46

Example: design a low pass filter, Butterworth, with 3dB bandwith of 500Hz
and 40dB attenuation at 1000Hz.
Solution:

c = 1000 ; s = 2000 ,

solve for N from the expression

1 1
= 2 N = 0.01
2

 s 
2N
1+ 2
1+  
 c 
| H ()|dB
log(104 − 1)
N= = 6.64  N = 7
2 log(2)

poles at
sk = 1000 e j ( / 2 +( 2 k +1) /14 ) , k = 0,...,6

 Chebychev Filters.
47
Based on Chebychev Polynomials: TN ( x ) = cos( Nt ) x = cos( t )
T0 ( x ) = cos(0) = 1, T3 ( x)

T1 ( x ) = cos(t ) x = cos( t ) = x ,
T2 ( x ) = cos(2t ) = 2 cos2 (t ) − 1 x = cos( t ) = 2 x 2 − 1,

TN +1 ( x ) = 2 xTN ( x ) − TN −1 ( x )
x
Property of Chenychev Polynomials:
within the interval − 1  x  +1 Chebychev polynomials have least maximum deviation
from 0 compared to polynomials of the same degree and same highest order coefficient

P( x) = 4 x 3 + ax 2 + bx + c

T3 ( x) = 4 x 3 − 3x

x
 Why? Suppose there exists P( x) = 4 x 3 + ax 2 + bx + c with smaller deviation then T3 ( x)
48
P  T3 = +1 P  T3 = +1
T3 ( x)
B D
P( x) − T3 ( x) +
A
P( x) + C
− D

−
A C
P  T3 = −1 P  T3 = −1 root
B root root

But: has degree 2 …

P( x) − T3 ( x) = 4 x 3 + ax 2 + bx + c − 4 x 3 + 3x
… and it cannot have three roots!!!

So: you cannot find a P(x) which does better (in terms of deviation from 0) then the
Chebychev polynomial.
 49

Chebychev Filter:
1
| H ()| =
2

 
1 +  TN 
2 2

 p 
1
Since TN (1) = 1 (easy to show from the definition), then | H ( p )|2 =
1+  2

| H ()|2 1
1+  2


p
 50
Design of Chebychev Filters:
Formulas are tedious to derive. Just give the results:

Given:  p the passband, and

 which determines the ripple in the passband,
compute the poles from the formulae

 2 1
1
 1+  + 1 N
r1,2 =
2

 =  
 2
  

 
 k= + (2 k + 1) , k = 0,..., N − 1
2 2N

s-plane
 51

Example: design a Chebychev low pass filter with the following specs:
• passband Fp = 500 Hz, with a 1dB ripple,
• stopband Fs = 1000Hz, with attenuation of at least 40dB.
Step 1: determine  , p . The passband frequency  = 1000
p

For 1dB ripple,

1
 1 

2
20 log
 1+  
2  = −10 log(1 +  2
) = −1dB  = 0.5088
| H ()|
Step 2: determine the order N. Use the formula

N=
log ( 1 −  + 1 −  (1 +  ) /   
2
2
2
2 2
2

log( /  ) + ( /  ) − 1
s p s p
2

with 2 = 0.01, to obtain N = 4 

p
 52

Frequency Transformations
We can design high pass, bandpass, bandstop filters from transformations of low pass filters.
Low Pass to High Pass: c2  c 2 
s → H
 j 

| H ()| s

− c c  − c c 

same value at c

 c2   c 
H  = H   = H ( − j c )
 j c   j 
 53

Low Pass to Band Pass:

− c c

− u − l l u
s2 + l u
The tranformation s →  c
s( u −  l )
maps
− l 2 + l u
s = j l → c = − j c
j l (  u −  l )
− u 2 + l u
s = j u → c = + j c
j u (  u −  l )
 54
Low Pass to Band Stop

− c c

− u − l l u

s( u −  l )
s → c 2
s + cl
 55
How to make the transformation:
Consider the transfer function
K ( s − z1 )( s − z2 )...( s − zm )
H ( s) = , mn
( s − p1 )( s − p2 )...( s − pn )
K ( F ( s) − z1 )... ( F ( s) − zm )
then with s → F ( s) HT ( s) = H ( F ( s)) =
we obtain
( F ( s) − p )...( F ( s) − p )
1 n

F ( s) = zk , k = 1,..., m
with zeros and poles solutions of F ( s) = p j , j = 1,..., n
also n-m extra zeros at s where F ( s) = 

 
 


 IIR filter design using Matlab 56

In Matlab there are three functions for each class of filters (Butterworth, Chebytchev1, Chebytchev2):
BUTTAP CHEB1AP CHEB2AP Poles and Zeros of Analog Prototype Filter
BUTTER CHEBY1 CHEBY2 Numerator and Denominator from N and c
BUTTORD CHEBY1ORD CHEBY2ORD N and c from specifications

Example. We want to design an IIR Digital Filter with the following specifications:
Pass Band 0 to 4kHz, with 1dB ripple;
Stop Band > 8kHz with at least 40 dB attenuation
Sampling frequency 40kHz
Type of Filter: Butterworth.
Using Matlab:
>> [N, fc]=butterord(fp, fs, Rp, Rs); % fp, fs=passband and stopband freq relative to Fs/2
>> [B, A]=butter(N, fc); % B, A vectors of numerator and denominator coefficients.
In our case:
[N, fc]=butterord(4/20, 8/20, 1, 40), would yield N=7, fc=0.2291;
[B, A]=butter(7, 0.2291), would yield the transfer function B(z)/A(z).
 57
Let’s verify these numbers:
Step 1: specifications in the digital frequency domain:

4 
 p = 2 = rad band pass freq
40 5
8 2
s = 2 = rad stop band freq
40 5

Step 2: specifications for analog filter from the transformation  = 2 Fs tan 
 2

 p = 2(40  10 ) tan  = 26  103 rad / sec
3
 10 

 s = 2(40  10 ) tan  = 58  103 rad / sec
3
 5
 58

2 1
H ( ) = 2N
Step 3: choose (say) Butterworh Filter 
2 

1+   
 p 
with  p = 26  103 rad / sec and  from the ripple specification
1
 1  2
20 log10  2  = −1   = 0.5
 1+  

Step 4: determine order N from attenuation of 40dB

1
10 log10 2N
=− 40
 s 
1 + 0.5 
2 
 
 p

 s 58
with = yields N=7
 p 26
 59

Step 5: finally the cutoff frequency, from the equation

2 1 1
H ( c ) = =
c 
14
2
2
1 + 0.5  
 26  103 

Which yields c = 28.7  103 rad / sec , corresponding to a digital frequency

 28.7 
c = 2atan  = 0.6889 radians
 2  40 

Step 6: the desired Filter is obtained by the function

[num, den] = butter( 7 , 0.6889/)
 60

Magnitude and Phase Plots:

 61

Example 3
• A filter is described by the following equation:
• y(n)=0.5x(n) + 0.2x(n-1) + 0.5y(n-1) + 0.2y(n-2),
with initial condition y(-1)=y(-2) = 0
• What kind of filter is it?
• Plot the filter’s transfer function on the z plane
• Is the filter stable?
• Plot the filter’s unit step response
• Plot the filter’s unit impulse response
 62

Software Implementation of
FIR filters
 63

Fixed-Point Binary Representation

• Representation of a number with integer and
fractional part: (d n −1d n − 2 ...d 0 .d −1d − 2 ...d − k ) 2
n −1
N= d r
i =− m
i
2

• This is denoted as Qnm representation

• The binary point is implied
• It will affect the accuracy (dynamic range and
precision) of the number
• Purely a programmer’s convention and has no
relationship with the hardware.
 64

EFFECTS OF FINITE FIXED-POINT

BINARY WORD LENGTH
• Quantization Errors
• ADC
• Coefficients
• Truncation
• Rounding
• Data Overflow
 65

Coefficient Quantization Error

• Effects on FIR filters

• Location of zeroes changes
• Therefore, frequency response changes
• Effects on IIR filters
• Location of poles and zeroes change
• Could move poles outside of unit circle, leading to unstable implementations
 66

Overflow error

• signals and coefficients normalized in the range of

−1 to 1 for fixed-point arithmetic, the sum of two B-
bit numbers may fall outside the range of −1 to 1.
• Severely distorts the signal
• Overflow handling
• Saturation arithmetic
• “Clips” the signal, although better than overflow
• Should only be used to guarantee no overflow, but should not be
the only solution
• Scaling of signals and coefficients
 67

Coefficient representation

• Fractional 2’s complement (Q) representation is

used
• To avoid overflow, often scaling down by a power of
two factor (S) (right shift) is used.
• The scaling factor is given by the equation:
S=Imax(|h(0)|+|h(1)|+|h(2)|+…)
• Furthermore, filter coefficient larger than 1, cause
overflow and are scaled down further by a factor B,
in order to be less than 1