Infinite Impulse Response

(IIR)
Digital Filters (III)
Bilinear mapping

Yogananda Isukapalli

1
• Mapping of integrals
Consider an analog integrator:
1
H (s) =
s
t

y (t ) = ò x(t )dt
0

Consider the nth sample:

nT

y (nT ) = ò x(t )dt

0

Consider now the (n+1)th sample:

nT +T

y (nT + T ) = ò x(t )dt

0
nT nT +T

= ò x(t )dt + ò x(t )dt

0 nT

nth sample

2
We now have a recursion formula:
nT +T

y (nT + T ) = y (nT ) + ò x(t )dt

nT

(n+1)th sample nth sample

nT +T

Consider the term: ò x(t )dt

nT

Assume trapezoidal integration:

x(t)

0 T 2T nT nT+T t

3
Follows:
nT +T

ò x(t )dt = [Triangle Area] + [Rectangle Area]

nT

1
= T [x(nT + T ) - x(nT )] + [Tx (nT )]
2
T
= [x(nT ) + x(nT + T )]
2

Therefore,
T
y (nT + T ) = y (nT ) + [x(nT ) + x(nT + T )]
2
Digital Integrator

Z-transform:
T
zY ( z ) = Y ( z ) + [ X ( z ) + zX ( z )]
2

Y ( z) T z + 1
H ( z) = =
X ( z) 2 z - 1

4
• Conclusion:

1 T z +1
H (s) = H ( z) =
s 2 z -1

Implies: s z mapping

T z +1
s -1 =
2 z -1 Bilinear
2 z -1 mapping
s=
T z +1

5
Bilinear transformation: General approach:
z -1
s= (T is ignored)
z +1
1+ s
z=
1- s

Analog filter mapping:

H ( z ) = H (s) z -1
s=
z +1

Evaluation of mapping:

(a) Let s = s + jw ¢ then

(1 + s ) + jw ¢
z=
(1 - s ) - jw ¢
Let also z = re jypolar form in the z-plane

then
(1 + s ) 2 + w ¢2
z =r=
(1 - s ) 2 + w ¢2
6
w¢ -1 - w ¢
y = tan (-1
) - tan ( )
1+ s 1-s
w¢ w¢
y = tan (
-1 -1
) + tan ( )
1+ s 1-s
(b) Consider the points on the imaginary axis of
the s-plane for all w ¢ .
s = 0 Þ z = r =1 all w ¢

Im(s)
s-plane

0 Re(s)

Im(z)
z-plane

Re(z)

z =1
7
(c) Consider the points in the left-half of the
s-plane.
Re( s ) = s < 0

then z <1 all w ¢

Im(s)
s <0 s-plane

Re(s)

Im(z)
|z|=1 z-plane

Re(z)
z <1

Stable poles are preserved.

8
(d) Consider the points in the right-half of the
s-plane.
Re( s ) = s > 0

then z >1 all w ¢

Im(s)
s >0
s-plane

Re(s)

Im(z)
z-plane

Re(z)

z >1

9
Conclusion: Stable analog filters are mapped from
s-plane to z-plane as stable digital filters. Aliasing
problem is eliminated.

Analog and Digital frequencies relationship.

Consider
z -1
s=
z +1
jw T jw T / 2 jw T / 2 - jw T / 2
e p - 1 e p (e p - e p )
jw ¢p = jw pT = jw p T / 2 jw p T / 2 - jw p T / 2
e +1 e (e +e )
j sin(w pT / 2)
= = j tan(w pT / 2)
cos(w pT / 2)

w ¢p = k tan(w pT / 2)

where w p = specified cutoff frequency

w ¢p = prewarped cutoff frequency

k = 1 or 2/T
T = sampling period
10
Frequency scale distortion:
w¢
w ¢ = k tan(wT / 2)

0 p p w
2

Digital frequencies may result into scaled values

different than the desired or specified values
except where analog amplitude functions are
piecewise constant. This problem is eliminated or
at least minimized by prewarping frequency
specifications

11
Summary of Bilinear Transformation method1

12
Example 1:

Low pass filter: Obtain the transfer function H(z) of the

digital low pass filter to approximate the following transfer
function:
1
H (s) =
s 2 + 2s + 1
Use Bilinear Transformation method and assume a 3 dB cut
off frequency of 150Hz and a sampling frequency of
1.28kHz.

Soln. The critical frequency, w p = 2p ´ 150 rad/s, and Fs

=1/T = 1.28kHz, giving a prewarped critical frequency of
w ¢p = tan(w p T/2) = 0.3857

13
Example 2:

Soln.

14
Example 3:

Soln.

15
Use of BZT and Classical Analog Filters

Analog Filters Review

1. Low pass Butterworth filter of order N
1
Frequency | H ( jw ¢) |2 =
w¢ 2N (8.24)
response 1+ ( )
w ¢p

Poles of a normalized ( w ¢p= 1) Butterworth LPF

é (2k + N - 1)p ù é (2k + N - 1)p ù
sk = e jp ( 2 k + N -1) / 2 N = cos ê úû + j sin êë úû
ë 2N 2N
k = 1,2,..., N
(8.25)

æ 10 As
ö
ç 10 - 1 ÷
logç Ap ÷
ç 10 10 - 1 ÷
Filter order N N³ è ø (8.26)
æ w s¢ ö
2 logç ÷
ç w¢ ÷
è pø

16
Analog Filters Review contd

2. Low pass Chebyshev Type 1 Filter of order N

Frequency K
| H ( jw ¢) | =
2
response 1 + e 2C N2 (w ¢ / w ¢p )
(8.27)

Poles of a normalized (w ¢p = 1) Chebyshev LPF

sk = sinh(a ) cos( b k ) + j cosh(a ) sin( b k ),

1 1
where a = sinh -1 ( );
N e (8.28)
(2k + N - 1)p
bk = , k = 1,2,..., N
2N

æ 10 As
ö
-1 ç 10 - 1 ÷
cosh ç Ap ÷
Filter order N ç 10 10 - 1 ÷
(8.29)
N³ è ø
æ ö
-1 ç w s¢ ÷
2 cosh
ç w¢ ÷
è pø

17
Analog Filters Review contd

3. Low pass Elliptic Filter of order N

Frequency K
response
| H ( jw ¢) |2 =
1 + e 2GN2 (w ¢)
(8.30)

Fig: Sketches of frequency response of some classical analog filters

(a) Butterworth response (b) Chebyshev type I © Chebyshev type II
(d) Elliptic

18
Analog Frequency Transformations

1. Desired Low pass Low pass prototype

The low pass to low pass transformation is:
s
s= From 8.21 (a)
w ¢p
Denote frequencies for
LPF wlp and Prototype wp

wlp wlp
jw = j
p
i.e. w =
p
(8.31)
w ¢p w ¢p

19
2. Desired High pass Low pass prototype
The low pass to high pass transformation is:

w ¢p
s= From 8.21 (b)
s
Denote frequencies for

HPF w hp and Prototype wp

w ¢p w ¢p
jw = - j
p
i.e. w = -
p (8.32)
w hp w hp

20
2. Desired Band pass Low pass prototype
The low pass to band pass transformation is:

s 2 + w o2
s= From 8.21 (c)
Ws
Denote frequencies for

BPF w bp and Prototype wp

( jw bp ) 2 + w o2 w 2
- w 2

jw p = i.e. w p =
bp o
(8.33)
jWw bp Ww bp

For a band pass filter,

w ¢p1 , w ¢p 2 = lower and upper pass band edge frequencies

w s¢1 , w s¢2 = lower and upper stop band edge frequencies

w0 = center frequency (w02 = w ¢p1w ¢p 2 )

21
From eqn (8.33)

22
2. Desired Band stop Low pass prototype
The low pass to band stop transformation is:
Ws
s= From 8.21 (d)
s 2 + wo2
Denote frequencies for

BSF wbs and Prototype wp

jWwbs Wwbs
jw =p
i.e. w = 2
p
(8.34)
( jwbs ) 2 + wo2 wo - wbs2

For a band stop filter,

- w ¢p1 , w ¢p 2 = lower and upper pass band edge frequencies

w s¢1 , w s¢2 = lower and upper stop band edge frequencies

w0 = center frequency (w02 = w ¢p1w ¢p 2 )

23
24
25
Design Examples

26
27
28
29
30
Problem
Design a low-pass digital filter operating at the rate of 20 kHz.
And having pass-band extending to 3 kHz. with a maximum
Pass-band attenuation of 0.5 dB, and stop-band starting at
5 kHz. with a minimum stop-band attenuation of 10 dB.
a) Use Butter-worth design b) Using Chebyshev design

Solution:

Digital frequencies in radians per sample:

2πf pass 2π
ωp = = x3 = 0.3π
fs 20
2πf stop 2π
ωs = = x5 = 0.5π
fs 20
For the bilinear transformation design,
wp¢ = tan(ω p /2) = 0.5095
ws¢ = tan(ωs /2) = 1.0

Since the filter specifications are not normalized,

1 / k = ws¢ / wp¢ = 1/0.5095 = 1.9627.

10-[AsdB /10 ] - 1 1010/10 - 1
1/k1 = = = 8.588
10
-[A pdB / 10 ]
-1 10 0.5/10
-1

31
a) Using the Butter-worth design:

Order of the filter is:

log10 (1/k1 ) log10 (8.588)

N = = = 3.1896 @ 4
log10 (1 / k ) log10 (1.9627)
Therefore,
1
H(s) =
s 4 + 2.6131s 3 + 3.4142s 2 + 2.6131s + 1
Hence, the digital low - pass filter' s transfer function is :
H(z) = H(s) s = z -1
z +1

1
= 4 3 2
æ z -1ö æ z -1ö æ z -1ö
ç ÷ + 2.6131ç ÷ + 3.4142ç ÷
è z +1ø è z +1ø è z +1ø
æ z -1ö
+ 2.6131ç ÷ +1
è z + 1 ø
b) Using the Chebyshev design:

Order of the filter is:

cosh -1 (1/k1 )
N = = 2.1929 @ 3
cosh -1 (1 / k )
Therefore,
1
H(s) =
s 3 + 1.253s 2 + 1.535s + 0.716
32
Hence, the digital low - pass filter' s transfer function is :
H(z) = H(s) s = z -1
z +1

1
= 3 2
æ z -1ö æ z -1ö æ z -1ö
ç ÷ + 1.253ç ÷ + 1.535ç ÷ + 0.716
è z + 1 ø è z + 1 ø è z + 1 ø

33