IIR Filter Design

(i) Ability to design analog Butterworth filters

(ii) Ability to design lowpass IIR filters according to

predefined specifications based on analog filter theory and
analog-to-digital filter transformation

(iii) Ability to construct frequency-selective IIR filters based

on a lowpass IIR filter
Steps in Infinite Impulse Response Filter Design
The system transfer function of an IIR filter is:

(11.1)

The task in IIR filter design is to find and such that

satisfies the given specifications.

Once is computed, the filter can then be realized in

hardware or software according to a direct, canonic,
cascade or parallel form
We make use of the analog filter design to produce the
required

analog-to-digital
analog lowpass frequency band
filter
filter filter design transformation
transformation
specifications

Fig.11.1: Steps in determining transfer function of IIR filter

Note that is the Laplace transform parameter and

substituting in yields the Fourier transform of
the filter, that is,

Main drawback is that there is no control over the phase

response of , implying that the filter requirements can
only be specified in terms of magnitude response
Butterworth Lowpass Filter Design

In analog lowpass filter design, we can only specify the

magnitude of . Typically, we employ the magnitude
square response, that is, :

passband transition stopband

Fig.11.2: Specifications of analog lowpass filter

Passband corresponds to where is the passband
frequency and is called the passband ripple

Stopband corresponds to where is the

stopband frequency and is called the stopband
attenuation

Transition band corresponds to

The specifications are represented as the two inequalities:

(11.2)
and
(11.3)
In particular, at and , we have:

(11.4)
and
(11.5)

Apart from and , it is also common to use their

respective dB versions, denoted by and :

(11.6)

and
(11.7)
The magnitude square response of a th-order Butterworth
lowpass filter is:

(11.8)

The filter is characterized by and , which represent the

cutoff frequency and filter order

 at and at for all

 is a monotonically decreasing function of
frequency which indicates that there is no ripple
 filter shape is closer to the ideal response as increases,
although the filter with order of is not realizable.
Fig.11.3: Magnitude square responses of Butterworth lowpass filter
To determine , we first make use of its relationship with
:

(11.9)

From (11.8)-(11.9), we obtain:

(11.10)

The poles of , denoted by , ,

are given as:

(11.11)
-plane -plane

Fig.11.4: Poles of Butterworth lowpass filter

 are uniformly distributed on a circle of radius with
angular spacing of in the -plane

 poles are symmetrically located with respect to the

imaginary axis

 there are two real-valued poles when is odd

To extract from (11.10), we utilize the knowledge that

all poles of a stable and causal analog filter should be on
the left half of the -plane. As a result, is:

(11.12)
Example 11.1
The magnitude square response of a Butterworth lowpass
filter has the form of:

Determine the filter transfer function .

Expressing as:

From (11.8), and

From (11.11):

Finally, we apply (11.12) to obtain:

To find and given the passband and stopband
requirements in terms of , , and , we exploit
(11.4)-(11.5) together with (11.6)-(11.7) to obtain

(11.13)

and

(11.14)
Solving (11.13)-(11.14) and noting that should be an
integer, we get

(11.15)

where rounds up to the nearest integer.

The is then obtained from (11.13) or (11.14) so that the

specification can be exactly met at or , respectively

From (11.13), is computed as:

(11.16)
From (11.14), is computed as:

(11.17)

As a result, the admissible range of is:

(11.18)

Example 11.2
Determine the transfer function of a Butterworth lowpass
filter whose magnitude requirements are ,
, dB and dB.
Employing (11.15) yields:

Putting in (11.18), the cutoff frequency is:

For simplicity, we select . Using Example 11.1, the

filter transfer function is:
 Magnitude Square Response
0

-8

-16

-50
0 4 6 20
/

Fig.11.5: Magnitude square response of Butterworth lowpass filter

The command freqs, which is analogous to freqz, is used to
plot

Analog-to-Digital Filter Transformation

Typical methods include impulse invariance, bilinear

transformation, backward difference approximation and
matched- transformation

Their common feature is that a stable analog filter will

transform to a stable system with transfer function .

Left half of -plane maps into inside of unit circle in -plane

Each has its pros and cons and thus optimal transformation
does not exist
 Impulse Invariance

The idea is simply to sample impulse response of the analog

filter to obtain the digital lowpass filter impulse
response

The relationship between and is

(11.19)

where is the sampling interval

Why there is a scaling of T ?

With the use of (4.5) and (5.3)-(5.4), is:

(11.20)

where the analog and digital frequencies are related as:

(11.21)

The impulse response of the resultant IIR filter is similar to

that of the analog filter

Aliasing due to the overlapping of which

are not bandlimited. However, corresponds to a
lowpass filter and thus aliasing effect is negligibly small
particularly when is chosen sufficiently small.
To derive the IIR filter transfer function from ,
we first obtain the partial fraction expansion:

(11.22)

where are the poles on the left half of the -plane

The inverse Laplace transform of (11.22) is given as:

(11.23)
Substituting (11.23) into (11.19), we have:

(11.24)

The transform of is:

(11.25)

Comparing (11.22) and (11.25), it is seen that a pole of

in the -plane transforms to a pole at in the -
plane:

(11.26)
Expressing :

(11.27)

where is any integer, indicating a many-to-one mapping

Each infinite horizontal strip of width maps into the

entire -plane

maps to , that is, axis in the -plane

transforms to the unit circle in the -plane

maps to , stable produces stable

maps to , right half of the -plane maps into the

outside of the unit circle in the -plane
 -plane -plane

Fig.11.6: Mapping between and in impulse invariance method

Given the magnitude square response specifications of
in terms of , , and , the design procedure
for based on the impulse invariance method is
summarized as the following steps:

(i) Select a value for the sampling interval and then

compute the passband and stopband frequencies for the
analog lowpass filter according to and

(ii) Design the analog Butterworth filter with transfer

function according to , , and

(iii)Perform partial fraction expansion on as in (11.22)

(iv)Obtain using (11.25)

Example 11.3
The transfer function of an analog filter has the form of

Use impulse invariance method with sampling interval

to transform to a digital filter transfer function .

Performing partial fraction expansion on :

Applying (11.25) with yields

Example 11.4
Determine the transfer function of a digital lowpass
filter whose magnitude requirements are , ,
dB and dB. Use the Butterworth lowpass filter
and impulse invariance method in the design.

Selecting the sampling interval as , the analog

frequency parameters are computed as:

and
Using Example 11.2, a Butterworth filter which meets the
magnitude requirements are:

Performing partial fraction expansion on with the use

of the MATLAB command residue, we get

Applying (11.25) with yields

The MATLAB program is provided as ex11_4.m.
 Bilinear Transformation

It is a conformal mapping that maps the axis of the -

plane into the unit circle of the -plane only once, implying
there is no aliasing problem as in the impulse invariance
method

It is a one-to-one mapping

The relationship between and is:

(11.28)
Employing , can be expressed as:

(11.29)

maps to , that is, axis in the -plane

transforms to the unit circle in the -plane

maps to , stable produces a stable

maps to , right half of the -plane maps into the

outside of the unit circle in the -plane
 -plane -plane

Fig.11.8: Mapping between and in bilinear transformation

H. C. So Page 34 Semester B 2011-2012

Although aliasing is avoided, the drawback of the bilinear
transformation is that there is no linear relationship
between and

Putting and in (11.28), and are related as:

(11.30)

Given the magnitude square response specifications of

in terms of , , and , the design procedure
for based on the bilinear transformation is
summarized as the following steps:
(i) Select a value for and then compute the passband and
stopband frequencies for the analog lowpass filter
according and

(ii) Design the analog Butterworth filter with transfer

function according to , , and .

(iii)Obtain from using the substitution of (11.28).

Example 11.5
The transfer function of an analog filter has the form of

Use the bilinear transformation with to transform

to a digital filter with transfer function .
Applying (11.28) with yields

Example 11.6
Determine the transfer function of a digital lowpass
filter whose magnitude requirements are , ,
dB and dB. Use the Butterworth lowpass filter
and bilinear transformation in the design.

Selecting , the analog frequency parameters are

computed according to (11.30) as:
and

Employing (11.15) yields:

Putting in (11.18), the cutoff frequency is:

For simplicity, is employed.

Following (11.11)-(11.12):

Finally, we use (11.28) with to yield

Frequency Band Transformation

The operations are similar to that of the bilinear

transformation but now the mapping is performed only in
the -plane:

(11.31)

where and correspond to the lowpass and resultant

filters, respectively, and denotes the transformation
operator.

To ensure the transformed filter to be stable and causal, the

unit circle and inside of the -plane should map into those
of the -plane, respectively.
Filter Transformation Operator Design Parameter
Type
Lowpass

Highpass

Bandpass
Bandstop

Table 11.1: Frequency band transformation operators

Example 11.7
Determine the transfer function of a digital highpass
filter whose magnitude requirements are , ,
dB and dB. Use the Butterworth lowpass filter
and bilinear transformation in the design.

Using Example 11.6, the corresponding lowpass filter

transfer function is:
Assigning the cutoff frequencies as the midpoints between
the passband and stopband frequencies, we have

With the use of Table 11.1, the corresponding value of is:

which gives the transformation operator:

As a result, the digital highpass filter transfer function is:

2. Analog Filter Design
 Decades of analysis of transistor-based
filters – sophisticated, well understood
 Basic choices:
 ripples vs. flatness in stop and/or passband
 more ripples  narrower transition band

Family PB SB
Butterworth flat flat
Chebyshev I ripples flat
Chebyshev II flat ripples
Elliptical ripples ripples
10
 CT Transfer Functions
 Analog systems: s-transform (Laplace)
Continuous-time Discrete-time
Transform H a (s) =  ha (t )e st
dt H d (z ) =  hd [n] z n
Frequency
response H a ( j) Hd e( )j

Im{s} Im{z}

j ej
Pole/zero Re{z}
diagram Re{s}
1
stable s-plane stable z-plane
poles poles
Dan Ellis 2005-11-10 11
 Butterworth Filters
Maximally flat in pass and stop bands
 Magnitude 1 filter
response (LP): H a ( j) =
2
order
( )
2N N

1+ c
 <<c,
|Ha(j)|2 1
  = c,

|Ha(j)|2 = 1/2
3dB point

12
Butterworth Filters 6N dB/oct
rolloff
 >>c, |Ha(j)|2 (c/)2

Log-log
magnitude
response

n
d
H a ( j) = 0
2
 flat  n
d
@  = 0 for n = 1 .. 2N-1
13
Butterworth Filters
 How to meet design specifications?
1 1

( ) 1+2
p 2N
1+ c
Design
Equation

( )
1 1
 2 1
2
A
1 log10  2
1+( )
 s 2N A N
c
( )
2 log10 s
p

 p
 k1 =  k=
A 1
2
s
=“discrimination”, <<1 =“selectivity”, < 1
14
Butterworth Filters
1
H a ( j) =
2
 but what is Ha(s)?
1+ ( )2 N
c

 Traditionally, look it up in a table

 calculate N  normalized filter with c = 1
 scale all coefficients for desired c
Im{s}
1
 In fact, H a ( s ) =  c
i(s  pi )  Re{s}

  s 2N
  = 1
j N +2 i1
where pi =  c e 2N
i = 1..N   c 

s-plane
15
 Butterworth Example
Design a Butterworth
filter with 1 dB cutoff
at 1kHz and a
minimum attenuation
of 40 dB at 5 kHz
1
1dB = 20 log10   = 0.259
2
1+  2
9999
40dB = 20 log10 A  A = 100
1 log
N  21 10 0.259
s log10 5
=5
p  N = 4  3.28
16
Butterworth Example
 Order N = 4 will satisfy constraints;
What are c and filter coefficients?
 from a table, -1dB = 0.845 when c = 1
 c = 1000/0.845 = 1.184 kHz
 from a table, get normalized coefficients for

N = 4, scale by 1184·2
 Or, use Matlab:
[b,a] =
butter(N,Wc,’s’);
17
M
 Chebyshev I Filter
 Equiripple in passband (flat in stopband)
 minimize maximum error

1
H a ( j)
2
=
1+  2TN2 (p )

Chebyshev
polynomial TN () = 
(
 cos N cos1  )  1
of order N (
cosh N cosh 1  )  >1
18
Chebyshev I Filter
 Design procedure:
 desired passband ripple  
 min. stopband atten., p, s  N :

1 1 1
= =
[
A 1+  TN ( p ) 1+  2 cosh N cosh 1  s
( )]
2 2 2 s 2
p

N
cosh 1
( A 2 1
 ) 1/k1, discrimination
1  s
cosh  p ( ) 1/k, selectivity

19
Chebyshev I Filter
 What is Ha(s)?
 complicated, get from a table
 .. or from Matlab cheby1(N,r,Wp,’s’)

 all-pole; can inspect them:

..like squashed-in Butterworth

20
 Chebyshev II Filter
 Flat in passband, equiripple in stopband

1
H a ( j) =
2

constant
T ( s ) 
 2
2  N p 
1+ 
 TN (s ) 
~1/TN(1/)   
zeros on imaginary axis

 Filter has poles and zeros (some )

 Complicated pole/zero pattern

21
 Elliptical (Cauer) Filters
 Ripples in both passband and stopband

1
H a ( j)
2
=
1+  2 RN2 (p )
function; satisfies
RN(-1) = RN()-1 very narrow
zeros for <1  poles for >1 transition band

 Complicated; not even closed form for 

22
Analog Filter Types Summary

N=6
r = 3 dB
A = 40 dB

23