Content

Chapter 4. Modern Filter Design

Filter specifications
Frequency Transformation
Butterworth Approximation

August 19, 2024 1 / 23

Introduction
Filters can be considered as frequency selective networks. A filter is required to
separate an unwanted signal from a mixture of unwanted and wanted signals.
In general, the overlapping of these signal is such that a complete separation of these
signals is not feasible, if not impossible. In such cases the transfer characteristic of
the filter is so shaped that the ratio of unwanted to wanted signal at the output of
the filter is minimized.
The filter specifications are generally given in terms of cut-off frequencies or
pass-band and stop-band regions.
Cut-off frequency:- The frequency associated with the boundary between stop
band and an adjacent pass-band. It is the frequency at which the output is 0.707
times the maximum value in the pass band.
Pass-band is the frequency band of wanted signals and stop-band is the frequency
band of unwanted signals.
August 19, 2024 2 / 23
Introduction

In contrast to the image parameter theory, modern filter design is based on the
selection of the filter transfer function to satisfy the specification, and then the
realization of this function by synthesis techniques.
Most of the widely used filter synthesis methods are for low pass filters only.
So how do we synthesize band-pass, high-pass and band-reject filters? We use
frequency transformation methods. By frequency transformation we mean:
1 Transform the filter specifications to normalized low-pass filter specification.
2 Obtain the normalized low-pass filter transfer function using modern filter synthesis
methods.
3 Transform the low-pass transfer function back to the required type of filter.
4 Synthesize the system function using realization techniques.

August 19, 2024 3 / 23

Filter Specification

An ideal filter should pass the wanted

signals with no attenuation and provide
infinite attenuation for the unwanted
signals.
Ideal filters impulse response is not
causal, hence, not realizable. In general,
a band limited frequency response leads
to non-causal impulse response.
In order to circumvent this limitation,
filter specifications are given for
low-pass filter as follows.

August 19, 2024 4 / 23

Filter Specification

Figure: High pass filter specification Figure: Band pass filter specification

α ≤ αp , for ω ≥ ωp α ≤ αp , for ω1 ≤ ω ≤ ω2
α ≥ αs , for ω ≤ ωs α ≥ αs , for ω ≤ ωs1 and ω ≥ ωs2

August 19, 2024 5 / 23

Filter Specification

Frequency Transformation:
High-pass filter:
Step 1: Transform the given HPF
specifications to normalized low pass
specifications with
ω0
ωn =
ω
Figure: Band Stop filter specification
Where ωn is the normalized low pass filter
frequency, ω0 is cutoff frequency of Highpass
α ≤ αp , for ω ≤ ω1 and ω ≥ ω2 filter and ω is the frequency of HPF.
α ≥ αs , for ωs1 ≤ ω ≤ ωs2
August 19, 2024 6 / 23
Frequency Transformation

Step 2: Design system function using

modern design techniques. (we will see)
1 ω02 − ω 2
Step 3: Transform the obtained normalized ωn = −
BW ω
low pass filter transfer function back to high
pass filter with Where BW is Bandwidth of BPF= ω2 -ω1 ,
√
ω0 ω0 is center frequency = ω1 ω2 and ω is the
sn = frequency of BPF. Step 3: Transform the
s
obtained normalized LPF transfer function
Band Pass Filter (BPF): back to BPF with
Step 1: Transform the given BPF
1 ω02 + s2
specifications to normalized low pass sn =
specifications with BW s

August 19, 2024 7 / 23

Frequency Transformation

Band Stop Filter (BSF):

Step 1: Transform the given BSF
specifications to normalized low pass
specifications with Step 3: Transform the obtained normalized
ω LPF transfer function back to BSF with
ωn =
ω02 − ω 2 s
sn =
√ ω02 + s2
Where ω0 is center frequency = ω1 ω2 and
ω is the frequency of BSF.
Step 2: Design system function using
modern design techniques.

August 19, 2024 8 / 23

Modern Design Techniques

The approximation methods solve the

problem of selecting a realizable rational be minimized. Here the criterion is to
function whose frequency response minimize the maximum error over the
approximates the low pass filter we have frequency band ω ≤ ω ≤ ω .
a b
seen.
We may also pick a mean-square error
Let f(ω) be the required frequency criterion, i.e. t(ω) is selected such that
response function and t(ω) the function it minimizes,
which approximates f(ω). We may
Z ωb
require that
ϵ(ω) = |f (ω) − t(ω)|2 dω
ωa
ϵ(ω) = M ax|f (ω) − t(ω)|
ωa ≤ ω ≤ ωb
August 19, 2024 9 / 23
Modern Design Techniques

In all the approximation methods the

normalized transfer function is selected
as The attenuation function will be
1 α(ω) = 10 log[1 + ϵ2 Kn2 (ω)] dB
|T (jω)|2 = , with
1+ ϵ2 Kn2 (ω)
0 ≤ Kn2 (ω) << 1, 0 ≤ ω ≤ ωp The constant ϵ determines the pass-band
Kn2 (ω) >> 1, ω ≥ ωs and/or stop band attenuation.
Now we will discuss Butterworth
With this selection, —T(jω)—is approximation technique.
approximately 1 in the pass band and 0
in the stop band.

August 19, 2024 10 / 23

Butterworth Approximation

In Butterworth approximation it
1
minimizes the maximum error at ω=0. |T (jω)|2 =
1 + c2 ω 2n
Kn (ω) is selected as
α(ω) =10 log[1 + c2 ω 2n ] dB
Kn (ω) = β0 + β1 ω + β2 ω 2 + ... + βn ω n

To be maximally flat, all the derivatives The normalized LPF become

of Kn (ω) must be zero at ω=0. This 1
reduces Kn (ω) = ω n . Hence, the |T (jω)|2 =
1+ c2 ( ωωp )2n
transfer function and the attenuation ω 2n
function are. α(ω) =10 log[1 + c2 ( ) ] dB
ωp
August 19, 2024 11 / 23
Butterworth Approximation

If the specifications arc barely satisfied, c2 =100.1αp − 1

c2
αp =10 log[1 + c2 ] dB =100.1αs − 1
ωs
2n k 2n s
αs =10 log[1 + c2 ] dB
ωp 100.1αp − 1
k1 =
ω 100.1αs − 1
The factor ωps , is called the selectivity
parameter and is represented by k
k1 is called discrimination parameter,
ωp
k= Solving for n from above equations,
ωs
log k1
In terms of k, n=
log k
August 19, 2024 12 / 23
Butterworth Approximation
Filter specification often includes the
cut-off frequency which is defined as the
The order of the Butterworth
-3dB point. If ωc is the cut-off frequency
polynomial selected an integer n≥ log k1
log k
then, the relation between pass-band
If n happens to be equal to log k1
log k , then and cut-off frequencies become
the value of c obtained from αp and
ωp
from αs will be equal. ωc = √
n
c
If n is an integer greater than log k1
log k ,
then the value of c is chosen to satisfy If the pass-band edge is the cut off
either αp or αs . √ frequency, i.e, αp = 3dB, then c=1.
In the first case: c= 10 0.1αp − 1,
√ Most of the time, for convenience we
In the second case: c= k 2n 100.1αs − 1 consider a normalized Butterworth
filter, i.e, ωp = l and c = 1.
August 19, 2024 13 / 23
Butterworth Approximation

In order to realize a filter we have to If n is even, it become 1 + s2n = 0

determine its transfer function. From reduces to s2n = −1 = ej(2k−1)π . The 2n
realizability conditions, the poles of the roots are [use complex roots solution
transfer function must be in the left half method]
of s-plane.
(2k−1)
To this end let, pk =ej 2n
π

2k − 1 2k − 1
2 1 1 = cos(j π) + j sin(j π)
|T (jω)| = 2
= 2n 2n
|H(jω)| H(s)H(−s)|s=jw
k = 1, 2, ..., 2n
|H(jw)|2 =1 + ω 2n = 1 + (−1)n s2n
If n is odd, it become 1 − s2n = 0
Solving 1 + (−1)n s2n = 0 and select the reduces to s2n = 1 = ej2kπ . The 2n roots
left hand poles from 2n roots give, are
August 19, 2024 14 / 23
Butterworth Approximation

k
pk = ej n π 1 2k − 1
T (s) = Qn/2 with θk = π
k k 2 2n
= cos(j π) + j sin(j π) k=1 (s + 2 cos θk s + 1)
n n
k = 0, 1, ..., 2n − 1
If n is odd,
1
T (s) =
The transfer function are (s + 1)
Q(n−1)/2
(s2 + 2 cos θk s + 1)
k=1
constructed from left hand
kπ
plane roots and is given by. For with θk =
n
n is even

August 19, 2024 15 / 23

Butterworth Approximation

In the preceding discussion c if n is even

was assumed to be 1. If it were
1 j (2k−1) π
not the case, i.e, when the pk = √
n
e 2n k = 1, 2, ..., 2n
c
pass-band edge is not the 3 dB
point, the radius of the If n is odd,
√
Butterworth circle, is 1/ n c
and the roots of H(s) are given 1 j (k) π
pk = √
n
e n k = 0, 1, ..., 2n − 1
by c

Most of the time getting roots are tiring. the following table gives you the denominator of
T(s).

August 19, 2024 16 / 23

 Butterworth Approximation

Example: Design band pass filter system

function
α ≤3dB, for 50k ≤ ω ≤72k rad/s
Now, the normalized LPF
α ≥ 40dB for 50k ≤ 30k and ω ≥ 120k rad/s
specification become
Solution: Step 1 change to normalized LPF
√ √ α≤3dB, ω≤1
specification
√ ω√
0 = ω1 ω2 = ωs1 ωs2
α≥40dB, ω≥4.095
= 50k ∗ 72k = 30k ∗ 120k = 60k rad/s Step 2 for this LPF design
Using frequency transformation, the stop band edge filter using Butterworth
of LPF is calculated using the formula, approximation
1 ω02 − ωs2
2 −1 (60k)2 − (120k)2
ωn = − = = 4.095
BW ωs2 72k − 50k 120k

August 19, 2024 17 / 23

Butterworth Approximation

100.3 − 1
k1 = = 0.01 Step 3: Transform from LPF to BPF
104 − 1
1 using
k= = 0.2442
4.095 1 ω02 + s2
log 0.01 sn =
n≥ = 3.27, n = 4 BW s
log 0.2442
1 (60k)2 + s2
=
From table for n=4 with c=1 and 22k s
ωp = 1 rad/s, (60k)2 + s2
=0.0455 ∗ 10−3
s
1
TLP (s) =
s4 + 2.613s3 + 3.414s2 + 2.613s + 1

August 19, 2024 18 / 23

Butterworth Approximation

step 4: Realize the network from system function. But we will not consider here.
you have already seen how to realize a network in previous lessons. It is enough if you do
till this.

August 19, 2024 19 / 23

 Butterworth Approximation
Example 2: Design low pass
filter transfer function

α ≤ 1dB, f or f ≤ 3M Hz For a normalized LPF, ωc = 1, from Table

α ≥ 60dB f or f ≥ 12M Hz 1
T (s) =
s6 + 3.86s5 + 7.46s4 + 9.14s3 + 7.46s2 + 3.86s + 1
100.1 − 1 However, αp ̸=3dB =⇒ ωc ̸= ωp let’s select to satisfy
k1 = = 0.5089 ∗ 10−3
106 − 1 pass band edge,
3
k = = 0.25
12
log 0.5089 ∗ 10−3
n≥ = 5.47, n = 6
log 0.25
August 19, 2024 20 / 23
Butterworth Approximation

For normalized LPF, ωp = 1

p ωp
c= 100.1 − 1 = 0.5089 ωc = √
n
= 1.12ωp = 1.12
c

The transfer function for the low-pass filter should be arranged in for ωc = 1.12. This can
s
be done by substituting using in place of s.
1.12
1
T (s) = s 6 s 5 s 4 s 3 s 2 s
( ) + 3.86( ) + 7.46( ) + 9.14( ) + 7.46( ) + 3.86( )+1
1.12 1.12 1.12 1.12 1.12 1.12

August 19, 2024 21 / 23

Butterworth Approximation

August 19, 2024 22 / 23

Butterworth Approximation

August 19, 2024 23 / 23