1

General Characteristics
of Filters

1.1 Introduction ................................................................................ 1-1

1.2 Characterization ......................................................................... 1-3
Laplace Transform . Transfer Function
1.3 Time-Domain Response........................................................... 1-6
General Inversion Formula . Inverse by Using Partial Fractions .

Impulse and Step Responses . Overshoot, Delay Time,

and Rise Time
1.4 Frequency-Domain Analysis ................................................. 1-10
Sinusoidal Response . Graphical Construction . Loss Function
1.5 Ideal and Practical Filters....................................................... 1-15
1.6 Amplitude and Delay Distortion.......................................... 1-16
1.7 Minimum-Phase, Nonminimum-Phase,
and Allpass Filters ................................................................... 1-17
Minimum-Phase Filters . Allpass Filters . Decomposition
of Nonminimum-Phase Transfer Functions
1.8 Introduction to the Design Process ..................................... 1-21
The Approximation Step . The Realization Step . Study
of Imperfections . Implementation
1.9 Introduction to Realization.................................................... 1-23
Passive Filters . Active Filters . Biquads . Types of Basic
Andreas Antoniou Filter Sections
University of Victoria References ............................................................................................ 1-29

1.1 Introduction
An electrical filter is a system that can be used to modify, reshape, or manipulate the frequency spectrum
of an electrical signal according to some prescribed requirements. For example, a filter may be used to
amplify or attenuate a range of frequency components, reject or isolate one specific frequency compon-
ent, and so on. The applications of electrical filters are numerous, for example,
. To eliminate signal contamination such as noise in communication systems
. To separate relevant from irrelevant frequency components
. To detect signals in radios and TV’s
. To demodulate signals
. To bandlimit signals before sampling
. To convert sampled signals into continuous-time signals
. To improve the quality of audio equipment, e.g., loudspeakers
. In time-division to frequency-division multiplex systems

1-1
1-2 Passive, Active, and Digital Filters

. In speech synthesis
. In the equalization of transmission lines and cables
. In the design of artificial cochleas
Typically, an electrical filter receives an input signal or excitation and produces an output signal or
response. The frequency spectrum of the output signal is related to that of the input by some rule of
correspondence. Depending on the type of input, output, and internal operating signals, three general
types of filters can be identified, namely, continuous-time, sampled-data, and discrete-time filters.
A continuous-time signal is one that is defined at each and every instant of time. It can be represented
by a function x(t) whose domain is a range of numbers (t1, t2), where
1  t1 and t2  1. A sampled-
data or impulse-modulated signal is one that is defined in terms of an infinite summation of continuous-
time impulses (see Ref. [1, Chapter 6]). It can be represented by a function

X
1
^x(t) ¼ x(nT)d(t
nT)
n¼
1

where d(t) is the impulse function. The value of the signal at any instant in the range nT < t < (n þ 1)T is
zero. The frequency spectrum of a continuous-time or sampled-data signal is given by the Fourier
transform.*
A discrete-time signal is one that is defined at discrete instants of time. It can be represented by a
function x(nT), where T is a constant and n is an integer in the range (n1, n2) such that
1  n1 and
n2  1. The value of the signal at any instant in the range nT < t < (n þ 1)T can be zero, constant, or
undefined depending on the application. The frequency spectrum in this case is obtained by evaluating
the z transform on the unit circle jzj ¼ 1 of the z plane.
Depending on the format of the input, output, and internal operating signals, filters can be classified
either as analog or digital filters. In analog filters the operating signals are varying voltages and currents,
whereas in digital filters they are encoded in some binary format. Continuous-time and sampled-data
filters are always analog filters. However, discrete-time filters can be analog or digital.
Analog filters can be classified on the basis of their constituent components as
. Passive RLC filters
. Crystal filters
. Mechanical filters
. Microwave filters
. Active RC filters
. Switched-capacitor filters
Passive RLC filters comprise resistors, inductors, and capacitors. Crystal filters are made of piezoelectric
resonators that can be modeled by resonant circuits. Mechanical filters are made of mechanical reson-
ators. Microwave filters consist of microwave resonators and cavities that can be represented by resonant
circuits. Active RC filters comprise resistors, capacitors, and amplifiers; in these filters, the performance of
resonant circuits is simulated through the use of feedback or by supplying energy to a passive circuit.
Switched-capacitor filters comprise resistors, capacitors, amplifiers, and switches. These are discrete-time
filters that operate like active filters but through the use of switches the capacitance values can be kept
very small. As a result, switched-capacitor filters are amenable to VLSI implementation.
This section provides an introduction to the characteristics of analog filters. Their basic characteriza-
tion in terms of a differential equation is reviewed in Section 1.2 and by applying the Laplace transform,
an algebraic equation is deduced that leads to the s-domain representation of a filter. The representation
of analog filters in terms of the transfer function is then developed. Using the transfer function, one can

* See Chapter 4 of Fundamentals of Circuits and Filters.

General Characteristics of Filters 1-3

obtain the time-domain response of a filter to an arbitrary excitation, as shown in Section 1.3. Some
important time-domain responses, i.e., the impulse and step responses, are examined. Certain filter
parameters related to the step response, namely, the overshoot, delay time, and rise time, are then
considered. The response of a filter to a sinusoidal excitation is examined in Section 1.4 and is then used
to deduce the basic frequency-domain representations of a filter, namely, its frequency response and loss
characteristic. Some idealized filter characteristics are then identified and the differences between
idealized and practical filters are delineated in Section 1.5. Practical filters tend to introduce signal
degradation through amplitude and=or delay distortion. The causes of these types of distortion are
examined in Section 1.6. In Section 1.7, certain special classes of filters, e.g., minimum-phase and allpass
filters, are identified and their applications mentioned. This chapter concludes with a review of the design
process and the tasks that need to be undertaken to translate a set of filter specifications into a working
prototype.

1.2 Characterization
A linear causal analog filter with input x(t) and output y(t) can be characterized by a differential equation
of the form

dn y(t) dn
1 y(t) dn x(t) dn
1 x(t)

bn n
þ bn
1 n
1
þ    þ b0 y(t) ¼ an n
þ an
1 þ    þ a0 x(t)
dt dt dt dt n
1

The coefficients a0, a1, . . . , an and b0, b1, . . . , bn are functions of the element values and are real if the
parameters of the filter (e.g., resistances, inductances, etc.) are real. If they are independent of time,
the filter is time invariant. The input x(t) and output y(t) can be either voltages or currents. The order of
the differential equation is said to be the order of the filter.
An analog filter must of necessity incorporate reactive elements that can store energy. Consequently,
the filter can produce an output even in the absence of an input. The output on such an occasion is
caused by the initial conditions of the filter, namely,
 
dn
1 y(t) dn
2 y(t)
, , . . . , y(0)
dt n
1 t¼0 dt n
2 t¼0

The response in such a case is said to be the zero-input response. The response obtained if the initial
conditions are zero is sometimes called the zero-state response.

1.2.1 Laplace Transform

The most important mathematical tool in the analysis and design of analog filters is the Laplace
transform. It owes its widespread application to the fact that it transforms differential into algebraic
equations that are a lot easier to manipulate. The Laplace transform of x(t) is defined as*

ð
1

X(s) ¼ x(t)e
st dt

1

where s is a complex variable of the form s ¼ s þ jv. Signal x(t) can be recovered from X(s) by applying
the inverse Laplace transform, which is given by

* See Chapter 3 by J. R. Deller, Jr. in Fundamentals of Circuits and Filters.

1-4 Passive, Active, and Digital Filters

ð
Cþj1
1
x(t) ¼ X(s)est ds
2pj
C
j1

where C is a positive constant. A shorthand notation of the Laplace transform and its inverse are

X(s) ¼ +x(t) and x(t) ¼ +
1 X(s)

Alternatively,

X(s) $ x(t)

A common practice in the choice of symbols for the Laplace transform and its inverse is to use upper case
for the s domain and lower case for the time domain.
On applying the Laplace transform to the nth derivative of some function of time y(t), we find that
 n   
d y(t) 
n
2 dy(t) dn
1 y(t)
+ ¼ s n
Y(s)
s n
1
y(0)
s
  

dt n dt t¼0 dt n
1 t¼0

Now, on applying the Laplace transform to an nth-order differential equation with constant coefficients,
we obtain
   
bn sn þ bn
1 sn
1 þ    þ b0 Y(s) þ Cy (s) ¼ an sn þ an
1 sn
1 þ    þ a0 X(s) þ Cx (s)

where
X(s) and Y(s) are the Laplace transforms of the input and output, respectively
Cx(s) and Cy(s) are functions that combine all the initial-condition terms that depend on x(t) and
y(t), respectively

1.2.2 Transfer Function

An important s-domain characterization of an analog filter is its transfer function, as for any other linear
system. This is defined as the ratio of the Laplace transform of the response to the Laplace transform of
the excitation.
An arbitrary linear, time-invariant, continuous-time filter, which may or may not be causal, can be
represented by the convolution integral

ð
1 ð
1

y(t) ¼ h(t
t)x(t)dt ¼ h(t)x(t
t)dt

1
1

where h(t) is the impulse response of the filter. The Laplace transform yields
2 3
ð
1 ð
1

Y(s) ¼ 4 h(t
t)x(t)dt5e
st dt

1
1
ð 1
1 ð
¼ h(t
t)e
st x(t)dt dt

1
1
ð 1
1 ð
¼ h(t
t)e
st  est  e
st x(t)dt dt

1
1
General Characteristics of Filters 1-5

Changing the order of integration, we obtain

ð
1 ð
1

Y(s) ¼ h(t
t)e
s(t
t)  x(t)e
st dt dt

1
1
ð 1
1 ð
¼ h(t
t)e
s(t
t) dt  x(t)e
st dt

1
1

Now, if we let t ¼ t0 þ t, then dt=dt0 ¼ 1 and t
t ¼ t0 ; hence,

ð
1 ð
1
0
Y(s) ¼ h(t 0 )e
st dt 0  x(t)e
st dt

1
1
ð
1 ð
1
0
st 0 0
¼ h(t )e dt  x(t)e
st dt

1
1
¼ H(s)X(s)

Therefore, the transfer function is given by

Y(s)
H(s) ¼ ¼ +h(t) (1:1)
X(s)

In effect, the transfer function is equal to the Laplace transform of the impulse response.
Some authors define the transfer function as the Laplace transform of the impulse response. Then
through the use of the convolution integral, they show that the transfer function is equal to the ratio of
the Laplace transform of the response to the Laplace transform of the excitation. The two definitions are,
of course, equivalent.
Typically, in analog filters the input and output are voltages, e.g., x(t) þ vi(t) and y(t) þ vo(t). In such a
case the transfer function is given by

Vo (s)
¼ HV (s)
Vi (s)

or simply by

Vo
¼ HV (s)
Vi

However, on occasion the input and output are currents, in which case

Io (s) Io
 ¼ HI (s)
Ii (s) Ii

The transfer function can be obtained through network analysis using one of several classical methods,*
e.g., by using

* See Chapters 18 through 27 of this volume.

1-6 Passive, Active, and Digital Filters

. Kirchhoff ’s voltage and current laws

. Matrix methods
. Flow graphs
. Mason’s gain formula
. State-space methods
A transfer function is said to be realizable if it characterizes a stable and causal network. Such a transfer
function must satisfy the following constraints:
1. It must be a rational function of s with real coefficients.
2. Its poles must lie in the left-half s plane.
3. The degree of the numerator polynomial must be equal to or less than that of the denominator
polynomial.
A transfer function may represent a network comprising elements with real parameters only if its
coefficients are real. The poles must be in the left-half s plane to ensure that the network is stable and
the numerator degree must not exceed the denominator degree to assure the existence of a causal network.

1.3 Time-Domain Response

From Equation 1.1,

Y(s) ¼ H(s)X(s)

Therefore, the time-domain response of a filter to some arbitrary excitation can be deduced by obtaining
the inverse Laplace transform of Y(s), i.e.,

y(t) ¼ +
1 fH(s)X(s)g

1.3.1 General Inversion Formula

If
1. the singularities of Y(s) in the finite plane are poles,* and
2. Y(s) ! 0 uniformly with respect to the angle of s as jsj ! 1 with s  C, where C is a positive
constant, then [2]
8
<0 for t < 0
y(t) ¼ Ð
Cþj1 Ð (1:2)
1
: 2pj Y(s)e ds ¼
st 1
2pj Y(s)e ds for t  0
st
C
j1 G

where G is a contour in the counterclockwise sense make up of the part of the circle s ¼ Re ju to the left of
line s ¼ C and the segment of the line s ¼ C that overlaps the circle, as depicted in Figure 1.1; C and R are
sufficiently large to ensure that G encloses all the finite poles of Y(s).
From the residue theorem [3] and Equation 1.2, we have
8
<0 for t < 0
y(t) ¼ Ð P
K
1
: 2pj Y(s)est ds ¼ res Y0 (s) for t  0
G i¼1 s¼pi

* Such a function is said to be meromorphic [2,3].