r \

\ <T
;

Band-elimination *n =
“o (\-
\
<
BW
0

-)
c°o + a /
*
v
v -«V 0

Problems PROB 13.9 . . ;

13.1 Find the transfer impedance Z21 = V2lli for the filter shown in the 13.10 Determine the asymptotic rate of falloff in the stop band of : ( a )
figure. What should L be in order for |Z21(/co)| to be maximally flat ? optimum filters ; (b ) linear phase filters.
13.11 Synthesize the n = 3 and n = 4 Butterworth responses as transfer
impedances terminated in a load of 600 ft with a cutoff frequency of 10* rad /sec.

o>.
L

1 V2
I 13.12 Synthesize a Chebyshev low-pass filter to meet the following speci-
fications:
{a ) load resistor, RL = 600 ft
(6) £-db ripple within pass band
(c) cutoff frequency = 5 x 10s rad /sec
PROB 13.1 . ( d ) at 1.5 x 106 rad/sec, the magnitude must be down 30 db.

.
 412 Network analysis and synthesis
13.13 Synthesize n = 3 Optimum and linear phase filters to meet the following
specifications :
( a ) load resistor = 10ISir- (QOC> OJ - ii n
( b) cutoff frequency = 106 rad/sec.
13.14 Design an n = 4 Butterworth amplifier with the following specifi-
T cations :
(a) impedance level = 500 Cl
(b) cutoff frequency = 103 rad/sec.
chapter 14
13.15 Synthesize a high-pass filter for a given transfer admittance terminated
in a 103- D load, whose amplitude characteristic is Optimum (L), with a cutoff
frequency of ro0 = 104 rad/sec .
The scattering matrix
13.16 Synthesize : (a) a band - pass filter ; (6) a band-elimination filter, with
maximally flat ( n = 4) amplitude response with a>c2 = 8 x 104 and cocl =
2 x 104.

14.1 INCIDENT AND REFLECTED POWER FLOW

In this chapter, we will devote our attention to certain power relation-

ships in one- and two- port networks. The characterization of a network in
terms of power instead of the conventional voltage-current description is a
helpful analytical tool used by transmission engineers. It is especially
important in microwave transmission problems where circuits can no
longer be given in terms of lumped R, L, and C elements. In the power-
flow description, we are concerned with the power into the network, which
we call the incident power and the power reflected back from the load,
which is the reflected power. A convenient description of the network in
terms of incident and reflected power is given by the scattering matrix, I
which is the main topic of discussion in this chapter. !
It is convenient to think of incident and reflected power when dealing
with transmission lines. Therefore, we will briefly review some concepts in :
transmission line theory. For a more comprehensive treatment of trans
mission lines, the reader is referred to any standard text on wave propa-
- ;

gation.1 Consider the transmission line shown in Fig. 14.1. The voltage
at any point down the line is a function of x, the distance from the source.
The parameters which describe the transmission line are given in the
following :
R = resistance per unit length
G = conductance per unit length
L = inductance per unit length
C = capacitance per unit length
1
See, for example, E. C. Jordan, Electromagnetic Waves and Radiating Systems,
Prentice-Hall, Englewood Cliffs, New Jersey , 1950.
413

414 Network analysis and synthesis The scattering matrix 415

Then we see that the reflected wave is zero.

+1
1

l( x )
V (x)
~1
1 2

Z
vyL = 0
Since eyL cannot be zero, we see that the coefficient VT is identically zero
(14.8)

Vi for this case. As a result, the reflected wave at any point x is zero. Also,
the impedance at any point x down the line is equal to Z0 as seen from
1' 2' Eq . 14.5 with VT = 0. With these brief thoughts of transmission lines in
mind , let us turn our attention to the main topic of this chapter, namely,
FIG . 14.1 . Transmission line. the scattering parameters .
Given these parameters, we can now define the impedance per unit
length as 14.2 THE SCATTERING PARAMETERS FOR A
( 14.1) ONE-PORT NETWORK
Z JR T jcoL
and the admittance per unit length as For the one- port network shown in Fig. 14.2a, consider the following
definitions. The incident parameter a is defined as
Y=G + jcoC (14.2)
The characteristic impedance Z0 of the line is given in terms of Z and Y as
z = Vz/ y
0 (14.3)
a=

and the reflected parameter b, is defined as

+ y /R j0
^ (14.9)

and the propagation constant is

sfZY
y= (14.4) (14.10)

With these definitions in mind , let- us turn to the general equations for where R0 is an arbitrary, positive, dimensionless constant called the
the current and voltage at any point x down the line reference impedance factor. For the transmission line described in
Section 14.1, if the characteristic impedance Z0 = R0 , then we can describe
V( x) = Vie~ yx + Vreyx the incident parameter in terms of the incident voltage as G>
I ( x) = Iie yx - ITeyx
~

= he-r* _ Lrer *
(14.5)
a= (14.11)
Z 0 Zo Similarly, b can be expressed in terms of the reflected wave as
The terms with the subscript i refer to the incident wave at point x and the
terms with subscript r refer to the reflected wave at x Solving Eqs. 14.5. b= -V f (14.12)
simultaneously, we obtain explicit expressions for the incident and reflected Vz 0
waves
~ yx
Vie = HV ( x ) + Z01( x )] (14.6) I
o-
Vreyx = UV ( x ) - Z0 /(*)]
Consider the case when a transmission line of length L is terminated in
V
One-port
network . 6
One-port
network

its characteristic impedance, that is,

( a) (b)
(14.7)
m FIG . 14.2. Scattering parameters of a one -port network .
 416 Network analysis and synthesis The scattering matrix 417
Thus we see that the parameters a and 6 do indeed describe an incident- „
resistor R , as depicted in Fig. 14.3, we Rg
reflected wave relationship in a one port network, as depicted in Fig. 14.26.
- will show that when we choose the -o-
To give further meaning to a and b, consider the power dissipated by the reference impedance R0 to be equal to +
-
One port
-
one port network R, V network
= £ Re VI* (14.13) K (14.20)
where I* denotes the complex conjugate of /. From Eqs. 14.9 and 14.10 JR,
we solve for V and I in terms of the incident and reflected parameters to The proof follows. By definition , the .
FIG 14.3
give incident parameter a is given as
V = ( a + b ) ffR0
(14.21)
b (14.14)
l
I =
^/
s Ro
Then the power dissipated by the one-port network is
From Fig. 14.3, we have V „ — IRg = V

K
(14.22)

(14.23)
and / =
P = \( aa* - bb* ) R0 +Z
(14.15)
= KM* - W ) Substituting these equations for V and I into the expression for a in Eq.
14.21, we obtain
where, again, the asterisk denotes complex conjugate. The term \aa* 1
can be interpreted as the power incident* while\bb* can be regarded as the •a =
power reflected. The difference yields the power dissipated by the one-port
network. (14.24)
The incident parameter a and the reflected parameter b are related by the
equation
b = Sa (14.16)

where S is called the scattering element or, more commonly, the reflection : 2 VK 0
a one-
coefficient. From the definitions of a and b we can make the following Consider once more the expression for the power dissipated in
substitution : port network :
p = \{ aa* - bb* ) (14.25)
( Jk VjR 7
° ) “
+ V*o/ ) (14.17)
Factoring the term aa* = \a\ from within the parentheses, P
2 becomes
\
S= -
R Z
Solving for S , we obtain 0
(14.18)
Z+R 0 V I « I «/
2
(14.26)
where Z is the impedance of the one-port network
= M!(i
_ ls| ) 2

Z = ^ I
(14.19)
When we choose the reference impedance to be equal to
, ?
the source resist-
?
ance, i.e., when R0 = R< , then S = 0 and
*2, ?
A further useful result is that when the impedance R0 is set equal to the
impedance Z, the reflected parameter 6 = 0. (14.27)
-
For the one port network excited by a voltage source V with a source „ Pa= 2 8R ,

418 Network analysis and synthesis The scattering matrix 419

Rg
dW
where PA represents the available gain or
available power of a voltage source VQ with
3. For an open circuit S = 1, and for a short circuit S' =
shown to be true from the equation
— 1. This is
a source resistance RQ For the case of a -. s = 2x - 1
6* Rg = Z
one-port network, the available gain is
defined as the power dissipated in the one-
port network when the impedance of the
+1
For an open circuit z = oo, so that 5 = 1. For a short circuit z = 0 ;
(14.34)

FIG. 14.4
network Z is equal to the resistance of the
source R„. As a result of this definition, we
therefore S = 1.
—
Before we proceed to the next property, let us consider the following
see that for the one-port network shown in definition.
Fig. 14.4, the power dissipated in Z with Z = R„ is DEFINITION . A bounded real function F( s) is defined by the con-

pKje*'
PA = -l !
8^
±Ra
2
~ (14.28)
ditions
( а ) |F(s)| K for Re s > 0 la 6 *Ju* s
<r*
- jib JUSl Jl
(б) F( s ) is real when s is real.
The available gain thus represents the maximum available power at the In (a), K denotes any positive real constant.
terminals of the voltage source. 4. If z = Z / R0 is a positive real function, then S is a bounded real
From this discussion , it is apparent that the value of the reference im - function .
pedance R0 should be chosen equal to the source impedance R A „. The proof follows from the equation,
-
standard procedure is to assume a 1 0 source impedance and denormalize
when necessary, that is, let 5 = (2 - l )/(2 + 1 ).
z -1
S= (14.29) From the positive real condition (a) Re z( s ) > 0, when Re s > 0, we see
z d- 1 that
0 J
_
Im 2(5) - [1 - Re z( s ) ]
where z
—* —Z 0
(14.30) j Im z(s) + [1 + Re z(s)]
(14.35) .

ez(s)]2)*
(S(s)| = Im2 s + 1 - R
{Im z(
2
z( ) [
Next, let us briefly consider some of the important properties of the so that <1 (14.36)
scattering parameter S for a one-port network. s)2
+ [1 + Rez(s)] /
1. The magnitude of S along the jco axis is always less or equal to unity
when Re J > 0. (6) Where s is real, z(s) is real. Then

for a passive network, that is, S = (z - 1 )l( x + 1)

Wl 1 (14.31) must be real. Thus the scattering parameter for a passive network is a
bounded real function .
This property follows from the fact that the power dissipated in a passive
network is always greater or equal to zero. Since the power can be
expressed as 14.3 THE SCATTERING MATRIX FOR A
-
TWO PORT NETWORK
P =1

^
2
(1 - |S| ) £ 0 (14.32)
In this section we will extend the concepts developed for one port -
networks discussed in Section 14.2 to two-port networks. In the two-port
we see that iwr < i (14.33)
network shown in Fig. 14.5, we are concerned with two sets of incident and
2. For a reactive network |S'(yw)| = 1. This property follows from the reflected parameters {a!, 6J at the 1-1' port, and (a 2, 62} at the 2-2' port.
fact that the power dissipated in a purely reactive network is zero. These parameters are defined in similar manner as for the one-port

i
 420 Network analysis and synthesis The scattering matrix 421
l 1

Two- port
network
02
* R2
b2
{>
. ..
FIG I 4 S Scattering parameters for a two port network - . I I
S2
network, that is,
FIG . 14.6

.
In Eqs. 14.41, the parameter S is called the input reflection coefficient ,
^ -
S2l is the forward transmission coefficient ; S12 is the reverse transmission
* 1 y / Roih ) -
coefficient , and S22 is the output reflection coefficient . Observe that all
four scattering parameters are expressed as ratios of reflected to incident
(14.37) parameters.
Now let us examine the physical meaning of these scattering parameters.
First, consider the implications of setting the incident parameters a and a2 ,
to zero in the defining relations in Eqs. 14.41. Let us see what the con -
dition a2 = 0 implies in the definition for the forward reflection coefficient
where 7?01 and R02 are the reference impedances at the input and output
ports respectively. Sa = h
tt
The scattering parameters Su for the two-port network are given by the 1 02 = 0

equations Figure 14.6 shows the terminating section of the two-port network of Fig.
b\ Sna + S12a2
b2 =
—
-f- S22a2
,
(14.38)
14.5 with the parameters a2 and b2 of the 2-2' port shown. If we treat
-
the load resistor R2 as a one port network with scattering parameter

In matrix form the set of equations of Eqs. 14.38 becomes S2 = R^ 2 — RQR 2

(14.42)
2 +
^ii rs,,
02
•S’i 2i r where R02 is the reference impedance of port 2, then a2 and b2 are related
_S 522J \_a _
21 2
(14.39)
by2
where the matrix
'
Sn Sa- a2 —Sb 2 2 (14.43)
[ S\ =
_S 21 S22 _ (14.40) When the reference impedance R02 is set equal to the load impedance R2 ,
is called the scattering matrix of the two-port network. From Eqs. 14.38, then S2 becomes

= Ro Roi = 0
~
we see that the scattering parameters of the two port network can be
expressed in terms of the incident and reflected parameters as
- S2
R
2

22 + i^ 02
(14.44)

.
*, = *al> OJ =0
S 12
a2 lai = 0 (14.41)
so that a2 = 0 under this condition. Similarly, we can show that, when
,
a = 0, the reference impedance of port 1 is equal to the terminating

s„=
°^ 1
* From the viewpoint of the load resistor Rt , the incident parameter is b 2 and the
ai parameter is a 2 .
reflected
(22
=0 2 21=0
(

A .
ELV

422 Network analysis and synthesis The scattering matrix 423

,
impedance (i.e., R0l = R ). We see as a result of this discussion that the then we can express a as ,
,
conditions a = 0 and a2 = 0 merely imply that the reference impedances Ki
R01 and R02 are chosen to be equal to the terminating resistors R2 and R ,
respectively.
2 2 JR , (14.51)
Since a2 = 0, we have the equation
Next, let us consider the relationship between the driving-point imped
ances at ports 1 and 2 and the reflection coefficients Sn and S22 . Let us
-
denote the driving-point impedances at ports 1 and 2 as a2 = 0 = -2
1
+ V*«*« ) (14.52)

Z .- S
h
z, = s
h
(14.45) from which we obtain V2
(14.53)

From the equation .

S, =-
ai 1 22 = 0
(
(14.46) Consequently, b2

We can write Su =
M(^i /V^oi ) ~ / Rofli ] \
(14.47) Y*_ (14.54)

which reduces easily to Sn

_ — IKVJjRn ) + yfR h ]
R01 ^ (14.48)
V* 2

Finally, we can express the forward transmission coefficient as

+ ZI R0 i V2 IJR 2

— — z2 R02 S2l =
Similarly, we have S 22 (14.49)
z2 + R02 2V ,
These expressions tell us if we choose the reference impedance at a given = (14.55)
port to equal the driving-point impedance at that port, the reflection Ki .
RZ= JRO 2 R i= Ro 1
coefficient of that port will be zero, provided the other port is terminated In similar fashion , we find that when port 1 is terminated in R01 = Rr and
in its reference impedance. when a voltage source Vg 2 with source impedance R2 is connected to port 2,
Next, let us derive some physically meaningful expressions for the then
forward and reverse transmission coefficients S2l and S12 Consider the . S , 2V , (14.56)
definition for S 21 2
KM
St1 = h (14.50) We see that both S12 and S21 have the dimensions of a voltage-ratio transfer
ai 1 12 =0 (
function . Indeed , if R01 = R02, then S12 and S21 are simple voltage ratios.
As we have just seen, the condition a2 = 0 implies that the reference im- It is seen that for a passive reciprocal network, .
pedance R02 is set equal to the load impedance R2 , as seen in Fig. 14.7. = Sl 2
Now let us consider as an example the scattering matrix of the n : 1 ratio
If we connect a voltage source Val with source impedance R
0l = R „ ideal transformer in Fig. 14.8a. Recall that for an ideal transformer

2
V1 = .nV2, /1 = - -n I ,1
(14.57)
Two-port b2 Assuming first that R01 = R02 = 1, let us find Sn by terminating the
network RO2 = R2
2-2' port in a 1-0 resistor, as shown in Fig. 14.86. Then
1'
L 7
(14.58)
FIG 14.7 . ~I2
 424 Network analysis and synthesis The scattering matrix 425
We obtain S21 by connecting a voltage source VgX with a source impedance
R01 = 1 O at the 1-1' port and terminating the 2-2' port with a resistance
R02 = 1 Q, as seen in Fig. 14.8c/. Since KJA = «2, the equivalent circuit
of the ideal transformer as seen from the voltage source as a l -£2 imped-
ance in series with an n2-ohm resistance (Fig. 14.8<?). Then Vx can be
expressed in terms of Val , as
*
K = n Kin
2
(14.62)
+1
Since V 2 = V1/ n , we have
V2 = Kin (14.63)
n2 +1
Since i?01 = R02 = 1 Q, S21 is
321
*
— Ki «2
2n
+1
(14.64)

We can show in similar fashion that

2n
S12 = n2 (14.65)
+1
Therefore the scattering matrix for the ideal transformer is given as

fn
n
2

2
—
+
1
1 n 2
2n
+1
1
S = 1 - n2
(14.66)

n2
2n
+ 1 n2 + 1 _
As a second example, let us find the scattering matrix for a lossless
transmission line of length L terminated in its characteristic impedance, as
shown in Fig. 14.9. If we assume that R01 = R02 = Z0, then the reflection
(e ) coefficients are
.
FIG. 14.8 Determination of scattering parameters for an ideal transformer.
- — —
Sn — ° ° 0 S 22 (14.67)

so that .
z = 5w
h
(14.59)
l f-
A> + A

1 2

Su = n 2 — 1
2
n 1
From Eq. 14.48, we have (14.60) \ Zo
+ 20 v2
Next we terminate the 1-1' port in a 1-Q resistor (Fig. 14.8c). We obtain, +,
Yn
as we did for Su ,
_
—
2
s22 (1 In2 ) - 1 1 - n2
2
(14.61) l' 2'
(1In ) + 1 1 + n . .
FIG 14.9 Lossless transmission line.

I
426 Network analysis and synthesis The scattering matrix 427
This result is not implausible, because a transmission line terminated in where T denotes the transpose operation , and
its characteristic impedance has zero reflected energy. To determine S21,
we terminate the line in Z0 at both ends and connect at the 1-1' port a fli
[« ] =
voltage source Vgl , as depicted in Fig. 14.9. Since" the transmission line U 2J
has zero reflected energy, that is, (14.78)

bx = a2 = 0 [h ] = KI
IK
then F* = Vgle ' L - (14.68) Since [6] = [ S ] [a ] , then
[ b* ] T = [n *]:r[S*] 2’ (14.79)
From the equation = (14.69) Equation 14.77 can now be rewritten as
W Ki
we obtain S 2l = 2e~ yL (14.70) 2 P = {[u *F[fl ] - [« *] r[S*]I'[S][a ]}
(14.80)
In similar fashion, we find that = [^]T[[« ] - [S*f [S]][fl ] > o
„ = 2e~ L
S /

Therefore the scattering matrix for the lossless transmission line is

(14.71) This then implies that the determinant of the matrix [[«]
must be greater or equal to zero, that is,
— [S*] r[S]]

0 Det [[« ] - [S*]T[S]] >0 (14.81)

S = (14.72)
L2e-* L Consider the special but, nevertheless, important case of a lossless
network. In this case P = 0, so that
14.4 PROPERTIES OF THE SCATTERING MATRIX
[S*f [S] = [« ] (14.82)
Having defined the scattering matrix of a two port network in - A matrix satisfying the condition in Eq. (14.82) is unitary. For a lossless
Section 14.3, let us consider some important properties of the scattering two-port network
matrix. From the general restriction for a passive network that the net
-_
' '

Sn* S2l* Sn S12 1 0

power delivered to all ports must be positive, we obtain the condition [S*] [S]
^Si2* S 22* S21 S 22 .0 1. __
(14.83)
_
P= + a2a2* - bxbx* - b2b2* ) 0 (14.73)
From this equation, we have the following conditions for the scattering
Equation 14.73 follows from the fact that the power delivered to the 1-1' matrix
port is
Pi = i(«x«i* - W) (14.74)
Sn *Su + S2X* S21 1 (14.84) —
and the power delivered to the 2-2' port is Su* sn + S22* S2i =0 (14.85)

P2 = i( a2a2* - b2b * ) 2 (14.75) „

S *s12 + S21* S22 = 0 (14.86)

The total power delivered to the network is then SI2*S12 + S22* S22 = 1 (14.87) :
:
Pi + P*
P= (14.76) Note that Eqs. 14.85 and 14.86 are conjugates of each other. If the L
network is reciprocal, then S21 = S12 and
which is exactly the expression in Eq. 14.73. In matrix notation, the power
delivered to the network is „
|s (/ C0)|2 + |S210)l 2
=1
(14.88)
p = - [ b* ]Tm 0 (14.77) |S22(;co) P + |S21( »| = 1 2

ii

::
 \
428 Network analysis and synthesis The scattering matrix 429

14.5 INSERTION LOSS

Two-port '
network Pv S **2 In Section 14.4, we described the forward and reverse transmission
coefficients in terms of voltage ratios . Perhaps a more appropriate
description of a transmission coefficient is in terms of a power ratio rather
FIG. 14.10
than a voltage ratio . In this section we will show that |5’21(yw)| 2 and |S12
(Jco )| 2 can be expressed in terms of power ratios . We will then introduce
from which it follows that for a lossless reciprocal network |5'11(/w)| = the very important concept of insertion loss and finally relate |S21(/co)|2 to
|S22(/ <w)| < 1 and |S21(yw)| < 1 . Also it is clear that when \ S21( jco ) \ = 0 the insertion power ratio .
(i .e . , when there is a zero of transmission), then | .Su( /a>)| = 1 . This _ Consider the equation for S21 in the two-port network shown in Fig.
condition states that all the power that has been delivered to the network 14.5,
from port 1-1' is reflected back to port 1-1'. 2V2
At this point, it might be profitable to discuss why we use scattering S 21 ( 14.91)
matrices. What are the advantages of the scattering description over Ki
conventional descriptions ? Let us discuss three major reasons for the
scattering formalism.
From the equation >) I = Sn( jco )S *( joj )
|S21( 2

2
21 ( 14.92)
V ( jco )\ l 2 R
1 . Many networks do not possess an impedance or admittance matrix . we obtain |S (»| = \
21
2 2 2

For example, an ideal transformer has no Z or Y matrix because its \ V, ( ja> )'\ l Ri
i
*
elements are not finite. However, as we have seen, the ideal transformer P2
( 14.93)
can be described by a scattering matrix . Carlin states3 that all passive PA
networks possess scattering matrices.
2. At high frequencies, incident and reflected parameters play dominant where P2 is the power dissipated by the load R2 , and PAl is the available
roles in problems of transmission, while voltage-current descriptions are gain of the generator VgX . Similarly , we have
relegated to the background. Then the scattering matrix is necessarily the
more powerful description of the system . Note that the voltage standing- ISisO^)!2 = — ( 14.94)
wave ratio ( VSIVR ) is given in terms of a reflection coefficient S as PA*
We see that both |S’21(/w)|2 and \ Sn( jw )\ 2 are power transfer ratios which
s
= + |\S|\
1 relate the power dissipated at a given port to the available power in the
VSWR ( 14.89)
1- other port .
3 . In networks where power flow is a prime consideration (e .g . , filters), Now let us examine the idea of insertion loss . Consider the network
the scattering matrix is very useful . For example, in the network given in shown in Fig. 14.1 la. Between the terminals 1-1' and 2-2' we will insert a
Fig. 14.10, if PA represents the available power from the generator and P2 two-port network, as shown in Fig. 14. lli. Let us denote by V20 the
is the power dissipated in the load R2 , then we can show that the magnitude- voltage across the load resistor R2 before inserting the two-port network,
squared forward transmission coefficient is and by V 2 the voltage across R2 after inserting the two-port network . A
measure of the effect of inserting the two-port network is given by the
|S2l0' )|2 ( 14.90) insertion voltage ratio IVR , which is defined as
" = TT
PA
We will discuss this point in more detail in Section 14.5 . IVR 4 (14.95)
V2
8
H. J. Carlin, “The Scattering Matrix in Network Theory,” Trans. IRE, CT 3, - Another method of gaging the effect of inserting the two-port network
No. 2, June 1956, 88-96 ; see his extensive bibliography . is to measure the power dissipated at the load before and after inserting the

:
£L
V
430 Network analysis and synthesis The scattering matrix 431
Ri l 2 The insertion power ratio can then be expressed as
+
+
P*0 l ll 8
R2 V20 A

*
p
_
^
\ V2\ ( Rt + R 2 ) 2 2 ( 14.101)

r 2' In the special case when the source and load impedances are equal , that
(a) is ,
Rl = R’i — Roi = RQ 2 ( 14.102)
the reciprocal of the squared magnitude of the forward transmission
2
+ coefficient in Eq . 14.93 is equal to the insertion power ratio
Inserted
two-port R2 V2 1 P 20
network ( 14.103)
\ Su jw )\'
( P2
1' T When Pj R2 , then
( b) P 20 4 R R2 1
~ ^ ( 14.104)
.
FIG 14, 11 P2 ( R* + R 2 f |S21( »| 2

In any event, we see that the magnitude- squared transmission coefficients

two-port network . If P 20 is the power dissipated at the load before the
two-port network is inserted, and if P 2 is the power dissipated after inser- 11 2(7“)12 and |S12( yco)|2 can be regarded physically as equivalent insertion
^
power ratios . In Section 14.6, we will use this relationship in the synthesis
tion, then the insertion power ratio of the two-port network is defined as
of double- terminated filter networks .
P 20
e 2* £ ( 14.96)
P2 14.6 DARLINGTON ’S INSERTION LOSS FILTER SYNTHESIS
If we take the logarithm of both sides, we obtain
In this section, we will consider a filter synthesis procedure first proposed
P by Darlington in a classic paper in 1939.4 We will use scattering matrix
‘
a = 10 log 20 ( 14.97)
P2 notation to describe the essence of Darlington’s original work . Our
where a is the insertion loss of the two-port network. In terms of the coverage will be restricted to the class of low-pass filters which are termi -
circuit given in Fig. 14.11 , we can calculate P20 from the relation nated in equal source and load impedances , P01 = R02 = R0 , as shown in
Fig. 14.12 . For normalizing purposes we will let R0 be equal to 1 Q .
Jk R2 ( 14.98)
^ 20 =
Ri + R 2 Ro

Then P20 is 20 =
\VJ 2 +
^ 2 R, Zi ( sK
Filter
network
R* \ Ki\*
( 14.99)
2(
* i + R*?
The power dissipated by the load after inserting the two-port network is FIG. 14.12
given by
l 2|-2
P = TZT ^ (14.100) 4
S. Darlington, “Synthesis of Reactance 4-Poles which Produce Prescribed Insertion
* 2R Loss Characteristics,” J . Math . Phys. , 18, 1939, 257-353.

I
!
 :

432 Network analysis and synthesis The scattering matrix 433

§3
Recall that when the source and load
impedances are equal, then the insertion
power ratio is equal, to the reciprocal of
—vw —w
u in
\
La Ln

!j 3 —
(a)
1 03
.
|5ii(y®) l * that is,
Ao
p2
1
IV»I 2
( 14.105)
O''.. ^ Zi( s )
C2
—
C' n 1"
— in v2

Expressed as a loss function, the insertion .

is ; db
power ratio is
A = 10 log ^ or
FIG 14.14. Canonical form for double terminated low

\ Sn( ja> )\2 = 1

-
1
+ e2 c„V)
-pass filters.

(14.112)

0 1 0) = -101og |S21(./ w) r db (14.106)

where C „( co ) represents an nth-order Chebyshev polynomial.
( b)
In circuit design, the specification of an
FIG. 14.13 insertion loss A ( Fig. 14.13a) is equivalent Example 14.1 . Let us synthesize a low-pass filter for the specification
to the specification of the amplitude-
1
squared transmission coefficient shown in Fig . 14.136. One of the most I W ») l 2 = 1 (14.113)
ingenious techniques given in Darlington’s synthesis procedure is the re- + to8
duction of insertion loss synthesis to an equivalent L- Csdriving-point -
which represents a third order Butterworth amplitude characteristic. The load
,
synthesis problem . This technique can be developed in terms of scattering and source impedances are R02 = R01 = 1 Cl . First we find |511( yft))|2 as
parameters. Our initial specification is in terms of |S21(/co)|. For an L C - 1
two-port network |Su(y«0 l 2 = 1 -
|SuO)l 2 = 1 - 1 0) 1* (14.107) 1 + CO6

Next,
^
( s ) is obtained from the magnitude-squared function
to8
Sn co8
(14.114)
1 +
„
S (s)Su(-s) = 1 - |S120 m)|2| :
= ; (14.108) Letting j<o = s in |Sn(yco)|2, we obtain
_ —R Sn( s )Sn(.-s ) = - 1
i8
(14.115)
Then from the equation
Zt + R0
0
(14.109) - 58
which factors into
we obtain the driving-point impedance

R- o + Sn
1 (s)
Sn( s) Sn( s ) —— (1 + 2s + 2s2 + *3)(1 - 2s + 2s - s ) 2 3
(14.116)

Zi(s) — 1 Sn(s)
( 14.110)
s3
so that Sn( s) is Sill*) = s3 + 2s2 + 2s + 1 (14.117)
shown in Fig. 14.12. We then synthesize the network from Z s ).
We will restrict our discussion here to low-pass filters given by the
lossless ladder structure terminated at both ports by l -f ) resistors in
^ Next, Zj(s) is obtained from the equation

Fig . 14.14. These low-pass filters can take the form of a Butterworth or 1 + ^11(5)
Chebyshev specification for |S21(/w)|2, that is,
ZM =1
— S1Y( s )

1 2s3 + 2s2 + 2s + 1
|S2iaco)|2 = (14.111) (14.118)
2s2 + 2s + 1
1 + co2n

434 Network analysis and synthesis

The scattering matrix 435
TABLE 14.1
Normalized Element Values for a Double Terminated -
Butterworth Filter ( Equal Terminations)
n Ci L2 C3 Li C6 L6 C7
1 2.000
2 1.414 1.414
.
FIG 14.15 3 1.000 2.000 1.000
4 0.765 1.848 1.848 0.765
We next perform a Cauer ladder expansion for Zx(s). 5 0.618 1.618 2.000 1.618 0.618
2s3 + 2s + 1)2 + 2s2 + 2s + l (s . 6 0.518 1.414 1.932 1.932 1.414 0.518
2s^3 + 2s2 + j
7 0.445 1.248 1.802 2.000 1.802 1.248 0.445
J + l )2s2 + 2s + I (2s
TABLE 14.2
2s2 + 2s
l)s + l ( s
Normalized Element Values for a Double Terminated
Chebyshev Filter with l -decibel Ripple ( Equal Terminations)
-
s n Cj L2 C3 Ln C5 Ts C7
w
1 1 1.018
= 3 2.024 0.994 2.024
-
The low pass filter is thus synthesized in the structure shown, in Fig. 14.15. 5 2.135 . 1.091 3.001 1.091 2.135
An equivalent realization for the double-terminated filter is obtained if 7 2.167 1.112 3.094 1.174 3.094 1.112 2.167
we use the equation
SuW = & Go
)~
* (14.119)
TABLE 14.3

Then, assuming G0 = 1 mho

YM + Go Normalized Element Values for a Double Terminated
Bessel Filter ( Equal Terminations)
-
ya(s) = 1 + Sn( s )
1 - Su(s)
(14.120)
n Ci X2 C3 -
Lx C5 LQ C7

1 2.000
The canonical realization for 7J (J) is shown in Fig. 14.16. Tables 14.1 , 2 1.577 0.423
14.2, and 14.3 list element values (up to n = 7) for double-terminated 3 1.255 0.553 0.192
Butterworth, Chebyshev ( 1 -db ripple) and Bessel filters, respectively. 4 1.060 0.512 0.318 0.110
These apply to the canonical realization for Y s ) given in Fig. 14.16. If 5 0.930 0.458 0.331 0.209 0.072
a Zi( s ) realization in Fig. 14.14 is desired, we simply replace all shunt
capacitors by series inductors and vice versa .
^ 6
7
0.838
0.768
0.412
0.374
0.316
0.294
0.236
0.238
0.148
0.178
0.051
0.110 0.038

Lt U U Note that the even orders for the double-terminated Chebyshev filters
'TKRP ISW' are not given. This is because the even-ordered Chebyshev filters do not
meet realizability conditions for minimum insertion loss at s 0.5 We
have only given tables for equal source and load terminations. For other —
7k Ci iCj C7 IQ vt- possible realizations, the reader should consult L. Weinberg’s excellent
book . 8
:

I
4L. Weinberg, Network Analysis and Synthesis, McGraw-Hill Book Company,
.
FIG 14.16 . Canonical form for filters in Tables 14.1, 14.2, and 14.3. New York, 1962, p 589 . .
• .
Ibid , Chapter 13.
;
 436 Network analysis and synthesis The scattering matrix 437
14.5 Find the insertion voltage ratio and insertion power ratios for each of
Problems the networks shown . These networks are to be inserted between a source
14.1 Determine the reflection coefficient S for the one- port networks shown impedance Rg = 2 Cl and a load impedance = 1 O. From the insertion
in the figure. power ratio, find |S21( /co)|2.
2h 2h
2h
—'TTOTT
°
1 r -^W?rv-° 2

T =»f
4 = if
l'c *2'
(a)
(a) (b) (c ) 1h
1» o2
PROB . 14.1

14.2 For the one-portnetwork in Fig. 14.3, let R = Rg . if the incident

0
parameter is a = VJlVRq > find the reflected parameter b .
=t= if :if

l'o- o 2‘
14.3 For the network in Prob. 14.1 , determine |S( <u)|. Show that the
/
scattering elements 5 for the networks in Prob. 14.1 are bounded real functions .
(b)

14.4 For each of the networks shown, find the scattering matrix for
7?02 = f • R0l = 2h

212 412
VA
if
1h
1 . 112 112
o2
JL
<c )
212
.
PROB 14.5

412
14.6 Synthesize low-pass filters for the specifications
l'o l'o
( a) 1
<b ) («) \ S21( jo> )\* = 1
+ to 4

h 1
-o .
( b) IWOI* = 1 + CO8
Vi )( v2 14.7 Synthesize an equal-ripple low-pass filter such that 20 log |S21( / <a)[ has
,

o- 2S at most i-db ripple in the pass band and an asymptotic falloff of 12 db/octave in
Gyrator the stop band .
(c)

Negative Pl = nV2
X i
impedance +
V2 Vi Z v2
converter

(d )
(e )
.
PROB 14.4

:j
Computer techniques in circuit analysis 439
The majority of circuit-analysis programs, however, perform their
calculations in the frequency domain. The program user is only required
to specify the topology of the network, the element values, and what
transfer functions he wishes to obtain . The computer does the rest. It
calculates the specified transfer functions in polynomial form, calculates
chapter 15 the poles and zeros of these functions, and can also provide transient
response and steady-state response, if desired. With versatile input-
output equipment, the output can also provide a schematic of the original
Computer techniques in network, as well as plots of time- and frequency-response characteristics.
Time- and frequency-domain analysis
circuit analysis The time- and frequency-domain analysis programs can be used in
-
conjunction with the circuit analysis programs or independently. The
-
time domain programs depend upon solving the convolution integral

15.1 THE USES OF DIGITAL COMPUTERS IN

CIRCUIT ANALYSIS

The advent of the high-speed computer has made routine many of the
formerly tedious and difficult computational aspects of circuit theory.
Digital computers have become widely used in circuit analysis, time and
frequency-domain analysis, circuit (filter) design , and optimization or
iterative design. We will discuss these aspects in general in this section .
In succeeding sections, we will discuss some specific circuit-analysis
computer programs.
Circuit analysis
w
The primary objective of a linear circuit-analysis program is to obtain
responses to prescribed excitation signals. These programs are based on
many different methods : nodal analysis, . mesh analysis, topological
formulas, and state variables. Most of them can handle active elements ;
such as transistors and diodes by means of equivalent circuit models.
The state-variable programs based upon Bashkow’s A matrix formula-
tion1 perform their calculations directly in the time domain via numerical
integration and matrix inversion. The outputs of these programs provide
impulse and step response in tabular form . If the excitation signal were
given in data form, the state-variable programs would calculate the
response directly in the time domain .
1
T. R . Bashkow, “The A Matrix—New Network Description,” IRE Trans, on Circuit
Theory , CT-4, No. 3, September 1957, 117-119. FIG . 15.1 . Frequency response of fifth-order Butterworth filter . ii;
FIG . 15.2. Phase response of fifth-order Butterworth filter .
438
 440 Network analysis and synthesis Computer techniques in circuit analysis 441
In Section 15.2 we will examine further details of a typical steady state
analysis program.
-
Circuit (filter) design
The filter design programs are probably the most convincing argument
for the use of computers in circuit design. Designing insertion loss filters2
to meet certain amplitude requirements requires considerable numerical
calculation even in the simplest cases. The use of digital computers in
insertion loss filter design is clearly a logical alternative. The amount of
programming time for a general filter synthesis program is considerable.
However, the ends certainly justify the means when large numbers of
filters must be designed to meet different specifications.
An outstanding example of a digital computer program for filter design
is the one written by Dr. George Szentirmai and his associates.3 The
program is complete in that it handles the approximation as well as the
synthesis problem . It is capable of dealing with low , high-, and band-pass -
filters with prescribed zeros of transmission (also called attenuation poles
or loss peaks ). There are provisions for either equal ripple or maximally -
- -
flat type pass band behavior, for arbitrary ratios of load to source im -
pedances, and for p'redistortion and incidental dissipation .
FIG . 15.3. Impulse response of fifth-order Butterworth filter evaluation by Laplace In addition, Dr. Szentirmai has a modified program that synthesizes
transform.
- -
low and band pass filters with maximally flat or equal ripple type delay - -
-
numerically. This approach obviates the necessity of finding roots of in their pass band , and monotonic or equal-ripple-type loss in the stop
high-order polynomials. It has the advantage that the excitation signal bands.
need not be specified analytically, but merely in numerical form. In the specifications, the designer could specify both the zeros of
The frequency-domain programs usually consist of finding transient transmission (loss peaks) and the network configuration desired.4 If .
and steady-state responses, given the transfer function in factored or his specifications include neither, the computer is free to pick both con-
unfactored form. The program user must specify the numerator and figuration and zeros of transmission . The computer’s choice is one in
denominator polynomials of the transfer functions, the types of transient which the inductance values are kept at a minimum. The program was
response he wishes (i.e., impulse or step response), and the types of written so that the same network could be synthesized from both ends.
steady-state responses he wishes (amplitude, amplitude in decibels, phase, Finally, the computer prints out the network configuration, its dual,
delay, etc.). In addition , he must specify the frequency and time data
points at which the calculations are to be performed. This may be done
.
in two ways If he requires evenly spaced data, he need only specify the
the normalized element values, and the denormalized ones. It also
provides information such as amplitude and phase response, as well as
plots of these responses obtained from a microfilm printer.
1
'
minimum point, the increment, and the number of points. If he wishes Figures 15.4, 15.5, 15.6, 15.7, 15.8, and 15.9 show the results of a
to obtain data at certain points, he must supply the list of data points at band -pass filter synthesis using Dr. Szentirmai’s program. Figure 15.4
[|
the input.
-
Examples of outputs of a steady state and transient analysis computer 2 R. Saal and E. Ulbrich, On the
“ Design of Filters by Synthesis,” Trans IRE on
.
.
program are shown in Figs. 15.1, 15.2, and 15.3. In Fig. 15.1, the magni- -
Circuit Theory , CT 5, December 1958, 287-327
3 G. Szentirmai, “Theoretical Basis of a Digital
!
tude of a fifth-order Butterworth filter is plotted via a microfilm plotting Computer Program Package for
Filter Synthesis,” Proceedings of the First Allerton Conference on Circuit and System i
subroutine. Figure 15.2 shows the phase of the filter, while Fig. 15.3 Theory , November 1963, University of Illinois.
shows its impulse response. 1
Saal and Ulbrich, op cit. .

442 Network analysis and synthesis

i Computer techniques in circuit analysis 443

CASE NUMBER 3.1 F0RWARO REALIZATION FROM A SHORT CIRCUIT AOMITTANC6

BAND PASS FILTER SYNTHESIS
0E6REE OF FILTER
- 13

CASE NUMBER
-
3 1
MULTIPLICITY OF PEAK AT ZERO
MULT I PLIC ITY OF PEAK AT INFINITY
EQUAL RIPPLE PASS BAND
-- 2
3

0.0500 0B “ ARBITRARY" STOP BAND

--
--
LOWER PASS BAND EDGE FREQU ENCY * 1.0000000E*04 CPS
DEGREE §F FILTER 13 UPPER PASS BAND EOGE FREQUENCY 1.80000006404 CPS
MULTIPLICITY OF PEAK AT ZERO
MULTIPLICITY OF PEAK AT INFINITY
3
2 -
MID BAND FREQUENCY 1.34164086*04 CPS

NUMBER OF FINITE PEAKS BELOW THE BAND *= CONFIGURATION SPEC IF I ED BY COM PUTER
803 501 801 100 802 4 0 0 200 100 300
NUMBER 0F FINITF PEAKS AB0VE THE BAND * 3
NORMALIZED UNNORMALIZED
EQUAL RIPPLE PASS BAND REQUESTED
PASS BAND RIPPLE MAGNITUDE 0.05000 DB . 1.00000006*00 ,R 6.0000000E 02
* TERMINATION
L0WER PASS 8AN 0 EDGE FREQUENCY 1.0000000E404 CPS .
-
MID BAND FREQUENCY . -
l 3416408E*04 CPS
1.8000000EF04 CPS
.
.
-
3.07043376 01 .C -
6.07061026 09 803
UPPER PASS BAND EDGE FREQUENCY
-
7.85273376 01 L C 5.58928I E-03
6 PEAK FREQUENCY
"ARBITRARY" ST0P BAND REQUESTED -
1.84997696 01 3.65762286-09 3.5200000E*04

NUMBER 0F SPECIAL P0LES

TERMINATI0NS
INPUT TERMINATI0 N
- 1 MA
- 1.0000000E400

6.0000000E402 0HMS
2.57230156*00
1.00853806*00
4.64298066 *00
..L.C...
c 5.085744OE 08
7.1783955 E 03
9.17972126 08
—
-
-
PEAK F REQUENCY
6.2000000 E *03
501

0UTPUT TERMINATION 0. 0HMS 1.364749 lE*00 C. 2.69827036 08 - 801

LOWER FINITE ST0P BAND PEAKS

FREQUENCY CPS NORMALIZED VALUE 0F M 5.3765272E 01
- - L C 3.8268107E 03
1.5447656E 08
-- PEAK FREQUENCV
2.0700000E*04
6.2000000E+03 4.6212071 E 01

FINITE STOP BAND PEAKS

- 0.6229266 I
7.81321836 01

7.56238946-01 l 1.4951737E 08 - 100

UPPER
FREQUENCY CPS NORMALIZED VALUE OF M 1.4438446 E*00 ...C.... -
2.85465146 08 802

2.07G0000E404 1.5428869E400 2.3788111 1

2.3100000E*04
3.5200000E +04
1.7217723E400
2.6236531 E*P0
1.9296560
1.4968756
2
3
2.71127466 01
1.2441566600 - L C 1.929783 -03
2.459844 -
76
56 00
PEAK FREQUENCY
2.31000006*04

COMPUTER WILL SPECIFY CONFIGURATION 1.01271976*01 .C. 2.00226656-07 400

LAST INDUCTOR IS A SERIES BRANCH 9.57702116 02
- 6.81656486 04
3.23828006 08
-
-
200
100
FIG . 15.4 1.63787896*00 C

6.29631566 01 - L 4.4814B16E 03 - 300

SH8RT R. SH8RT TERMINATIiN

gives the specification of the problem. The pass-band magnitude is to be
>

equal ripple with ripple magnitude of 0.05 db, and the degree of the filter FIG . 15.5
is to be 13. As we indicated in Chapter 14, odd -degree, equal-ripple
filters are nonrealizable. In this example the designer utilized an ingenious The terminations are: input = 60 0, output = 0 O which means the
— —
device an extra pole to accomplish the synthesis. The program logic
then provided two extra zeros : one to cancel the extra pole and the other
filter terminates in a short circuit. In this example, the filter configuration
is chosen by the computer, and is a minimum inductance configuration.5
to provide the odd degree. The extra pole is called a special pole in Fig. In Fig. 15.4 there are, in addition, listings of the finite zeros of trans-
15.4, and is located at .s = 1.0. —
Further specifications call for the lower band -edge frequency to be 104
mission (loss peaks), which the designer specified.
Figure 15.5 is a printout of the configuration of the filter as shown by
cycles ; the upper, 1.8 X 104 cycles ; and the midband frequency to be the dotted lines flanked by the associated element values, both normalized
Vl .8 x 104 = 1.3416408 X 104 cycles. In the stop band , there are to be (left column) and unnormalized (right column). Since there are four
zeros of transmission at / = 0 (three), / = oo (two), and four finite zeros finite zeros of transmission, there must be associated four L C tank circuits. -
of transmission : one below the pass band and three above the pass band.
The positions of these finite zeros of transmission are chosen by the
5
W. Saraga, “ Minimum Inductance or Capacitance Filters,” Wireless Engineer,
30, July 1953, 163-175.
designer as indicated by the notation “arbitrary” stop band requested. i

k
 Computer techniques in circuit analysis 445
444 Network analysis and synthesis

FREQUENCY V 0 L TAGE RA PHASE REAL Z IN IMAG Z IN REAL Y IN I MAG Y IN

IN CYCLES T I 0 IN DB IN DEGREES IN 0HMS IN 0HMS •IN MILMH 0 S IN MILMH0 S

56.981 6.2973 -
2.6641 6
---3.5756
16000 5.718532 145.549717 134.691
6.2766 2.9395
16250
16500
5.704379
5.707644
128.958888
111.534102 .
130.664
125 883
61.193
64.762
67.928
6.2014
6.3129
3.2315
5
16750
17000
17250
5.729113
5.750516
5.743140
92.934808
72.763852
50.659000
119.931
112.687
104.489
71.154
74.696
6.3445
6.3337
-4.0061
5277
17500 5.708979 26.219204 95.717 77.919 6.2835 -5.1151
---9.4759 \
17750 5.720983 358.194819 05.165 79.138 6.3011 5.8552 4
18000 5.702947 322.320795 67.620 78.767 6.2748 7.3083
87.140 4.3533 X
18250 4.131556
.
272.913485 40.032
-8.7828
18500
18750
- 1.119811
-7.753694
225 6 8 9 5 8 9
195.922 9 8 8
16.386
5.462
111.450
139.624
...
1.2913
2797 - 7.151 ?
-6.0967 13
---14.148566
0641

..167
19000 177.085862 1.725 164.004
--5.4176
150
192 20.210264 163.748654 541 184.583 0158
. ^1 2
...0010
195
500 26.118735 153.507 8 0 3 202.605 0040 4.9357
-32.100833 - 4 . 5672
...013
750 145.227 0 2 9 • 049 218.952
20000 --45.857895
38.473164 138.296917 234.147 0002 - 4 . 2708
--4.0241
20250
20500 -56.325030
132.354567
127.166241
002
000
.000
248.501
262.211
. .0000
0000
0000 3.8137
- 3.6310
20750 --59.444189
71.701988 302.572268
. 000
275.407
.
21000
21250 - 57.483949
298.458S 82
294.740796 •000
288.178
300.592 ..0000
0000
4701
3268
0
-57.587501
--- 2.9750
3.1980
21500
217 0 --60.546362
58.703992
291.354706
288.250341
•000
.000
312.698
324.536 ..0000
0000 3.0813 -1
22000^ 285.388013 .000
.. 336.138 0000
0.000 0.400 0.800 1.200 1.600 2.000 2.400 2.800 3.200 3.600 4.000
---71.867460
22250 63.101161 282.735723 000 347.527 ..0000 - 2.8775
22500 66.602572 280.267304
. 000 358.726 0000
.0000 - 2.7076
- 2.7045 Add l.OOOOOiS + 03 to abscissa Frequency, cycles x 104
..
22750 277.961066 000 369.753

--
23000 83.403713 275.798866 .
. 000 380.621 0000 - 2.6273
FIG . 15.8
23250 80.597817 93.765362 000 391.345 0000 -2.5553
.
FIG 15.6

20 0.600

0.400
0
'o 0.200
— 20 x
16 0.000

J—
L?
40 N* -0.200
I -0.400
5
-80 -0.600
-0.800
-100 -1.0000.000
0.400 0.800 1.200 1.600 2.0002.400 2.800 3.200 3.600 4.000
-1200.000 0.400 0.800 1.200 1.600 2.000 2.400 2.800 3.200 3.600 4.000
Add I.00000.E + 03 to abscissa Frequency , cycles x 104
Add 1.000002? + 03 to abscissa Frequency, cycles x 104 .
FIG 15.9

.
FIG 15.7

446 Network analysis and synthesis Computer techniques in circuit analysis 447
The peak frequencies are listed under the column entitled “termination,” conventional manner. Moreover, the designs are completed in minutes
and the associated L-C tank circuits are on the same line two columns to rather than days and at a typical cost of twenty dollars rather than two
the left. thousand (not including initial programming costs, of course).
Figure 15.6 is a listing of a small portion of the frequency response, Optimization
which was calculated after the filter had been synthesized. Figure 15.7 Network design cannot always be accomplished by way of analytical
is a plot of voltage ratio in decibels versus frequency. Note that 1000 means. Quite often networks are designed through a trial-and-error
cycles must be added to every frequency value in the abscissa. This process. The network designer begins with a set of specifications. He
merely reflects an idiosyncrasy of the plot routine . then selects a network configuration, and makes an initial guess about
Figures 15.8 and 15.9 show the real and imaginary parts of the input the element values. Next he calculates or measures the desired responses
admittance of the filter. Note particularly the shapes of these character- and compares them with the specifications. If the measured responses
istics. If a number of these band-pass filters were connected in parallel differ by a wide margin from the specified responses, the designer changes
and if the pass bands of the filters were adjacent but nonoverlapping, the values of the elements and compares again. He does this a number
then the input conductance of the filter system would be essentially of times until (hopefully) the measured responses agree with the specified
constant over the entire pass band , while the input susceptance would responses to within a preset tolerance.
cancel out , as shown in Figs. 15.10a and b . This then means that the This process of cut and try can be made to converge, sometimes quite
input admittance of the filter system would be real and could then be rapidly, if one uses a method of steepest descent. 6 To get a rough idea
driven by any arbitrary source impedance without fear of reflections. of the steepest descent method, suppose there are n parameters in the
Filters obtained using computer programs of the type we have just network . Let us regard each parameter x( as a dimension in an n-
described are rapidly supplanting conventional hand- and handbook- dimensional euclidean space . We define a function /(xlt x2 , x3, . . . , xn)
designed filters. The filter synthesis programs provide filters that are that assigns a functional value to each point of the euclidean space. We
orders of magnitude more sophisticated than filters designed by the then ask , at point pit what direction of motion decreases the value of
f ( xi , x2 , x3 xn
) most rapidly ? The function f ( x{ ) may be defined as
a least squared error, or as an absolute error between calculated response
IE and specified response. The direction of steepest descent is the direction
defined by the gradient
df df df
<v
ce
grad / = —. x2 -|
— *1 + dx + — *„ (15.1)
2 dxn
Therefore, incremental value of change for each parameter is
Frequency
(a) 5 x{ = -C ?l (15.2)
dxi
.. where C is a constant . After all parameters have been changed by the
o
incremental value in Eq. 15.2, a new gradient and new incremental values
£ r are obtained . The process continues until the optimum is obtained .
£ Frequency An excellent paper by C. L . Semmelman7 describes a steepest descent
a :
1 •Charles B. Tompkins, “ Methods of Steep Descent,” in Modern Mathematics for the
Engineer (Edwin F. Beckenbach, ed. ), Chapter 18, McGraw-Hill Book Company, New
York , 1956.
( b) 7
C. L . Semmelman, “ Experience with a Steepest Descent Computer Program for
FIG. 15.10. Effects of paralleling several band-pass filters with adjacent pass bands. Designing Delay Networks, ” IRE International Convention Rec., Part 2, 1962, 206-210.

:
:
1
li
 448 Network analysis and synthesis Computer techniques in circuit analysis 449
Fortran program used for designing delay networks. The specifications
are the delay values R, given at the frequency data points /}. The program Memory Data Remote
successively changes the parameters x{ so that the squared error bank link
terminal
1

*(*,) =2
- j i
,
[ R - TM )]' , (15.3)

is minimized . In Eq . 15.3, T ( xitfJ) represents the delay, at frequency ft , Buffer Data Remote
memory terminal
of the network with parameters xt . link 2
In order to use the program, the network designer must first select the Memory Central
initial values for the parameters xt . He must also provide the specified bank processor
delays R, and the frequency data points /}. The program provides for 128 Queueing Data Remote
match points and 64 parameter values. It is capable of meeting require - logic link
terminal
3
ments simultaneously in the time and frequency domains. The designer
is not restricted to equal-ripple approximations or infinite Q requirements.
He is free to impose requirements such as nonuniform dissipation and
range of available element values on the design . For related methods of Memory
optimization, the reader should refer to a tutorial paper by M. R. Aaron.8 bank

Machine- aided design

-
The concept of real time interaction between man and computer holds
-
much promise in the field of network design. Multiple access computing
systems, such as the project MAC of Massachusetts Institute of Tech-
nology, make cut and try design procedures practicable. The initials Memory
MAC could describe either the term multiple-access computer or the term bank
Data Remote
machine-aided cognition . In an article describing the MAC computer terminal
link
system, 9 Professor R. M. Fano states “The notion of machine-aided cog -
nition implies an intimate collaboration between a human user and a FIG . 15.11 . Block diagram showing typical multiple-access computing system .
-
computer in a real time dialogue on the solution of a problem, in which
the two parties contribute their best capabilities.” before he gets the results through the teletypewriter or oscilloscope
A simple multiple-access computer system is shown in Fig. 15.11. display. He examines the results, changes some parameters, and feeds the
There are n data links and n terminals connected to the central processor. program into the central processor again . The queueing time and proc-
Located at each terminal is input- output equipment, such as teletype - essing time on the multiple-access computer may amount to only a
writers, teletypes, and oscillographic displays. The sequence in which minute or two, which, for the user, is probably not significantly long.
the central processor accepts programs from the terminals is controlled by Dr. H. C. So has written a paper on a hybrid description of a linear
a built-in queueing logic. n-port network10 in which he has shown that the hybrid matrix is ideally
The main reason a multiple-access computer is effective is that the suited for such problems such as multielement variation studies and
central processor computes thousands of times faster than the user’s iterative design . Suppose there are n variable elements in the network.
reaction time. When a user feeds in a program, it seems only a “ moment ”

8 M. R . Aaron , “The Use of Least

—
These can be “extracted ” and the rest the bulk of the network can be
described by the n- port hybrid matrix.
—
Squares in System Design,” IRE Trans , on Circuit
Theory , CT-3, No. 4, December 1956, 224-231 .
' R. M. Fano, “The MAC System : The Computer Utility Approach,” IEEE -
10 H . C. So, “On the Hybrid Description of a Linear n Port Resulting from the

Extraction of Arbitrarily Specified Elements, ” Trans. IRE on Circuit Theory , CT-12,

Spectrum, 2, No. 1 , January 1965, 56-64.
No. 3, September 1965, pp. 381-387.

450 Network analysis and synthesis Computer techniques in circuit analysis 451
over a set of frequencies 0);. In Eq . 15.4, Cn and Cd are real constants,
Command Initialize and E( s ) and F( s) are polynomials in s = o + ja> The program described .
here computes the amplitudes M( coand phase over a set of
frequencies wk , where
Frequency jo>
range C „ E( jcok )\
M ( cok ) = \ (15.5)
Control Computer
Engineer I Ca F( jcok )\ w;
Parameter and
Jp
adjust
= arctan EtUa >k
Er( j“>k )
)
— arctan Fi( jcok )
Fr( juk )
(15.6)
Display Output
where in Eq . 15.6 the r and i subscripts
indicate real and imaginary parts,
—a a 0

respectively. Note that in Eq . 15.6,

is limited to the range TT <
FIG . 15.12. Iterative design of network using machine-aided cognition . <f> < tr radians. In order to circumvent

Dr. So has written a computer program to formulate the hybrid matrix

for the n-port network automatically.11 The inputs to this program are
this restriction on the phase, we use the
following method to compute amplitude
FIG . 15.13. Calculation of
— jo)
magni-
and phase. tude and phase at <o, due to zero z .
(1) the node connections specifying the network topology ; (2) the im - Method used
pedance functions of the elements ; and ( 3) the specifications of the
Let H{ s ) be factored into poles and zeros such that
special elements to be extracted . This program was written with man-
machine interaction in mind . The process is described in Fig. 15.12.
( s zi )( s ~ gg) • • • (s - )
First the computer reads in the n port program with initial specifications. = Cn( s - )( s - p ) • (s znpj
~
- H( s) (15.7)
It performs the calculations and feeds information, such as transient or cd Pi 2 -
steady-state responses, back to the engineer via a visual display console. Consider the amplitude and phase due to any pole or zero, for example,
The engineer assesses the data and then changes the parameter values of
the extracted elements, and, in some instances, the frequency range over
—
z = a + jfi , shown in Fig. 15.13. The magnitude is
which the calculations are made. The process is repeated a number of M&t ) = [a8 + ( cot - m'A (15.8)
times until the engineer has obtained the desired results. Such a program
could also be controlled by a steepest descent steering program if the
engineer wished to obtain his results with less “eyeballing.”
and the phase is <f> Xa>i ) = arctan (15.9)
Now let us examine some details of specific network analysis programs.
The amplitude and phase of the overall function is
15.2 AMPLITUDE AND PHASE SUBROUTINE

Purpose 1 k=1
Our purpose is to compute the amplitude and phase of a rational MK) = (15.10)
function ca IT
1= 1
(s)
H ( s ) = CnE (15.4)
11
H . C. So, unpublished memorandum .
Cd s )
F (
;
and < .
£(" ) = k2= 1&*(«>») - i2=l &>*(",) (15.11)

i :
. 1
 452 Network analysis and synthesis Computer techniques in circuit analysis 453
where the subscripts p and 2 indicate the contribution due to a pole and A flow chart listing the input instructions is given in Fig. 15.14. Def-
zero, respectively. initions of the symbols in the flow chart follow.
Input N1 = number of frequency points
1. The numerator E( s ) and denominator F( s ) can be read in either as
polynomials (high degree to low) or in terms of their roots. If numerator
ND = degree of denominator
and / or denominator are read in as polynomials, their roots will be printed.
NN = degree of numerator
KK = 0 indicates equally spaced data points. We then read in only
If they are given in terms of their roots, the program could generate
wMIN, Aft), and number of points.
polynomials from these roots. = 1 indicates unequally spaced data points. We then read in all
2. The data points can be read in if unequally spaced ; if equally spaced ,
Mi .
only the minimum point and the spacing need be read in . LI =0 Read in zeros (if complex conjugate, both zeros must be
read in) .
= 1 Read in numerator polynomial, high degree to low.
L2 = 0 Read in poles (if complex conjugate, both poles must be
'Read Nl , ND , NN read in ) .
KK , LI , L2 , L3
= 1 Read in denominator polynomial, high degree to low.
I CN = numerator multiplier
Read CN , CD CD = denominator multiplier
Output
KK = 0 KK = 1 The main output of the program is the frequency response consisting
of the following columns : frequency in radians W( I) ; amplitude,
M( W(I)) ; magnitude in decibels, 20 log10 M ; and phase in degrees,
Read W (l) Read WMIN, DW </> ( W ( I )) . A typical printout of an amplitude and phase program is given
in Fig. 15.6, which shows the magnitude and phase of the band - pass filter
designed using Szentirmai’s program.

LI =0 Test LI
LI = 1 15.3 A FORTRAN PROGRAM FOR THE ANALYSIS OF
LADDER NETWORKS

Read A (I), B (I) Read E (I) In this section we will discuss the methods and organization for a
Fortran computer program for the analysis of linear ladder networks.
The analysis proceeds by computing the voltage and current at the input
of successive L sections, beginning with the terminating section .
L2 = 0 L2 = 1
Test L2 The program calculates the branch impedances of the ladder by com-
bining R , L , C series impedance arms according to the instructions of the
individual programmer.
Read C (l), D (l) Read F (I)
The poles and zeros and the frequency responses of the input im-
I pedance and voltage transfer ratio at the input of each L section are
T
To main program
found . In addition, the program provides for the analysis of short- and
open-circuited networks as well as for those with normal terminations.
FIG . 15.14. Flow chart of input instructions for magnitude and phase program. Separate problems may be run consecutively if so desired .

, hi

454 Network analysis and synthesis Computer techniques in circuit analysis 455
For the normal load and open circuit termination, the frequency
V„ 4-1, Vo
Vm responses of the input impedance, and voltage-transfer ratio are computed ;
for short-circuit termination , the input impedance and current gain are
Zn -1
computed . Provision is made for either a linear or logarithmic increment
Yn 4-1 Yl in frequency. The frequency boundaries and increments are read by the
Ym
program along with the input data.
According to instructions given by the programmer, the various cal-
culations are printed for each L section or only for the network as a whole.
.
FIG 15.15
Program operation
The ladder network-analysis program requires two sets of input data.
The network is initially decomposed into separate L sections as in The first set consists of the control instructions. These tell the computer
Fig. 15.15. These L sections are then added successively to the terminating which of the various options, such as information concerning polynomial
L section to form the complete network . With the addition of an L roots or frequency responses, are to be exercised .
section , the voltage and current at the input of the resulting network are In addition , the control instructions provide the computer with necessary
computed by the equations parameters such as normalizing factors, and the number of L sections in
the ladder. The second set of data determines the branch impedances of
4 = Vn-\ Yn + 4-1 ( 15.12) the network by specifying the R , L, C impedance arms and their combining
Vn = 424 + K l -
instructions.

which were originally discussed in Chapter 9 of this book . Thus the Impedance data
program proceeds toward the front of the ladder requiring only the branch The data specifying the branch impedances consist of an ordered series
immittances of the present L section and the voltage and current of the of instruction cards, each determining an impedance arm and / or a com-
previous L section to make its calculations. bining instruction. The first arm of any branch impedance requires no
Suitable assumptions for the initial voltage and current V0 and I0 allow combining instruction . The branch impedance is set equal to the im-
for analysis of short - and open-circuited networks as well as for those pedance of this arm. Subsequent impedance arms are added in parallel
with normal terminations. These initial values of voltage and current are or series with the existing branch impedance, according to the combining
determined by control instructions. instructions of the arms. A blank card tells the computer that a complete
For the calculation of the voltage and current at an L section , the branch impedance has been formed and that the next arm begins a new
branch impedances of the section are required. To simplify the prepara- branch impedance. If blank cards are not properly inserted , the computer
tion of the input data specifying these impedances, the following procedure will calculate a single giant branch impedance.
is used. Each branch impedance is formed by the addition in series or The order of the data determining the branch impedances is Z1 1/ Fj,(

parallel of basic R , L, C impedance arms. A basic R , L , C impedance Z2, 1 / Y 2

arm is a series combination of a resistance, an inductance, and a capac- The values of R, L , C and the combining instruction XIN are written
itor any or all of which may be absent. The impedance arms are speci-
— on one card in the order XIN, 'R , L, C .
fied by the element values R , L , C and an instruction indicating how the The values of XIN and the associated operation are shown in the
arm is to be combined. table at the bottom of page 456.
The elements of the impedance arms are frequency and impedance Although the instructions XIN = 1.0, 2.0 will suffice for the con-
normalized to aid computation, and then the arms are combined as struction of many branch impedances, they are inadequate for other
specified to form the appropriate branch impedance. impedances.
The roots of the voltage and current polynomials for each L section For example, suppose the series combination of two tank circuits ( Fig.
are computed by a root-finding routine. 15.16) is desired as a branch impedance, this impedance may only be

Ill
456 Network analysis and synthesis Computer techniques in circuit analysis 457
50 MMf 1 Mf

2 mh .05 h
Tank circuit A Tank circuit B

XIN R L C

1.0 5.00000£ - 11
2.0 2.00000F - 03
3.0 .
05
2.0 .000001
4.0
Blank Card

.
FIG 15.16 Construction of branch impedances using instructions for XIN
. .

constructed with use of three more instructions (XIN = 3.0, 4.0, and 5.0)
and an additional set of computer storage locations. . .
FIG 15.17 Branch impedances that cannot be formed by program.
Tank circuit A is formed in the usual manner, but must then be tem-
porarily stored during the calculation of tank circuit B. The two tank Output
circuits are then added to form the final branch impedance. Examples of The coefficients of the voltage and current polynomials are printed in
impedances that cannot be formed are shown in Fig. 15.17. Observe that frequency normalized form . To calculate the denormalized coefficients,
a series branch impedance of zero ohms often simplifies the formation of
‘
divide by the normalizing frequency raised to the power of the corre-
branch impedances. Two blank cards will indicate a short circuited - sponding exponent.
The poles and zeros of the voltage and current polynomials are in
impedance.
frequency normalized form to facilitate zero-pole plots.
XIN Operation Frequency responses appear in denormalized form . The frequency
variable is in radians per second. The impedance level is in decibels,
Blank card A complete branch impedance has been calculated, and the phase in degrees. The gain and phase of the transfer ratios are
the next arm begins a new branch impedance. expressed similarly.
1.0 This arm is added in series with the existing branch
impedance. 15.4 PROGRAMS THAT AID IN DARLINGTON
This arm is paralleled with the existing branch impedance.
FILTER SYNTHESIS
2.0
3.0 This arm begins a new internal branch imped ance . Two double precision Fortran programs have been written that help
Subsequent XIN = 1.0 and 2.0 refer to this new in - in the synthesis of double-terminated filters using the Darlington pro-
ternal branch. cedure. The mathematics of the programs is described here.
Add the two internal branches in series to form the final Given a forward transmission coefficient

1
4.0
branch impedance.

5.0 Parallel the internal branches.

^> (
2l S) — KN ( s )
D( s )
(15.13)

458 Network analysis and synthesis Computer techniques in circuit analysis 459

we generate an input reflection coefficient Sn( s ) from the equation Program descriptions
We next proceed with brief descriptions of the two programs The .
Su(s) Sn(-s) = 1 - Sn( s ) Sn(
D( s ) D( - s) - X
— s)
2
N ( s ) N(
.
— s) (15.14)
first program finds the roots of 5n(s) Sn( s ) given the roots of N( s ),
— —
N( .s ), D( s ), and D( s ) A flow chart for the input of the program is
. —
given in Fig. 15.18. Note that we must read in both the zeros of N( s ) and
D(s) D(-s)
—
N( s ), although only half of a complex conjugate pair need be read in.
The same applies for the roots of D( s) and D( s ). The constant CN is —
The denominator of Sa( s ) is D( s ) because all the poles of Sn( s ) must be K 2 in Eq. 15.14.
in the left-half plane. The zeros of Su(s) are not so restricted . If Su($)
—
5n( J) has zeros at a ± jb and a ± jb , the zeros of Su(.s) can be chosen
—
-
as either a ± jb or a ± jb. The only difference is that certain choices
lead to filters with unity-coupled coils and others without .
The second program performs the computations in Eq. 15.15 for
different combinations of zeros of Sn( j), given the fact that the zeros of
£u(y) need not be in the left-hand plane . The previous program finds all
—
12
the zeros of Sn(s) Sn( s ). We must choose only half the number of
Once the zeros of Sa( s ) are chosen, the input impedance of the filter
is then
zero pairs for S'nCO. Certain combinations of zeros of Su( j) Sn( s )
lead to filters with coupled coils ; others do not. If one wishes to try a
—
1 ~ Sn (s) (15.15) number of combinations of zeros for Su( j), the program has the facility
ZiOO = 1
+ Sn( s ) to enable him to do so. He need only supply those zeros of Su( s ) Sn( s )
in the right-hand plane, C(I) + y‘B(I), and a list of constants S(I), which
—
i —
are either 1.0 or 1.0 so that the zeros of Sn( s ) may be represented as
S(I) • C(I) + y’B(I) = A(I) + y'B(I). The routine forms numerator and
Read CN Numerator multiplier denominator polynomials of Su( s )
I F( s )
(15.16)
Read CD Denominator multiplier

I Number of zeros
1 Number of pole pairs of Sii
Read LN (complex conjugate zeros Read LD (real poles count as one pair)
count as one zero)

I
Read
Read real and imaginary
I
Read Real and imaginary parts of poles
parts of zeros and
A (I), B (I), N (l) multiplicity, N ( l) H ( I ) , G ( I ), M ( I ) of Su and multiplicity, M ( I)

I Number of poles I
Read LD (conjugate poles Number of right-hand plane
Read LN
count as one pole) zeros of Su (s;Sii ( s) -
Read
I Read real and imaginary I
parts of poles and Read Real and imaginary parts of right-hand
H (I), G (I), M (I) multiplicity, P (l) C (I), G (I), N (I) plane zeros and multiplicity, N (l)
1
To program I
. .
FIG 15.18 Input to
—
Sn( s ) Sn( s) program.
Read S (l)

1
Multiplicative constants ( + 1.0) (each
combination of zeros has LN cards)

13
T. Fujisawa, “Realizability Theorem for Mid-Series or Mid-Shunt Low Pass To program
Ladders without Mutual Induction,” Trans. IRE on Circuit Theory , CT-2, December
1955, 320-325. . .
FIG 15.19 Input to Z [ a( s ) program.

hi i
 460 Network analysis and synthesis
and then computes the numerator and denominator of Zfs ) as

Zi( s ) = 1
1 - Sn _ F( s ) - E( s)‘
(15.17)
F( s ) + E ( s )
+ Sn
The input instructions are summarized on the flow chart in Fig. 15.19.

Problems
A appendix
15.1 Organize a flow-chart for computing magnitude and phase given only
the unfactored numerator and denominator polynomials . Do not use a root- Introduction to matrix algebra
finding subroutine.
15.2 Write a program to calculate the delay of
3( s + 1)
H ( s) = -
s + Is + 5
at the points co = 0, 1 , 2, . . . , 10 . A. I FUNDAMENTAL OPERATIONS
15.3 Repeat Prob. 15.2 for a general transfer function given in terms of its
unfactored numerator and denominator polynomials. The frequency points toi Matrix notation is merely a shorthand method of algebraic symbolism
are to be read in by the computer. that enables one to carry out the algebraic operations more quickly.
15.4 Suppose you have calculated the phase of a transfer function at points The theory of matrices originated primarily from the need (1) to solve
<o = 0, 1 , 2 50. Devise an algorithm to test for phase linearity. The simultaneous linear equations and (2) to have a compact notation for
deviation from phase linearity is to be called phase runoff. linear transformations from one set of variables to another.
15.5 Write a program to analyze a two-mesh network made up of only As an example of (2), consider the set of simultaneous linear equations
R, L, and C elements.
15.6 Repeat Prob. 15.5 for nodal analysis . «11*1 + «12*2 + • • + alnxn = Vl
15.7 Write a program to calculate the step and impulse response of a linear «21*1 + «22*2 + • • • + =
«2n*n Vi (A I ) .
system whose system function contains only simple, real poles and zeros .
15.8 Repeat Prob. 15.7 if simple complex poles and zeros are allowed . «ml*l + «m2*2 + «mn*n = Vm
15.9 Write a program to calculate the residues of a transfer function with These may express a general linear transformation from the xt to the yv
multiple as well as simple poles . The poles and zeros are to be real . In general, m n. An example where the numbers of variables in the
two sets are unequal is that of representing a three-dimensional object in
two dimensions (in a perspective drawing). Here, m = 2 and n 3. =
Definition
A matrix is an ordered rectangular array of numbers, generally the
coefficients of a linear transformation. The matrix is defined by giving
all its elements, and the location of each.
The matrix of the equations in Eq. A.I written as
«11 «12 '’' «ln
«21 «22 ' ' «2n

L« ml «m2 ’‘’ «mnj

is of order m X n. (The first number here is the number of rows ; the
461

Appendix B 471
physicists. It was not until 1953, when George Temple produced a
more
elementary (although no less rigorous) theory through the use of general
ized functions,3 that this new branch of analysis received the -
attention it
deserved. Our treatment of generalized functions will be limited to
the
definition of the generalized step function and its derivative, the
unit
appendix B impulse. The treatment of these functions follows closely the work of
Temple4 and Lighthill.ft
To get an idea of what a generalized function is, it is convenient to use
Generalized functions and as an analogy the notion of an irrational number a beng a
of rational numbers oc„ such that
sequence {<*„}

the unit impulse a = lim a„

-
n * co
where the limit indicates that the points an on the real line converge
to
the point representing a. All arithmetic operations performed on
the
irrational number a are actually performed on the sequence {a „} defining
B. l GENERALIZED FUNCTIONS a.6 We can also think of a generalized function as
being a sequence of
functions, which when multiplied by a test function and integrated over
The unit impulse, or delta function, is a mathematical anomaly. P. A.
M. Dirac, the physicist, first used it in his writings on quantum mechanics.1
He defined the delta function d ( x ) by the equations
—
( oo, co ) yields a finite limit. Before we formally define a
generalized
function, it is important to consider the definition of (1) a testing function
and (2) a regular sequence.
DEFINITION B. l A function <}>( t ) of class C [<f>( t ) £ C ] is one that (
1)
J °°
d { x ) dx =1
(B. l)
is differentiable everywhere, any number of times and that (2) when
any of its derivatives are multiplied by t raised to any power, the limit is
it or

<5(*) =0 for x
^O lim- [tm
-
t * ± ob
-* 0 for all m & k 0 ( B . 3)
Its most important property is
Any testing function is a function of class C.

/ " ( ) <5( ) d
/ * * * = /(0) (B 2) Example B.l . The Gaussian function e~ti / nt is a function
of class C . It is
obvious that if a function is of class C then all of its derivatives belong to class .
C
—
where /( x ) is continuous at x 0. Dirac called the delta function an
improper function, because there existed no rigorous mathematical justi-
5
G. Temple, “Theories and Applications of Generalized Functions ”
, J . London
fication for it at the time. In 1950 Laurent Schwartz 2 published a treatise Math. Soc ., 28 , 1953, 134-148 .
4
G . Temple, “The Theory of Generalized Functions,” Proc.
entitled The Theory of Distributions, which provided , among other things, 1955, 175-190. Royal Society, A, 228,
a fully rigorous and satisfactory basis for the delta function . Distribution 5
M . J . Lighthill , Fourier Analysis and Generalized Functions,
Cambridge University
theory, however, proved too abstract for applied mathematics and Press, 1955. Lighthill dedicated his excellent book to “Paul Dirac,
true , Laurent Schwartz, who proved it , and George
who saw it must be
Temple, who showed how simple
it could be made ."
1 P. A . M . Dirac, The
Principles of Quantum Mechanics, Oxford University Press, * This defines an irrational number according to the Cantor definition
1930. . For a more
detailed account see any text on real variables such as E. W. Hobson, The Theory
‘ L. Schwartz, Theorie des Distributions, Vols. I and II, Hermann et Cie, Paris, 1950 of
Functions of a Real Variable, Vol. I, third edition, Chapter 1 , Cambridge University
and 1951 . Press, Cambridge, England, 1927.
470

1.
 472 Network analysis and synthesis Appendix B 473
DEFINITION B.2 A sequence {/„( /)} of functions of class C is said DEFINITION B.7 The product ag of a generalized function g {#„} ~
to be regular if for any function 4( 0 belonging to C, the limit and a constant a is defined by the representation ag {ag„}.
DEFINITION B.8 The derivative g of a generalized function
~
lim ( fn , 4 ) = Hm
-
n * oo «-» oo
/»( 0 4( 0 dt ( B - 4) {g„} is defined by the representation g' { g' n } ~ -
Example B.4. For the generalized function gx ~ {e t 2 / n2 }
_ 2 that the sequence converge pointwise.
exists. Note that it is not necessary
~
the derivative is
For example, the sequence {e 7!< (n / 7r) H} approaches infinity as n oo at represented by
the point / = 0. However, the limit lim ( fn , <>j ) exists.
DEFINITION B.3 Two regular
71 yCO

sequences {/„} and {£„} are equiv -

g\ ~ (- H
alent if for all 4£ C lim (/„, 4> ) = lim ( g „, <f> ) ( B . 5) and ( g\, 4 ) = lim j (\ e tVn
~
-\] ct ( t ) dt
> (B.8)
-
« * oo 71 v 00 co n
Example B 2 .. The regular
2 2 2
-
sequences {e-n < (n /7r)14} and {e * /a « ( i / V 2TI7J)} are In Definitions B.6, B.7, and B.8 we have defined the operations of addition,
equivalent . multiplication by a scalar and differentiation. It must be pointed out that
DEFINITION B.4 A generalized function g is defined as a total , or the operation of multiplication between two generalized functions is not
complete, class of equivalent regular sequences. The term total implies defined in general .
here that there exists no other equivalent regular sequence not belonging We next consider an important theorem , whose proof is given in
to this class. Any member of the class, for example, { gn } , is sufficient to Lighthill, 7 which will enable us to represent any ordinary function, such
represent both g and the total class of equivalent regular sequences as a step function by a generalized function equivalent.
defining g . Wc denote this symbolically by the form g ^ { gn }.
Theorem B.l . Given any ordinary function /(/ ) satisfying the condition
Example B.3 . All of the equivalent, regular sequences
{ e~ % {«- /»'}, . . . ,
represent the same generalized function g
.
DEFINITION B 5 The inner product ( g ,
g and a function 4( 0 E C is defined as
^ “
~ {e ‘ /n }.
“ 2 2

<j> ) of a generalized function,

for some N
J
—>
r ( l i /coi
+ tON
dt

> 0, there exists a generalized function8 / ~ {/„(0} such that

( f t) =
/
<

) 4( 0 dt
co (B.9)

(B.10)

J
*
“
( g , 4> ) = lim _ g„( 0 4( 0 d t ( B.6)
-
« » oo for all <j> £ C. In other words, an ordinary function satisfying Eq. B.9 is equiva-
lent in terms of inner products to a generalized function. Symbolically, we
The inner product is often given the following symbolic representation .
write / = / If, in addition , / is continuous in an interval, then lim fn = f
pointwise in that interval .
I—" g( 0 4( 0 dt
*

( g , 40 = ( B.7) Furthermore, it can be shown that all the operations of addition, scalar
J 00
multiplication , and differentiation performed on both / and f yield equivalent
Note that the integral here is used symbolically and does not imply results, that is,
actual integration. ( «/i + //> )' = ( «/i + // )' ( B. 11)
.
DEFINITION B 6 If g and h are two generalized functions rep-
'

when differentiation is permitted on the ordinary function .

~
resented by the sequences g { gn } and h { hn } , the sum g + h is ~
defined by the representation g + h { gn + hn }. ~
Note that the set of sequences { gn + hn ) represents a total class of Lighthill , op. cit ., Section 2.3.
7
.
Note that when we represent an ordinary function by generalized function equivalent,
8
equivalent regular sequences made up of the sum of sequences defining g
we use a bold face italic letter to denote the generalized function.
and h ; therefore g + h is a properly defined generalized function .

:
:
474 Network analysis and synthesis
DEFINITION B.9 The generalized step function u is defined as the
:
'
It should be stressed that 8 ( t ) is merely the symbolic representation for a
Appendix B 475

total class of equivalent regular sequences { un( t ) } such that total class of equivalent regular sequences represented by {«'„(0). Thus
when we write the integral
(H , 40 = «lim
»
— f— un
GO J CO
( t ) <f>( t ) dt
£% 0 4 ( ( t ) dt

=
>
j ) 4( 0 dt ( B.12)
we actually mean

f— <3(0 4( 0 dt = ( <5, 4 ) = «lim f u' Jt ) 4( t ) dt ( B.14)

\
~
o
4( 0 dt
Example B.6. The
J co

sequence
- oo J »
— CO

where u{ t ) is the unit step defined in Chapter 2. That {w „(0} exists is

guaranteed by the previous theorem allowing representations of ordinary
functions by generalized functions. Hence, we write u = u.
' n( 0 t >0
“ (B. l 5)
.
Example B 5. The sequence =0 t 0

!i
=0
«»(0 = exP
H(M]
t 0
t >0 ( B.13) in Fig. B. 2 is one member of the class of equivalent regular sequences which
represents the unit impulse. Other members of the class are the sequences
{ e-nt\nl 0
- ' } and { e-t2l2\
n i V 2 wri) } .
which is plotted in Fig. B. l , is one member of the class of equivalent regular
^ /
sequences which represents the generalized step function .

j:
1.0
I
6

0.8
5

s °- 6 4
s
0.4
3
;
0.2
n=1 2

0
0 1 2 3 4 5 6 7 8 t 1

. .
FIG B. l The generalized step sequence, «„(() n= 1
1.0 1.2 1.4 1.6
t
0 0.2 0.4 0.6 0.8
DEFINITION B.10 The unit impulse, or Dirac delta function <5(0 >
is defined as the derivative of the generalized step function S(t) {u'„(t)}- ~ ..
FIG. B 2 The generalized sequence, u' Jt ).

t-
476 Network analysis and synthesis Appendix B 477
where/ '(0 must exist over [a, /?]. Note that when the limits of integration
B.2 PROPERTIES OF THE UNIT IMPULSE are infinite, we actually mean
Sifting
o dt s^iim [ mm dt
The most important property of the unit impulse is the sifting property
represented symbolically by
p>0
V =°”
In the sifting property, if both a, /3 > 0 or a, (} < 0, then
- <
( B.20)

f
J x<0
d( t ) f ( t ) dt = m;
where / is any function differentiable over [a, /9]. The left hand side of
(|a|, |/?| < co ) ( B .16)
n
mm dt = o ( B. 21)
Eq. B.16 is defined formally by
The proof of this property is similar to the original proof of the sifting

J
p>o

Joc < 0
d ( t ) f ( t ) dt = lim
~
p »o
J aflt < 0
« »( 0 / ( 0 dt ( B.17)
property, and will be left as an exercise for the reader.
71 * GO Integration
The proof of the sifting property is obtained by simply integrating by The defining equations of the delta function according to Dirac are
parts, as follows. p> 0
S ( x ) dx -1
J\
fi > 0
lim
n
f
- Ja <0
« »(0 / ( 0 dt = lim u „( 0 /( 0
oo
- lim
-
n * oo
(0 / '(0 dt «<0 ( B. 22)
KX )
d( x ) = 0; x 0
= /(/?) - £ lim un( t ) f ' ( t ) dt
71 00 (B.18)
These are actually properties of the delta function as viewed from the
generalized function standpoint . The proof can be obtained directly from
the sifting property. Suppose we have the integral
= m - nodt [ p> o
= m - i m - /(o)] = /(o)
Pictorially we represent S ( t ) by a spike as shown in Fig. B.3. If the and we let f ( t ) = 1. Then we have
J
Jx < 0
mm dt = m -
( B 23)

impulse is centered at t = a , then the sifting property is given symbolically

I
fi > 0
as
rp > a J <x < 0
<5(0 dt = /(0) = 1 ( B. 24)
8( t - a ) f ( t ) dt = f ( a ) ( l « l , 1 /51 < oo) ( B.19)
J <i < a If both a, /3 are greater than zero or both are less than zero, then

m
’j m dt = o (B.25)

This property is stated symbolically by the conditions <5(0 = 0 for t ^ 0.

Differentiation across a discontinuity
0 t Consider the function f ( t ) in Fig. B.4. We see that /(0 has a discon-
tinuity of A at t = T. If we let fx( t ) = f { t ) for t < T , and /x(0 = /(0 -
for t > T , then we have ^
.
FIG. B.3 The unit impulse . m = / (0 + A u{t - T ) i (B.26)

478 Network analysis and synthesis Appendix B 479

A
f( Q
A -
^ h( t )
0 t
0 T t (a)

h’( t )

FIG. B.4. Function with discontinuity.

A

Since /(0» /i (0 > ancl w(0 satisfy the condition

i: (i + ty
MOL
< oo dt
A
0 t

for some N ; we can represent these ordinary functions by generalized T

functions
(/0 = /i(0 + «(0 ( B. 27) *

( b)

Taking derivatives on both sides of Eq . B.27 yields . .

FIG B 5. Differentiation across a discontinuity.

no = /' (o + A «'(o i ( B.28) Differentiation '

The derivative of a delta function , which we call a doublet , is defined
which symbolically can be written as
~
symbolically as <5'(0 {«"„(0}- 11 has the following property, where
no = /' (o + A «5(0 i ( B.29) /*(0 exists over [a, /?].
p>o
We thus see that whenever we differentiate across a discontinuity, we
obtain a delta function times the height of the discontinuity.
I.
a<0
8’( t )mdt = -/'(0) ;
The proof is obtained through successive integration by parts.
( W, \P\ < CO ) ( B.32)

mm dt = lim Jar«, " (0 /(0 dt

.
Example B 7. The step response of an R C network is given as -
I
0>O
»
K0 = Ae -WuO ) (B.30) a <0 n ~* oo <0 ( B.33)
p>0 CP > 0
shown in Fig. B.5u. The impulse response is = lim « ' (0 /( 0|ot < 0- lim
71“* CO
B
— co Ja < 0 7i ^
mono dt
A'(0 = /1 8(0 - je Wu( t )
~
(B.31)
We see that since lim u'„(/3)
—
71 CO
— lim
- Tt ^ CO
w'n(a) = 0,
p>0
and is shown in Fig. B.5A. lim M'«(0 /(0 =0 ( B.34)
71 CO I« < 0

i 1
 480 Network analysis and synthesis
S' ( t ) -

o t appendix C
Elements of complex variables
. ..
FIG B 6 The doublet <5'(0 -
We then integrate by parts again so that
C. l ELEMENTARY DEFINITIONS AND OPERATIONS
- lira
71 ~* co
f fi > 0

Ja <0
« '*(0 / X 0 dt = —— 71
lim unf ' \
* CO
*
+ lim
71 O0 J\ 0/ 0
( "( d t
( B. 35)
A complex variable 2 is a pair of real variables ( z , y ) written as
= -iw + /'Xo at
= - fw + rm - no ) = -no)
JVo 2 = x + jy (C. l )

where j can be thought of as 1 V— .

In general, the derivative-sifting property can be stated symbolically as x
The variable is called the real part of 2, and y is the imaginary part of 2.
n> > o Written in simpler notation , we have
6 n( t - a ) f ( t ) d t = ( - i y f M(a) ( B.36)
, Ja < 0 x = Re (2), y = Im (2) (C.2)
where / ( n +1 (0 exists over [a, /?]. *
The generalized function <5 X0 is sometimes called a doublet . The The variable can be plotted on a pair of rectangular coordinates. The
2

pictorial representation of a doublet is given in Fig. B.6. abscissa represents the x or real axis, and the ordinate represents the y or
imaginary axis. The plane upon which x and y are plotted is called the
Other properties of the unit impulse complex plane. Any point on the complex plane, such as 2 = 3 + j2 , can
Dirac and others have obtained a host of identities concerning the unit be represented in terms of its real and imaginary parts, as shown in Fig. C.1.
impulse. We will merely give these here without proof. From the origin of the complex plane, let us draw a vector to any point 2.
d( -t ) = <5 (0 ( 1) The distance from the origin to 2 is given by
<5'( 0 =
t «5( 0 - 0
— -m (2)
( 3)
|z| = (a2 + y2 )'A
and is known as the modulus of 2. The
(C.3)
72
jy

t <5'( = - o m (4) angle which the vector subtends is known

d ( at ) = |a|-i ,5(0 ( 5) as the argument of 2 or
= \|a|-i {<5(t - a ) + 6( t + a) }
3 *
(5 ( / 2 - a 2) (6)
arg 2 = tan-iV ( C.4)
/(0 <5(f - a ) = f ( a ) 6( t a ) ( 7) - x
The proofs of these properties are obtained through the inner product Letting =0 arg 2 and r = |z|, we can jy -
with a testing function <f>{ t ) 6 C. represent z in polar coordinates as
FIG. C.l
•The condition on / , + 1 is sufficient, but not necessary.
B l z = re* (C.5)
481

482 Network analysis and synthesis Appendix C 483

Expanding this last equation by Euler’s formula, we obtain The following rules deal with operations involving the conjugate definition :
z = r cos 6 + jr sin 0, (C.6) zi + H = 2i* + Z2*
so that x = r cos 0 .
{C l )
Z1Z2 = zl* ' z 2 * (C.l 7)
y = r sin 0 zjzz = Z /Zj*
The rule for addition for two complex numbers is given as Finally, if 2 has a modulus of unity, then
^
(a + jb ) + (c + jd ) = ( a + c ) + j( b + d ) (C.8)
1
(C .18)
When two complex numbers are multiplied , we have z
(a + jb ){ c + jd ) = ( ac — bd + j( ) ad + be ) (C.9) The operations of raising a complex number to the nth power, or taking
where j 2 =
obtain
—. 1 If we express the complex numbers in polar form, we
the nth root of a complex number, can be dealt with most readily by using
the polar form of the number. Thus, we have zn = ( rei6 )n = rneinB ,
_ rVneil
(a + jb ) = rie* i
(C.10) and zVn ( B+2k !r ) / n ]
k , 0, 1, ..., n — 1 (C.19)
and ( c + jd ) = rtei9z (C.ll) .
C 2 ANALYSIS
When we multiply the two numbers in polar form, then
If to each 2 = x + jy, we assign a complex number w = u + jv , then w
r#** = rirtemi+H ) (C.12)
r
If we divide these two numbers in polar form, then ^ is a function of 2 or
w = /(z) (C.20)

rieiBl __ £ 1 1 -02 ) (C.13)

The following are examples of complex functions, i.e., functions of a
complex variable :
r 2 c / 9i r2 w = 2z
In rectangular coordinates , the operation of division can be expressed as w = log, z
a + jb _ (a + jb )( c — jd ) w = 1/z (C.21)
c + jd (c + jd )( c — jd )
(C.14)
H = z2 + 4
;

ac + bd . . be
c* + d 2
+ j c2 —+ dad8
w = \ z\
We see that w may be complex, pure real, or pure imaginary, depending
In connection with the modulus of a complex number, it is useful to note upon the particular relationship with z. In general, the real and imaginary
the following rules: parts of w are both functions of x and y. That is, if we let w = u + jv ,
lziz« l = lzrl • N then
u = f {* , y ) and v = f ( x, y ) (C.22)
lvi*l = lzil •IVI = N2 (C. l 5)
As an example, let us find u and v for the function w = z2 + 4.
2 - 2* =|| Z 2
w = z 2 + 4 = ( x + jy )2 + 4 (C.23)
where z* is the complex conjugate of z and is defined as Simplifying, we obtain
z* = x + jy = x - jy (C.l 6) w = ( z2 - y2 + jlxy ) + 4 (C.24)

• •
484 Network analysis and synthesis Appendix C 485
Jy
we finally arrive at
jy
AM + j Av
f' ( z ) = limO Aylim
As - - 0 A + j Ay
* >
(C. 33)
— Alim
An + / Av = du , . dv~
<c -* 0 A# Tx + Jd x
For path 2, we have
o X 0 X
An j Av
/'(*) = Aylim-* o AKlimO AX + j Ay
Path 1 Path 2
+ - A

-= lim
An + j Av
( C . 34)
Av -* o 7 Ay
-/y -jy
dv du
FIG. C 2 . . .
FIG C 3
dy dy
so that u = x 1 — y2 + 4 and p = 2a;y (C.25)
Since we assume that the function /(z) is differentiable, the derivatives
The derivative of a complex function / (z) is defined as must be independent of path . Thus , we have
+ Az ) — f ( z )
rWAz - lim /
(z
(C. 26)
A « -+ 0
dv_
dy
_ . du _ du
dy d%
. dv_
d%
(C . 35)
If one restricts the direction or path along which Az approaches zero, then
we have what is known as a directional derivative . However, if a complex From this last equation, we obtain the Cauchy- Riemann equations , which
function is to possess a derivative at all , the derivative must be the same are
at any point regardless of the direction in which Az approaches zero. In dv du
other words , in order for /(z) to be differentiable at z = z0, we must have dy dx
(C . 36)
dm = constant (C.27) du do
dz \t=zo dy dx
for all directions of approach of Az . We have just seen that in order for a function to have a derivative, the
Consider the two directions in which Az approaches zero in Figs. C.2 Cauchy- Riemann equations must hold . A function which is single valued
and C. 3. For path 1 , we have and possesses a unique derivative is called an analytic function. A set of
f\z ) = lim lim / +
(z Az) /(z)
(C . 28) — sufficient conditions for analyticity is that the Cauchy-Riemann equations
A « -» 0 Ay -» 0 Az are obeyed . For example, consider the function
If we substitute

into Eq . C. 28 , we obtain
Az = Ax + j Ay (C.29)
f ( z ) is analytic because
m = z° + 4 (C. 37)

f [ x + Ax + j( y + Ay )] - f ( x + jy ) dv du
— = 2x =
f ( z ) = lim lim
Ax -» 0 Ay -* 0 Ax + j Ay
(C . 30)
dy dx ^ (C. 38)
Since = u + jv m (C. 31 ) du dv
and f(z + Az) = u + AM + j{ v + Av ) (C. 32) dy dx

486 Network analysis and synthesis Appendix C 487

On the other hand, /(z) = z * is not analytic because Example C.3. Find the residues of the poles at s
= 0 and s = - 1 of the
function
U = x and v — —y
du
—= 1 dv (C. 39) /» - s
s2( s
+2
+ l )2
(C.43)
dx +
and
dy To find the residues, we simply perform a partial fraction expansion

.
C 3 SINGULARITIE S AND RESIDUES 2
/ ( s) = ? - ~s + ( s
3 1
+
3
(C.44)
+ l )2 s + l
Iff ( z ) is analytic within a region or domain in the complex plane except Thus the residue of the pole at s = 0 is - 3, and the residue of the pole
at a point z0 , then /(z) has an isolated singularity at z0 . Suppose /(z) has a at
s = — 1 is + 3 .
singularity at z„, then we can expand /(z) about z0 in a Laurent series
C4. CONTOUR INTEGRATION
m = (* n „*o)” + ~
• •
+Z—ZQ
+ a0( z - z0)° + flx(z - z0) + • • •

In complex integration the integral is taken over a piecewise smooth path

+ am( z - zor + • •
C and is defined as the limit of an infinite summation
(C.40)
In the expansion, if m is finite, then z0 is called a pole of order m.1 The f /(z) dz = lim- 2= /0,) Az,-
Jc
(C.45)
term a_ j is called the residue of the singularity .
Example C. l . Consider the Laurent series for the function (z) = ezjz about
_ 71 * CO j \
where z, lies on C. Unlike the process of differentiation, the path along
/ which we take the integral makes a difference as to the ultimate value of
the pole at the origin. We can expand ez in a power series to give the integral . Thus the integral
e* 1
f 7( ) dz
=- +1 +
z
1
1 +Z +
1
2!
Z
^ 2

+
+ Ji 23 +
1 2
3!
z + •••
' "
) (C.41 )
Jzi
z

in general , has different values depending upon whether we choose to inte-

grate along path C 1 or path C2 , as shown in Fig. C.4. If we integrate along
a closed path, say from a to b and then to a again, we are integrating along
(C . 46)

According to the definition, the residue of the pole at 2 = 0 is equal to 1 . a closed contour . The path shown in Fig. C. 5 is an example of a closed
contour. The following theorem , known as Cauchy' s residue theorem
Example C.2. Expand the function f ( z ) = l /z(z
and find the residue of the pole at z = 1 .
l )2 about the pole at z = 1 , — a method for rapid evaluation of integrals on closed paths.
gives

1 1 1
b
z( z - l )2 (z — l ) 1 + (z — 1)
2 Jy Jy
Closed
= ?“[ j2 [1 - (z - 1) + (Z - l )2 - (Z - l )3 + • • •] (C.42) "contour

1 1
1 - (z - 1 ) + ( z - l )2 +
-1 +
••• 0 x 0
(z - l )2 z
for 0 < |z — 1 | < 1 . Here, the residue of the pole at z =1 is equal to — 1. . .
FIG C 4 FIG C.5 .
1
Note that if we have an infinite number of nonzero terms with negative exponents, Theorem C.l . If C is a simple closed curve in a domain D , within which (z) is
then z 0 is an essential singularity . analytic except for isolated singularities at z1; z2 /
zn , then the integral along
 Appendix C 489
488 Network analysis and synthesis
the closed path C is A partial fraction expansion off ( s ) shows that

where
i/ 00 dz = 2ij{ Kx + K2 +
Kt represents the residue of the singularity z .
•
+ KJ (C.47)
s +1 S+2 /(*) - (C.51)

the integral along the closed

so that the residues are 1 and 2. The value of
(

Example C 4 .. Consider the integral path within which both the singularities lie is then
(C.48)
j
< > /(a) ds = 27rj(1 + 2) = 6 TRJ ' (C.52)

along the circle |s| = 2, as given in Fig. C.6. Since there are two singularities , then the integral
'

and + 3, then the integral along the circle is

—
within the circle, at s = 0 and at 5 = 1, whose residues are respectively 3 — If f ( z ) is analytic in a domain with no singularities
along any closed path is zero, that is ,

2
s+2
s s 4 l )2
(
ds = 2ij( 3 + 3) = 0
- — (C.49)

theorem.
j
< ) /(«) dz = 0 (C.53)

This result is known as Cauchy' s integral

. Find the integral off ( s ) along the closed contour shown in Fig.
Example C 5.
C.7. The function f ( s ) is given as
3s + 5 (C.50)
/00 = (s + l )(s + 2)
Jy

-2 -l
"

- jy
FIG. C.7

Appendix D 491
Re Z( s )

27r + <f> 6

appendix D
Proofs of some theorems on FIG D. l .
positive real functions where n is real and finite. In the neighborhood of the pole s0 , Z( s ) can be
approximated by
Z( s )
— k-n
( s - S0 )n

We can represent Z( s ) in polar form by substituting each term by its polar form ;
Theorem D l .. If Z( s ) and W( s ) are both positive real, then Z( W( sj) is also —
i.e., let (J s0)+n = rnein0 and k _ „ = Ke * so that ’
positive real. JZ K
Proof. When Re s 0, both Re Z( s ) and Re W ( s) 0, then Re Z( W ( s)) > 0, Z( s ) = rn nO ) Re Z( s ) = — cos ( — nd )
<f>

also. When s is real, both Z( s) and W ( s) are real, hence Z( W ( s)) is real . Since
Z( lV( s)) satisfies both conditions of positive realness, it is positive real. which is represented in Fig. D. l . When 6 varies from 0 to 277, the sign of Re Z( s )
will change 2n times. Since Re Z( s ) 0 when Re s 0, it is seen that any
..
Theorem D 2 If Z( s ) is positive real, then Z{\ js ) is positive real. -
change of sign of Re Z( s) in the right half plane will show that the function is not
Proof. W ( s ) = 1 )s is positive real, hence Z( tV ( s )) = Z(\ / s ) is positive real. positive real. Therefore, we cannot have a pole in the right -half plane. Since the
.
Theorem D.3 If W ( s ) is positive real, then l / lV ( s ) is also positive real. function 1 / Z( s ) is positive real if Z{ s ) is positive real, it is obvious that there
.
Proof Z{ s ) = 1/s is positive real, hence Z( W { s )) = 1/W ( s ) is positive real by cannot be any zeros in the right half plane also. -
Theorem D.l . Theorem D 6 .. Cinly simple poles with positive real residues can exist on the
Theorem D 4 .. The sum of positive real functions is positive real. jco axis.
Proof. Suppose Zf s ) and Z2(s) are both positive real. When Re s 0, then Proof. As a consequence of the derivation of Theorem D.5, it is seen that
poles may exist on the jco axis if n = 1, and <f> = 0. The condition n = 1 implies
ReZj 0 and Re Z2 0 that the pole is simple and the condition <j> = 0 implies that the residue is positive
and real. It is readily seen that zeros on the jco axis must also be simple.
so that ReZi 4- ReZ2 = ReZ ^: 0. Theorem D 7. The poles and zeros of Z( s ) are real or occur in conjugate pairs.
.
Also, when s is real, both Zx and Z2 are real. The sum of two real numbers Proof If a complex pole or zero exists without its conjugate, Z( s ) cannot be
is a real number. Therefore, Zx + Z2 is positive real. real when s is real. As a result of this theorem and Theorem D.5, it is seen that
both the numerator and denominator polynomials of Z{ s ) must be Hurwitz.
..
Theorem D 5 The poles and zeros of Z(s) cannot have positive real parts
(i.e., lie in the right half of the s plane). ..
Theorem D 8 The highest powers of the numerator polynomial and the
.
Proof Suppose there is a pole s0 in the right-half plane Let us make a . denominator polynomial of Z( s ) may differ by at most unity .
Laurent series expansion about s0 so that .
Proof Let Z{ s ) be written as

= bmsm +
ansn + an . sn ^ 1
1
+ • + ty + a0 ~
P( s )
Z( s ) = (7
=^r +
<* -T)"-
1
1 +" +*
490
i(* -'
«) + •• + fcr( s - S0 y + Z( s )
+ •• + V + b0 Q( s )
 492 Network analysis and synthesis

ju>

* 0
0 r = oo
0 appendix E
0 An aid to the improvement
0
0 of filter approximation
. .
FIG D 2

—
If m n > 2, when s = oo , Z( s ) will have a zero of order 2 or greater at s = oo ,
which is on the jw axis . Similarly , if n — m 2, then, at s = oo , Z( s ) will have a .
E l INTRODUCTION
pole of order 2 or more at s = oo . Since Z( s) cannot have multiple poles or
zeros on the jio axis, these situations cannot exist ; therefore \ n m\ <, 1 . — The introduction of an additional pole and zero in the second quadrant
Theorem D.9. The lowest powers of P( s ) and Q( s) may differ by at most unity , of the complex frequency plane, and at their conjugate locations, can give
Proof . The proof is obtained as in Theorem D. 8 by simply substituting l / s amplitude or phase corrections to a filter approximant over some desired
for s and proceeding as described . band of frequencies without significantly changing the approximant at
-
other frequencies. However, a cut-and try procedure for finding the best
Theorem D.10. A rational function F( s ) with real coefficients is positive real if :
( a ) F( s ) is analytic in the right -half plane .
positions for such a pole zero pair can be tedious. A visual aid is presented
-
(b) If F( s ) has poles on the jco axis, they must be simple and have real, positive herein which reduces the amount of labor required to make modest
residues. corrections of this type.
(c) Re F( jai) > 0 for all to . Constant phase and constant logarithmic gain contours for the correc-
Proof . We need only show that these three conditions fulfill the same require- tion by a pole-zero pair1 are plotted on transparent overlays. One of these
ments as Re Z( s) > 0 for Re s > 0 . We will make use of the minimum modulus may be placed over a suitably scaled sheet of graph paper representing the
theorem which states that if a function is analytic within a given region the complex frequency plane. Then the pair-shaped phase and gain correc-
minimum value of the real part of the function lies on the boundary of ^that tions along the jco axis are indicated by the intersections of the overlay
region. The region with which we are concerned is the right-half plane which is contours with this axis. Corrections which best reduce the errors in the
bounded by a semicircle of infinite radius and the imaginary axis with small original approximant are then sought by variation of the overlay position
indentations for the jcn axis poles. If the minimum value on they cuaxis is greater and orientation . Either phase or amplitude may be corrected . However,
'

than zero, then Re Z( s ) must be positive over the entire right-half plane ( Fig. D. 2).
it is not always possible to simultaneously improve both the phase and the
amplitude characteristics of an approximant by a single pair shaped -
correction.

,
F. F. Kuo and M . Karnaugh, reprinted from the IRE Transactions on Circuit Theory
CT-9, No . 4, December 1962 pp
, . 400-404 .
1 This will be called, hereafter, a pair-shaped correction .

493
-

494 Network analysis and synthesis Appendix E 495

.
E 2 CONSTANT LOGARITHMIC GAIN CONTOURS Let
Suppose we begin with a transfer function G( s) with certain deficiencies ( E .9 )
in its amplitude or phase response. Let us consider ’the transfer function q =
of a corrective network Gfs) such that the product
Then (E .10)
Gt(s) = Gfs) G(s) (E.l )
will have better gain or phase characteristics. For the purposes of this
sc = So - l ±i!M ( E.l 1)
paper, we will restrict ( (s) to have the form
^ ( s - q )( s - q)
1 - k2 \ 2/
Here, s0 is the midpoint between the pole and zero, and their separation
Gx( s ) = C (s - p )( s (E.2) is e. It is easy to see from Eq. E.11 that the center of each circle of constant
- p) gain is externally collinear with the pole and zero. For the purpose of
where C is a constant , and q and p are a zero and a pole, respectively, in drawing the family of constant gain contours, we may let s0 be the origin
the second quadrant of the complex frequency plane. If the correction of coordinates and the scale factor e may be set equal to unity.
is to be applied at sufficiently high frequencies such that s q s p
then — ^ — Furthermore, only half the pattern need be drawn because the function
D has negative symmetry with respect to a reflection about the perpen-
C( s - q ) dicular bisector of the pole-zero pair.
Gfs ) (E.3)
s- p ^
Let us consider the effect when the pole-zero pair in Eq . E.3 is used to .
E 3 CONSTANT PHASE CONTOURS
augment any given rational function in the complex frequency plane.
Figure E.l a represents a pole at p , a zero at q and an arbitrary point
The added gain, in decibels, due to this pole-zero pair is
s in the complex frequency plane. When the pole and zero are used to
correct a given phase characteristic, the added phase at s is
D = 20 log10 2 (E.4)
Is -p 4> = oq - ev (E.12)
where we have neglected the effect of the constant C in Eq. E.3. If we let where dQ and 0 P are measured with respect to a single arbitrary reference.
In Fig. E. lh, a circle has been drawn through p , q, and s. Angle <f> is
k = (10)D/ ° 2
(E.5)

then k = \ Lz± (E.6)

Is -P Bq Dp

For fixed k , this is the equation of a circle with inverse points 2 at q and
p. Its radius is
_
fc IP g \ - (E.7)
U - k2l
and its center is at
q - k2
s« = l - k2P (E.8)

2 E. C.
Titchmarsh, The Theory of Functions, Oxford University Press, Oxford ,
England, second edition 1939, pp . 191-192. FIG. E. l . Derivation of constant phase contours.
 Appendix E 497
496 Network analysis and synthesis
equal to one half the subtended arc ps'q. Therefore, the arc qsp is a
constant phase contour. The angle at the center of the circle between cp
and the perpendicular bisector of chord pq is also equal to <f>. All of the
circular contours of constant phase have their centers on this perpendicular 40“
bisector.
Note that minor arc ps' q is also a contour of constant phase, but the
50®
phase angle
a = <f> IT — (E.13)

is negative, as indicated by the clockwise rotation from s’ p to s’q. The

60 ®

convention for positive rotation is herein taken to be counterclockwise.

70"
80°

.
E 4 CONTOUR DRAWINGS
120"
Sets of constant phase and constant logarithmic gain contours are 130"
to ®

drawn in Figs E.2 and E.3 The curves are symmetric about the zero-
1
. . iso®
160 ®

decibel line, except that the gain curves are of opposite sign . Therefore,
only one half of each figure has been drawn. 150 “
140'
130°
120°
110”
100°

20“
70'

. ..
FIG E 3 Constant amplitude and phase contours .

The zero-decibel line is the perpendicular bisector of the line segment

joining the pole and zero. It is a gain contour of infinite radius.
The line through the pole and zero is a zero phase contour. Together
,
these perpendicular axes form a useful reference system . The signs of the
shown in
phase and gain in the four quadrants formed by these axes are
Fig. E.4.
The corrections in Figs. E .2 and E.3 do not carry algebraic signs
because
. ..
FIG E 2 Constant amplitude and phase contours .

498 Network analysis and synthesis Appendix E 499

Zero gain zero phase axis should be rotated clockwise with respect to the a (real)
axis.
G - G+
We thus see that by varying the frequency scale of the complex frequency
-
plane, the position of the center of the pole zero pair, and the orientation
t-
<> t-
<> of the pole-zero pair, the pair-shaped correction can be made to approxi-
mate the desired correction.
9
X
P
—-
> Zero phase It must be emphasized that the method suggested herein is an aid to
cut-and -try correction. As such, it is easier to use the method than to
G - G+
precisely set down rules for applying it. However, a few rather general
t
< >+ <P +
statements may be helpful.
Unless the pole and zero both lie on the real axis, one must remember
that another pole and zero are located at conjugate positions. The con-
.
FIG. E 4. Signs of the gain G and phase
$ corrections. tributions from both pole-zero pairs may be added algebraically. In most
practical cases, the desired correction will have a band-pass character.
only half of the symmetrical pattern has been drawn. The signs may be Therefore, only one pole-zero pair will normally contribute significantly
obtained from Fig. E.4, and are important in selecting the orientation of at any frequency.
the pole-zero pair.
The shape and magnitude of the desired correction will dictate the way
in which the jco axis must intersect the correction contours. The broadness
E.5 CORRECTION PROCEDURE of the desired correction will dictate the proper scaling of the jco axis.
Usually, only a few trials are needed to fix the pole and zero locations for
In correcting a given approximant, it is first necessary to plot the desired the best fit.
magnitude and/or phase corrections versus frequency. When the contours It will be found that a worthwhile correction can be made in either the
.
of Fig. E.2 or E 3 are overlaid on a second sheet representing the
complex phase or the gain characteristic. Only fortuitously can they be improved
frequency plane, the contour intersections with the jco axis indicate the simultaneously by a single pair-shaped correction.
corrections that will actually be realized .
There are several variables at the designer’s disposal. The first is the Example E.l. The amplitude response of a third-order Butterworth filter is
distance between the correcting pole and zero. Since the scale of the
contours in Figs. E.2 and E.3 is not specified, the frequency scale of
-
given by the solid curve in Fig. E.5. It is desired to steepen the gain roll-off
near the cutoff frequency eoc = 1 . This is done by increasing the gain just below
the underlying complex plane determines the distance between co = coc and decreasing it above that frequency. Figure E.6 illustrates the type
the pole
and zero. The second design variable is the location of the center of the of correction desired. Figure E.7 shows a pole-zero pair that achieves this type
-
pole zero pair. The distance of the center from a given band on the jco of correction. The gain at co = 1 remains unchanged by this particular choice.
Other pole-zero pairs that “aim” the zero-decibel line at co = 1 but give asym-
axis determines the magnitude of the correction.
The third variable is the orientation of the pole-zero pair. Figure E.4 metrical corrections about that point might also be used. The dashed curve in
shows how the orientation affects the gain and phase corrections , As a Fig. E.5 shows the corrected gain .
simple example, suppose one wishes to have zero phase correction at A different pole-zero pair, also shown in Fig. E.7, has been chosen to minimize
the deviation of the slope of the phase response from its slope at co = 0. This
co = 1.0, negative phase correction above and positive correction below
pair-shaped correction decreases the phase for co < 0.7 and increases it for
.
that frequency The attack would be to point the zero phase axis at co > 0.7 .
co = 1.0 with the pole nearest to the jco axis. If it is desired to have Figure E.8 shows the deviation of the phase responses from linear phase.
equal
phase correction above and below co = 1.0, the zero phase axis should be It is clear from Figs. E.5 and E.8 that the gain corrected approximant has a
oriented parallel to the real axis of the complex frequency plane. If one poorer phase response than the original, while the phase corrected approximant
wishes to have more phase correction above co 1.0 and less below, the has a gentler gain roll-off than the original.
=
 Appendix E 501
500 Network analysis and synthesis
2
- 44 -4-- --l- Loss 1.2
With amplitude correction ^
^^ K correction
o
1.0

—M
0 \ x
-
~
-
''Third order Butterworth \
\
Butterworth
poles 0.8
I
1 x o
Phase corrected response -^ S
N ’
Phase
correction
0.6
g \
-
" -4 0.4

-
D

0.2
I 0 jo
'1 -1.0t -0.8 - 0.6 -0.4 -0.2
-8 \ -0.2
l \
\ j \
- 10
\\ \
\
-0.4
\
V
l -0.6
- 12 i X O
0.1 0.2 0.3 0.5
Frequency , o
0.7 1.0 1.5 2.0 N -0.8
o
.
FIG E.5. Amplitude response of corrected and uncorrected Butterworth filters. x -1.0
-I 1.2
correction equalizers.
-
This will not surprise the experienced filter designer. It is possible, however, . ..
FIG E 7 Poles and zeros of the original filter and
to achieve moderate corrections in both gain and phase by using two pair-
shaped corrections. A useful approach to this objective lies in localizing the
gain correction further out of band , and the phase correction further in- band 20 7
than in the separate corrections just discussed. This can be done by shifting
the pole-zero centers to higher or lower frequencies and also by experimenting 10 / /
with nonsymmetrical corrections. With phase correction / /
0
1
. XX Third-order /
.5 -10
_o>
=S
I
I -2 0
f
\N
\
\
Butterworth

^/

/ Withcorrection
amplitude
__
<3
0
- 30 V- JL
\
-40 0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
Frequency, rad /sec
uc
O -
> FIG E 8 Phase deviation from slope at zero frequency.
. ..
.
FIG. E.6 Amplitude correction to steepen fall-off of third degree Butterworth filter - .

Appendix E 503
502 Network analysis and synthesis
jo
E.6 CORRECTION NETWORK DESIGN

A few words are in order concerning the synthesis of the equalizer from
the pole-zero pair obtained by the method described . s plane
In order to provide optimum power transfer and to facilitate cascading /
np
the correction network with the original filter, the correction network j nz
should be of the constant-resistance type. We will restrict our discussion I
I /
to the bridged-T network in Fig. E. 9, whose voltage transfer function is fo?
0
given as
V (s) Ro
— a
G (s ) = * ( E . 14)
VM R0 + z,(s)
provided the network is of the constant resistance type as given by the
equation
Z ,( s ) Z2( s ) = R0* (E. 15)
For normalization purposes, we will let R0 = 1 Q .
Let the pole- zero correction be written in general as - jo
S + a ,S + f l0
2 . . .
FIG E I 0 Poles and zeros of Z ( s) , .
G( s ) = 2
( E . 16 )
fc(s + + b0 ) Moreover, in order for a biquadratic driving-point immittance to be

-
Since the correction has minimum phase, we know that { ait bt ]> 0}. In
positive real, the following condition must apply.
3

addition, since the d-c gain cannot exceed unity, kb0 > a0 .

•ptff 'MPtffir 'T

We can express the impedance Z j) in terms of G( s ) in Eq . E . 14 as
1 ( k - l )s ( kb , - qjs + ( kb0 - a0 )
8 ^+ ( E . 19)

X_ j=
2,(3) =
s2 + ats + a 0
(E .17)
G( s ) Details concerning the synthesis of Z s ) are also given in Seshu s paper.
’

Since Z ,( s ) must be positive real , the coefficients must all be nonnegative Let us plot the pole-zero pair of Z ,( s ) as in Fig. E. 10. We can represent
angles <f> P and 0 Z
^
so that
k-1>0 the locations of the poles and zeros in terms of the polar 4 has shown that in
and their distances from the origin, nv and nz . Rack
kb , a, > 0 — (E. 18)
order for in- band loss to be finite
kb0 - a0 > 0
— (E. 20)
n, <
2

In addition he has shown that for an unbalanced bridged-T circuit, if

Ro Ro a = max [02, 4>v ] then the larger the angle a, the larger the in- band
loss.
WWA vw In particukr, a should be less than 70° to restrict the in-band loss to
V , Ro z2 Ro
t
v2
reasonable proportions . If one considers other network configurations
3S. Seshu, Minimal Realizations of the Biquadratic Minimum Function
, IRE
Trans , on Circuit Theory , CT-6, December 1959, 345-350.
4 A. J. Rack, private communication .
.
FIG. E.9 Bridged T network. -

£
504 Network analysis and synthesis
such as lattice networks, it is conceivable that Rack’s restrictions might
be relaxed . For lattice network design, Weinberg’s method is applicable,6
although here again, the in-band loss restrictions are severe.

E.7 CONCLUSION

We have presented here a simple visual aid to the correction of the

amplitude or phase response of filters. The method has the advantage of
facilitating the commonly used cut-and-try approach to this problem. It
has the disadvantage, in many cases, of failing to provide simultaneous
amplitude and phase corrections in a single step.
Bibliography
5 .
L. Weinberg, “ RLC lattice networks,” Proc IRE, 41, September 1953, 1139-1144.

SIGNAL ANALYSIS

Goldman , S., Frequency Analysis , Modulation and Noise , McGraw- Hill ,

New York , 1948.
Javid , M. and E. Brenner, Analysis, Transmission and Filtering of Signals,
McGraw- Hill , New York, 1963.
Lathi, B. P., Signals, Systems and Communications, John Wiley and Sons,
New York, 1965.
Lighthill , M . J., Introduction to Fourier Analysis and Generalized Functions,
Cambridge University Press, New York , 1958.
Mason , S. J. and H. J. Zimmerman, Electronic Circuits , Signals and
Systems, John Wiley and Sons, New York, 1960.
Papoulis, A., The Fourier Integral and Its Applications, McGraw-Hill, New
York , 1962.
Rowe, H . E., Signals and Noise in Communications Systems, D. Van
Nostrand , Princeton, N.J. 1965.
Schwartz, M., Information Transmission , Modulation and Noise , McGraw-
Hill, New York, 1959.

NETWORK ANALYSIS

Balabanian , N., Fundamentals of Circuit Theory , Allyn and Bacon, Boston ,

1961.
-
Bohn, E. V., The Transform Analysis of Linear Systems, Addison Wesley,
Reading, Massachusetts, 1963.
Brenner, E. and M. Javid , Analysis of Electric Circuits, McGraw- Hill , New
York , 1959.
505