Network Handout

AS&TU, EE Department Network Analysis & Synthesis – EEng 3209
Chapter 1
1. Introduction to Network Analysis & Synthesis

Network is a combination of electrical elements connected in any manner (conductively, inductively
or capacitively), whose impedance may be lamped or distributed or both. Further, it can be defined as
a collection of elements connected together to perform a certain task governed by a certain laws.
Based on the component from which network developed we have two classes: passive and active
networks.
In generally accepted definitions of network analysis and synthesis, there are three key words: the
excitation, the network, the response as shown by fig. 1.1.
Excitation Response
Network
Fig.1.1. Key terms in network analysis & synthesis
A network in which the electrical properties are unaffected by interchanging input and output
terminals is called balanced network. In such networks, the elements are symmetrical with respect to
ground potential. To mention some examples of networks:
 Filters  electrical networks which pass selective frequency signals.
 Amplifiers/ Attenuators  electrical networks which are used to magnify or reduce signal power
level from input to output.
 Equalizers  electrical networks used to counteract frequency or phase distortions.
 Matching networks  electrical networks that match source and load impedances.
The study of network analysis and synthesis is sometimes referred to as network theory. Network
theory is based on fundamental electrical parameters. Network theory can be categorized into three
classes
1
i. Mathematical modeling of electromagnetic phenomena

ii. Mathematical analysis of models of individual elements and systems
iii. Synthesis and design of systems & devices
Modeling of a system should adequately describe the physical behavior of the system i.e. model is
the compromise between reality and theory (simplicity). Most electrical system analysis depends on
two fundamental theoretical concepts:
a. The lumped parameter circuit theory based on laws: KVL and KCL, ohms laws, etc…
b. Field theory based on Maxwell‟s equations for networks composed of distributed
elements.
An electric network (amplifier, filter or equalizer circuits and the like) can be modeled using certain
lumped circuit elements. A basic problem faced by an engineer is to design a network model to
satisfy certain signal processing specifications and then fabricate physical components which
approximate the idealized elements. A prelude to design (and synthesis) is analysis, which is mainly
an algebraic problem. In this portion we discuss the characterization and modeling of lumped circuit
elements. A network consisting of lumped elements exhibits certain basic properties depending on
the type of elements used which themselves are classified according to their properties.
1.1.Lumped Circuit elements
A circuit element is said to be lumped if the instantaneous current entering one terminal is equal to
the instantaneous current leaving at the other terminals otherwise it is distributed elements. A
television antenna is not a lumped element as the current at the foot of the antenna is not the same as
that at the tip of the antenna. If the two were same there would be no radiation. The current through
and the voltage across a lumped element are well defined quantities and satisfy Kirchhoff‟s laws.
If the physical dimension are small compared with the wave length of the highest signal frequency
applied to the network, then the element will be consider to be lumped. For example a frequency of
operation of 10 kHz (audio frequency) corresponds to a wavelength of 30km and any circuit built in
a laboratory has negligible physical dimension compared to this wavelength. If the frequency of
operation is 1GHz (microwave frequency) then the wave length is 0.3m and a circuits dimension
becomes comparable to the wavelength.
2
A large part of network theory deals with the study of lumped circuits. Lumped circuit is a result of
interconnection of lumped elements. Typical lumped elements are: resistors, capacitors, inductors,
and voltage and current sources.
Resistors: An element which can be characterized by a curve in the V–I plane is called a resistor (V
represents voltage and I represents current). There are two types‟ i.e. linear and non-linear resistors.
a. Linear resistors: I = GV or V=RI
I
I = GV or
slope=G=1/R
Fig.1.2. Linear resistor (a)characteristic curve (b) Circuit symbol
Another characterization of linear network is that the excitation and response of the
network are related by a linear differential equation.
b. Non linear resistors: i-v curve is not straight line, two nonlinear resistors are tunnel
diode and SCR(silicon controlled rectifier). The current through tunnel diode is single
valued function of the voltage, which makes it a voltage controlled resistor. This is
depicted in fig. 1.3(a). In SCR, the voltage is single valued function of the current, which
makes it act as the current controlled resistor as shown in fig. 1.3(b). These nonlinear
resistors V-I characteristic have unique property of negative resistance in some ranges.
I
I=f(V)
Negative resistor
region
Fig.1.3. Non-linear characteristic curves
3
Evidently, nonlinear resistors cannot be characterized by a single valued resistance R, as in the case
of linear resistors. The resistance value of nonlinear resistor depends on the operating point and
hence is characterized by incremental resistance which is defined as the slope of the v-i characteristic
at the particular operating point. Nonlinear resistors find use in rectification, frequency
multiplication, current and voltage limiting, and many other electronic applications. Fig.1.4 shows
the symbol and the v-i characteristic of a semiconductor diode (continuous curve). In analyzing a
network with a nonlinear element, such as a diode, one often resorts to piece-wise linearization. The
piece-wise linear model of a diode (broken line) is indicated in fig. 1.4.
Fig.1.4. Piece-wise linearization of characteristic curve
A distinct property of a linear resistor not usually possessed by a non-linear resistor is that the v-i
characteristic is symmetric with respect to the origin. An element exhibiting such a symmetric
property is called bilateral element. All linear circuits are bilateral but not all nonlinear circuits are
bilateral.
Capacitors: An element which can be characterized by a curve in the v-q plane is called a capacitor
(q represents the charge). A capacitor is linear if its characteristic is a straight line passing through
the origin of the v-q plane.
Fig.1.5. Linear capacitor (a) characteristic curve (b) Circuit symbol
4
A linear capacitor is represented by the symbol shown above and is described analytically by
q  CV where the constant C is the slope of the straight line, and is called the capacitance. The unit
of capacitance is farad. (A farad is a very big unit for measuring capacitance; a more practical unit in
a network being F -microfarad, or pF microfarads).
The current charge relation being i  dq dt , for a linear capacitor, we have
dv(t )
iC
dt
If a capacitor is characterized by a v-q curve other than a straight line through the origin, it is called a
nonlinear capacitor. Fig.1.6 shows the characteristic of a nonlinear capacitor (metal oxide
semiconductor capacitor). A nonlinear capacitor is characterized by q  f (v) (for a voltage-
controlled capacitor) and by v   (q) (for a charge-controlled capacitor).
For a voltage controlled capacitor the current through the capacitor is

dq(t ) df (v) dv(t )
i(t )   
dt dv dt
dv(t )
Or i(t )  C (v) where C (v)  df (v) / dv is called the incremental capacitance of
dt
the capacitor.
Fig.1.6. Non-linear capacitor characteristic curve
The instantaneous power entering an element is given by

p(t )  v(t )i(t )
where v is the voltage in volts and i is the current in ampere. Hence, the energy in the element
from t 0 and t is
t
W (t0 , t )   v( )i( )d
to
5
Let a capacitance be characterized by c  vˆ(q) the energy in the capacitor, from the above equations
as
q (t )
t
W (t 0 , t )   v( )i( )d   vˆ(q)dq
t0
q ( t0 )
If the initial charge on the capacitors, q(t 0 ), is zero, then the energy stored in the capacitor is
q (t )
W (t )   vˆ(q)dq
0
For a linear capacitor q  Cv  it reduces to
1 q 2 (t ) 1 2
W (t )   Cv (t )
2 C 2
Inductors: An element which can be characterized by a curve in the i   plane is called an
inductor ( represents the flux). An inductor is called a linear inductor, if its characteristic is a
straight line passing through the origin of the i   plane as shown by fig.1.7.
Fig.1.7. Linear inductor (a) characteristic curve (b) Circuit symbol
A linear inductor is represented by the symbol shown in fig.1.7.(b) and described analytically by
  Li
where the constant L is the slope of the straight line and is called inductance. The unit of inductance
is Hennery. (Since inductance of one Henry is quite large, typical values of inductors are given mH-
milli-henrys.)
The voltage flux linkage relation being v  d / dt for a linear inductor, we have
di(t )
vL
dt
6
If an inductor is characterized by a i   curve other than a straight line through the origin, then it is
called nonlinear inductor. A nonlinear inductor is characterized by
  f (i) (For a current controlled inductor) and by i   ( ) (for a flux-controlled inductor) for a
current-controlled inductor the voltage across the inductor is
d (t ) df (i) di(t )
v(t )   
dt dt dt
di(t ) df (t )
Or v(t )  L(i) where L(i)  is called the incremental inductance.
dt dt
Nonlinear inductors can be used in a frequency conversion, memory and storage.
A special type of nonlinear inductor, such as a ferromagnetic-core inductor, has a characteristic that
exhibits the phenomenon of hysteresis.
Independent sources: Fig.1.8 shows an arbitrary network N excited by a source of electric

energy. If we change the network N, then v and i in general also change. If the prescribed voltage is
maintained then the source of energy is called a voltage source. On the other hand if the prescribed
current is maintained it is called current source.
Fig.1.5. Independent source
A two terminal element is called an ideal independent voltage source, if it is capable of supplying
any current at the same prescribed voltage, i.e. the voltage across the source is independent of the
current drawn from the source as shown by fig.1.9. If the voltage of a voltage source is identically
zero, the voltage source is effectively a short circuit. The internal resistance of an ideal voltage
source is considered to be zero.
7
Fig.1.9. Voltage source:.symbol and the i-v characteristic curve
A two-terminal element is called an ideal independent current source, if it is capable of supplying a

prescribed current at any voltage, i.e. the current drawn from a current source is independent of the
voltage across the source. Fig.1.10. shows the symbol and the i-v characteristic of current source. If
the current drawn from a current source is identically zero, the current source is effectively an open
circuit. The internal shunt resistance of an ideal current source is considered to be infinite. In the
physical world there is no ideal independent current and voltage sources.
Fig.1.10. Voltage source:.symbol and the i-v characteristic curve
A dependent voltage (or current) source is a source the voltage (or current) of which depends on
another voltage (or current). A dependent or controlled source is said to be voltage or current
controlled if its terminal behavior is controlled by another voltage or current. This leads to four
different controlled sources which are:
1. Voltage-controlled voltage source
2. Voltage-controlled current source
3. Current-controlled voltage source
4. Current-controlled current source
Some physical devices operate almost like ideal dependent sources. For example, an operational
amplifier is a voltage controlled voltage source, a field effect transistor a voltage controlled current
source.
8
1.2. Classification of Networks
The overall behavior of an electrical network can be predicated by the constituent element and their
interconnection. The behavior of the network, considered as a black box, leads to a number of
classifications like linear, nonlinear: time-invariant and time variant; passive, active.
Linear and Nonlinear Networks
In a linear network, the relationship between the voltage and current is described by a linear
equation. Consider two networks N1, and N2 as shown in fig.1.11(a) and (b) respectively. Network
(a) is made up of a linear resistor R, while (b) is made up of a semi-conductor diode and linear
resistor R. let the cut-in voltage of the diode be 0.6volt. In network (a) current I1 is given by V/R and
exists for all values of V. In network (b) if V is less than 0.6 volt the current I 2 is zero and for
voltages higher than 0.6volts I2 is given by (V-0.6)/(R+RF), where RF is the forward resistance of the
diode. Obviously, in network (a) the current response is linear in contrast to that in (b).
A system (network) is linear if (i) the principle of superposition and (ii) the principle of
proportionality hold. By the superposition principle, if, for a given network, e1 (t ), r1 t  and
e2 (t ), r2 t  are excitation-response pairs, then if the excitation were e(t )  e1 (t )  e2 (t ), the response
would be r (t )  r1 (t )  r2 (t ) . By the proportionality principle, if the excitation were C1e1 (t ), where
C1 a constant is, then the response would be C1r1 (t ), i.e. the constant proportionality C1 is preserved
by the linear network.
Let both the networks be excited by two serially connected voltage sources V1 and V2. Then it can be
seen that I1 (V1  V2 )  I1 V1   I1 (V2 ) And I 2 (V1  V2 )  I 2 V1   I1 (V2 ) .
Fig.1.11. Linear and Non-linear networks
9
Where I i (V j ) is the current into the terminals of network N i when excited be a voltage source V j .
We say that network N 1 is linear and N 2 is nonlinear as the principle of superposition holds for N 1
but not for N 2 .
Linearity of a network can be defined as follow: let a network be characterized by F ( xi )  yi where
xi is the input and y i the output, and F (.) denotes some function. Then the network is linear if, and
only if,
F (1 x1   2 x2 )  1 F ( x1 )   2 F ( x2 )  1 y1   2 y2
where  1 and  2 are arbitrary constants, and x1 and x 2 are any two allowable inputs. The principle
expressed by this equation is called the principle of superposition and homogeneity. Hence, we
conclude that a network is linear if it satisfies this principle; other-wise, it is nonlinear.
Time-invariant and Time-variant Networks
Let a linear resistor be characterized by v(t )  R(t )i(t ) where R(t ) is a prescribed time function. This
can be achieved, for example, by the sliding contact of potentiometer being moved back and forth by
a motor. Such a resistor is called a time-varying resistor. Similarly, it is possible to build time-
varying capacitors and inductors. The elements we considered previously were all time-invariant in
that they were characterized by parameters which were not dependent on time.
A time – invariant network is characterized by a constant coefficient equation whereas the time-
variant one by a time-variant coefficient equation. Mathematically, we can describe a time-invariant
network by
F x(t  t 0 )  y(t  t 0 )
when the network is characterized by F x(t )  y(t ) i.e. the response (output) depends on the shape
of the excitation (input) but not on the time of application. A network composed of time-invariant
elements is necessarily time-invariant whereas network composed of time-variant elements may
exhibit time-invariant terminal behavior.
10
Passive and Active Networks
Consider a network made up of a single linear resistor. The energy supplied to (or dissipated by) the
resistor, from the previous relation is
t
W (t , t 0 )   i 2 (t ) Rdt
to
If the resistor has to deliver power to the external world, R has to be negative. As long as R is
positive the resistor will consume power, and such a resistor is called a passive resistor.
Let v(t) and i(t) be the two voltage and current at the terminals of a network. Then the energy
delivered to the network is given by
t
W (t , t0 )   v( )i( )d
to
A network is said to be passive if, and only if, W (t , t0 )  E (t0 )  0 for all t and t 0 , and for all
v(t ) and i (t ) , where E (t 0 ) is the energy in the network at t  t 0 . Otherwise, the network is said to
be active. In other words, if the energy delivered to the network is non-negative for all time and
input, the network is said to be passive. The conditions for activity of an element can also be
obtained by a study of its characteristics. For example, we can state that a nonlinear resistor is
passive if, and only if, its characteristic, for all time, is in the first and third quadrants of v-i plane.
Similarly, a capacitor (inductor) is passive if, and only if, its characteristic is in the first and third
quadrants of the v  q(  i) plane.
11
Chapter 2
2. Network Transform Representation and Analysis
2.1. Introduction
A concept of transform impedance and transform admittance is discussed in this chapter.

Furthermore, a function relating currents or voltages at a different parts of the network, called a
transfer function, is found to be mathematically similar to the transform impedance function. These
functions are called network functions.
2.2. Network Functions
In an electrical network, the word port has special meaning. It is a pair of terminals in which the
current into one terminal is equal to the current out of the other. The port at which the
input/excitation is given is called driving point or input port. The port at which output is taken is
called output port. If the network has only driving port, it is called one-port network and if the
network has input and output ports, then it is called two-port network. In fig.2.1., a general two-port
network and a standard convention adopted in the designation of voltages and currents are shown.
A one port network is completely specified when a voltage-current relationship at the terminals of
the port is given. The four variables (I1, I2, V1 and V2) of the two-port network, taken two at a time,
results in six set of equations that describe the two-port network. These equations are called network
functions.
A network function is defined as the ratio of the zero – state response to the input, both the response
and the input expressed in Laplace domain, i.e.
( )
( ) (2.1)
( )
where s the Laplace variable is the complex frequency variable .
12
Two sets of network functions can be defined: (1) Driving point functions (2) Transfer functions.
These are defined and illustrated by using the network shown in fig.2.1.
Fig.2.1. Network used to define network functions

1. Driving Point Functions
If the excitation and the response are measured at the same set of terminals (terminal 1-1‟, or 2-2‟),
the network function is called driving point function. There are two driving point functions:
impedance and admittance.
( )
(driving point impedance)
( )
( )
(driving point admittance) (2.2)
( )
Because of the similarity of impedance and admittance, these two quantities are assigned one name
“Immitance” (a combination of impedance and admittance).
From the two relations, we can observe that the reciprocal of a driving point function is another
driving point function.
2. Transfer functions
If the excitation and response are measured at different sets of terminals, then the corresponding
network function is called a transfer function.
( )
( ) Transfer impedance
( )
( )
( ) Transfer Admittance (2.3)
( )
( )
( ) Transfer voltage ratio
( )
( )
( ) Transfer current ratio
( )
It can be observed from the definition of DP and transfer functions that for an impedance function,
excitation is a current source and response a voltage, and for an admittance function excitation a
13
voltage source and response a current. For simple networks DP function can be obtained by
inspection. For example, consider the ladder network shown in fig.2.2.
Fig.2.2. Ladder network

( ) ( ) (2.4)
( )
( )
i.e. the DP impedance of a ladder can be written in the form of a continued fraction expansion and
such an expansion is very useful in network analysis.
Exercise: Find the impedance and admittance for some simple networks containing,
a) Series b) Parallel
i. R – L i. R - L
ii. R – C ii. R - C
iii. L – C iii. L - C
iv. R – L – C iv. R – L - C
2.3. Poles and Zeros
Consider a linear time invariant network. Let x(t) be excitation and y(t) the response. Thus, in
general, we can express the relationship between x(t) and y(t) in the form of differential equation,
( ) ( ) ( ) ( ) (2.5)
The coefficients in this differential equation depend on the elements and the topology of the network.
Laplace transforming and rearranging (with all initial conditions assumed to be zero) we have
( ) ( )
( ) , i.e., ( ) (2.6)
( ) ( ) ( )
where p and q are polynomials in the complex frequency variable s. Thus we conclude that a network
function is a rational function of s with real coefficients. Alternatively, we can write equation (2.6) as
∏ ( )
( ) where ⁄ (2.7)
∏ ( )
Note: zi is called zero of H(s) and pi is called poles of H(s). The poles and zeros of a network need
not be distinct. The coefficients of the polynomials p(s) and q(s) being real any complex zeros and
poles must appear in conjugate pairs. From equation (2.7), it is clear that any network function is
completely specified by its poles and zeros and the scale factor k.
14
Exercise:
Consider the network shown in fig.2.3. Let x(t) be the input and y(t) (mesh current of the mesh II) be
the output. Find H(s).
Fig.2.3.
The poles and zeros of a network function can be represented by a pole –zero plots. The pole – zero
plot of the network function is shown in fig.2.4.
Fig.2.4. pole-zero plot
If x(t) and y(t) are measured at the same set of terminals in equation (2.5), then H(s) would be a
driving point function. Without loss of generality, we assume that x(t) is current, and the response
y(t) is voltage,. Then
( ) ( ) ( ) ( ) (2.8)
If we make the current x(t) go to zero, i.e., the network is being tested under open circuit conditions,
then the zero input voltage response is governed by the equation:
( ) ( ) (2.9)
Hence the natural frequencies of the voltage response are determined by the roots of the equation,
( ) (2.10)
Thus the roots of q(s) are the natural frequencies of the network under open circuit natural
frequencies (OCNF). On the other hand if we make the voltage y(t) go to zero, i.e., the network is
being tested under short circuit conditions, then the zero input current response is governed by the
equation
( ) ( ) (2.11)
And the natural frequencies of the current response are determined by the roots of the equation
( ) (2.12)
Hence the roots of p(s) are called the short circuit natural frequencies (SCNF).
15
 Necessary conditions for driving point Immitance Functions (with common factors in
N(s) and D(s) cancelled):
1. The coefficients in the polynomials N(s) and D(s) must be real and positive.
2. Poles and zeros must be conjugate if imaginary or complex.
3. The real part of all poles and zeros must be negative or zero, if the real part is zero, then
that pole or zero must be simple, i.e, all the roots of N(s) = 0 and D(s) = 0 lie on the left
half of s- plane and simple roots may lie on the imaginary or jw – axis.
4. The polynomials N(s) and D(s) may not have missing terms between those of highest and
lowest degrees, unless all even or all odd terms are missing.
5. The highest degree of N(s) and D(s) may differ by either zero or one only.
6. The lowest degree of N(s) and D(s) may differ by either zero or one only.
Exercise:
Check whether given functions are suitable in representing the driving point Impedance functions.
a) ( )
b) ( )
 Necessary conditions for transfer functions (with common factors in N(s) and D(s)
cancelled)
1. The coefficients in the polynomials N(s) and D(s) of T = N/D must be real and those for
D(s) must be positive.
2. Poles and zeros must be conjugate if complex.
3. The real part of poles must be negative or zero, if the real part is zero, then that pole must
be simple. This includes the origin.
4. The polynomial D(s) may not have any missing term between that of highest and lowest
degrees, unless all even or all odd terms are missing.
5. The polynomials N(s) may have terms missing, and some of the coefficients may be
negative.
6. The degree of N(s) may be as small as zero independent of the degree of D(s).
7. (a) for G and α: the maximum degree of N(s) is equal to the degree of D(s).
(b) for Z and Y: the maximum degree of N(s) is equal to the degree of D(s) plus one.
Exercise
Check whether the given functions are suitable in representing the transfer functions.
a) ( )
b) ( )
16
2.4. Stability of a Network
Among the many forms of performance specifications used in design, the most important
requirement is that the system be stable. An unstable system is generally considered to be useless.
For analysis and design purposes, we can classify stability as absolute stability and relative stability.
Absolute stability refers to the condition of whether the system is stable or unstable; it is a yes or no
answer. Once the system is found to be stable, it is of interest to determine how stable it is, and this
degree of stability is a measure of relative stability.
A system is said to be stable if its output (response) cannot be made to increase indefinitely by the
application of a bounded input excitation.
Relationship between impulse response and stability

The stability of a network function T(s) can be conveniently determined by considering its response
to an impulse function which is obtained by taking inverse laplace transform of the partial fraction
expansion of the function.
Stable system: A system is said to be stable if the impulse response approaches to zero for
sufficiently large time.
Ustable system: A system is said to be unstable if the impulse response grows without bound i.e,
approaches infinity for sufficiently large time.
Marginally stable system: A system is said to be marginally stable if the impulse response
approaches a constant non zero value or a constant amplitude oscillation for sufficiently large time.
Relationship between Poles positions and stability

The necessary and sufficient condition for the system to be stable is that all roots of characteristics
equation of the system lie in the negative half of the s–plane. In other words, we can say the poles of
the transfer function T(s) = N(s)/D(s) (i.e, roots of D(s) =0) lie in left half of the s–plane. On the
other hand, a system having any pole in the right half (or positive half) of the s–plane will be
unstable.
Routh Hurwitz stability criterion

Routh Hurwitz stability criterion is method of determining number of roots of characteristic equation
i.e., poles of transfer function with negative real parts, zero real parts and positive real parts and
hence system stability.
Consider that the characteristic equation of a system is of the form:
( ) (2.13)
where all the coefficients ai are real. In order that equation (2.13) not have roots with positive real
parts, it is necessary that the following conditions hold:
1. All the coefficients of the equation have the same sign.
2. None of the coefficients vanishes, i.e, no missing term between that of highest and lowest
degrees.
17
Routh’s Tabulation
The first step in the Routh – Hurwitz criterion is to arrange the coefficients of the equation (2.13)
into two rows. The first row consists of the first, third, fifth…, coefficients and the second row
consists of the second, fourth, sixth,…, coefficients, all counting from the highest order term
an an-2 an-4 …
an-1 an-3 an-5 …
The next step is to form the array of numbers by the indicated operations
sn an an-2 an-4 …
sn-1 an-1 an-3 an-5 …
n-2
s bn bn-1 bn-2 …
sn-3 cn cn-1
. .
. .
. .
s1
s0
where
| | | |
| | | |
| |
The array is called the Routh‟s tabulation or Routh‟s array. The last row of the Routh‟s tabulation
should always be the s0 row.
The necessary and sufficient condition that all roots of equation (2.13) have negative real parts if all
the elements of the first column of the Routh’s tabulation are of the same sign. The number of the
changes of signs in the elements of the first column equals the number of roots with positive real
parts.
Exercise:
1. Determine the stability of the systems whose characteristic equations are:
a) s4 +2s3+3s2+4s+5 = 0
b) s4+8s3+18s2+16s+5 = 0
2. Calculate the range of K for which the system given by characteristics equation
S3+7s2+10s+10K = 0 is stable.
18
Special cases when Routh’s tabulation terminates prematurely

Depending on the coefficients of the equation, sometimes the following difficulties may occur that
prevent Routh‟s tabulation from completing properly:
1. The first element in any one row of Routh‟s tabulation is zero, but the others are not.
2. All elements in one row of Routh‟s tabulation are zero.
Case 1: If a zero appears in the first element of a row, the elements in the next row will all become
infinite, and Routh‟s tabulation cannot be continued. To remedy this situation, we replace the zero
elements in the first column by an arbitrary small positive number ε, and then proceed with Routh‟s
tabulation.
Exercise:
Check the stability of the system with characteristic equation 2s5+s4 +6s3+3s2+s+1=0.
Case 2: If all the elements in one row of Routh‟s tabulation are zero before the tabulation is properly
terminated, it indicates that one or more of the following conditions may exist:
1. The equation has at least one pair of real roots with equal magnitude but opposite signs.
2. The equation has one or more pairs of imaginary roots with equal magnitude opposite
signs. (e.g., s = ± j1, s= ±j2, ±j3).
3. The equation has pairs of complex conjugate roots forming symmetry about the origin of
the s–plane (e.g., s = -1 ± j1, s = 1 ± j1).
The situation with the entire row of zeros can be remedied by using the auxiliary equation A(s) = 0,
which is formed from the coefficients of the row just above the row of zeros in Routh‟s tabulation.
The auxiliary equation is always even polynomials. The roots of the auxiliary equation also satisfy
the original equation. Thus, by solving the auxiliary equation, we also get some of the roots of the
original equation. To continue with Routh‟s tabulation when a row of zeros appears, we conduct the
following steps:
1. Form the auxiliary equation A(s) = 0 by use of the coefficients from the row just
preceding the row of zeros.
( )
2. Take the derivative of the auxiliary equation with respect to s; this gives = 0.
( )
3. Replace the row of zeros with coefficients of = 0.
4. Continue with Routh‟s tabulation in the usual manner with the newly formed row of
coefficients replacing the row of zeros.
5. Interpret the change of signs, if any, of the coefficients in the first column of Routh‟s
tabulation in the usual manner.
Exercise: check the stability of the system with characteristic equation S5 + 4s4+8s3 + 8s2+7s+4 = 0.
19
Deficiencies of Routh Hurwitz Criterion

1. It is valid only if the characteristic equation is algebraic with real coefficients. If any of
the coefficients is complex, or if the equation is not algebraic, such as containing
exponential or sinusoidal functions of s, the Routh Hurwitz criterion is simply cannot be
applied.
2. It gives an information about roots of characteristic equation only with respect to the left
half or right half of the s plane i.e., it doesn‟t give the information about the roots on the
jw axis (stability boundary).
3. It cannot be applied to discrete time systems.
Exercise: Determine the range of K so that the system given by characteristic equation
S3+3Ks2+(K+2)s+4 = 0 is stable.
Tutorial Exercises
1. Obtain the pole zero location for the function
( )( )
( )
( )( )
2. Calculate the driving point impedances of the networks shown in the fig.2.5 and plot their
pole zero diagrams.
Fig 2.5
3. The pole zero plot of a voltage transfer function is shown in fig.2.6. Find the transfer function
if the gain is to be 10.
Fig.2.6
20
4. Fig.2.7shows an infinite resistive ladder. Find the input resistance of the ladder.
Fig.2.7
5. Determine the stability of the systems whose characteristic equations are given by
a)
b)
6. Using Routh‟s Hurwitz criterion, determine the number of roots in the right half of s–plane
for the characteristic equations:
a)
b)
7. Determine the range of K for which the systems given by the following characteristic
equations are stable:
a)
b) ( )
21
Chapter 3
3. Elements of Realizability Theory
3.1. Causality and Stability

In the preceding chapters we have been primarily concerned with the problem of determining the
response, given the excitation and he network; this problem lies in the domain of network analysis.
The starting point for any synthesis problem is the system function
Rs 
H s   3.1
E (s)
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  can be realized as a physical passive network. There are two
important considerations: causality and stability. By causality we mean that the voltage cannot
appear between any pair of terminals in the network before a current is impressed, or vice versa. In
other words, the impulse response of the network must be zero for t  0 , that is,
h(t )  0 for t  0 . As an example, the impulse response h(t )  e  atu(t ) is causal, where as
a t
h(t )  e 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. 3.1(a) is not realizable. If we delay the response by T
seconds, we find that the delayed response h(t  T ) as shown in fig. 3.1(b) is realizable.
(a) (b)
Fig.3.1: (a) Non-realizable impulse response. (b) Realizable impulse response
22
In the frequency domain, causality is implied when the paley-wiener criterion is satisfied for the
amplitude function H ( j ) . The Paley-wiener criterion states that a necessary and sufficient
condition for an amplitude function H ( j ) to be realizable (causal) is that
 log H ( j )

 1 2
d   (3.3)
The following conditions must be satisfied before the Paley-Wiener criterion is valid:
1. h(t ) must possess a Fourier transform H ( j ) .
2. The square magnitude function H ( j ) must be integrable that is

2

 H ( j ) d  
2

3.4
The physical implication of the Paley-Wiener criterion is that the amplitude H ( j ) of a realizable
network must be zero over a finite band of frequencies. Another way of looking at the Paley-wiener
criterion is that the amplitude function cannot fall off to zero faster than exponential order. For
example, the ideal low pass filter in fig. 3.2 is not realizable because beyond  C the amplitude is
drops to zero.
Fig. 3.2: Ideal filter characteristic
The Gaussian shaped curve H ( j )  e  shown by fig. 3.3 is not realizable because
2
 2
log H ( j )   and the integral
2
 1   2 d is not finite.
23
Fig 3.3: Gaussian Filter charactersitc

On the other hand, the integral, the amplitude function
H ( j ) 
1
3.5
1 2
does represent a realizable network.
In fact the voltage ratio transfer function of the R-C network in the figure below has an amplitude
characteristic given H ( j ) by equation (3.5).
For the ideal filter, the inverse transforms h(t ) has the form
A0 sin C t
h(t )  3.6
t
sin x
where A0 is constant. From the curve, we can observe that h(t ) is nonzero for t less than zero.
x
In fact, in order to make h(t ) causal, it must be delayed by an infinite amount. In practice, however,
if we delay h(t ) by a large but finite amount of t d such that for t  0 the magnitude of h(t  t d ) is
less than a very small quantity  , that is h(t  t d )   t  0 .
We then can approximate h(t  t d ) by a causal response h1 (t ) which is zero for t  0 . For a more
detail discussion of the Paley-Wiener criterion, then you have to refer to other materials focusing to
24
this issue. If a network is stable, then for a bounded excitation e(t ) the response r (t ) is also bounded.
In other words, if e(t )  C1 for 0  t   then r (t )  C 2 for 0  t   where C1 and C 2 are real,
positive, finite quantities. If a linear system is stable, then from the convolution integral we obtain

r (t )  C1  h( ) d  C2 3.7
0
Equation 3.7  requires that the impulse response be absolutely integrable, or


0
h( ) d   3.8
One important requirement for h(t ) to be absolutely integrable is that the impulse response
approach zero as t approaches infinity, that is,
lim h(t )  0
t 
3.9
Generally, it can be said that with the exception of isolated impulses, the impulse response must be
bounded for all t, i.e., h(t )  C for all t where C is real, positive, finte numer.
Observe that our defination of stability precludes such terms as sin 0 t from the impulse response
because sin 0 t is not absolutely integrable. These undammed sinusoidal terms are associated with
simple poles on the j axis. Since pure L  C network have system functions with simple poles on
the j - axis, and since we do not wish to all these networks unstable, we say that a system is
marginally stable if its impulse response is bounded according to 3.8 , but does not approach zero
as t approaches infinity.
In the frequency domain, the stability criterion requires that the system function possess poles in the
left-half plane or on the j axis only. Moreover, the poles on the j axis, if H (s) is given as
a n s n  a n1 s n1  a n2 s n2  ....  a1 s  a0

H ( s)  3.10
bm s m  bm1 s m1  bm2 s m2  ....  b1 s  b0
then the order of the numerator n cannot exceed the order of the denominator m by more than unity,
that is n  m  1. If n exceeded m by more than unity, this would imply that s  j   , and there
25
would be a multiple pole. To summarize, in order for a network to be stable, the following three
conditions on its system function H (s) must be satisfied:
1. H (s) cannot have poles in the right-half plane.
2. H (s) cannot have multiple poles on the j  axis.
3. The degree of numerator of H (s) cannot exceed the degree of the denominator by more than
unity.
Finally, it should be pointed out that a rational function H (s) with poles in the left half plane only
has an inverse transform h(t ) , which is zero for t  0 . In this respect, stability implies causality. Since
system functions of passive linear networks with lumped elements are rational functions with poles
in the left-half plane or j axis only, causality ceases to be a problem when we deal with the system
functions of this type. We are only concerned with the problem of causality when we have to design
a filter for given amplitude characteristic such as the ideal filter. We know we could never hope to
realize exactly a filter of this type because the impulse response would not be causal.
3.2. Hurwitz Polynomials

In section 3.1 we saw that in order for a system function be stable, its poles must be restricted to the
left-half plane or on the j -axis. Moreover, the pole on the j -axis must be simple. The
denominator polynomial of the system function belongs to a class of polynomials known as Hurwitz
polynomials. A polynomial P(s) is said o be Hurwitz if the following conditions are satisfied:
a. P(s) is real when s is real
b. The roots of P(s) have real parts which are zero or negative.
As a result of this conditions, if P(s) is a Hurwitz polynomial given by
P(s)  an s n  an1s n1      a1s  a0 3.11

then all the coefficients a i must be real and if s i   i  j is the root of P(s) , then  i must be
negative.
26
The polynomial P(s)  (s  1)(s  1  j 2 )(s  1  j 2 ) is Hurwitz because all of its roots
have negative real parts. On the other hand, G(s)  (s  1)(s  2)(s  3) is not Hurwitz
because of the root s=1, which has positive real part.
Hurwitz polynomials have the following properties:
i. All the coefficients ai are non negative. This is readily seen by examining the three types
of roots that a Hurwitz polynomial might have. These are s   i , s   ji and Real,
s   i  ji where  i ,  i & i real and positive.
The polynomial P(s) which contains these roots can be written as

P(s)  (s   i )(s 2  i2 )[(s   i ) 2  i2 ]    3.12
Since P(s) is the product of terms with only positive coefficients, it follows that the
coefficients of P(s) must be positive. A corollary is that between the highest order
terms in s and the lowest order term, none of the coefficients may be zero unless the
polynomial is even or odd. In other words, an1 , an2 ,  , a2 , a1 must not be zero if the
polynomial is neither even nor odd. This is readily seen because the absence of term ai
implies cancellation brought about by a root s   i with a positive real part.

ii. Both the odd and even parts of a Hurwitz polynomial P(s) have roots on the j -axis
only. If we denote the odd part of P(s) as n(s) and the even part as m(s) , so that
P(s)  n(s)  m(s) then m(s) and n(s) both have roots on the j -axis only.
Refer books for the proof.
iii. As a result of property (ii) above, if P(s) is either even or odd, all its roots are on the
j -axis.
The continued fraction expansion of the ratio of the odd to even parts or the even to odd parts of a
Hurwitz polynomial yields all positive quotient terms. Suppose we denote the ratios as
27
 (s)  n(s) / m(s) or  (s)  m(s) / n(s) , then the continued fraction expansion of  (s)
can be written as
 s   q1s 
1
1
q2 s 
1 3.13
q3 s 
1
q4 s 
.... 1

qn s
where the quotients q1, q2 , ..., qn must be positive if the polynomial P(s)  n(s)  m(s) is
Hurwitz.
To obtain the continued fraction expansion, we must perform a series of long divisions. Suppose
 s  is  s  
m( s )
where m(s) is one higher degree than n(s) . Then if we divide n(s) into m(s) , we
n( s )
obtain a single quotient and a remainder
R1 s 
 ( s)  q1s  3.14
ns 
The degree of the term R1 s  is one lower than the degree of ns  . Therefore if we invert the
remainder term and divide, we have
ns  R s 
 q2 s  2 3.15
R1 s  R1 s 
Inverting and dividing again, we obtain
R1 s  R s 
 q3 s  3 3.16
R2 s  R2 s 
We see that the process of obtaining the continued fraction expansion of  s  simply involves
division and inversion. At each step we obtain quotient term qi s and a remainder term,
Ri1 (s) / Ri (s) . We then invert the remainder term and divide Ri1 ( s) into Ri (s) to obtain a new
28
quotient. There is a theorem in the theory of continued fractions which states that the continued
fraction expansion of the even to odd or odd to even parts of a polynomial must be finite in length.
Another theorem states that, if the continued fraction expansion of the odd to even or even to odd
parts of polynomial yields positive quotient terms, then the polynomial must be Hurwitz to within a
multiplicative factor W (s) . That is, if we write
F (s)  W (s) F1 (s) 3.17

Then F (s) is Hurwitz, if W (s) and F1 ( s) are Hurwitz.
For example, let us test whether the polynomial F (s)  s 4  s 3  5s 2  3s  4 is Hurwitz or not.
The even and the odd parts of F (s) are m(s)  s 4  5s 2  4 and n(s)  s 3  3s .
We now perform a continued fraction expansion of  (s)  m(s) / n(s) by dividing n(s) by m(s) , and
then inverting and dividing again, as given by the operation
s 3  3s s 4  5 s 2  4 ( s
s 4  3s 2
2 s 2  4 s 3  3s ( s / 2
s 3  2s
s 2s 2  4 (2s
2s 2
4) s ( s / 4
s
So that the continued fraction expansion of  (s) is

m( s ) 1
 ( s)  s
n( s ) s 1

2 1
2s 
s/4
Since all the quotient terms of the continued fraction expansion are positive, F (s) is Hurwitz.
Example 1: Let us test whether the polynomial G(s)  s  2s  3s  6 is Hurwitz. The continued
3 2
fraction expansion of n(s) / m(s) is obtained from the division
29
2 s 2  6 s 3  3s s / 2
s 3  3s
0
We see that the division has been terminated abruptly by a common factor s 3  3s . The polynomial
can then be written as
G( s)  s 3  3s 1  
 2
 s
We know that the term 1 2 / s is Hurwitz. Since the multiplicative factor s 3  3s is also Hurwitz,
then G(s) is Hurwitz. The term s 3  3s is the multiplicative factor W (s) , which we referred to earlier.
Example 2: Next consider a case where W (s) is non-Hurwitz
S 7  2s 6  2s 5  s 4  4s 3  8s 2  8s  4
The continued fraction expansion of F (s) is now obtained.
n( s ) s 1
 
m( s ) 2 4 1
s
3 3 4
2
s s 4  
s4  4  
We thus see that W ( s)  s  4 , which can be factored into
4
 
W ( s )  s 2  2 s  2 s 2  2s  2 . It is clear that F (s) is not Hurwitz since W(s) is not
Hurwitz.
Example 3: Let us consider a more obvious non-Hurwitz polynomial
F (s)  s 4  s 3  2s 2  3s  2
The continued fraction expansion is
30
s 3  3s s 4  2 s 2  2 s
s 4  3s 2
 s 2  2 s 3  3s  s
s 3  2s
5s   s 2  2 s / 5
 s2
2 5s
5
s
2
5s
We see that F (s) is not Hurwitz because of the negative quotients.

Example 4: Consider the case where F (s) is an odd or even function. It is 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) . If the 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
F (s)  s 7  3s 5  2s 3  s Then F ' ( s) is F ' (s)  7s 6  15s 4  6s 2  1 without going

into the details, it can be shown that the continued fraction expansion of F (s) / F ' (s) does not yield
all positive quotients. Therefore F (s) is not Hurwitz.
3.3. Positive Real Functions
In this section we will study the properties of a class of functions known as positive real functions.
These functions are important because they represent physically realizable passive driving-point
immittances. A function F (s) is positive real (p.r.) if the following conditions are satisfied:
i. F (s) is real for real s ; that is, F ( ) is real.
ii. 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 (s)]  0 , for Re s  0 .
Let us consider a complex plane interpretation of a p.r. function. Consider the s -plane and F (s)
plane in Fig. 3.4. If F (s) is p.r., then a point  0 on the positive real axis of the s -plane would
correspond to, or map into, a point F ( 0 ) which must be on the positive real axis of the F (s) plane.
31
In addition, a point s i in the right half of the s plane would map onto a point F ( si ) 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 right half of the F (s) -plane. The real axis of the s plane maps onto the real axis of the F (s)
plane.
Fig. 3.4: Mapping of s -plane onto F (s) -plane

A further restriction we will impose is that F (s) be rational. Consider the following examples of p.r.
functions:
i. F (s)  Ls (where L is a real, positive number) is p.r. by definition. If F (s) is an impedance
function, then L is an inductance.
ii. F (s)  R (where R is a real and positive) is p.r. by definition. If F (s) is an impedance
function, then R is a resistance.
iii. F (s)  K / s (where K is a real and positive) is p.r. because, when s is real, F (s) is real. In
addition, when the real part of s is greater than zero i.e., Re( s)    0 , then
K K
Re   2  0 . Therefore, F (s) is p.r. If F (s) is an impedance function, then the
 s   
2
corresponding element is a capacitor of 1/K farads.
We thus see that the basic passive impedances are p.r. functions. Similarly, it is clear that the
admittances are positive real.
We now show that all driving point immittances of passive networks must be p.r. The proof depends
upon the following assertion: for a sinusoidal input, the average power dissipated by a passive
32
network is nonnegative. For the passive network in the fig. 3.5, the average power dissipated by the
1
Re[ Z in ( j )] I  0 . We then conclude that, for any passive network
2
network is
2
Re[ Z in ( j )]  0 3.18
We can now prove that for Re s    0 Re Z in (  j )  0 . Consider the network in fig. 3.5,
whose driving point impedance is Z in (s) . Let us load the network with incidental dissipation such
that if the driving-point impedance of the uniformly loaded network is Z i (s) , then
Zi (s)  Zin (s   ) 3.19

where  , the dissipation constant, is real and positive.
Fig. 3.5: Passive one-port network
Since Z i (s) is the impedance of a passive network, Re Zi ( j)  0 So that:
Re Z in (  j )  0
Since  is an arbitrary real positive quantity, it can be taken to be  . Thus the theorem is proved.
Next let us consider some useful properties of p.r. functions.
1
1. If F (s ) is p.r. then is also p.r. This property implies that if a driving point impedance is
F (s)
p.r., then its reciprocal, the driving point admittance is also p.r.
2. The sum of p.r. function is p.r. from an impedance standpoint; we see that if two impedances
are connected in series, the sum of the 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
p.r.; for example, F (s)  s  1 / s is not p.r.
33
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 s plane.
4. Only simple poles with real positive residues can exist on the j -axis.
5. The poles and zeros of a p.r. function are real or occur in conjugate pairs. We know that the
poles and zeros of a network function are functions of the elements in the network. Since the
elements themselves are real, there cannot be complex poles or zeros without conjugates
because this would imply imaginary elements.
6. The highest powers of the numerator and denominator polynomials may differ at most by
unity. This condition prohibits multiple poles and zeros at s   .
7. The lowest powers of the denominator and numerator polynomials may differ at most by
unity. This condition prohibits the possibility of multiple poles or zeros at s  0 .
8. The necessary and sufficient conditions for a rational function with real coefficients F (s) to
be p.r. are
a. F (s) must have no poles in the right-half plane.
b. F (s) may have only simple poles on the j -axis with real and positive residues.
c. Re[ F ( j )]  0 for all  .
Let us compare this new definition with the original one which requires the two conditions:
i. F (s) is real when s is real.
ii. Re[ F (s)]  0 when Re[ s]  0
In order to test condition ii  of the original definition, we must test every single point in the right-
half plane. In the alternate definition, condition (c) merely requires that we test the behavior of F (s)
along the j axis. It is apparent that testing a function for the three conditions given by the alternate
definition represents a considerable saving of effort, except in simple cases as F (s)  1 / s .
34
Let us examine the implications of each criterion of the second definition. Condition a  requires that
we test the denominator of F (s) for roots in 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
expansion of the odd to even or even to odd parts of the denominator. The second requirements-
condition b  is tested by making a partial fraction expansion of F (s) and checking whether the
residues of the poles on the j -axis are positive and real. Thus, if F (s) has a pair of poles at
*
K1 K1
s   j , a partial fraction expansion gives terms of the form  .
s  j1 s  j1
The residues of complex conjugate poles are themselves conjugates. If the residues are real-as they
*
K1 K1 K s
must be in order for F (s) to be p.r. then K1  K1 so that
*   2 2 2.
s  j1 s  j1 s  1
If K1 is found to be positive, then F (s) satisfies the second of the three conditions.
In order to test for the third condition for positive realness, we must first find the real part of F ( j )
from the original function F (s) . To do this, let us consider a function F (s) given as a quotient of two
polynomials
F ( s) 
P( s )
3.20
Q( s )
We can separate the even parts from the odd parts of P(s) and Q(s) so that F (s) is
M 1 ( s)  N1 ( s)
F ( s)  3.21
M 2 ( s)  N 2 ( s)
Where M 1 ( s) is an even function and N1 ( s) is an odd function. F (s) is now decomposed into its
even and odd parts by multiplying both P(s) and Q(s) by M 2  N 2 so that
M 1  N 1 M 1  N1 M 1 M 2  N 1 N 2 M 2 N 2  M 1 N 2
F ( s)    3.22
M 2  N2 M 2  N2 M 2  N2
2 2
M 2  N2
2 2
We see that the products M 1 M 2 and N 1 N 2 are even functions, while M 1 N 2 and M 2 N1 are odd
functions. Therefore, the even parts of F (s) is
M 1 M 2  N1 N 2
EvF (s)  3.23
M 2  N2
2 2
35
And the odd part of F (s) is

M 2 N2  M1N2
Odd F (s)  3.24
M 2  N2
2 2
If we let s  j , we see that the even part of any polynomial is real, while the odd part of the
polynomial is imaginary, so that if F ( j ) is written as F ( j)  ReF ( j)  j Im[ F ( j)] , it is clear
that Re[ F ( j )]  Ev[ F (s)] s  j and j Im[ F ( j )]  Odd[ F (s)] s  j .
Therefore, to test for the third condition for positive realness, we determine the real part of F ( j )
by finding the even part of F (s) and then letting s  j . We then check to see whether
Re F ( j)  0 for all  .
The denominator of Re[ F ( j )] is always a positive quantity because
M 2  j   N 2  j   M 2    N 2    0 .
2 2 2 2
That is, there is an extra j or imaginary term in N 2  j  , which, when squared, gives -1, so that the
denominator of Re F  j  is the sum of two squared numbers and is always positive. Therefore, our
task resolves into the problem of determining whether A()  M 1 ( j)M 2 ( j)  N1 ( j) N 2 ( j)  0
Note that A( ) must not have positive real roots of the type shown in fig. 3.5(a); i.e., A( ) must
never have single real roots of  . However, A( ) may have double roots as shown in fig. 3.5(b),
because A( ) need not become negative in this case.
(a) b)
Fig. 3.5 A( ) can have a double root
As an example, consider the requirements for
sa
F ( s) 
s  bs  c
2
36
To be p.r. first, we know that, in order for the poles and zeros to be in the left-half plane or on the
j -axis, the coefficients a, b, c must be greater or equal to zero. Second, if b  0 then F (s) will
possess poles on the j axis we can then write F (s) as
s a
F ( s)   2
s c s c
2
We will show later that the coefficient a must also be zeros when b  0 . Let us proceed with the
third requirement, i.e., Re[ F ( j )]  0 . From the equation M 1 ( j)M 2 ( j)  N1 ( j) N 2 ( j)  0
We have a( 2  c)  b 2  0 which simplifies to A( j )  (b  a) 2  ac  0 .

It is evident that in order to prevent A( ) from having positive real roots of  , b must be greater than
or equal to a , that is b  a . As a result, if b  0 , then a  0 . To summarize, the conditions that must
be fulfilled in order for F (s) to be real are
1. a, b, c  0
2. b  a
s2 s 1 s4
We see that F ( s)  is p.r., while the functions F ( s)  2 & F ( s)  are
s  3s  2
2
s 2 s  2s  1
2
not p.r.
As a second example, let us determine the conditions for the biquadrate function
s 2  a1 s  a0
F ( s) 
s 2  b1 s  b0
To be p.r. we will assume that the coefficients a1 , a0 , b1 , b0 are all real positive constants. Let us test
whether F (s) is p.r. by testing each requirement of the second definition.
First, if the coefficients of the denominator b1 and b0 are positive, the denominator must be Hurwitz.
Second, if b1 is positive, we have no poles on the j -axis. Therefore we can ignore the second
condition.
The third condition can be checked by first finding the even part of F (s) which is
( s 2  a0 )(s 2  b0 )  a1b1 s 2 s 4  [(a0  b0 )  a1b1 ]s 2  a0 b0

Ev[ F ( s)]  
( s 2  b0 ) 2  b1 s 2
2
s 2
 b0 
2
 b1 s 2
2
The real part of F ( j ) is then

37
 4  [(a0  b0 )  a1b1 ] 2  a0 b0
Re[ F ( j )] 
( 2  b0 ) 2  b1  2
2
We see that denominator of Re[ F ( j )] is truly always positive so it remains for us to determine
whether the numerator of Re[ F ( j )] ever goes negative. Factoring the numerator, we obtain
(a0  b0 )  a1b1 1
2 1, 2   [(a0  b0 )  a1b1 ]2  4a0 b0
2 2
There are two situations in which Re[ F ( j )] does not have a simple real root.
1. When the quantity under the radical sign of equation above is zero (double, real root) or
negative (complex roots). In other words,
[(a0  b0 )  a1b1 ]2  4a0 b0  0 Or [(a0  b0 )  a1b1 ]2  4a0 b0
If (a0  b0 )  a1b1  0 , then (a0  b0 )  a1b1  2 a0 b0 or a1b1  ( a0  b0 ) 2
If (a0  b0 )  a1b1  0 , then a1b1  (a0  b0 )  2 a0 b0
but (a0  b0 )  a1b1  0  a1b1  (a0  b0 )
So again a1b1   a0  b0 
2
2. The second situation in which Re[ F ( j )] does not have a simple real root is when  2 1, 2 in
equation above is negative so that the roots are imaginary. This situation occurs when
[(a0  b0 )  a1b1 ]2  4a0 b0  0 and (a0  b0 )  a1b1  0
From the above equation we have

a1b1  (a0  b0 )  2 a0 b0  (a0  b0 )  a1b1 Thus a1b1  ( a0  b0 ) 2
We thus see some of the above equation is a necessary and sufficient condition for a biquadrate
function to be positive real. If we have a1b1  ( a0  b0 ) 2 , then we will have double zeros for
ReF ( j ).
s 2  2s  25
Consider the following example; F ( s)  we see that
s 2  5s  16
a1b1  2  5  10  ( a0  b0 ) 2  ( 25  16 ) 2  1
So that F (s) is p.r. The examples just given are, of course, special cases. But they do illustrate the
procedure by which functions are tested for the p.r. property. Let us consider a number of other
38
helpful points by which a function might be tested quickly. First, if F (s) has poles on the j -axis, a
partial fraction expansion will show if the residues of these poles are positive and real.
3s  5
For example, F ( s)  has a pair of poles at s   j1 . The partial fraction expansion of

s s2 1 
 2s 5
F (s) , F ( s)  
s2 1 s
shows that the residue of the poles at s   j1 is negative. Therefore F (s) is not p.r.
Since impedances and admittances of passive time-invariant networks are p.r. functions, we can
make use of our knowledge of impedances connected in series or parallel in our testing for the p.r.
property. For example, if Z1 ( s) and Z 2 ( s) are passive impedances, then Z 1 connected in parallel
with Z 2 gives an overall impedance
Z ( s) 
Z1 ( s) Z 2 ( s)
3.25
Z1 ( s)  Z 2 ( s)
Since the connecting of the two impedances in parallel has not affected the passivity of the network,
we know that Z (s) must also be p.r. We see that if F1 ( s) and F2 ( s) are p.r. functions, then
F1 ( s) F2 ( s)
F ( s)  must also be p.r. Consequently, the functions
F1 ( s)  F2 ( s)
Ks K
F ( s)  and F ( s)  where  and K are real and positive quantities, must be p.r. We
s  s 
then observe that functions of the type
s s 
F ( s)     ,   0 must be p.r. also.
s  s  s 
Ks
Finally, let us determine whether F ( s)   , K  0 is p.r.
s 
2
If we write F (s) as
1
F ( s) 
s / K   / Ks
we see that the terms s / K and  / Ks are p.r. Therefore, the sum of the two terms must be p.r. Since
the reciprocal of a p.r. function is also p.r. we conclude that F (s) is p.r.
39
Tutorial Exercise
1. Test the following polynomials for the Hurwitz property.
a. s 3  s 2  2s  2 b. s4  s2  s 1
c. s7  s5  s3  s d. s 3  4s 2  5s  2
e. s 5  2s 3  s f. s  2s  2s  s  4s  8s  8s  4
7 6 5 4 3 2
2. Determine whether the following functions are p.r. For the following with the denominator
already factored, perform a partial fraction expansion first.
a. F ( s) 
s2 1
c. F ( s) 
s  2s  4
s 3  4s ( s  1)( s  3)
2s 2  2s  4 s2  4 5s 2  s
b. F ( s)  d. F ( s)  e. F ( s) 

s  1 s 2  2  s 3  3s 2  3s  1 s2 1
s 2  Xs
3. Given Z ( s)  2 ;
s  5s  4
a. What are the restrictions on X for Z (s) to be a p.r. function?
b. Find X for ReZ ( j ) to have a second-order zero at   0 .
c. Choose a numerical value for X and synthesize Z (s) .
2s 2  s  2
4. Z ( s)  is p.r. determine min Re Z ( j) and synthesize Z (s) by first removing
s2  s 1
min Re Z ( j) .
5. Perform a continued fraction expansion on the ratio
s 3  2s 2  3s  2
Z ( s) 
s 3  s 2  2s  1
What does the continued fraction expansion if Y (s) is the driving-point admittance of a
passive network? Draw the network from the continued fraction.
40
6. The following functions are impedance functions. Synthesize the impedances by successive
removals of j axis poles or by removing min Re Z ( j) .
s 3  4s s 1
a. Z ( s)  b. Z ( s) 
s2  2 s( s  2)
2s  4 s 2  3s  1
b. . Z ( s)  d. Z ( s) 
2s  3 s2 1
7. Check whether the following equations are p.r. functions or not.
s 2  10s  4
a. F ( s) 
s2
s 3  5s 2  9 s  3
b. F ( s) 
s 3  4s 2  7 s  9
41
3.4. Elementary Synthesis procedures

The basic philosophy behind the synthesis of driving point functions is to break up a p.r. function
Z (s) into a sum of simpler p.r. functions Z1 ( s), Z 2 (s),  , Z n (s) and then to synthesize these
individual Z i (s) as elements of the overall network whose driving-point impedance is Z (s) .
Z ( s )  Z1 ( s )  Z 2 ( s )      Z n ( s )
First, consider the “breaking-up” process of the function Z (s) into the sum of functions Z i (s) . One
important restriction is that all Z i (s) must be p.r. Certainly, if all Z i (s) were given to us, we could
synthesized a network whose driving-point impedance is Z (s) by simply connecting all the Z i (s) in
series. How if we were to start with Z (s) to give us the individual Z i (s) ? Suppose Z (s) is given in
general as
an s n  an1 s n1      a1 s  a0
Z ( s) 
bm s m  bm1 s m1      b1 s  b0
Consider the case where Z (s) has a pole at s  0 (that is, b0  0 ). Let us divide P(s) and Q(s) to
give a quotient D / s and a remainder R(s) , which we can denote as Z1 ( s) and Z 2 ( s) .

D
Z ( s)   R( s ) 0
s
 Z1 ( s)  Z 2 ( s)
Are Z 1 and Z 2 p.r.? From previous discussions, we know that Z1  D / s is p.r. Is Z 2 ( 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 residues must be real and
positive.
3. Re[ Z 2 ( j)]  0 for all 
Let us examine these cases one by one. Criterion 1 is satisfied because the poles of Z 2 ( s) are poles of
Z (s) . Criterion 2 is satisfied by this same argument. A simple partial fraction expansion does not
affect the residues of the other poles. When s  j , ReZ ( j)  D / j   0 . Therefore, we have
Re Z 2 ( j)  Z ( j)  0
42
From the foregoing discussion, it is seen that if Z (s) has a pole at s  0 , a partial fraction expansion
can be made such that one of the terms is of the form K / s and the other terms combined still remain
p.r. A similar argument shows that if Z (s) has a pole at s   (that is, 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
Z1 ( s) and Z 2 ( s) . Then Z (s)  Ls  R(s)  Z1 (s)  Z 2 (s) . Here Z 2 ( s) is also p.r.
If Z (s) has a pair of conjugate imaginary poles on the imaginary axis, for example, poles at s   j1 ,
then Z (s) can be expanded into partial fractions so that
2 Ks 2 Ks  j 2K 
Z ( s)   Z 2 ( s) Here Re( 2 )  Re   0 So that Z 2 ( s) is p.r.
2 
s  1
2 2
s  1 2
    1 
2
Finally, if ReZ ( j ) is minimum at some point  i , and ReZ ( ji )  K i as shown in fig.3.6, we can
remove a constant K  K i from ReZ ( j ) so that the remainder is still p.r. This is because
ReZ ( j ) will still be greater than or equal to zero for all values of  .
Fig.3.6
43
CHAPTER 4
4. Synthesis of One-Port Networks with Two Kinds Of Element
4.1. Introduction
In this chapter we will study methods for synthesizing one-port network with two kinds of elements.
Since we have three elements to choose from, the networks to be synthesized is either
R  C, R  L or L  C network. We first discuss the properties of a particular type of one-port
network and then we will synthesize it.
A single-port network is a load on the source that excites it. As we have discussed already, the ratio
of source voltage to current drawn by the network is called driving-point impedance, and equation
below expresses how the driving-point impedance is obtained. The ratio of source current to source
voltage is called as the driving-point admittance. Both the driving-point impedance and the driving-
point admittance are the driving-point functions. The term, the driving-point immittance, is used to
either the driving-point impedance or the driving-point admittance.
V ( s) I ( s)
Z ( s)  Y (s) 
I ( s) V ( s)
Fig. 4.1. A single port or two terminal network

4.2. Properties and Synthesis of L-C Driving-point Immittances
Before we proceed to synthesis of L-C network, we will first examine some properties of L  C
driving point immittances.
4.1.1. Properties of L-C Immittance Functions
Consider the impedance Z s  of a passive one-port network. Let us represent Z s  as
44
M 1 s   N1 s 
z s   where M 1 , M 2 are even parts of the numerator and denominator and
M 2 s   N 2 s 
N1 , N 2 are odd parts respectively. The average power dissipated by the one–port network is
Rez  j I 
1
Average power 
2
2
where I is the input current. For a pure reactive network, it is know that the power dissipated is zero.
We therefore conclude that the real part of Z(s) is zero. That is the same as
ReZ  j   EvZ  j   0
M 1 s M 2 s   N1 s N 2 s 
where EvZ s  
M 2 s   N 2 s 
2 2
In order for EvZ  j   0  M 1  j M 2  j   N 2  j N1  j   0 , either of the following case must

hold
(a) M1  j  =0=N2  j 
(b) M2  j  =0=N1  j 
N1 ( s)
In case (a), Z s   and in case (b), Z s  
M2
.
M 2 ( s) N2
We see from this development the following two properties of L-C function
1. Z LC s  or YLC s  is the ratio of even to odd or odd to even polynomials
2. Since both M i s  and N i s  are Hurwitz, they have only imaginary roots with real parts
equals to zero and it follows that the poles and zeros of Z LC s  or YLC s  are on the j -axis.
Consider the example of an L-C immittance function given by

a 4 s 4  a 2 32  a a
Z s  
s2
b2 3  b2  b1
s s
Let us examine the constraints on the coefficients ai and bi. We know first of all that in order for the
impedance to be positive real, the coefficients must be real and positive. We also know that an
impedance function cannot have multiple poles or zeros on the j -axis. Since  is defined to be
on the j -axis the highest powers of the numerator and the denominator polynomials can differ by
at most unity. For example, if the highest order of the numerator is 2n then the highest order of the
denominator can either be 2n-1 so that there is simple zero at s   or the order can be 2n  1 so that
45
there is a simple zero at s   . Similarly, the lowest order of numerator and denominator can differ
by at most unity, or else there would be multiple poles or zeros of Z s  at s   .
Another property of the numerator and denominator polynomial is that if the highest power of the
polynomial is 2n for example the next highest order tem must be 2n  2 and the succeeding power
must differ by two orders all the way through. There cannot be any missing terms, i.e. no two
adjacent terms of either polynomial may differ by more than two power. For Z (s) given above, if
b3  0 , then Z (s) will have poles when b5 s 5  b1 s  0 so that the poles be at s  0 and at
1/ 4
b 
sk   1  e j ( 2 k 1) / 4 k  0,1,2,3
 b5 
It is clearly seen that none of the pole are even on the j -axis thus violating one of the basic
properties of an immittance function. From the properties given in the above equation, we can write a
general L-C impedance or admittance as
Z (s) 
   
K s 2  12 s 2  32    s 2  i2    
   
s s 2  22 s 2  42    s 2   2j    
Expanding into partial fraction, we obtain
K0 K s K s
Z ( s)   2 2 2  2 4 2      K s
s s  2 s  4
where K i are the residues of the poles. Since poles are all on the j -axis, the residues must be real
and positive in order for Z (s) to be positive real. Letting s  j , we see that Z  j  has zero real
part, and can thus be written as a pure reactance jX ( ) . Thus we have
 K K 
Z  j   j   0  2 2 2      K    jX  
  2  2 
Differentiating with respective to  , we have
dX   K 0
 2  K 

2 K 2  2   22 
 
d  
 22   2
2

Since all the residues K i are positive, it is seen that for an L-C function,
46
dX  
d
0
A similar development show that the derivative of ImY ( j )  B( ) is also positive. That is
dB 
d
0
Consider the following example Z(s) is given as

Ks s 2  3
2

s 
Z ( s) 
s 2
 2
2 2
 4
2
Letting s  j we obtain X ( )
Z  j   jX     j

K   2  32 
 
  2   22   2   42 
Let us draw a curve of X ( ) versus  beginning with the zero at   0, and examine the sequence
of critical frequencies encountered as  increase. Since the slope of the X ( ) curve is always
positive the net critical frequency we encounter is when X ( ) becomes infinitely large or the pole is
at  2 . As we pass  2 , X ( ) change sign and goes from +ve to -ve. In general, whenever we pass
though any critical frequency, there is always a change of sign as seen from the way jX ( ) is
written in the last equation. After we pass through  2 , with the slope of X ( ) always positive it is
easy to see that the next critical frequency is the zero at  3 . Thus, if impedance function is an L-C
imittance, the poles and zeros of the function must alternate. The particular X ( ) under discussion
takes the form shown in fig. 4.2. Since the highest power of the numerator and the denominator
always differ by unity and the lowest power also differ by one we observe that at s  0 and at s  
there is always a critical frequency whether a zero or a pole.
For the example just discussed there is a zero at s  0 and a zero at s   the critical frequencies at
s  0 and at s   are called external critical frequencies whereas the remaining unite critical
frequencies are referred to as internal thus in the previous example 1 , 2 & 4 , and internal critical
frequencies.
47
Fig. 4.2. Poles and zeros of an L-C immittance function alternate
Finally let us summarize the properties of L-C impedance or admittance functions.

1. Z LC (s) or YLC (s) is the ratio of odd on even or even to odd polynomials.
2. The pole and zeros are simple and lie on the axis
3. The pole and zero interlace on the axis
4. The highest power of numerator and denominator differ by unity; the lowest power also differ
by unity
5. There must be either a zero or a pole at the origin and infinity.
Exercises: Check whether the following functions are L-C or not. If not give the reason.
1. z ( s) 

Ks s 2  4  2. Z ( s) 
s 2  4 s 2  5s
s2  1s2  3 3s 2  6s 2
3. Z ( s) 

K s2 1 s2  9   4. Z ( s) 
 
2 s2 1 s2  9 
 
s 2  2 s 2  10 
s s2  4 
4.1.2. Synthesis of L-C Immittance Functions
We saw in the above section that an L-C immittance is a positive real function with poles and zeros
on the axis only. The partial fraction expansion of an L-C function is expressed in general terms as
K0 Ks K s Ks
F ( s)   2 1 2  2 2 2 .    2 i 2  K s
s s  1 s  2 s  i
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
K 0 / s represents a capacitor of 1 / K 0 farads: the term K  s is an inductor of K  henrys, and the
Ki s
term is a parallel tank circuit that consists of a capacitor of 1 / K i farads in parallel with an
s  i
2 2
inductor of K i / i2 . Thus a partial fraction expansion of general L-C impedance would yield the
48
network shown in fig. 4.3. This method of synthesis that is based on partial fraction expansion is
called Foster synthesis.
Fig. 4.3. Foster Synthesis of an L-C impedance function
For example, consider the following L-C function.
Z s  
 
2 s 2  1 s 2  9

s s2  4
A partial fraction expansion of Z(s) gives
15
s
Z s   2s 
9/2 2
 2
s s 4
We then obtain the synthesized network shown in fig. 4.4.
Fig. 4.4.
The partial fraction expansion method is based upon the elementary synthesis procedure of removing
poles on the j axis. The advantage with L-C functions is that all the poles of the function lie on the
j -axis so that we can remove all the poles simultaneously. Then the partial fraction expansion of
Y(s) which is given by the equation below gives us a circuit consisting of parallel branches as shown
in fig. 4.5. This is Foster II synthesis method of an LC network.
K0 K s K s
Y ( s)   2 2 2  2 4 2      K s
s s   2 s  4
49
Fig. 4.5. Foster II synthesis
For example consider the admittance function Y s  

 
s s 2  2 s2  4  .
s 2  1s 2  3
The partial fraction expansion of Y(s) is
1 3
s s
Y ( s)  s  2
 22
s  3 s 1
2
from which we synthesize the network shown in fig. 4.6.
Fig. 4.6
The L.C networks synthesized by a partial fraction expansions are, as mentioned above, sometimes
called Faster–type networks the impedance from is sometimes called a foster series networks(Foster
I) and the admittance form is a foster parallel network (Foster II).
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 zero or pole at s   . Suppose we consider the case of
an L-C impedance Z(s) whose numerator is of degree 2n and denominator is of degree 2n-1, giving
Z(s) a pole at s   . We can remove this pole by removing impedance L1s so that the remainder
function Z2(s) is still L-C.
Z 2 (s)  Z (s)  L1 s
The degree of the denominator of Z 2 ( s) is 2n  1but the numerator is of degree 2n-2 because the
numerator and denominator must differ in degree by 1. Therefore we see that Z 2 ( s) has a zero at
50
s   . If we invert Z 2 ( s) to give Y2 (s)  1 / Z 2 (s) , then Y2 ( s) will have a pole at s   which we

can again remove to give a capacitor C2s and a remainder Y3(s), which is
Y3 (s)  Y2 (s)  C2 s.
We readily see that Y3 ( s) has a zero at s   , which we can invert and remove. This process
continues until the remainder is zero. Each time we remove a pole, we remove an inductor or a
capacitor depending upon whether the function is impedance or admittance. Note that the final
structure of the network synthesized is a ladder whose series arms are inductors and whose shunt
arms are capacitors, as shown in fig. 4.7.
Fig. 4.7 Cauer Synthesis of an L-C impedance function

Consider the following example,
2s 5  12s 3  16s
Z ( s) 
s 4  4s 2  3
We see that Z (s) has a pole at s   , which we can remove by first dividing the denominator into the
numerator to give a quotient 2s and a remainder Z 2 ( s) , as shown in fig. 4.8.
Fig. 4.8
Then we have
4s 3  10s
Z 2 ( s)  Z ( s)  2s  4
s  4s 2  3
51
Observe that Z 2 ( s) has a zero at s   . Inverting Z 2 ( s), we again remove the pole at infinity. Then
1
we realize a capacitor of farad and remainder Y3 ( s) , as may be seen in fig. 4.9.
4
3 2
s 3
1 2
Y3 ( s)  Y2 ( s)  s  2
4 4s  10s
8
Removing the pole at s   of Z 3 ( s)  1 / Y3 ( s), gives a series inductor of h and
3
8 2s
Z 4 ( s)  Z 3 ( s)  s  as shown in fig. 4.10.
3 3 2
s 3
2
Fig. 4.9 Fig. 4.10

3
The admittance Y4 (s)  1 / Z 4 (s) has a pole at s   , which we remove to give a capacitor of farad
4
2
and a remainder Y5 (s)  3 / 2s, which represents an inductor of h . Removing this inductor gives us
3
zero remainder. Our synthesis is therefore complete and the final network is shown in fig. 4.11.
Fig. 4.11
52
Since we always remove a pole at s   by inverting the remainder and dividing, we conclude that
we can synthesis an L-C ladder network by a continued fraction expansion. The quotients represent
the poles at s   , which we remove, and we invert the remainder successively until the remainder is
zero. For the previous example, the continued fraction expansion is
s 4  4 S 2  3 2 s 5  12 s 3  16 s 2 s  Z
2 s 5  8s 3  6 s
1
4s 3  10 s ) s 4  4 s 2  3 s  Y
4
5
s4  s2
2
3 2  8
s  3 4 s 3  10 s s  Z
2  3
4 s 3  8s
3
2 s  s 2  3 s  Y
3
2 4
3 2
s
2
2
3 2 s s  Z
3
2s
We see that the quotients of the continued fraction expansion give the elements of the ladder
network. Because the continued fraction expansion always invites each remainder and divide, the
successive quotients alternate between Z and Y and then again Z, as shown in the proceeding
expansion. If the initial function is impedance, the first quotient must necessarily be impedance.
When the first function is an admittance the first quotient is an admittance.
Since the lowest degrees of numerator and denominator of an L-C admittance must differ by unity, it
follows that there must be a zero or a 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 realization in a ladder
structure. To do this by continued fraction, we arrange both numerator and denominator in ascending
order and divide the lowest power of the denominator into the lowest power of the numerator: then
we invert the remainder and divide again.
53
For example in the case of the impedance we have
Z s  
s 2
 
 1 s 2  3 3  4s 2  s 4


s s2  2  2s  s 3
The continued fraction expansion to give the alternate realization is
 3
2s  s 3 3  4s 2  s 4   z
 2s
3
3  s2
2
5 2   4
s  s 4 2 s  s 3   Y
2   5s
4
2s  s 3
5
1 35 2  25
s  s  s4 Z
5 2  2s
5 2
s
2
 1
s4  15 s 3
   Y
  5s
1 3
s
5
The final synthesized network is shown in fig. 4.12. The ladder networks realized are called cauer
ladder network because W. Cauer discovered the continued fraction method for synthesis of a
passive network.
Fig. 4.12
Note that for both the foster and the Cauer-form realizations the number of element is one greater
than the number of intern critical frequents which we defined previously as being all the poles and
zeros of the function excluding those at s  0 and s   . Without going into the proof of the
54
statement, it can be said that both the foster and the Cauer forms give the minimum number of
elements for a specified LC driving .point function. These realizations are sometimes known as
canonical forms.
4.3. Properties and Synthesis of R-C Impedances and R-L Admittances

As we have done for synthesis of impedance functions of L-C networks, we will first discuss
properties of R-C driving point impedance and then we proceed to its synthesis.
4.3.1. Properties of R-C Impedances

The properties of R-C driving point impedances can be derived from known properties of L-C
functions by a process of mapping the j -axis on to the   -axis. Here we will assume that all
driving point function that can be realized with two kinds of elements can be realized in a foster
form. Based upon this assumption we can derive all the pertinent properties or R-C or R-L driving
point functions.
Let us consider first the properties or R-C driving point impedance functions.
Referring to the series foster form for an L-C impedance given in the fig. 4.3, we can obtain a foster
realization of an R-C impedance by simply replacing all the inductances by resistance so that a
general R-C impedance could be represented as shown in Fig. 4.13.
Fig. 4.13 Foster realization of an R-C Impedance function
K0 Ki
Z s  
K1 K2
 K    ... 
s s  1 s   2 s i
Ki
where C0  1 , R  K , Ci  1 , Ri 
K0 Ki  i.
From this development two major properties of R-C impedance are obtained and are listed in the
following
55
1. The poles of R-C driving point impedance are on the negative real    -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 conclude that the zeros of an R-C impedances are also on the    -axis.
2. The residues of the poles K i are real and positive. We shall see later that this property does
not apply to R-C admittances.

Since the poles and zero of R-C impedances are on the    -axis, let us examine the slope of Z  
dZ ( )
along the -  -axis. To find the slope, , we first let s   in Z (s) and then we take the
d
derivative of Z (  ) w. r. t  . Thus we have,
k0 K2 K2
Z ( )   K  
   1    2
dZ ( )  K0  K1  K2
and     ...
d 2
   1 2
(   2 ) 2
dZ ( )
It is clear that 0
d
Let us now look at the behavior of Z(s) at the two points where the real axis and the imaginary axis
interest, namely at     0 and at      . This is readily done by examining the general R-C
network in fig. 4.13 at these two frequencies. At   0 (dc), if the capacitor Co is in the circuit, it is
an open circuit and there is a pole of Z(s) at   0 . If Co is not in the circuit, then Z (0) is simply the
sum of all the resistances in the circuit because all of the capacitors are open circuits at   0 .
. Z (0)  R1  R2  ...  R
At    all the capacitors are short circuits. Thus if R is in the circuit, Z   R . If R is
missing then Z   =0. To summarize these last two statements we have
 C 0 Pr esent

1. Z (0)   
 Ri C 0 mis sin g
 i 1
0 R mis sin g
2. Z ()  
R R Pr esent
If we examine the two cases for Z 0 and Z   , we see that Z (0)  Z  .
56
Next let us see whether the pole and zeros of an R-C impedance function alternative. We have
already established that the critical frequency nearest the origin must be a pole and the critical
frequency nearest    must be a zero. Therefore if Z(s) is given as
Z ( s) 
s  2 s   4 
s   1 s   3 
Then, if Z s  is R-C, the singularity nearest the origin must be a pole which we will assume to be at
s   1 ; the singularity furthest from the origin must be a zero, which we will take to be s   4 . Let
us plot Z ( ) 
  2    4  versus   beginning at   0 and extending to    . At
   1    3 
  0, Z (0) is equal to a positive constant.
 2 4
Z (0) 
 1 3
Since the slope of Z   is always positive as   increase, Z   must increase until the poles
s   1 is reached. At    1 , Z   change sign and is negative until the next critical frequency is
reached. We see that this next critical frequency must be the zero, s   2 . Since Z ( ) increase for
increasing   , the third critical frequency must be the pole, s   3 . Because Z ( ) change sign at
  3 , the final critical frequency must be the zero, s   4 . Beyond    4 , the curve becomes
asymptotic to Z ()  1 . From this analysis we see that the poles and zeros of R-C impedance must
 2 4
alternate. For the case being considered    4   3   2   1  0 . In addition we see that 1
 1 3
which shows that z (0)  z ()
57
Fig. 4.14. Poles and zeros alternate on   -axis for an R-C impedance
To summarize the three properties we need to recognize an R-C impedance are

1. Pole and zeros lie on the negative real axis and they alternate.
2. The singularity nearest to (or at) the origin must be a pole whereas the singularity nearest to
(or at)    must be zero.
3. The residues of the poles must be real and positive
Example: Check if the following impedance functions are R-C impedance.
1. Z ( s) 
s  1s  4s  8 3. Z s  
s  2s  4
3s  2s  6 s  1
2. Z s  
s  1s  8 4. Z s  
s  1s  2
s  2s  4 ss  3
Let us re-examine the partial fraction expansion of a general R-C impedance
F s  
K0 K1 K2 Ki
 K     
s s  1 s   2 s i
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 net
work shown in Fig. 4.15.
58
Fig. 4.15. R-C impedance function can be realized as R-L admittance
We see that an R-C impedance ZRC(s) can also be realized as R-L admittance YRL(s). All the
properties of R-C impedances it is therefore important to specify where a function is to be realized as
an R-C impedance or an R-L admittance.
4.3.2. Synthesis of R-C Impedances or R-L Admittances

We postulated in section 4.3.1 that the foster form realization exists for an R-C impedance or an R-L
admittance. Since foster network are synthesized by partial fraction expansions the synthesis is
accomplice with ease. An important point to remember is that we must remove the minimum real
part of z ( j ) in the partial fraction expansion . It can be shown that min Re Z  j   Z  so that
we have to remove Z   as a resistor in the partial fraction expansion. In cases where the numerator
is of lower degree than the denominator, Z ()  0 . When the numerator and the denominator are of
the same degree, then Z () can be obtained by dividing the denominator to numerator. The quotient
is then Z () .
Consider the following example.

3s  2s  4
F s  
3s  3
The partial fraction expansion of the remainder function is obtained as
8 1
F ( s)   3
s s3
Here F ()  3 . If F (s) is an impedance Z (s) , it must be an R-C impedance and it is realized in the
series foster form as shown in fig. 4.16.
59
Fig. 4.16.
On the other hand if F (s) represents an admittance, we realize Y (s) as an R-L network in the parallel
foster form as in fig. 4.17.
Fig. 4.17.
An alternate method of synthesis is based on the following fact. If we remove min ReZ  j   Z 
from Z (s) , we create a zero at s   for the remainder Z 2 ( s) . If we invert Z 2 ( s) , we then have a pole
at s   which we can remove to give Z 3 ( s) . Since min ReY3  j   Y3  , if we remove Y3  
we would have a zero at s   again which we again invert and remove. The process of extracting
Z () or Y () and the removal of a pole of the reciprocal of the reminder involve dividing the
numerator by the denominator. Consequently, we see that the whole synthesis process can be
resolved by a continued fraction expansion. The quotients represent the elements of a ladder network.
3s  2s  4
For example the continued fraction expansion of F s  
3s  3

s 2 3s 3s 2  18s  243  Z / Y
3s 2  9s
s
9s  24s 2  3s  Y / Z
9
60
8
s2  s
3
1 
s 9s  2427  Z / Y
3 
9s
 1
24  13 s s  Y / Z
 72
1
3s
If F(s) is impedance Z(s) the resulting network is shown in Fig. 4.18. If F(s) is admittance Y(s), we
have the R-L network of fig. 4.19.
Fig. 4.18. Fig. 4.19.
4.4. Properties and Synthesis of R-L Impedances and R-C Admittances
The immittance that represents series foster R-L impedance or a parallel foster R-C admittance is
given as.
Ki s
F s   K  s  K 0 
K1 s K2s
  ... 
s  1 s   2 s i
The significant difference between an R-C impedance and R-L impedance is that the partial fraction
expansion term for the R-C tank circuit is K i / s   i  ; whereas for the R-L impedance the
corresponding term must be multiplied by an s in order to give an R-L tank circuit consisting of a
resistor in parallel with an inductor.
The properties of R-L impedance or R-C admittance function can be derived in much the same
manner as the properties of R-C impedance functions. Without going into the derivation of the
properties, the more significant ones are given in the following
61
1. Poles and zeros of an R-L impedance or R-C admittance are located on the negative real axis
and they alternate.
2. The singularity nearest to or at the origin is a zero .The singularity nearest to or at s   must
be a pole.
3. The residues of the pole must be real and negative.
Because of the third property a partial fraction expansion of an R-L impedance function would yield
Ki
terms as 
s i
This does not present any trouble, as the term above does not represent an R-L impedance at all. To
obtain the foster from of an R-L impedance we expand Z ( s) / s into partial fraction. If Z(s) is and R-
L impedance, we will state without proof that the partial fraction expansion of Z ( s) / s yields
positive residues. Thus we have.
Z ( s) K K1 K2 Ki
 K  0    ... 
s s s  1 s   2 s i
where K 0, K1 , ... K  0 . If we multiply both sides by s we obtain Z (s) in the desired form for
synthesis.
2( s  1)(s  3)
Consider the following function F ( s)  .
( s  2)(8  6)
F(s) represents an R-L impedance or an R-C admittance because it satisfies the fires two criteria
cited. The partial fraction Expansion of F (s) is
1 15
F ( s)  2  2  2
s  2 (8  6)
So we see that the residues are negative.
The partial fraction expansion of F ( s) / s on the other hand is
1 1 5
F ( s)
 2 4  4
s s s2 s6
If we multiply both sides by s we obtain
1 5
s s
1 4 4
F (s)   
2 s2 s6
62
If F (s) represents an impedance Z (s) it is synthesized in series foster form giving the R-L networks
in fig. 4.20. If F (s) is an admittance Y(s), then it is synthesized in parallel foster form resulting in the
R-C networks shown in fig. 4.21.
Fig. 4.20 Fig. 4.20
To synthesize an R-L impedance in ladder form, we make use of the fact that min ReZ  j   Z 0 .
If we remove Z(0) from Z(s), the remainder function Z1(s) will have a zero at s  0 . After inverting
Z1(s), we can then remove the pole at s  0 . Since the value Z (0) is obtained by dividing the lowest
power term of the denominator into the lowest power term of the numerator the synthesis could be
carried out by a continued fraction expansion by arranging the numerator and denominator
polynomials in ascending order and then dividing.
For example the following function is either an R-L impedance or an R-C admittance.
2( s  1)(s  3) 6  8s  2s 2
F ( s)  
( s  2)(s  6) 12  8s  s 2
The continued fraction expansion of F (s) is
63
1
12  8s  s 2 6  8s  2 s 2 (
2
1 2
6  4s  s
2
3 2 3
4s  s 12  8s  s 2 
2  s
9
12  s
2
9  3 8
s  s 2 4s  s 2 
2  2 7
8
4s  s 2
7
5 27  49
s  s  s2 
14  2  5s
7
s
2
s2 5 2 5
s 
14  14
5 2
s
14
0
If F (s) is an impedance function, the resulting network is shown in fig. 4.22. If, on the other hand,
F (s) is an R-C admittance Y (s) , then the network is synthesized as fig. 4.23.
Fig. 4.22. Fig. 4.23.
4.5. Synthesis of an R-L-C Functions

Under certain conditions R-L-C driving point function may be synthesized with the use of either
partial fractions (Foster Form) or continued fractions (Cauer Form). For example the function,
s 2  2s  2
Z ( s)  2
s  s 1
64
Z (s) is neither L-C, R-C nor R-L. Nevertheless the function can be synthesized by continued fraction
expansion as shown below.
s 2  s  1 s 2  2s  2 1  Z
s2  s 1
s  1s 2  s  1s  Y
s2  s
1s  1s  1  Z
s 1
0
The network derived from this expansion is given in fig. 4.24.
Fig. 4.24
Consider the following admittance function. The poles and zeros of the admittance function are all on
the negative real axis but they do not alternate.
s  2)( s  3)
Y ( s) 
( s  1)( s  4
The partial fraction expansion for Y(s) is,
2
3
Y (s)  1  s

s 1 s4
Since one of residues is negative we cannot use this expansion for synthesis. An alternate method
would be to expand Y ( s) / s and then multiply the whole expansion by s .
3 2 1
Y (s) 2
  3  6
s s s 1 s  4
Multiply by s we obtain,
65
2 1
s s
3 3 6
Y (s)   
2 s 1 s  4
Note that Y(s) also has a negative term. If we divide the numerator of this negative term to the
denominator we can rid ourselves of any terms with negative signs.
 2  1 2 1

3 2  s
5
s
Y (s)    3   6   3  6
2  3 s 1 s  4 6 s 1 s  4
 
 
The network that is realized from the expanded function is given in Fig. 4.25 below.
Fig. 4.25
Considering the same example. To synthesis in cauer form if we try to expand Y (s) by continued
fractions, we see that negative quotients result. However we can expand Z (s)  1 / Y (s) by continued
fraction. Although the expansion is not as simple or straight forward as in the case of an R-C
functions, because we sometimes have to reverse the order of division to make the quotients all
positive. The continued fraction expansion of Z (s) is
66
6  5s  s 2 4  5s  s 2 2 / 3  Z
10 s 2 2
4  s
3 3
5 1   18
s  s 2 6  5 s  s 2   Y
3 3   5s
6
6 s
5
19 1 5 1
s  s 2  s 2  s  Z
5 3 3 3
1 2 19
s  s
3 15
6  19  19
s s  s2  Y
15  5 2
19
s
5
s2 6  6
s
15  15s
Z
6
s
15
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 on fig. 4.26.
Fig. 2.26
At 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 use of a ladder form or the foster forms. These conditions are
not given here because they are rather involved instead. When a positive real function is given and it
is found that the function is not synthesizable by using two kinds of element only, it is suggested that
a continued fraction expansion or a partial fraction expansion be tried first.
67
Tutorial Problems
1. a. Which of the following functions are L-C driving point impedances? Why?
s( s 2  4)( s 2  16)
i. Z1 ( s) 
( s 2  9)( s 2  25
( s 2  1)(s 2  8
ii. Z 2 ( s) 
s( s 2  4)
b. Synthesize the realizable impedances in a foster and Cauer forms.
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.

( s 2  1)( s 2  9) s 2  25
Z ( s) 

s( s 2  4)(s 2  16)
There is no need to calculate the element values of the network.
6s 4  42s 2  48
3. Synthesize the L-C driving point impedance Z ( s)  in the form shown the
s 5  18s 3  48s
figure below i.e., determine the element values of the network
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 two elements. Find this network.
68
5. The input impedance for the network shown is

2s 2  2
Z in  2
s  2s 2  2s  2
If Z 0 is an L-C network,
(a) Find the expression for Z 0 .
(b) Synthesize Z 0 in a foster series form.
6. Indicate which of the following function is either R-C, R-L of L-C impedance functions
s 3  2s s 2  6s  8
a) Z ( s)  4 b) Z ( s ) 
s  4s 2  3 s 2  4s  3
s 2  4s  3 s 2  2s  6
c) Z ( s)  2 d) Z ( s) 
s  6s  8 s2  s
s 4  5s 2  6
e) Z ( s) 
s3  s
7. An impedance function has the pole zero pattern shown in the figure below. If Z (2)  3
synthesize the impedance in a foster form and a Cauer form.
8. From the following functions pick out the ones which are R-C admittances and synthesize in
one foster and one Cauer form.
2( s  1)(s  3) 4( s  1)( s  3)
a) Y ( s)  c) Y ( s) 
( s  2)(s  4) s( s  2)
s( s  4)(s  8) ( s  1)( s  4)
b) Y ( s)  d) Y ( s) 
( s  1)( s  6) s( s  2)
9. Find the networks for the following function. Both foster and ladder forms are required.
( s  1)(s  4) 3( s  1)( s  4)
a) Z ( s)  b) Y ( s) 
s( s  2) ( s  3)
69
10. For the network shown, find Y when

V 1 s( s 2  3)
G(s)  2   2
V0 2  Y 2s  s 2  6s  1
Synthesize Y as L-C admittance
11. Synthesize by continued fractions the function

s 3  2s 2  3s  1)
Y ( s)  3
s  s 2  2s  1)
12. synthesis the following functions in Cauer form.
s 3  2s 2  s  1
a) Z ( s)  ,
s3  s2  2
s 3  s 2  2s  1
b) Z ( s) 
s 4  s 3  3s 2  s  1
4s 3  3s 2  4s  2
c) Z ( s) 
2s 2  s
13. Of the three pole-zero diagrams shown below, pick the diagram that represents an R-L
impedance function and synthesize in series Foster form.
70
CHAPTER 5
5 Two Port Networks
5.1. Introduction
In most cases we encountered that networks with terminals are connected in pairs to other networks.
A two-port network (a kind of four-terminal network or quadripole) is an electrical circuit or
device with two pairs of terminals connected together internally by an electrical network. Two
terminals constitute a port if they satisfy the essential requirement known as the port condition: the
same current must enter and leave a port. Examples include small-signal models for transistors (such
as the hybrid- model), filters and matching networks.
A two-port network makes possible the isolation of either a complete circuit or part of it and
replacing it by its characteristic parameters. Once this is done, the isolated part of the circuit becomes
a "black box" with a set of distinctive properties, enabling us to abstract away its specific physical
buildup, thus simplifying analysis. Any linear circuit with four terminals can be transformed into a
two-port network provided that it does not contain an independent source and satisfies the port
conditions.
There are a number of alternative sets of parameters that can be used to describe a linear two-port
network, the usual sets are called z, y, h, g, and ABCD parameters, each described individually
below. These are all limited to linear networks since an underlying assumption of their derivation is
that any given circuit condition is a linear superposition of various short-circuit and open circuit
conditions. They are usually expressed in matrix notation, and they establish relations between the
variables which are shown in fig. 5.1.
Fig. 5.1 Two port network

V1  Input voltage, V2  Output voltage, I1  Input current, I 2  Output current
71
These current and voltage variables are most useful at low-to-moderate frequencies. At high
frequencies (e.g., microwave frequencies), the use of power and energy variables is more
appropriate, and the two-port current–voltage approach is replaced by an approach based upon
scattering parameters.
A pair of terminals can be called a port only if the current entering one terminal is equal to the
current leaving the other; this definition is called the port condition. A four-terminal network can
only be properly called a two-port when the terminals are connected to the external circuitry in two
pairs both meeting the port condition.
By analogy with transmission network, one of the ports is called the input port, while the other
termed as output port. There are certain properties of two-ports that frequently occur in practical
networks and can be used to greatly simplify the analysis. These include:
Reciprocal networks: A network is said to be reciprocal if the voltage appearing at port 2 due to a
current applied at port 1 is the same as the voltage appearing at port 1 when the same current is
applied to port 2. In general, a network will be reciprocal if it consists entirely of linear passive
components (that is, resistors, capacitors and inductors).
Symmetrical networks: A network is symmetrical if its input impedance is equal to its output
impedance. Most often, but not necessarily, symmetrical networks are also physically symmetrical.
Sometimes also antisymmetrical networks are of interest. These are networks where the input and
output impedances are the duals of each other.
Lossless network: A lossless network is one which contains no resistors or other dissipative
elements.
5.2. Two-port Parameters

A two port network is a special case of multiport network. Each port consists of two terminals one
for entering the current and the other for leaving. In order to describe the relationship between the
port voltage and port current of a linear two port network, two linear equations are required among
the four variables.
A general two-port network, shown in fig. 5.1, has two pairs of voltage current relationships. The
variables are V1 ,V2 , I1 , I 2 . Two of these are dependent variables; the other two are independent
72
variables. The number 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.
A one-port network is completely characterized by its driving-point impedance or admittance

function. In contrast a two-port network requires in general, the specification of four parameters. In
the realization of filter networks an often specified quantity is the open circuit voltage ratio
T (s)  V2 / V1 , (V2 is the output or the load end voltage and V1 the input or the source end voltage).
Figure  2' . equivalent circuit of a two port network in terms of Z-parameters.

5.2.1. Open circuit Impedance parameters (Z- Parameters)
Expressing two-port voltages in terms of two-port currents, i.e., V1 ,V2   f ( I1 , I 2 )
V1  Z11I1  Z12 I 2
V2  Z 21I1  Z 22 I 2
V1   Z11 Z12   I1 

V    Z       
or V  Z I
 2   21 Z 22   I 2 
 
Where Z is the open circuit impedance matrix of the two port network and impedances Z ij are the
open circuit impedance parameters and the equations of the two port networks given above are called
the z-parameter equations. In these equations the variables V1 and V2 are dependent, and I 1 , I 2 are
independent.
The individual z parameters are defined by
V1 V1 V2 V2
z11  , z12  , z 21  , z 22 
I1 I 2 0
I2 I1 0
I1 I 2 0
I2 I1 0
All these parameters are measured under open circuit conditions as shown in the fig.5.2. Hence these
circuit parameters are called open circuit impedance parameters. z11 relates the current and voltage in
the 1-1‟ port only; whereas z 22 gives the current-voltage relationship for the 2-2‟ port. Such
parameters are called open-circuit driving-point impedances. On the other hand, the parameters z12
73
and z 21 relate the voltage in one port to the current in the other. These are known as (open-circuit)
transfer impedances. If the two port network is reciprocal, then z12  z 21 .
Fig. 5.2 Determination of the Z-parameters
The equivalent circuit model of the above equations is shown in fig. 5.3.
Fig. 5.3 equivalent circuit model of Z-parameter
It is observed that all the -parameters have the dimensions of impedance. Moreover, the individual
parameters are specified only when the current in one of the ports is zero. This corresponds to one of
ports being open circuited, from which the z parameters also derive the name open-circuit
parameters.
As an example, let us find the open-circuit parameters for the T circuit in fig. 5.4. We can obtain the
z  parameters by inspection.
Fig. 5.4
74
V1 V2 V1 V1
z11   Za  Zb , z 22   Z b  Z c , z12   Z b , z12   Zb
I1 I 2 0
I2 I 2 0
I2 I1 0
I2 I1 0
Observe that z12  z21 and hence the network is reciprocal. In general, most passive time-invariant
networks are reciprocal.
Most two-port networks, whether passive or active, can be characterized by a set of open-circuited
parameters. Usually, the network is sufficiently complicated so that we cannot obtain the z-
parameters by inspection, as we did for the T circuit in fig. 5.4. The question now, „how do we obtain
the z-parameters for any circuit in general? The procedure is as follows. We write a set of node
equations with the voltages at the ports V1 and V2 , and other node voltages within the two-port
V3,V4, ,  ,Vk as the dependent variables. The independent variables are the currents I 1 and I 2 ,
which we will take to be current sources. We then proceed to write a set of node equations.
I1  n11V1  n12V2  n13V3      n1kVk
I 2  n21V1  n22V2  n23V3      n2 kVk
0  n31V1  n32V2  n33V3      n3kVk
  
0  nk1V1  nk 2V2  nk 3V3      nkkVk
1
where nij represents the admittance between the i th and j th nodes, that is, nij  Gij  sCij 
sLij
If the circuit is made up of R-L-C elements only, then it is clear that nij  n ji . As a result, the ji th
cofactor of the determinant of the node equations  ij , must be equal to the ji th cofactor  ji , that is,
 ij   ji . This result leads directly to the reciprocity condition z 21  z12 .
11  12 
V1  I1  21 I 2 , V2  I1  22 I 2
   
In relating this last set of equations to the defining equations for the z parameters, it is clear that
11  21 12 
z11  z12  z 21  and z 22  22
   
Since for a passive network 12   21 , it follows that z12  z 21 , the network is then reciprocal.
75
As an example, let us find the z parameters of the circuit in fig. 5.5. First, the node equations are
I1  YA  YC V1  YCV2
I 2  YB  YC V2  YCV1
Fig. 5.5
The determinant for this set of equations is   YAYB  YAYC  YBYC in terms of  , the open-circuit
parameters for the circuit are
YB  YC YC YC YA  YC
z11  z 21  z12  z 22 
   
5.2.2. Short circuit admittance parameters (Y- Parameters)
If we choose V1 and V2 as independent variables, we can characterize the network by
 I1   y11 y12  V1 

I   y y 22  V2 
 2  21
Where y11 , y12 , y 21 and y 22 are called y-parameters or short circuit admittance parameters. The y 
parameters are expressed explicitly as
I1 I1 I2 I1
y11  , y12  , y 21  and y 22 
V1 V2 0
V2 V1 0
V1 V2 0
V1 V 0
1
The reason that the y  parameters are also called short-circuit admittance parameters is now
apparent. In obtaining y11 and y 21 , the 2  2' port must be short circuited, and when we find y 22 and
y12 , the 1 1' port must be short circuited, as shown in fig. 5.6. When y11  y22 or z11  z 22 , the
network is symmetrical.
76
y11 and y 22 are short circuit driving point admittances at the two port networks and y 21 & y12 are
short circuit transfer admittances. In particular y 21 is the forward transfer admittance i.e. the ratio of
current response in port-2 to voltage response in port 1 and y12 is reverse transfer admittance.
Fig. 5.6 Determination of y-parameters
Suppose we were to write a set of mesh equations for the general two-port network. Then the
voltages V1 and V2 would become independent sources, and the currents I 1 and I 2 would be just
two of the dependent mesh currents. The general set of mesh equations are
V1  m11I1  m12 I 2  m13 I 3      m1k I k
V2  m21I1  m22 I 2  m23 I 3      m2 k I k
0  m31I1  m32 I1  m33 I 3      m3k I k
  
0  mk1 I1  mk 2 I 2  mk 3 I 3      mkk I k
Where mii represents the sum of the impedances in the i th mesh and mij is the common impedance
between mesh i and mesh j , we note here again that for an R-L-C network, mij  m ji for all i and
j . Thus reciprocity holds. Solving the set of meshes equations for I 1 and I 2 using Cramer‟s rule, we
obtain the following equations.
11   
I1  V1  21 V2 and I 2  12 V1  22 V2
   
This equation define the short-circuit admittance parameters as
I1  y11V1  y12V2
I 2  y 21V1  y 212V2
 ij
where yij  for all i,& j .

77
Fig. 5.7 shows an equivalent circuit model of y-parameter equations.
Fig. 5.7Equivalent circuit model for y-parameter equation
As an example, let us obtain the y  parameters of the circuit depicted in fig. 5.5. To obtain y11 and
y 21 , we short circuit terminals 2  2' . We then have
y11  YA  YC , y21  YC

We next short-circuit terminals 1 1' to obtain y 22  YB  YC , y12  YC
Let us find the y parameters for the bridged-T circuit given in fig.5.8. The mesh equations for the
circuit are
1  1
V1    1 I1  I 2  I 3
s  s
1  1
V2  I1    1 I 2  I 3
s  s
1 1 1 
0   I1  I 2  2  1 I 3
s s s 
In straightforward fashion we obtain
22s  1 2s 2  4s  1 2s 2  2s  1
 , 11   22  , 12   21  
s2 s2 s2
78
Fig. 5.8
The short-circuit parameters
2s 2  4s  1 2s 2  2s  1
y11  y 22  y 21  y12  
s2s  1 s2s  1
5.2.3. Hybrid Parameters ( h -parameters)
A set of parameters that is extremely useful in describing transistor circuits are the h parameters
given by the equations
V1  h11I 1  h12V2
I 2  h21I 1  h22V2
The individual parameters are defined by the relationships
V1 V1
h11  h12 
I1 V2 0
V2 I1 0
I2 I2
h21  h12 
I1 V2 0
V2 I1 0
It can be seen that h- parameters are interpreted under mixed set of terminal conditions, some of them
under open circuit and some under short circuit conditions. h11 and h21 are short-circuit type
parameters, and h12 and h22 are open-circuit type parameters. The parameter h11 can be interpreted as
the input impedance at port 1 with port 2 short circuited. It is easily seen that h11 is merely the
1
reciprocal of y11 i.e., h11  . The parameter h22 is an open-circuit admittance parameter and is
y11
1
related to z 22 by h22  .
z 22
79
Both the remaining h -parameters are transfer functions; h21 is a short-circuit current ratio, and h12 is
an open-circuit voltage ratio. Their relationships to z and y-parameters is discussed later in this
chapter.
Fig. 5.7 shows an equivalent circuit model of y-parameter equations.
Fig. 5.9 Equivalent circuit of h-parameter equation

For the circuit in fig. 5.5, the h -parameters are
1 YC YC YAYC
h11  , h12  , h21   , h22  YB 
YA  YC YA  YC YA  YC YA  YC
Observe that for the circuit, h21  h12 . This is the reciprocity condition for the h -parameters and
can be derived from their relationships to either the z or y parameters.
Next let us consider the h -parameters of an ideal device called the negative impedance converter
(NIC), which converts positive load impedance into a negative impedance at its input port. Consider
the NIC with load impedance Z L shown in fig, 5.10. Its input impedance is
V V
Z in  Z L , which can be rewritten as 1  2
I1 I 2
The following voltage-current relationships hold for the NIC.
V1  kV2 , I1  kI2 .
If we interpret these equations using h parameters, we arrive at the following conditions.
1
h11  h22  0 , h12  k
h21
Fig. 5.10 Negative impedance converter with load impedance

80
We see that since h12  h21, the NIC is nonreciprocal. In matrix notation, the h matrix of the NIC is
 h11 h12   0 k 
h    1 0
 21 22   k
h 
The NIC is a convenient device in the modeling of active circuits. It is not, however, a device that
exists only in the imagination. Practical realizations of NIC‟s have been achieved using transistors.
I1  g11V1  g12 I 2 
 g  parameters
V2  g 21V1  g 22 I 2 
5.2.4. Transmission Parameters (ABCD Parameters)
Let us take as the dependent variables the voltage and current at the port 1, and define the following
equation.
V1   A B   V2 
 I   C C   I 
 1   2 
This matrix equation defines the A, B, C, D parameters, whose matrix is known as the transmission
matrix because it relates the voltage and current at the input port to their corresponding quantities at
the output. The reason the current I 2 carries a negative sign is that most transmission engineers like
to regard their output current as coming out of the output port instead of going into the port, as per
standard usage.
In explicit form, the ABCD parameters can be expressed as
V1 V1 I1 I1
A , B , C , D
V2 I 2 0
I2 V2 0
V2 I 2 0
I2 V2 0
From these relations we see that A represents an open-circuit voltage transfer function; B is a short-
circuit transfer impedance; C is an open-circuit transfer admittance; and D is a short-circuit current
ratio. Note that all four parameters are transfer functions so that the term transmission matrix is a
very appropriate one.
81
Let us describe the short-circuit transfer functions B and D in terms of y parameters, and the open-
circuit transfer functions A and C in terms of z parameters. Using straight forward algebraic
operations, we obtain
Z11 1 1 Y11
A , B , C , D
Z 21 Y21 Z 21 Y21
For the ABCD parameters, the reciprocity condition is expressed by the equation
A B
det    AD  BC  1
C D
Let us find, as an example, the ABCD parameter for the ideal transformer in fig. 5.11, whose
defining equations are
V1  nV2 I1 
1
 I 2 
n
Fig. 5.11
If we express the above equation in matrix form, we have
V1  n 0  V 
V   0 1  2 
I
 2   n   2 
So that the transmission matrix of the ideal transformer is
 A B  n 0
C D   0 1
   n 
Note, incidentally, that the ideal transformer does not possess an impedance or admittance matrix
because the self- and mutual inductances are infinite.
For the ideal transformer terminated in load impedance shown in fig. 5.12, the following set of
equations apply.
82
V2
V1  nV2 I1 
nZ L
Taking the ratio of V1 to I 1 , we find the input impedance at port 1 to be
V1
Z1   n2Z L
I1
Thus we see that an ideal transformer is an impedance transformer. If the load element were an
inductor L (fig. 5.12 (b)), at port 1 we would see an equivalent inductor of value n 2 L. Similarly, a
capacitor C at the load (fig. 5.12 (c)) would appear as a capacitor of value C / n 2 at port 1.
Fig. 5.12
As a second example indicating the use of the transmission matrix in network analysis, consider the
ABCD parameter of the circuit in fig. 5.5.
YB  YC 1 YAYB  YBYC  YAYC Y Y

A , B , C , D A C
YC YC YC YC
If we check for reciprocity of the above equations, we see that
AD  BC 
YA  Yc YB  Yc   YAYB  YBYC  YAYC  YC2
 1
YC2 YC2
So that it is reciprocal.
83
To conclude this section, the possible combinations generated by the four variables, taken two at a
time is six. We have discussed four of them in the previous section. The remaining two are the
inverse h-parameters or g-parameters and the inverse ABCD-parameters or A' B' C ' D' -parameters.
The six two-port parameters are summarized in table 5.1 below.
Table 5.1: Summary of two-port parameters

Name Function
Matrix Equation
Express In terms of
Open circuit impedance [Z ] V1 ,V2 I1 , I 2 V1   Z11 Z12   I1 

V   Z  
 2   21 Z 22   I 2 
Short-Circuit admittance Y  I1 , I 2 V1 ,V2  I1  Y11 Y12  V1 
 I  = Y  
 2   21 Y22  V2 
Transmission or Chain T  V1 , I1 V2 , I 2 V1   A B   V2 
 I   C D  I 
 1   2 
Inverse Transmission T  V2 , I 2 V1 , I1 V2   A' B'   V1 
 I   C ' D'  I 
 2   1 
Hybrid h V1 , I 2 I1 ,V2 V1   h11 h12   I 1 
 I  = h  
 2   21 h22  V2 
Inverse Hybrid g  I1 ,V2 V1 , I 2  I 1   g11 g12  V1 
V  =  g  
 2   21 g 22   I 2 
5.3. Relationships between two port Parameters
The relationships between two-port parameters are quite easily obtained because of the simple
1 1
algebraic nature of the two-port equations. For example, we have seen that h11  and h22  .
y11 z 22
To drive h22 in terms of open circuit parameters, consider the z parameter equations when port 1 is
V1 z12
open circuited I1  0 : V1  z12 I 2 , V2  z 22 I 2 Therefore, we have h12  I1 0 
V2 z 22
84
Similarly, since h21 is defined as short-circuit type parameter, we drive h21 in terms of y parameters
y21
as h21  We can express all the h parameters as functions of the z parameters or y parameters
y11
alone. An easy way to accomplish this task is by finding out what the relationships are between 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. The z and y relationships can be obtained very easily
by using matrix notation. If we define the z matrix as
y y12 
Z   
z11 z12 
 And the y matrix as Y    11
 z 21 z 22   y 21 y 22 
In simplified notation we can write the two sets of equations as V   Z I  and I   Y V 
Replacing I  by Y V  from equation above, we obtain V   Z Y V  so that the product Z Y 
must yield the unit matrix U  . The matrices Y  and Z  must therefore be inverses of each other,
that is, Z   Y  and Y   Z 

1 1
From the relationship, we can find the relations between the individual z and y parameters.
z22 z11 z12 z 21
y11  y 22  y12   y 21  
z z z z
where  z  z11z 22  z12 z 21 ; and

y 22 y11 y12 y 21
z11  z 22  z12   z 21  
y y y y
where  y  y11 y22  y12 y21 .
Using these identifies, we can drive the h or ABCD parameters in terms of either the z or y
parameters. Table 5.2 provides a conversion table to facilitate the process. Note that in the table
 T  AD  BC.
85
Table 5.2: Parameter Conversion Matrix
5.4. Transfer Functions Using Two-Port Parameters
In this section we will examine how to determine driving point and transfer functions of two-port by
use of two-port parameters. These functions fall into two broad categories. The first applies to two-
ports without load and source impedances. These transfer functions can be described by means of z
or y parameters alone. For example, let us derive the expressions for the open-circuit voltage ratio
V2 V1 by using z -parameters first and y parameters next. Consider the z parameter equations for
two-port network. When port 2 is open circuited,
V2  z 21I1 V1  z11I1
If we take the ratio of V2 to V1 , we obtain
V2 z 21

V1 z11
By letting I 2 of the second y parameter equation go to zero, we derive the open-circuit voltage ratio
as
V2 y
  21
V1 y 22
86
In similar manner we can derive the short-circuit current ratio of two-port as

I 2 y 21 I2 z
 and   21
I1 y11 I1 z 22
The open and short-circuit transfer functions are not those we usually deal with in practice, since
there are frequently source and load impedances to account for. The second category of tow-port
transfer functions are those including source or load impedances. These transfer functions are
functions of the two-port parameters z, h, or y and the source and /or load impedance.
I2
For example, let us derive the transfer admittance of two-port network that is terminated
V1
in resistor of R ohms, as given in fig. 5.13. For this two port network, the following equations
apply.
I 2  y21V1  y22V2 V2   I 2 R
Fig 5.13 Two-port network terminated in load resistance R
By eliminating the variable V2 , we obtain the transfer admittance as
I2 y 21 R
Y21  
V1 y 22  1 / R
Note that Y21 and y 21 are not the same. Y21 is the transfer admittance of the two-port network
terminated in a resistor R, and y 21 is the transfer admittance when port 2 is short circuited . We must
be careful to make this distinction in other cases of a similar nature.
In order to solve for transfer functions of two-ports terminated at either port by an impedance Z L , it is
convenient to use the equivalent circuit of the two port network given in terms of its z parameter. The
equivalent voltage sources z12 I 2 and z 21I1 are called controlled voltage sources because they
87
depend upon a current or voltage somewhere in the network. Similarly the current sources
y12V2 and y 21V1 are controlled sources. For the circuit in fig. 5.24, let us find the transfer impedance
Z 21  V2 / I1 , with port 2 terminated in load impedance, Z L .
Fig 5.14 Two-port equivalent circuit terminated in Z L
If we write the mesh equation for the I 2 mesh we have
 z 21I 1 ( z 22  Z L ) I 2 but V2   I 2 Z L
V2 z 21Z L
Thus, Z 21 I  z  Z
1 22 L
It also is clear that the current- ratio transfer function for the terminated two-port network is
I2  z 21
 . In similar fashion, we obtain the voltage-ratio transfer function for the circuit
I 1 z 22  Z L
represented in the fig. 5.15 as:
V2 y 21

V1 Y 2 y 22
Fig. 5.15 Two port equivalent
88
V2
Next, suppose we are required to find the transfer function for the two –port network terminated
Vg
at both ends, as shown in fig. 5.16. We first write the two mesh equations
Vg  ( R1  z 11) I 1 z 12 I 2
0  z 21I 1( z 22  R 2 ) I 2
Next, we solve for I2 to give
V2 z 21
I 2
( R1  z11 )( R2  z 22 )  z12 z 21
From the equation V2  R2 I 2 , we may now arrive at the following solution.

V2 RI z 21R2
 2 2 
Vg Vg ( R1  z 11)( R 2  z 22 )  z 21z12
Fig. 5.16 Two-port network terminated at both ports
Note that the equivalent circuits of the two-port networks are not unique. Two other examples are
given in fig. 5.17 and fig.1.18. Observe that the controlled sources are nonzero in these equivalent
circuits only if the circuit is nonreciprocal.
Fig. 5.17 Two-Port equivalent circuit with one controlled-voltage source
89
Fig. 5.18 Two port equivalent circuit with one controlled-current source
Finally, let us consider the hybrid equivalent circuit shown in fig. 5.19. Observe the voltage–
controlled source h21V2 at input port and the current controlled source h11I1 at the output port. Let us
find the input impedance Z in . The pertinent equations are
V 1h11I 1h12V2
V 2 Z LI 2 h 21I 1h 22V2 )Z I 
Solving this equation for V2 , we find
h Z I
V 2  21 L 1
1  h 22 Z L
By substitution, we have
 h h Z 
V 1  h11  12 21 L  I 1
 1  h 22 Z L 
So that
V h h Z
Z in  1  h11  12 21 L
I1 1 h22 Z L
Note that h11 has the dimensions of impedance h22 is an admittance, and h12 , h21 are dimensionless
since they represent voltage and current ratios, respectively.
Fig. 5.19 Hybrid Equivalent Circuit

90
5.5. Interconnection of Two-Port Networks
In this section we will consider various interconnections of two–port networks. We will see that
when a pair of two-ports is cascaded, the overall transmission matrix is equal to the product of the
individual transmission matrices of the two-port networks. When two-port networks are connected in
series 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 N a and N b in cascade or in
tandem, as shown in fig. 5.20.
Fig. 5.20 Cascade connection of two port networks

V 2 a  V1b 
We see that I   I 
 2 a   1b 
The transmission matrix equation for Na is
V 1   Aa Ba  V 2 a 
 I   C D a   I 2 a 
 1  a
Correspondingly for Nb we have
V 2 a  V1b   Ab Bb  V 2 
 I    I   C Db   I 2 
 2 a   1b   b
Substituting the second matrix into the first, we obtain
V 1   Aa Ba   Ab Bb  V 2 
I     .
D b   I 2 
 1  Ca Da  C b
We see that the transmission matrix of the overall two two-port networks connected in cascade is
simply the product of the transmission matrices of the individual two ports.
As an example, let us calculate the overall transmission matrix of a gyrator in tandem with a T
network shown in fig. 5.21. The ideal gyrator is an impedance inversion device whose input
impedance Z in is related to its load impedance Z L by
91
a2
Z in a YL 
2
ZL
The constant a in above is defined as the gyration resistance. If we regard the gyrator as a two-port
network, its defining equations are
V2
V1  a( I 2 ) , I1 
a
0 a
So the transmission matrix of the gyrator is  1 
 a 0

We see that for the gyrator AD  BC  1. Therefore, the gyrator is a nonreciprocal device, although
it is passive.
Fig. 5.21 Gyrator in tandem with T circuit

 z a  zb z a zb  zb z c  z a z c 
 z zb 
For the T-network the transmission matrix is  b  . Therefore, the overall
 1 zb  z c 
 zb zb 
transmission matrix of the configuration in fig. 5.21 is obtained by the product of the individual
transmission matrices
 z a  zb z a zb  zb zc  z a zc   a a( z b  z c ) 
A B 0 a  z zb   z zb 
C D    1  b
 b

   a 0  1 zb  zc   z  zb z a zb  zb zc  z a zc 
 a
 z b zb  
  az b az b 
If we check the configuration in fig. 5.21 for reciprocity, we see that for its transmission matrix
AD  BC  1 i.e., AD  BC  1 and hence it is not reciprocal. However, the T-network is
reciprocal. We thus see that any reciprocal network connected in tandem with a gyrator yields a
configuration that is nonreciprocal.
92
Next consider the situation in which a pair of two-port networks N a and N b are connected in
parallel, as shown in fig. 5.22.
Fig. 5.22 Cascade connection of two port networks
Let us find the y parameters for the overall two-port network. The y-matrix equations for the
individual two-ports are
 I1a   y11a y12a  V1a  I   y y12b  V1b 
I    y    and  1b    11b  
 2 a   21a y 22a  V2 a   I 2b   y 21b y 22b  V2b 
From fig. 5.22 see that the following equations must hold.
V1  V1a  V1b , I1  I1a  I1b
V2  V2a  V2b , I 2  I 2a  I 2b
In matrix notation, the sum of the individual I i  matrices of two-port networks in parallel must equal
the I  matrix of the overall two-port network. Thus we have
 I1   I1a   I1b   y11a y12a   y11b y12b  V1 

 I    I    I    y 
y 22a   y 21b

y 22b  V2 
 2   2 a   2b   21a
So that the y parameters of the overall two-port network can be expressed in terms of the y
parameters of the individual two-port as
 y11 y12   y11a y12a   y11b y12b   y11a  y11b y12a  y12a 
  
y
 21 y 22   y 21a y 22a   y 21b y 22b   y 21a  y 21b y 22a  y 22b 
If we connect two-port networks in series as in fig. 5.23, in similar way as derived above, we can
express the z -parameters of the overall two-port network in terms of z -parameters of individual
two-port networks as
 z11 z12   z11a  z11b z12a  z12a 


z
 21 z 22   z 21a  z 21b z 22a  z 22b 
93
Fig. 5.23 Series connection of two-port networks
Note that in connecting two-port networks in series or in parallel, we must be careful that the
individual character of a two-port network is not altered when connected in series or parallel with
another two-port network. For example, when we connect the two-port networks in parallel as shown
in fig. 2.24, the impedances Z 7 and Z 8 will be short circuited. Therefore, to ensure that a two
port network does not interfere with the internal affairs of the other, ideal transformers are used to
provide the necessary isolation.
Fig. 2.24
We may summarize the interconnection of two port networks by the following three points:
1. When two-ports are connected in parallel, find the y parameters first, and, from y
parameters, derive the other two-port parameters.
2. When two-ports are connected in series, it is usually easiest to find the z parameters.
3. When two-ports are connected in tandem, the transmission matrix is generally easier to
obtain.
As a final example, let us find the y parameters of the bridge-T circuit in fig. We see that the bridge-
T-circuit would be decomposed into a parallel connection of two-port networks as shown in 5.25.
Our task is to first find the y-parameters of the two-port networks N a and N b . Since both
Na and N b are symmetrical networks, y11  y 22 and y12  y 21 .
94
Fig. 5.25
The y parameters of N b are obtained by inspection and are
1
y12b  y 21b  
2
1
y11b  y 22b 
2
N a is a T-circuit so that the z-parameters can be obtained by inspection.
s 1
z11a  z 22a 
s
z12a  z 21a 1
Using the conversion matrix, the y-parameters can be found as
z 22a ss  1
y11a  y 22a  
z a 2s  1
z12a s2
y12a  y 21a  
z a 2s  1
Since both N a and N b are symmetrical networks, the overall network is also symmetrical and its
y-parameters y11  y 22 and y12  y 21 . We know that for parallel connection two two-port networks,
the y-parameters of the overall network is equal to the sum of the respective y-parameters of
constituent two-port networks, i.e.,
 y11 y12   y11a  y11b y12a  y12a 

y
 21 y 22   y 21a  y 21b y 22a  y 22b 
ss  1 1 2s 2  4s  1
y11  y11a  y11b     y 22
2s  1 2 22s  1
s2 1 2s 2  2s  1
y12  y12a  y12b      y 21
2s  1 2 22s  1
95
5.6. Synthesis of Transfer Functions (two-port networks)
Auxiliary functions - transmission and reflection coefficients; terminated two port networks
96
Chapter 6
6. Topics in Filter Design
6.1. Introduction
Filters are network which passes signals with a specific frequency range and rejects or attenuates the
signals whose frequencies are outside of the desired frequency range. Filter may be categorized as
low pass, high pas, band pass and band rejection (elimination) filters depending on the frequency
range they pass. They are also classified as passive and active filters depending on constituent circuit
components.
6.2. The Filter Design Problem
In the preceding chapter we examined different methods of synthesizing a driving point or transfer
function H(s). Most problems have as their initial specification amplitude or phase characteristic, or
an impulse response characteristic instead of the system function H(s). The 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  C , 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. 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.
6.3. The Approximation Problem in Network Theory
The essence of the problem is the approximation of a given function f(x) by another function
f a ( x;1 ,...., n ) in an interval x1  x  x2 the parameter 1 ,...., n in the approximating function are
fixed by the particular error criterion chosen. When we let  f (x)  f a ( , 1 ,....,  n  , the following
error criteria most common:
x2 2
1. Least Squares. The value of I (1 ,....,  n is minimized where I (1 ,...., n  = x   ( x)dx
1
And  (x) is a weighting function which stresses the error in certain subintervals.
2. Maximally flat. The first n-1 derivatives of E are made to vanish at x=xo.
3. Chebyshev .the value of  is minimized in the interval x1  x  x 2 where   max
4. Interpolation. The value of  is made to vanish at a set of points in the interval x1  x  x 2 .
After an error criterion is chosen, we must determine the particular form of the approximating
function. This depends up on whether we choose to approximate in the time or frequency domain
97
suppose f(x) represents a magnitude function in the frequency domain and the approximating
function is to be rational in  2 ; then
1   3 x   5 x 2    
f a ( ,  1 ,....,  n  Where x   2 in addition, the values of  k must be
 2   4 x   6 x 2  ....
restricted to insure that f a ( ;1 ,....,  n   0 in the f(x) might represent an impulse response of a
system to be synthesized. In the case of an R-C transfer function, we have
f a ( ;1 ,...., n   0  1  x  3  x ...
2 4
Where x=t. since an R-C transfer function must have its poles on the negative real axis , the values of
 k , k even, are restricted to negative real numbers. The keystone of any approximation problem lies
in the choice of a suitable error criterion subject to reliability restrictions. The problem can be
simplified when some of the  ' s are assigned before applying the error criteria. All the error criteria
cited, except the Chebyshev, can then be reduced to a set of linear algebraic equations for the
unknowns 1 ,...,  n .
Time–domain approximation
The principal problem of time-domain approximation consists of approximating an impulse response
h(t ) by an approximating function h * (t ) such that the squared error    ht   h * t  dt is a
 2
0
minimum. A generally effective procedure in time –domain approximation utilizes orthonormal
n
function k (t ) . The approximating function h*(t) takes the form h * (t )    k  k (t )
k 1
2
  n

So that the error    h(t )    k  k (t ) dt is minimized when

 k 1 

 k   h(t )k (t )dt k  1,2,..., n
0
sk t
If the orthonormal set is made up of a sum of exponentials e , then the approximate impulse
n n
k
response. h * (t )    k e sk t Has a transform H * ( s)  
k 1 k 1 s  sk
Reliability is insured if in the orthonormal set  e , Re s
k
sk t
k  0 ; k  1,2,  , n . Synthesis then
proceeds from system function H*(s).
Frequency-Domain Approximation
In frequency-domain approximation the principal problem is to find a rational function H(s) whose
magnitude H  j  approximates the ideal low–pass characteristic in figure below according to a
predetermined error criterion. In the next few sections we examine several 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, elliptic or Bessel.
98
Figure: ideal low-pass filter characteristic.

Another major problem is that of obtaining transfer function H(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 on
the maximally flat method/the joint problem of approximating both magnitude and phase over a
given frequency range is possible.
6.3 The Maximally Flat Low-Pass Filters Approximation.
In previous chapter we saw that the ideal low–pass filter in figure above is not realizable because its
associated impulse response is 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 approximations. 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.
K0
M ( )  Where K 0 is the d-c gain constant and f ( 2 ) is the polynomial to be
[1  f ( 2 )]1 / 2
selected to give the desired amplitude response. For example, if f ( 2 )   2 n , then the
amplitude function can be written as
K0
M ( ) 
 1   2n ) 1
2
We see that M (0)  K 0 , and that M ( ) is monotonically decreasing with  . In addition, the 0.707 or
K
3-decible point is at   1 for all n, that is, M (1)  o for all n
2
The cut off frequency is thus seen to be   1 the parameter n controls the closeness of
approximation in both the pass band and the stop band. Curves of M ( ) for different n are shown in
figure below. Observe that the higher n , is the better the approximation.
99
Figure: Amplitude response of Butterworth low-pass filters

The amplitude approximation of the above equation is called Butterworth or maximally flat response.
The reason for the term „maximally flat” is that when we expand M ( ) in a power series about   0
we have
35 8 n
M ( )  K 0 (1  12  2 n  83  4 n  165  6 n 
    )
128
We see that the first 2n-1 derivatives of M ( ) are equal to zero at   0 for   1, the amplitude
response of a Butterworth function can be written as (with K 0 normalized to be unity)
1
M ( )  n   1

We observe that asymptotically, M ( ) falls off as   n for a Butterworth response in terms of
decibels, the asymptotic slope is obtained as.
20 log M ()  20n log 
Consequently, the amplitude response falls asymptotically at a rate 6ndb / octave or
20n db / decade .
One question remains. How do we obtain a transfer function H (s) from only the amplitude
characteristics M ( ) ? The procedure is as follow. We first note that the amplitude response M ( )
and the complete system function H ( j ) are related by M 2 ( )  H ( j ) H ( j )
If we define a new function h( s 2 ) such that h(s 2 )  H (s) H (s)
We see that M 2 ( )  h( 2 )
From h( 2 ) all we need to do is to substitute s 2   2 to give h( s 2 ) . Then we factor h( s 2 ) in to
the product H (s) H (s) . Since the poles and zeros of H (s) are the mirror images of the poles and
zeros of H (s) , we simply choose the Hurwitz factors of h( s 2 ) as H (s) . An example will serve to
clarify this discussion. Consider the third order (n=3) Butterworth response given by
100
1 1 1
M 2 ( )   We see that h( s 2 ) is h( s 2 ) 
1  6
1  ( 2 ) 3 1  (s 2 ) 3
1 1
Factoring h( s 2 ) , we obtain h( s 2 )   H ( s) H ( s)
1  2s  2s  s 1  2s  2s 2  s 3
2 3
1 1
We then have H ( s)  3 
s  2s  2s  1 ( s  1)(s  2  j 2 )(s  12  j 3 / 2 )
2 1 3
The poles of H(s) and H (-s) are shown in figure below. Observe that the poles of H (-s) are mirror
images of the poles of H(s), as given by the theorem on Hurwitz polynomials.
Figure: Poles of H (s) H (s) for an n=3 are Butterworth filter

j ( 2 k 1)
For a Butterworth response, the poles of H (s) H (s) are the roots of (1) n s 2n  1  e
s k  e j ( 2 k 1)  n even
k  0,1,2,  ,2n  1 . The poles s k are then given by
 e j ( k / n )  n odd
Or simply by sk  e j 2 k n1/ 2 n  k  0,1, ,2,  ,2n Expressing sk as
sk   k  jk , the real and imaginary parts are given by
2k  n  1  2k  1   2k  n  1  2k  1  
k  sin   cos 2  k  cos   sin 2
2n  n  2n  n 
It is seen from the above equations that all the poles of H ( s) H (s) are located on the unit circle in
the s plane, and are symmetrical about both the  and j axes to satisfy reliability conditions, we
associate the poles in the right-half plane with H (s) , and the poles in the left–half plane with H (s)
.
As an example, consider the construction of H (s) that gives an n=4 Butterworth responses from the
above equations it is seen that the poles are given by sk  e j[( 2k 3) / 8] H (s) is then given as
1
H (s) 
s  e   s  e   s  e   (s  e
j 5
8
 j 7
8
 j 9
8
 j ( 11
8
)
)
101
If we express s k in complex form and expand, we obtain

1
H ( s) 
s  0.76536s  1(s 2  1.84776s  1)
2
To simplify the use of Butterworth function, H(s) is given in tables below for n=1 to n=8 in factored
form as in equation above or multiplied out as
1
H ( s)  n 1
an s  an1 s
n
 ...  a1 s  1
Table 1: Butterwort Polynomials (Factored Form)
1. s  1
2. s 2  2s  1

3. s 2  s  1 s  1 

4. s 2  0.76536s  1 s 2  1.84776s  1  

5. s  1 s 2  0.6180s  1 s 2  16180s  1  
6. s 2

 0.5176s  1 s  2s  1) s  19318s  1
2
 2

7. s  1s 
 0.4450s  1 s  1.2456s  1 s  1.8022s  1
2 2
 2

8.   
s  0.3986s  1 s 2  1.1110s  1 s 2  1.6630s  1 s 2  1.9622s  1
2
 
Table 2: Butterworth polynomials
n a1 a2 a3 a4 a5 a6 a7 a8
1. 1
2. 2 1
3. 2 2 1
4. 2.613 3.414 2.613 1
5. 3.236 5.236 5.236 3.236 1
6. 3.864 7.464 9.141 7.464 3.864 1
7. 4.494 10.103 14.606 14.606 10.103 4.494 1
8. 5.126 13.138 21.848 25.691 21.848 13.138 5.126 1
6.4 Other Low-Pass Filter Approximations

In section 6.3 we examined the maximally flat approximation to a low–pass filter characteristic we
will consider other low pass filter approximants in this section.
6.4.1 The Chebyshev or Equal- Ripple Approximation
We have seen that the maximally flat approximation to the ideal low pass filter is best at   0
whereas as we approach the cutoff frequency   1 , the approximation which “ripples” about unity in
the pass band and falls off rapidly beyond the cutoff   1 the approximation is equally good at
  0 and   1 and, as a result is called an equal-ripple approximation. The equal–ripple property is
brought about by the use of Chebyshev cosine polynomials defined as
102
C n ( )  cos(n cos 1  )  1
 cosh(n cosh 1  )  1
For n=0 we see that C0 ( )  1 and for n=1, we have C1 ( )  
Higher order Chebyshev polynomials are obtained through the recursive formula
Cn ( )  2Cn1 ( )  Cn2 ( )
Thus for n=2 we obtain C2 ( ) as C2 ()  2 ( )  1  2 2  1
In the table below Chebyshev polynomials of orders up to n= 10 are given.
Table 3:
n Chebyshev polynomials Cn ( )  cos(n cos 1  )
0 1
1 
2 2 2  1
3 4 3  3
4 8 4  8 2  1
5 16 5  20 3  5
6 32 6  48 4  18 2  1
7 64 7  112 5  56 3  7
8 128 8  256 6  160 4  32 2  1
9 256 9  576 7  432 5  120 3  9
10 51210  1280 8  1120 6  400  50 2  1
Figure: C3 ( ) and C4 ( ) 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   1 as seen by the plots of C3(  )
and C4(  )in figure above.
2. within the interval   1 , the absolute value of Cn ( ) never exceeds unity, that is Cn ( )  1
for   1
3. beyond the interval   1, Cn ( ) increases rapidly for increasing values of 
103
Now, how do we apply the Chebyshev polynomials to the low-pass filter approximation? Consider
the function 2 Cn ( ), where  real and small is compared to 1. It is clear that 2 C n ( ), will vary
2 2
between 0 and 2 in the interval   1 . Now we add 1 to this function making it 1 2 Cn ( ), . This
2
new function varies between 1 and 1 2 , , a quantity slightly greater than unity, for   1 . Inverting
this function, we obtain the function which we will associate with H ( j ) 2 ; thus
H  j  
2 1
1  C n ( )
2 2
Within the interval   1 , H ( j ) oscillates about unity such that the maximum value is 1 and the
2
 
minimum is 1 / 1 2 . Outside this interval Cn2   becomes very large so that as  increase, a point
will be reached where 2 Cn2    1 and H ( j )

2
approaches zero very rapidly with further
increase in  .
Figure below shows a Chebyshev approximation to ideal low-pass filter. We see that within the pass
band and 0    1 , H ( j ) ripples between the value 1 and 1 2  
 12
. The ripple height or distance
between
Figure: Chebyshev approximation to low-pass filter

Maximum and minimum in the pass band is given as
1 1
Ripple  1  At   1, H ( j ) is H ( j1)  Because C n (1)  1
2
(1  ) 2
2 1
(1  ) 2
2 1
In the stop band that is for   1, as increase, we reach a point  , where C n ( )  1 so that
2
1
H ( j )    
 C n ( )
The loss in decibels is given as Loss  20 log10 H  j   20 log  20 log Cn ()
104
But for large , Cn ( ) can be approximated by its leading term 2 n1  n , so that
Loss  20 log  20 log 2 n1 n  20 log   6(n  1)  20n log 
We see that the Chebyshev response also falls off at the rate of 20ndb/decade after an initial drop of
20 log  6(n  1) decibels. However in most applications,  is a very small number so that the
20 log  term is actually negative. It is necessary, therefore, to compensate for this decrease in loss
in the stop band by choosing a sufficiently large n.
From the preceding discussion, we see that a Chebyshev approximation depends up on two variables,
 and n which can be determined from the specifications directly the maximum permissible ripple
puts a bound on  . Once  is determined any desired value of attenuation in the stop band fixes n .
The derivation of the system function H (s) from a Chebyshev amplitude approximation H  j  is
somewhat involved and will not be given here instead, we will simply give the results of such a
derivation. First we introduce a design parameter.
Figure: Locus of Poles of Chebyshev filters

  1n sinh 1 1
Where n is the degree of the Chebyshev polynomial and  is the factor controlling ripple width. The
poles, sk   k  j , of the equal- ripple approximant H (s) are located on an ellipse in the s plane,
given by.
 k2 k2
 1
sinh 2  cosh 2 
The major semi axis of the ellipse is on the j -axis and has a value    cosh  k , the minor semi
axis has a value    sinh  k , and the foci are at   1 . The half power point of the equal ripple
amplitude response occurs at the point where the ellipse intersects the j axis i.e. at   cosh  k .
Recall that for the Butterworth response, the half-power point occurs at   1 let us normalize the
Chebyshev poles s k such that the half-power point also
105
falls at   1 instead of at   cosh  k i.e. let us choose a normalizing factor cosh  k Such that the
s' k 
sk k j 
normalized pole locations s' k are given by     ' k  j ' k
cosh   cosh   cosh  
 2k  1  
The normalized pole locations can be derived as  '  tanh   sin 2
 n 
 2k  1  
 '  cos 2
 n 
Comparing the normalized Chebyshev pole locations with the Butterworth pole locations, we see that
the imaginary parts are the same, while the real part  ' k of the Chebyshev pole location is equal to
the real part of the Butterworth poles times the facto tanh   . For example, with n=3 and
tanh    0.444, the butterwort poles are.
s1  1  j 0 s2,3  0.5  j 0.866

So that the normalized Chebyshev poles are given by
s1  1(0.444)  j 0  0.444  j 0
s2,3  0.5(0.444)  j 0.866  0.222  j 0.866
Finally, to obtain the demoralized Chebyshev poles, we simply multiply s' k by cosh   ' that is,
sk   'k  j 'k cosh 
There is an easier geometrical method to obtain the Chebyshev poles, given only the semi axis
information and the degree n . First we draw two circles, the smaller of radius sinh   and the larger
of radius cosh   as shown in figure below. Next we draw radial lines according to the angles of
Butterworth poles. Finally we draw vertical dashed lines from the intersection of the smaller circle
and the radial lines and horizontal dashed lines from the intersection of the large circle and the radial
lines. The Chebyshev poles are located at the intersection of the vertical and horizontal dashed lines
as shown in the next figure.
106
Figure: n=3 Chebyshev filter poles

Consider the following example we would like to obtain a system function H (s) that exhibits a
Chebyshev characteristic with not more than 1- decibel ripple in the pass band and is down at least
20 decibels at   2 . When we design for 1- decibel ripple, we know that at   1 , H ( j1) is down 1
20 log H  j1  20 log

1
decibel so that  1 We than obtain
(1 2 ) 2
1
1
1
 0.891, and   0.509
(1 2 ) 2
Our next task is to find n from the 20 decibels at   2 specification. From the previous relation the
loss can be given as approximately. 20  20 log 0.509  6(n  1)  20n log 2
Solving for n, =2.65 since n must be an integer, we let n=3 with the specification of n and  , the pole
locations are completely specified. Our next task is to determine these pole locations. First we must
1 1
find      1n sinh
1
 sinh 1 1.965  0.476
 3
In order to find the normalized Chebyshev poles from the Butterworth poles, we must first determine
tan   Here we have tanh  k  tanh 0.476  0.443
Fro table 1, the n=3 Butterworth poles are s1  1.0, s 2 , 3  0.5  j 0.866
Multiplying the real parts of these poles by 0.443, we obtain the normalized Chebyshev poles.
s1  0443, s2,3  0222  j 0.866
Finally, the denormalized Chebyshev poles are obtained by multiplying the normalized ones by
cosh   1.1155 so that the denormalized poles are s1  0494 and s2,3  0249  j 0.972
0.502
H(s) is then H (s) 
( s  0.494)( s  0.249  j 0.972)( s  0.249  j 097)
0.502
 3
s  0992s  1.255s  0.502
2
In figure below, the amplitude responses of the Chebyshev and n =3 Butterworth filter are shown.
Monotonic filters with optimum cutoff
107
In comparing Butterworth filters with Chebyshev filters, the following can be said. The Butterworth
response is a maximally flat, monotonic response, whereas the Chebyshev response is equal ripple in
the pass band. In the stop band, the Chebyshev response falls off more rapidly than the Butterworth
(except when e is very, very small). In this respect, the Chebyshev filter is a better filter than the
Butterworth. However, as we shall see in the next section, the transient response often Chebyshev
filter is very poor. If we require sharp cutoff characteristics for a given degree n, however, the
Butterworth filter is quite unsatisfactory.
Figure: Amplitude response of n=3 Chebyshev filter with 1.0-decibel ripple in pass band and Butterworth
response (n=3)
A class of filters called optimum or „L‟ filters, which have the following properties:
1. The amplitude response is monotonic
2. The fall –off rate at  cutoffs the greatest possible, if monotonic is assumed
3. The zeros of the system function of the L filter are all at infinity.
Recall that the magnitude response of a low –pass filter with all zeros at infinity can be express as
k0
M ( )  Let us denote the polynomial generating the L filter by
1  f  
1
2 2
 
f ( 2 )  Ln  2
The polynomial Ln  2  has the following properties.
a. Ln 0  0 b. Ln 1  1 c.
 
dLn  2
0 d.
 
dLn  2
 M (max imum)
d d  1
108
Properties a, b and c are the same as for the Butterworth generating polynomial f ( 2 )   2 n .
Property c insures that the response M ( ) is monotonic and property„d‟ requires that the slope of
 
Ln  2 at   1 be the steepest to insure sharpest cutoff. Papoulis originally derived the generating
equation for the polynomials Ln  for n odd  to be
2
2 2 1 k 
Ln ( )    ai Pi ( x) dx
2
1
 i 0 
Where n  2k  1 and the Pi (x) are the Legender polynomials of the first kind
P0 ( x)  1 P1 ( x)  x, P2 ( x) 
2

1 2
 1
3x  1 , P3 ( x)  (5 x 3  3x)   
2
a1 a2 a 1
And the constants at are given by ao      k 
3 5 2k  1 2 k  1
Later Papoulis and, independently, showed that the even order Ln polynomials can be given by.
2
2 2 1 k 
L 2 k  2 ( )  
2
( x  1)  ai Pi ( x) dx , n  2k  2 Where the constants a1 are given by:
1
 i o 
a2 a 1
Case 1(k even): ao   ... k  a1  a3  ...  ak 1  0
5 2k  1 k  1k  2
a1 a3 a 1
Case 2(k odd):   ... k  ao  a2  ...  a2k  0
3 7 2k  1 k  1k  2
Fukada tabulated the Ln   polynomials up to n=7 together with
2 dLn  2
evaluated at   1 to give
 
d
an indication of the steepness of the cut off. This is shown in table 4. To obtain the system function
 
H (s) for the "L" filter, we must factor the equation for h s 2 and choose the Hurwitz factors as
H (s) .
1
h( s 2 )  H ( s ) H (  s ) 
1  Ln ( s 2 )
For example for n  3 the magnitude response squared is
1 1
M 2 ( )  
1  L3 ( ) 1    3 4  3 6
2 2
1
Substituting   2  s 2 , we obtain h( s 2 )  H ( s ) H (  s ) 
1  s 2  3s 4  3s 6
Table 4 Ln  2  polynomials
dLn 1
N  
Ln  2
d
109
2 4 4
3 3  3  
6 4 2
8
4 6 8  6 6  3 4 12
5 2010  40 8  28 6  8 4   2 18
6 5012  12010  105 8  40 6  6 4 24
7 175  525  615  355  105  15  
14 12 10 8 6 4 2
32
0.577
After we factor h( s 2 ) , we obtain H ( s) 
s  1.31s  1.359s  0.577
3 2
Where the numerator factor, 0.577, is chosen too let the d-c gain be unity. The poles of
H ( s) are s1  0.62; s 2,3  0.345  j 0.901. The amplitude response of third –order
optimum
(L) and Butterworth filters are compared in figure below. Not that the amplitude response of the
optimum filter is not maximally flat, although still monotonic. However, the cut off characteristic of
the optimum filter is shaper than the cutoff of the Butterworth filter.
Figure: Amplitude response of optimum Versus Butterworth filters

Linear phase filters
Suppose a system function is given by H (s)  Ke sT 
Were K is a positive real constant. Then the frequency response of the system can be expressed as
H ( j )  Ke  jT
110
So that the amplitude response M ( ) is a constant K , and the phase response is linear in  .
    T
 sT
The response of such a system to an excitation denoted by R( s)  KE ( s)e
So that the inverse transform r (t ) can be written as r (t )  
1
Rs   Ke(t  T )u(t  T )
We see that the response r (t ) is simply the excitation delayed by a time T, and multiplied by a
constant. Thus no signal distortion results from transmission through a system described by H(s) in
equation () . We note further that the delay T can be obtained by differentiating the phase response
d ( )
   by  ; that is, Dealy   T
d
Consequently, in a system with linear phase, the delay of the system is obtained by differentiating the
phase response    . A system with linear phase and constant amplitude is obviously desirable from
a pulse transmission viewpoint. However, the system function H(s) in Eq () is only realizable in
terms of a lossless transmission line called 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 due to Thomson. We can write H (s) as
K0 K0
H ( s)  
e sT
sinh sT  cosh sT
Where K 0 is chosen such that H (0)  1. Let the delay T be normalized to unity and let us divide both
numerator and denominator of H (s) by sinh s to obtain.
K 0 / sinh s
H ( s)  If sinh s and cosh s are expanded in power series, we have
coth s  1
s2 s4 s6 s3 s5 s7
cosh s  1     ... sinh s  s     ...
2! 4! 6! 3! 5! 7!
From these series expansions, we then obtain a continued fraction expansion of coth s as
1 1
coth s  
s 3 1

s 5 1

s 7
 ...
s
Ko
If the continued fraction is terminated in n terms, then H (s) can be written as H ( s) 
Bn ( s )
Where Bn (s) is Bessel polynomials defined by the formulas
111
Bo  1
B1  s  1

Bn  (2n  1) Bn 1  s 2 Bn 2
From these formulas, we obtain B2  s  3s  3 , B3  s 3  6s 2  15s  15

2
Higher order Bessel polynomials are given in table 5 and the roots of Bessel polynomials are given in
table 6. Note that the roots are all in the left-half plan. The amplitude and phase response of a system
function employing unnormalized third-order Bessel polynomial are given by the solid lines in figure
below.
15
H ( s)  3
s  6s  15s  15
2
Table 5coeffcients of Bessel polynomials

N b0 b1 b2 b3 b4 b5 b6 b7
0 1
1 1 1
2 3 3 1
3 15 15 6 1
4 105 105 45 10 1
5 945 945 420 105 15 1
6 10,395 10,395 4,725 1,260 210 21 1
7 135,135 135,135 62.370 17,325 3,150 378 28 1
These are compared with the amplitude and phase of an un-normalized third-order Butterworth
function given by the dotted lines. Note that the phase response of the constant-delay function is
more liner than the phase of the Butterworth function. Also, the amplitude cutoff of the constant-
delay curve is more gradual than that of the Butterworth.
Table 6 Roots of Bessel Polynomials

n Roots of Bessel Polynomials
1 -1.0+j0
2  1.5  j 0.866025
112
 2.32219  j 0
3 
 1.83891  j1.75438
 2.89621  j 086723
4 
 2.10379  j 265742
 3.64674  j 0
5 
 3.35196  j1.74266
 4.24836  j 0.86751

6  3.73571  j 2.62627
 2.51593  j 4.4967

 4.97179  j0
 4.75829 
 ji.73929
7 
 4.07014  j 3.51717
 2.68568  j 5.42069
Figure: Amplitude response of n=3 Bessel and Butterworth filters.
113
Figure: Phase responses of low-pass filters

6.5 Transient Response of Low –Pass Filters
In this section we will compare the transient response of the filters discussed in section above in
particular, we will compare the step response of the filters according to the following figures of
merit:
1. Rise time t R : The rise time of the step response is defined here as the time required for the step
response to rise from 10% to 90% of its final value as depicted below.
2. Ringing: Ringing is an oscillatory transient occurring in the response of a filter as result of a
sudden change in input (such as a step) .A quantitative measure of the ringing in a step response is
given by its settling time.
3. Settling time: The settling time is that time t s beyond which the step response does not differ from
the final value by more than 2% as depicted in fig. below.
4. Delay time, t D : Delay time is the time which the step response requires to reach 50% of its final
value as shown in fig. below.
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 expressed as a percentage of the final value.
Most of the foregoing figures of merit are related to frequency response particularly bandwidth and
phase linearity. Some of the quantities such as rise time and delay time are intimately related to each
other but have rather tenuous ties with overshoot, Let us examine qualitatively the relationship
between the transient response criteria just cited and frequency response.
114
Figure: figure of merit for step response. t R  Rise time; t s  setting time; t D  delay time
Rise time and bandwidth have an inverse relationship in a filter. The wider the bandwidth, the
smaller the rise time; the narrower the band width, the longer the rise time .physically, the inverse
relationship could be explained by noting that the limited performance of the filter at high
frequencies slows down the abrupt rise in voltage of the step and prolongs the rise time. Thus we
have.
TR  BW  cons tan t
Rise time is a particularly important criterion in pulse transmission. In an article on data transmission
it was shown that in transmitting a pulse of width T1 though a system with adjustable bandwidths, the
following results wee obtained:
Bandwidth ( f   1 / T1) Rise time (milliseconds)
fC 0.5
2 fC 0.25
3 fC 0.16
4 fC 0.12
5 fC 0.10
The table shows a definite inverse relationship between rise time and bandwidth. A definition of time
delay is given by Elmore as the first moment or centroid of the impulse response

TD   t  h(t )dt

Provided the step response has little or no overshoot .Elmore‟s definition or rise time if given as the
second moment.
115
1/ 2
TR  2  (t  TD ) 2 h(t )dt 

  
These definitions are useful because we can obtain rise time and delay time directly from the
coefficients of the system function H (s) without going into the roof, which is in Elemore and sands,
1  a1s  a2 s 2  ...an s n
if H (s) is given as H ( s) 
1  b1s  b2 s 2  .....  bm s m
 
The time delay TD is TD  b1  a1 . And the rise time is TR  (2 b1  a1  2(a2  b2 )
2 2

1/ 2
For the R-C network in Figure below, H (s) is

v( s ) R
H ( s)   , TD  RC , TR  2 RC
I ( s) 1  sRC
R-C network
Is should be emphasized that Elmore‟s definitions are restricted to step response without overshoot
because to the moment definition. The more general definition of rise time is the 10-90% on cited
earlier, which has no formal mathematical definition. Overshoot is generally caused by “excess” gain
at high frequencies .By excess gain we normally mean a magnitude characteristic with a peak such as
the shunt peaked response shown by the dashed curve in fig below. A magnitude characteristic with
no overshoot is the magnitude characteristic of an R-C interstage shown by the solid curve
.
Figure: Comparison of shunt-peaked and simple R-C magnitudes
The step response of the n=3, n=7, and n=10 Butterworth filters are shown in fig. below. Note that
as n increases, the overshoot increase, this is because the higher order Butterworth filters have flatter
magnitude characteristics (i.e. there is more gain at frequencies just below the cutoff).
116
Figure: Step response of normalized Butterworth low-pass filters

Ringing is due to sharp cutoff in the filter magnitude response, and is accentuated by a rising gain
characteristic preceding the discontinuity. The step response of an n=3 Bessel (liner phase) filter is
compared to the response of an n=3 Chebyshev filter with 1- decibel ripple in Fig below. We cannot
compare their rise times since the bandwidths of the two filters have not been adjusted to be equal.
However, we can compare their ringing and settling time‟s .The Chebyshev filter has a sharper
cutoff, and therefore has more ringing and longer settling time than the Bessel filter. Note also the
negligible overshoot of the Bessel filter that is characteristic of the entire class of Bessel filters.
Figure: Comparison of filter transient responses.

The decision as to which filter is best depends upon the particular situation. In certain applications,
such as for transmission of music, phase is not important. In these cases, the sharpness of cutoff may
be the dominate factor so that the Chebyshev or the optimum filter is better than the other s, suppose
we were dealing with a pulse transmission system with the requirements that the output sequence
117
have approximately the same shape as the input sequence, except for a time delay of T  T2  T1 , as
shown in fig “a” below. It is clear that a filter with a long rise time is not suitable, because the pulse
would: smear” over each other.
As seen in fig “b” below, the same can be said for long settling times. Since a pulse transmission
system must have linear phase to insure undistorted harmonic reconstruction at the receiver, the best
filter for the system is a linear phase filter with small rise and settling times.
Figure: smearing of pulses in systems with long rise and settling times
6.6 A Method of 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 is shown in fig below. It is seen that the overshoot is
about 18% we note that after the first peak, the ringing of the step response has an approximate
sinusoidal wave shape. Let us now consider the second derivative of the step response shown by the
dashed curve in below. Beyond the first peak of the step response, the second derivative is also
(approximately)sinusoidal, and is negative when the step response is less than unity if we add the
second derivative to the step response, we reduce the over shoot and ringing.
118
Suppose  (t ) is the step response and H (s) is the system function of the filter. The corrected step
d 2 (t )
response can be written as.  1 (t )   (t )  K Where K is a real, positive constant
dt 2
Taking the Laplace transform, we have
 H ( s)   ( Ks 2  1) H ( s)
1 (t ) 
H ( s)
 Ks 2 
s  s  s
We see that by adding a pair of zeros on the j -axis at s   j / k , the overshoot and ringing are
reduced. For low –pass filters, the factor 1 / k , must in general be greater than the bandwidth of the
system. For normalized Butterworth filters the bandwidth is   1 so that k  1 . The factor k also
controls the amount of overshoot reduction. If K is too small, adding the zeros on the j axis will
have negligible effect.
119
Therefore the zeros should be added some where near the band edge. Figure above shows the effects
of adding zeros at   1 (i.e. right at the band edge) and at  =1.5 we see that the further away the
zeros are placed from the band edge the less effect they will have. The addition of the zeroes will
decrease the 3-decible band width of the filter, however as seen in figure above. Therefore a
compromises must be reached between reduction band width and reduction overshoot.
6.7 A Maximally Flat Delay and Controllable Magnitude Approximation.

In this section we will examine an interesting result, which is due to Budak. The result deals with
linear phase approximation with controllable magnitude. In previous section, we discussed
approximation of a flat delay using Bessel polynomials. The resulting rational approximant was an
all- pole function (all transmission zeros at s   ) whose denominator was a Bessel polynomial.
There was no control of the magnitude using the all –pole approximant. In Budak‟s method the
magnitude is controllable, while the phase is as linear as the standard Bessel approximation.
Budak‟s approximation is obtained by introducing the parameter k to split e  s in to two parts such
s e  ks
that. e  ( k 1) s
0  k 1
e
And then approximate independently e ks , and e( k 1) s with all–pole Bessel polynomial
s
approximations. Thus the resulting approximation for e we have Bessel polynomials for both
numerator and denominator. The poles of the e ( k 1) s approximant will be the zeros in the final
approximate, while the poles of the e  ks approximate remain as poles in the final approximate, For
 ( k 1) s
reliability, the degree of the e approximant should be less than the degree of the e  ks
approximant.
As an example, consider the approximation with three zeros and four poles
120
105
e ks 
(ks)  10(ks)  45(ks) 2  105(ks)  105
4 3
15
e ( k 1) s 
(k  1) s   6(k  1s 2  15(k  1) s   15
3
We then perform the operation as indicated above to obtain

s
e 
 3 2

7 (k  1) s   6(k  1) s   15(k  1) s   15
(ks) 4  10(ks) 3  45(ks) 2  105(ks)  105
In Fig “a” below the magnitude characteristic of the above equation is plotted with k as a parameter.
The phase characteristic is given in terms of deviation of phase from linearity      ( ) and is
shown in figure “b”. The improvement in phase linearity over the all- pole Bessel approximation
k=1 is shown by these curves. Note as the bandwidth is increased (k decreasing) phase linearity is
improved.
Figure below shows the step response of the same equation also with k as a parameter. Since the
effect of decreasing k increases band width, the corresponding effect in the time domain is to
decrease rise time. Budak also observes that as k decrease from unity, the poles and zeros migrate to
keep the phase liner. The zeros move in ward from infinity along radial lines, while the poles move
outward along radial lines.
121
6.8 Synthesis of Low –Pass Filters

Given the system function of the low–pass filter as derived by the methods described in previous
section, we can proceed with the synthesis of the filter network. If we consider the class of filters
terminated in a 1   load, and if we let the system function be transfer impedance,
z 21 y 21
Z 21 ( s)  Or a transfer admittance Y21 ( s) 
1  z 22 1  y 22
We can synthesize the low–pass filter, consider then n=3 Optimum (L) filter function.
Given as a transfer impedance
0.577
Z 21 ( s)  3
s  1.31s  1.359s  0.577
2
We see that the zeros of transmission are all at infinity. Since the numerator of Z21 is even, we divide
both numerator and denominator by the odd part of the denominator s 2+1.359s . Thus we have
0.577 1.31s 2  0.577
z 21  , z 
s 3  1.359s s 3  1.359s
22
The structure of the low –pass filter with three zeros of transmission at infinity is given in chapter 12
we must synthesize z 22 to give the  reactance structure. This we accomplish through the following
continued fraction expansion 1 / z 22 :
122
1.31s 2  0.577 s 3  1.359s (0.763s

s 3  0.440s
0.919s )1.3s 2  0.577(1.45s
1.31s 2
0.577)0.919s (1.593s
0.919s
The optimum filter is shown in figure below for the n= 3 Butterworth filter given by the transfer
impedance.
1
z 21 ( s) 
1
Z ( s ) 
1
We have z ( s ) 
s 3  2s
22
s 3  2s s 3  2s 2  2s  1
21
we then synthesize z22(s) by a continued fraction expansion to give the filter shown in figure below
Figure:   
Figure: Canonical forms for filters described in tables 7, 8, and 9 below.

In table 7, 8 and 9 are listed element values (up to n=7) for single-terminated Butterworth Chebyshev
(1-decible ripple) and Bessel filters, respectively. These apply to the canonical realization for transfer
impedance Z 21 ( s) shown in the immediate above figure. If Y21 ( s) realization is desired, we simply
replace all shunt capacitors by series inductors and vice versa. Some common interstage structures
are shown in figure below. In this figure a structure known as the shunt peaked network is shown.
The transfer impedance of the shunt-peaked network is
1 s  R/ L
Z 21 ( s) 
C s  sR / L  (1 / LC )
2
Table 7 Normalized Element Values for a single Terminated Butterworth filter

N C1 L2 C3 L4 C5 C6 C7
1 1.000
2 0.707 1.414
3 0.500 1,333 1.500
123
4 0.383 1.082 1.577 1.553

5 0.309 0.894 1.382 1.694 1.531
6 0.259 0.758 1.202 1.553 1.759 1.553
7 0222 0.656 1.055 1.397 1.659 1.799 1.558
Table 8 Normalized Element values for single Terminated Chebyshev filter with 1 db Ripple
N C1 L2 C3 L4 C5 C6 C7
1 0.509
2 0.911 0.996
3 1.012 1.333 1.509
4 1.050 1.413 1.909 1.282
5 1.067 1.444 1.994 1.591 1.665
6 1.077 1.460 2.027 1.651 2.049 1.346
7 1.083 1.496 2.044 1.674 2.119 1.649 1.712
Table 9 Normalized Element values for a single Terminated Bessel Filter
N C1 L2 C3 L4 C5 C6 C7
1 1.000
2 0.333 1.000
3 0.167 0.480 0.833
4 0.100 0.290 0.463 0.710
5 0.067 0.195 0.310 0.422 0.623
6 0.048 0.140 0.225 0.301 0.382 0.560
7 0.036 0.106 0.170 0.229 0.283 0.349 0.511
We see that Z 21 ( s) has a real zero and a pair of poles which may be complex depending upon the
values of R, L, & C. In figure “b” below a simple R-C interstage is shown whose transfer impedance
1 1
is Z 21 ( s) 
C S  1 / RC
Observe that all the filter transfer function considered up to this point is made up of pairs of
conjugate poles and simple poles on the j axis. It is clear that if we cascade shunt-peaked stages
and R-C stages, we can 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. for example , if we whish to
design an amplifier with an n =3 low –pass Butterworth characteristic , we first break up the system
function into complex pole pairs and real pole terms ,a s given by
124
1 s 1 1 1
Z 21 ( s)   2
( s  s  1( s  1) s  s  1 s  1 s  1
2
(a) Shunt-peaked interstage (b) R-C interstage

We then associate the individual factors with shunt-peaked or simple R-C stages and solve for the
element values. The n=3 Butterworth amplifier is given in figure below.
Figure: Butterworth Amplifier

6.9 Magnitude and Frequency Normalization
Filters designed with these restrictions are considered to be normalized in both cutoff frequency and
impedance level. we will now discuss methods whereby the normalized filters can be converted in to
filters which meet arbitrary cutoff frequency and impedance level specification. Let us denote by a
subscript n the normalized frequency variable s n and the normalized element values Ln , Rn , and C n ,
The normalized frequency variables s n is related to the actual frequency s by the relation
s
sn  Where 0 the normalizing constant, is dimensionless and is often taken to be the actual
0
cut off frequency. Since the impedance of an element remains invariant under frequency
normalization, we obtain the actual element values from the normalized values by setting the
impedances in the two cases equal to each other. For example, for an inductor, we have;
sn Ln  sL  0 sn L from this equation we then obtain the denormalized value of inductance as
Ln
L
0
125
1
Similarly, from the impedance of a frequency normalized capacitor C n we obtain the
sn Cn
1 1
denormalized value of capacitance thorough the equation. 
s n C n sC
Cn
So that actual value of the capacitance is C 
0
Since resistances, ideally, are independent of frequency, they are unaffected by frequency
normalization.
Consider, next, impedance denormalization. Suppose the actual impedance level should be R0 ohms
instead of 1 then denormalized impedance Z is related to normalized impedance Z n by Z  R0 Z n
Where R0 is taken to be dimensionless here. Thus for a normalized resistor Rn the denormalized
(actual) resistance is R  R0 Rn
For an inductance, the corresponding relationship is sL  R0 sLn 
So that the actual inductance value is L  R0 Ln
1 R
Similarly, for a capacitor we have  0
sC sCn
Cn
So that the actual capacitances is C 
R0
For combined frequency and magnitude denormalization, we simply combine the two sets of
equations to give R  R0 Rn C
Cn
L
R0 Ln
R00 0
Let us consider an actual example in design. In section above, we synthesized transfer impedance
Z 21 with an n  3 Butterworth amplitude characteristics with cutoff frequency of 1 rad/sec and a load
impedance of 1 . Let us redesign this filter for a cutoff frequency of 10 4 rad/sec to work in to a load
of 500  . From the original network in Fig    above, we take the element values and
denormalize with the normalizing factors, 0  104 and R0  500
Then the denormalized element values are:
R  500RL  500
126
1 4 3
(500)
C1  2
 0.1f , L 3
 0.0067h , C2  2
 0.3f
500(10 4 ) 10,000 500(10 4 )
The final design is shown in the next figure:
Figure: Denormalized Low-pass Filter
6.10 Frequency Transformations

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-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 filter, we can generate
any other form of filter. Using frequency transformation, the elements of the normalized low –pass
filter are changed in to elements of a high-pass, band –pass or band –elimination filter.
Analytically, a frequency transformation simply changes on L-C driving–point function in to another
L-C function themselves. Also, since we proceed from normalized low-pass filter, the transformation
equation include built- in frequency denormalization factors so that the resulting networks need only
be scaled for impedance level. Consider the simplest transformation equation that of low–pass to
0
high pass, which is s
sn
Where s n represents the normalized low-pass frequency variable, s is the regular frequency
variable, and  o is the cutoff frequency of the high-pass filter. In terms of real and imaginary parts,
we have
0 0  n  j n 
  j  
 n  j n  n2   n2
Since we are interested principally in how the j n axis maps into the j axis we let  n  0 so
0
that 
n
127
Which is the equation that transforms normalized low-pass filters to denormalized high-pass filters.
From immediate above equation, we see that the point   1corresponds to the point   o it is
also clear that the transformation maps the segment  n  1 on to the segments defined by
0    , as shown in figure below.

Now let us see how the frequency transformations change the network elements. For convenience, let
us denote the normalized low–pass network element with a subscript n, the high–pass element with a
subscript h, the band-pass elements with a subscript b, and the band-elimination elements with a
subscript e. For the low –pass to high-pass case; let us first consider the changes for the capacitor C n
Figure: Low-pass to high-pass Transformation

The transformation is given by the equation.
1 s
Ln , we have 0 1
 Lh s For the inductor Ln sn  Ln 
Cn sn oCn s Ch s
We have observed that a capacitor changes in to an inductor and an inductor changes in to a capacitor
in a low-pass to high-pass transformation (figure below). The element values of the high-pass filter
are given in terms of the normalized low –pass filter elements as
1 1
Lh  And Ch 
0 Cn 0 Ln
Consider the following example. From the normalized third –order Butterworth filter given in Fig
  , let us design a corresponding high pass filter with its cutoff frequency  o  10 6 rad / sec
and the impedance level of 500  .
128
Figure: Element changes resulting from frequency transformations
Figure: transformation of low-pass filter in figure   into high –pass filter.

From the low-pass filter, we can draw by inspection the high –pass –filter circuit shown in figure
above. Its element values are:
500 1 500
RL  500 , L1h  6 1
 10 3 h , Cn  6 4
 1.5  10 9 f , L2 h  6 2
 0.333  10 3 h
10 ( 2 ) (500)10 ( 3 ) 10 ( 3 )
Next, let us examine the low –pass to band –pass transformation (also an L-C function):
0  s 0 
sn    
BW  0 s 
Where, if C 2 and C1 denote the upper and lower cutoff frequencies of the band –pass filter, BW is
the bandwidth BW  C 2  C1
And  0 is the geometric mean of C 2 and C1  0   C 2  C1
129
The low-pass to band –pass transformation map the segment  n  1 to the segments
 C 2     C1 shown in figure below. The normalized low-pass elements are then modified
according to the following equations.
Ln 02 Ln  1
Ln sn  s  Lb1s 
BW BWs Cb1s
Figure: Low-pass to band-pass transformation

We not the inductor Ln is transformed in to a series-tuned tank, shown in fig. above, whose elements
are given as
Ln Cn
Lb1  , Cb1 
BW 02 Ln
The capacitor C n is transformed into a parallel –tuned tank (figure above) whose elements are
BW Cn
Lb 2  , Cb 2 
o 2Cn BW
Let us transform the third-order Butterworth low-pass filter in figure   in to a band –pass
filter with a 1 impedance level, whose bandwidth is , BW  6 x104rad / sec and its band –pass is
“centered” at o =4x104rad/sec . We draw the band-pass filter shown in figure below by the rule
given above.
130
Figure: Band-pass filter transformed from low-pass filter in figure   

The element values of the band-pass filter are given in the following equation.
6 x10 4 4
1
1 4
L1   o.75  10  h , C 1 2
  10 4 f , L2  3
 92  10 4 h
4 10  (
4 2 1
2 ) 6  10 4
12 6  10 4
6  10 4 9 6  10 4 3
C 2   10 4 f , L3   0.25  10 4 h , C 3 2
 0.25  10 4 f
(4  10 ) ( 2 ) 32
4 2 3
(4  10 4) 2( 2 )
3
6  10 4
BW
Finally, the band –elimination filter is obtained through the transformation s n   s 0 
 0   
0 s 
Where BW and 0 are defined in a manner similar to that for the band –pass filter .The
transformation maps the segment of the jn  axis in figure “a” below onto the segments shown on
the j  axis in figure “b” below.
Figure: Low-pass to band-elimination transformation

For the low –pass to band-elimination transformation we, therefore, have the following element
changes.
1  1 1 0  s 0   1
Ln sn        Le 2 s 
( s / Ln BW )  (02 / L n BWs ) Ce1 s  (1/ Le1 s) Cn sn Cn BW  0 s  Ce 2 s
Observe that the normalized low-pass inductor goes in to a parallel tuned circuit and the capacitor Cn
goes into a series –tuned circuit.
Table 6.10: Table of various Frequency Transformations s
Transformation Low –pass to Equation
131

High –pass sn 
s
  s  o 
sn    
BW   o s 
Band-pass
BW
Band –elimination sn 
 s o 
 o   
 o s 
6.11 Active Low Pass Filter

Limitations of passive filters:
1. At low frequency (approximately 100Hz to few hundred kHz) increases the size and weight
of inductors for better device value, which deteriorated sharpness at cut off frequency.
2. Impossibility of integrating inductors with practical value of inductances in the micro-
miniaturization circuit element.
3. Necessity of buffer (isolation) amplifiers to prevent loading while cascading sections of
filters.
4. Need external amplifier to adjust the required gain.
Merits are: They are stable, non frequency limited and operate under unlimited power
Active filters are networks which use active device together with RC elements. In the Passive Filter
discussion, we saw how a basic first-order filter circuits, such as the low pass and the high pass filters
can be made using just a single resistor in series with a non-polarized capacitor connected across a
sinusoidal input signal. We also noticed that the main disadvantage of passive filters is that the
amplitude of the output signal is less than that of the input signal, i.e., the gain is never greater than 1
and that the load impedance affects the filters characteristics. With passive filter circuits containing
multiple stages, this loss in signal amplitude called "Attenuation" can become quiet severe. One way
of restoring or controlling this loss of signal is by using amplification through the use of Active
Filters.
As their name implies, Active Filters contain active components such as operational amplifiers,
transistors or FET's within their circuit design. They draw their power from an external power source
and use it to boost or amplify the output signal. Filter amplification can also be used to either shape
or alter the frequency response of the filter circuit by producing a more selective output response,
making the output bandwidth of the filter narrower or even wider.
An active filter generally uses an operational amplifier (op-amp) within its design and in the
Operational Amplifier we can realize that as an Op-amp has high input impedance, low output
132
impedance and a voltage gain determined by the resistor network within its feedback loop. Unlike a
passive high pass filter which has in theory an infinite high frequency response, the maximum
frequency response of an active filter is limited to the Gain/Bandwidth product (or open loop gain) of
the operational amplifier being used. Still, active filters are generally easier to design than passive
filters; they produce good performance characteristics, very good accuracy with a steep roll-off and
low noise when used with a good circuit design.
Active Low Pass Filter
The most common and easily understood active filter is the active Low Pass Filter. Its principle of
operation and frequency response is exactly the same as those for the previously seen passive filter;
the only difference this time is that it uses an op-amp for amplification and gain control. The simplest
form of a low pass active filter is to connect an inverting or non-inverting amplifier, similar to the
basic RC low pass filter circuit as shown.
First Order Active Low Pass Filter
This first-order low pass active filter consists simply of a passive RC filter stage providing a low
frequency path to the input of a non-inverting operational amplifier. The amplifier is configured as a
voltage-follower(Buffer) giving it a DC gain of one, Av = +1 or unity gain as opposed to the
previous passive RC filter which has a DC gain of less than unity. The advantage of this
configuration is that the op-amps high input impedance prevents excessive loading on the filters
output while its low output impedance prevents the filters cut-off frequency point from being
affected by changes in the impedance of the load. While this configuration provides good stability to
the filter, its main disadvantage is that it has no voltage gain above one. However, although the
voltage gain is unity the power gain is very high as its output impedance is much lower than its input
impedance. If a voltage gain greater than one is required we can use the following filter circuit.
133
Active Low Pass Filter with Amplification
The frequency response of the circuit will be the same as that for the passive RC filter, except that
the amplitude of the output is increased by the pass band gain, AF of the amplifier. For a non-
inverting amplifier circuit, the magnitude of the voltage gain for the filter is given as a function of the
feedback resistor (R2) divided by its corresponding input resistor (R1) value and is given as:
R1
DC  gain 
R1  R2
Therefore, the gain of an active low pass filter as a function of frequency will be:
Gain of a first-order low pass filter
Vout AF
Voltage  gain, ( Av )   Where:
Vin  f 
2
1   
 fC 
 AF = the pass band gain of the filter, (1  R2 / R1 )

 f = the frequency of the input signal in Hertz, (Hz)
 f C = the cut-off frequency in Hertz, (Hz)
Thus, the operation of a low pass active filter can be verified from the frequency gain equation above
as:
Vout
1. At very low frequencies, f  fC ,  AF 2. At the cut-off frequency, f  fC ,
Vin
Vout AF
  0.707 AF
Vin 2
Vout
3. At very high frequencies, f  f C ,  AF
Vin
134
Thus, the Active Low Pass Filter has a constant gain AF from 0Hz to the high frequency cut-off
point, f C . At f C the gain is 0.707 AF , and after f C it decreases at a constant rate as the frequency
increases. That is, when the frequency is increased tenfold (one decade), the voltage gain is divided
by 10. In other words, the gain decreases 20dB (= 20log 10) each time the frequency is increased by
10. When dealing with filter circuits the magnitude of the pass band gain of the circuit is generally
expressed in decibels or dB as a function of the voltage gain and this is defined as:
Magnitude of Voltage Gain in (dB)
V   2Vout 
Av dB   20 log10  out   3dB  20 log10  

 Vin   Vin 
Example1: Design a non-inverting active low pass filter circuit that has a gain of ten at low
frequencies, a high frequency cut-off or corner frequency of 159Hz and an input impedance of
10KΩ.
R
The voltage gain of a non-inverting operational amplifier is given as: AF  1  2  10
R1
Assume a value for resistor R1 of 1kΩ rearranging the formula above gives a value for R2 of
R2  10  1  R1  9  1k  9k Then, for a voltage gain of 10, R1 = 1kΩ and R2 = 9kΩ. However,
a 9kΩ resistor does not exist so the next preferred value of 9k1Ω is used instead.
Converting this voltage gain to a decibel dB value gives: Gain in dB 20 log A  20 log 10  20dB
The cut-off or corner frequency ( f C ) is given as being 159Hz with an input impedance of 10kΩ.
This cut-off frequency can be found by using the formula:
1
fC  Hz Where f C = 159Hz and R = 10kΩ. Then, by rearranging the above formula we can
2RC
1 1
find the value for capacitor C as: C   100nF
2Rf C 2  159  10k
Then the final circuit along with its frequency response is given below as:
135
Low Pass Filter Circuit
Frequency Response Curve
If the external impedance connected to the input of the circuit changes, this change will also affect
the corner frequency of the filter (components connected in series or parallel). One way of avoiding
this is to place the capacitor in parallel with the feedback resistor R2. The value of the capacitor will
change slightly from being 100nF to 110nF to take account of the 9k1Ω resistor and the formula used
to calculate the cut-off corner frequency is the same as that used for the RC passive low pass filter.
1
fC  Hz An example of the new Active Low Pass Filter circuit is given as.
2R2 C
136
Simplified non-inverting amplifier filter circuit Equivalent inverting amplifier filter

circuit
Applications of Active Low Pass Filters are in audio amplifiers, equalizers or speaker systems to
direct the lower frequency bass signals to the larger bass speakers or to reduce any high frequency
noise or "hiss" type distortion. When used like this in audio applications the active low pass filter is
sometimes called a "Bass Boost" filter.
Second-order Low Pass Active Filter
As with the passive filter, a first-order low pass active filter can be converted into a second-order low
pass filter simply by using an additional RC network in the input path. The frequency response of the
second-order low pass filter is identical to that of the first-order type except that the stop band roll-
off will be twice the first-order filters at 40dB/decade (12dB/octave). Therefore, the design steps
required of the second-order active low pass filter are the same.
137
Second-order Active Low Pass Filter Circuit
When cascading together filter circuits to form higher-order filters, the overall gain of the filter is
equal to the product of each stage. For example, the gain of one stage may be 10 and the gain of the
second stage may be 32 and the gain of a third stage may be 100. Then the overall gain will be
32,000, (10 x 32 x 100) as shown below.
Cascading Voltage Gain
AV  AV1  AV2  AV3  AV  10  32  100  32,000  AV dB   20 log10 32,000
AV dB   90dB Or 90dB  20dB  30dB  40dB

Second-order (two-pole) active filters are important because higher-order filters can be designed
using them. By cascading together first and second-order filters, filters with an order value, either
odd or even up to any value can be constructed. In the next tutorial about filters, we will see that
Active High Pass Filters, can be constructed by reversing the positions of the resistor and capacitor
in the circuit.
6.12 Active High Pass Filter

The basic operation of an Active High Pass Filter (HPF) is exactly the same as that for its
equivalent RC passive filter circuit, except that this type of circuit has an operational amplifier or op-
amp included within its design for amplification and gain control. Like the previous active low pass
filter circuit, the simplest form of an active high pass filter is to connect a standard inverting or non-
inverting operational amplifier to the basic RC high pass passive filter circuit as shown.
First Order Active High Pass Filter
138
Technically, there is no such thing as an active high pass filter. Unlike Passive High Pass Filters
which have an "infinite" frequency response, the maximum pass band frequency response of an
active high pass filter is limited by the characteristics or bandwidth of the operational amplifier being
used, making them appear as if they are band pass filters with a high frequency cut-off determined by
the selection of op-amp and gain.
In the Operational Amplifier we saw that the maximum frequency response of an op-amp is limited
to the Gain/Bandwidth product or open loop voltage gain ( AV ) of the operational amplifier being
used giving it a bandwidth limitation, where the closed loop response of the op amp intersects the
open loop response. A commonly available operational amplifier such as the uA741 has a typical
"open-loop" (without any feedback) DC voltage gain of about 100dB maximum reducing at a roll off
rate of -20dB/Decade (-6db/Octave) as the input frequency increases. The gain of the uA741 reduces
until it reaches unity gain, (0dB) or its "transition frequency" ( Ft ) which is about 1MHz. This causes
the op-amp to have a frequency response curve very similar to that of a first-order low pass filter and
this is shown below.
Frequency response curve of a typical Operational Amplifier
Then the filters performance at high frequencies is limited by this unity gain crossover frequency
which determines the overall bandwidth of the open-loop amplifier. The gain-bandwidth product of
the op-amp starts from around 100 kHz for small signal amplifiers up to about 1GHz for high-speed
139
digital video amplifiers and op-amp based active filters can achieve very good accuracy and
performance provided that low tolerance resistors and capacitors are used. Under normal
circumstances the maximum pass band required for a closed loop active high pass or band pass filter
is well below that of the maximum open-loop transition frequency. However, when designing active
filter circuits it is important to choose the correct op-amp for the circuit as the loss of high frequency
signals may result in signal distortion.
Active High Pass Filter

A first-order (single-pole) Active High Pass Filter as its name implies, attenuates low
frequencies and passes high frequency signals. It consists simply of a passive filter section
followed by a non-inverting operational amplifier. The frequency response of the circuit is the
same as that of the passive filter, except that the amplitude of the signal is increased by the gain
of the amplifier and for a non-inverting amplifier the value of the pass band voltage gain is
given as 1 + R2/R1, the same as for the low pass filter circuit.
Active High Pass Filter with Amplification
This first-Order high pass filter, consists simply of a passive filter followed by a non-inverting
amplifier. The frequency response of the circuit is the same as that of the passive filter, except that
the amplitude of the signal is increased by the gain of the amplifier. For a non-inverting amplifier
circuit, the magnitude of the voltage gain for the filter is given as a function of the feedback resistor
(R2) divided by its corresponding input resistor (R1) value and is given as:
Gain for an Active High Pass Filter

 f 
AF  
Voltage  gain, ( Av ) 
Vout
  fC 
Where:
Vin  f 
2
1   
 fC 
 AF = the pass band gain of the filter, (1  R2 / R1 )

 f = the frequency of the input signal in Hertz, (Hz)
140
 f C = the cut-off frequency in Hertz, (Hz)
Just like the low pass filter, the operation of a high pass active filter can be verified from the
frequency gain equation above as:
Vout
1. At very low frequencies, f  f C ,  AF
Vin
Vout AF
2. At the cut-off frequency, f  f C ,   0.707 AF
Vin 2
Vout
3. At very high frequencies, f  f C ,  AF
Vin
Then, the Active High Pass Filter has a gain AF that increases from 0Hz to the low frequency cut-
off point, f C at 20dB/decade as the frequency increases. At f C the gain is 0.707 AF , and after f C all
frequencies are pass band frequencies so the filter has a constant gain AF with the highest frequency
being determined by the closed loop bandwidth of the op-amp. When dealing with filter circuits the
magnitude of the pass band gain of the circuit is generally expressed in decibels or dB as a function
of the voltage gain and this is defined as:
Magnitude of Voltage Gain in (dB)
V   2Vout 
Av dB   20 log10  out   3dB  20 log10  

 in 
V V
 in 
For a first-order filter the frequency response curve of the filter increases by 20dB/decade or
6dB/octave up to the determined cut-off frequency point which is always at -3dB below the
maximum gain value. As with the previous filter circuits, the lower cut-off or corner frequency ( f C )
1
can be found by using the same formula: f C  Hz
2RC
The corresponding phase angle or phase shift of the output signal is the same as that given for the
passive RC filter and leads that of the input signal. It is equal to +45o at the cut-off frequency f C
value and is given as:
 1 
Phase Shift   tan 1  
 2fRC 
A simple first-order active high pass filter can also be made using an inverting operational amplifier
configuration as well, and an example of this circuit design is given along with its corresponding
frequency response curve. A gain of 40dB has been assumed for the circuit.
141
Inverting Operational Amplifier Circuit
Frequency Response Curve
Example: A first order active high pass filter has a pass band gain of two and a cut-off corner
frequency of 1 kHz. If the input capacitor has a value of 10nF, calculate the value of the cut-off
frequency determining resistor and the gain resistors in the feedback network. Also, plot the expected
frequency response of the filter.
With a cut-off corner frequency given as 1 kHz and a capacitor of 10nF, the value of R will therefore
be:
1 1
R   15.92k
2f C C 2  1000  10  10 9
R2 R R
The pass band gain of the filter, AF is given as being, 2. AF  1  ,  2  1  2 And 2  1
R1 R1 R1
As the value of resistor, R2 divided by resistor, R1 gives a value of one. Then, resistor R1 must be
equal to resistor R2, since the pass band gain, AF  2 . We can therefore select a suitable value for the
142
two resistors of say, 10kΩ's each for both feedback resistors. So for a high pass filter with a cut-off
corner frequency of 1 kHz, the values of R and C will be, 10kΩ's and 10nF respectively. The values
of the two feedback resistors to produce a pass band gain of two are given as: R 1 = R2 = 10kΩ's
The data for the frequency response bode plot can be obtained by substituting the values obtained
above over a frequency range from 100Hz to 100 kHz into the equation for voltage gain:
 f 
AF  
Voltage  gain, ( Av ) 
Vout
  fC 
This then will give us the following table
Vin  f 
2
1   
 fC 
of data.
Frequency, ƒ Voltage Gain Gain, (dB) Frequency, ƒ Voltage Gain Gain, (dB)
( Hz ) ( Vo / Vin ) 20log( Vo / Vin ) ( Hz ) ( Vo / Vin ) 20log( Vo / Vin )
100 0.20 -14.02 3,000 1.90 5.56
200 0.39 -8.13 5,000 1.96 5.85
500 0.89 -0.97 10,000 1.99 5.98
800 1.25 1.93 50,000 2.00 6.02
1,000 1.41 3.01 100,000 2.00 6.02
The frequency response data from the table above can now be plotted as shown below. In the stop
band (from 100Hz to 1 kHz), the gain increases at a rate of 20dB/decade. However, in the pass band
after the cut-off frequency, f C = 1 kHz, the gain remains constant at 6.02dB. The upper-frequency
limit of the pass band is determined by the open loop bandwidth of the operational amplifier used as
we discussed earlier. Then the bode plot of the filter circuit will look like this.
The Frequency Response Bode-plot for our example
Applications of Active High Pass Filters are in audio amplifiers, equalizers or speaker systems to
direct the high frequency signals to the smaller tweeter speakers or to reduce any low frequency
143
noise or "rumble" type distortion. When used like this in audio applications the active high pass filter
is sometimes called a "Treble Boost" filter.
Second-order High Pass Active Filter
As with the passive filter, a first-order high pass active filter can be converted into a second-order
high pass filter simply by using an additional RC network in the input path. The frequency response
of the second-order high pass filter is identical to that of the first-order type except that the stop band
roll-off will be twice the first-order filters at 40dB/decade (12dB/octave). Therefore, the design steps
required of the second-order active high pass filter are the same.
Second-order Active High Pass Filter Circuit
Higher-order high pass filters, such as third, fourth, fifth, etc are formed simply by cascading
together first and second-order filters. For example, a third order high pass filter is formed by
cascading in series first and second order filters, a fourth-order high pass filter by cascading two
second-order filters together and so on. Then an Active High Pass Filter with an even order number
will consist of only second-order filters, while an odd order number will start with a first-order filter
at the beginning as shown.
144
Cascading Active High Pass Filters
Although there is no limit to the order of a filter that can be formed, as the order of the filter
increases so to does its size. Also, its accuracy declines, that is the difference between the actual stop
band response and the theoretical stop band response also increases. If the frequency determining
resistors are all equal, R1 = R2 = R3 etc, and the frequency determining capacitors are all equal, C1
= C2 = C3 etc, then the cut-off frequency for any order of filter will be exactly the same. However,
the overall gain of the higher-order filter is fixed because all the frequency determining components
are equal. In the next tutorial about filters, we will see that Active Band Pass Filters, can be
constructed by cascading together a high pass and a low pass filter.
6.13 Active Band Pass Filter

As we saw previously in the Passive Band Pass Filter tutorial, the principal characteristic of a Band
Pass Filter or any filter for that matter is its ability to pass frequencies relatively unattenuated over a
specified band or spread of frequencies called the "Pass Band". For a low pass filter this pass band
starts from 0Hz or DC and continues up to the specified cut-off frequency point at -3dB down from
the maximum pass band gain. Equally, for a high pass filter the pass band starts from this -3dB cut-
off frequency and continues up to infinity or the maximum open loop gain for an active filter.
However, the Active Band Pass Filter is slightly different in that it is a frequency selective filter
circuit used in electronic systems to separate a signal at one particular frequency, or a range of
signals that lie within a certain "band" of frequencies from signals at all other frequencies. This band
or range of frequencies is set between two cut-off or corner frequency points labeled the "lower
frequency" (ƒL) and the "higher frequency" (ƒH) while attenuating any signals outside of these two
points. Simple Active Band Pass Filter can be easily made by cascading together a single Low Pass
Filter with a single High Pass Filter as shown.
145
The cut-off or corner frequency of the low pass filter (LPF) is higher than the cut-off frequency of
the high pass filter (HPF) and the difference between the frequencies at the -3dB point will determine
the "bandwidth" of the band pass filter while attenuating any signals outside of these points. One way
of making a very simple Active Band Pass Filter is to connect the basic passive high and low pass
filters we look at previously to an amplifying op-amp circuit as shown.
Active Band Pass Filter
This cascading together of the individual low and high pass passive filters produces a low "Q-factor"
type filter circuit which has a wide pass band. The first stage of the filter will be the high pass stage
that uses the capacitor to block any DC biasing from the source. This design has the advantage of
producing a relatively flat asymmetrical pass band frequency response with one half representing the
low pass response and the other half representing high pass response as shown.
The higher corner point (ƒH) as well as the lower corner frequency cut-off point (ƒL) are calculated
the same as before in the standard first-order low and high pass filter circuits. Obviously, a
reasonable separation is required between the two cut-off points to prevent any interaction between
the low pass and high pass stages. The amplifier provides isolation between the two stages and
defines the overall voltage gain of the circuit. The bandwidth of the filter is therefore the difference
between these upper and lower -3dB points. For example, if the -3dB cut-off points are at 200Hz and
600Hz then the bandwidth of the filter would be given as: Bandwidth (BW) = 600 - 200 = 400Hz.
The normalized frequency response and phase shift for an active band pass filter will be as follows.
146
Active Band Pass Frequency Response
While the above passive tuned filter circuit will work as a band pass filter, the pass band (bandwidth)
can be quite wide and this may be a problem if we want to isolate a small band of frequencies. Active
band pass filter can also be made using inverting operational amplifiers, and by rearranging the
positions of the resistors and capacitors within the circuit we can produce a much better filter circuit
as shown below. The lower cut-off -3dB point is given by ƒC2 while the upper cut-off -3dB point is
given by ƒC1.
Inverting Band Pass Filter Circuit
147
 R2 1 1
Voltage  gain  , f C1  , f C2 
R1 2R1C1 2R2C2
This type of band pass filter is designed to have a much narrower pass band. The centre frequency
and bandwidth of the filter is related to the values of R1 , R 2 , C1 and C 2 . The output of the filter is
again taken from the output of the op-amp.
Multiple Feedback Band Pass Active Filter
We can improve the band pass response of the above circuit by rearranging the components again to
produce an infinite-gain multiple-feedback (IGMF) band pass filter. This type of active band pass
design produces a "tuned" circuit based around a negative feedback active filter giving it a high "Q-
factor" (up to 25) amplitude response and steep roll-off on either side of its centre frequency.
Because the frequency response of the circuit is similar to a resonance circuit, this centre frequency
is referred to as the resonant frequency, (ƒr). Consider the circuit below.
Infinite Gain Multiple Feedback Active Filter
This band pass filter circuit uses the full gain of the operational amplifier, with multiple negative
feedbacks applied via resistor, R2 and capacitor C2. Then we can define the characteristics of the
IGMF filter as follows:
fr 1 R2
fr 
1 QBP   Maximum Gain,
2 R1 R2 C1C 2 BW(3dB) 2 R1
R2
AV    2Q 2
2 R1
We can see then that the relationship between resistors, R1 and R2 determines the band pass "Q-
factor" and the frequency at which the maximum amplitude occurs, the gain of the circuit will be
equal to  2Q 2 . Then as the gain increases so to does the selectivity. In other words, high-gain and
high-selectivity
148
Example: An active band pass filter that has a gain Av of one and a resonant frequency, f r of 1
kHz is constructed using an infinite gain multiple feedback filter circuit. Calculate the values of the
components required to implement the circuit.
Firstly, we can determine the values of the two resistors, R 1 and R2 required for the filter using the
gain of the circuit to find Q as follows.
2
 
1 1 R2 R2  0.7071 
AV  1  2Q 2 QBP   0.7071 Q  0.7071     2
2 2 R1 R1  1 
 
 2 
Then we can see that a value of Q = 0.7071 gives a relationship of resistor, R2 being twice the value
of resistor R1. Then we can choose any suitable value of resistances to give the required ratio of two.
Then resistor R1 = 10kΩ and R2 = 20kΩ. The centre or resonant frequency is given as 1 kHz. Using
the new resistor values obtained, we can determine the value of the capacitors required assuming that
C = C1 = C2.
1
f r  1000 Hz 
2C R1 R2
1 1
C    11.2nF
2f r R1 R2 2 1000 10,000  20,000
The closest standard value is 10nF.
Resonant Frequency
The actual shape of the frequency response curve for any passive or active band pass filter will
depend upon the characteristics of the filter circuit with the curve above being defined as an "ideal"
band pass response. An active band pass filter is a 2nd Order type filter because it has "two"
reactive components (two capacitors) within its circuit design and will have a peak response or
Resonant Frequency ( f r ) at its "centre frequency", f C . The centre frequency is generally
calculated as being the geometric mean of the two -3dB frequencies between the upper and the lower
cut-off points with the resonant frequency (point of oscillation) being given as:
fr  fL  fH Where:
 f r is the resonant or Centre Frequency

 f L is the lower -3dB cut-off frequency point
 f H is the upper -3db cut-off frequency point and in our simple example above the resonant
centre frequency is given as: f r  200  600  120,000  346HZ
149
The "Q" or Quality Factor
In a Band Pass Filter circuit, the overall width of the actual pass band between the upper and lower -
3dB corner points of the filter determines the Quality Factor or Q-point of the circuit. This Q Factor
is a measure of how "Selective" or "Un-selective" the band pass filter is towards a given spread of
frequencies. The lower the value of the Q factor the wider is the bandwidth of the filter and
consequently the higher the Q factors the narrower and more "selective" is the filter. The Quality
Factor, Q of the filter is sometimes given the Greek symbol of Alpha, (α) and is known as the alpha-
1
peak frequency where: 
Q
As the quality factor of a band pass filter (Second-order System) relates to the "sharpness" of the
filters response around its centre resonant frequency ( f r ) it can also be thought of as the Damping
Factor or Damping Coefficient because the more damping the filter has the flatter is its response and
likewise, the less damping the filter has the sharper is its response. The damping ratio is given the
Greek symbol of Xi,   where:   2

The "Q" of a band pass filter is the ratio of the Resonant Frequency, ( f r ) to the Bandwidth, (BW)
between the upper and lower -3dB frequencies and is given as:
Re sonant  Frequency
Q
Bandwidth
Then for our simple example above the quality factor "Q" of the band pass filter is given as:
346Hz / 400Hz = 0.865. Note that Q is a ratio and has no units.
When analysing active filters, generally a normalized circuit is considered which produces an "ideal"
frequency response having a rectangular shape, and a transition between the pass band and the stop
band that has an abrupt or very steep roll-off slope. However, these ideal responses are not possible
in the real world so we use approximations to give us the best frequency response possible for the
type of filter we are trying to design. Probably the best known filter approximation for doing this is
150
the Butterworth or maximally-flat response. In the next tutorial we will look at higher order filters
and use Butterworth approximations to produce filters that have a frequency response which is as flat
as mathematically possible in the pass band and a smooth transition or roll-off rate.
151

Network Handout

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Network Handout

Uploaded by

Copyright:

Available Formats

AS&TU, EE Department Network Analysis & Synthesis – EEng 3209

1. Introduction to Network Analysis & Synthesis

Fig.1.1. Key terms in network analysis & synthesis

i. Mathematical modeling of electromagnetic phenomena

1.1.Lumped Circuit elements

Fig.1.2. Linear resistor (a)characteristic curve (b) Circuit symbol

Fig.1.3. Non-linear characteristic curves

Fig.1.4. Piece-wise linearization of characteristic curve

Fig.1.5. Linear capacitor (a) characteristic curve (b) Circuit symbol

The current charge relation being i  dq dt , for a linear capacitor, we have

For a voltage controlled capacitor the current through the capacitor is

Fig.1.6. Non-linear capacitor characteristic curve

The instantaneous power entering an element is given by

For a linear capacitor q  Cv  it reduces to

Fig.1.7. Linear inductor (a) characteristic curve (b) Circuit symbol

Independent sources: Fig.1.8 shows an arbitrary network N excited by a source of electric

Fig.1.5. Independent source

Fig.1.9. Voltage source:.symbol and the i-v characteristic curve

A two-terminal element is called an ideal independent current source, if it is capable of supplying a

Fig.1.10. Voltage source:.symbol and the i-v characteristic curve

1.2. Classification of Networks

Linear and Nonlinear Networks

Fig.1.11. Linear and Non-linear networks

but not for N 2 .

Linearity of a network can be defined as follow: let a network be characterized by F ( xi )  yi where

Time-invariant and Time-variant Networks

Passive and Active Networks

quadrants of the v  q(  i) plane.

2. Network Transform Representation and Analysis

A concept of transform impedance and transform admittance is discussed in this chapter.

2.2. Network Functions

where s the Laplace variable is the complex frequency variable .

Fig.2.1. Network used to define network functions

Fig.2.2. Ladder network

2.3. Poles and Zeros

Fig.2.4. pole-zero plot

2.4. Stability of a Network

Relationship between impulse response and stability

Relationship between Poles positions and stability

Routh Hurwitz stability criterion

Special cases when Routh’s tabulation terminates prematurely

Deficiencies of Routh Hurwitz Criterion

1. Obtain the pole zero location for the function

3. Elements of Realizability Theory

3.1. Causality and Stability

condition for an amplitude function H ( j ) to be realizable (causal) is that

2. The square magnitude function H ( j ) must be integrable that is

Fig. 3.2: Ideal filter characteristic

Fig 3.3: Gaussian Filter charactersitc

less than a very small quantity  , that is h(t  t d )   t  0 .

Equation 3.7  requires that the impulse response be absolutely integrable, or

a n s n  a n1 s n1  a n2 s n2  ....  a1 s  a0

2. H (s) cannot have multiple poles on the j  axis.

3.2. Hurwitz Polynomials

As a result of this conditions, if P(s) is a Hurwitz polynomial given by

P(s)  an s n  an1s n1      a1s  a0 3.11

Hurwitz polynomials have the following properties:

s   i  ji where  i ,  i & i real and positive.

The polynomial P(s) which contains these roots can be written as

implies cancellation brought about by a root s   i with a positive real part.

F (s)  W (s) F1 (s) 3.17

So that the continued fraction expansion of  (s) is