Chapter 7

Finite Impulse Response(FIR)

Filter Design
 1. Features of FIR filter

 Characteristic of FIR filter

N
y ( n)   h(k ) x (n  k )
k 0
N
H ( z )   h( k ) z  k
k 0

– FIR filter is always stable

– FIR filter can have an exactly linear phase response
– FIR filter are very simple to implement. Nonrecursive FIR
filters suffer less from the effect of finite wordlength than IIR
filters

2/74
 2. Linear phase response
 Phase response of FIR filter
N
H (e jT
)   h(k )e  jkT (1)
k 0

= H (e jT ) e j ( )

where  ( )  arg  H (e jT )  (2)

– Phase delay  p and group delay  g

 ( )
p 

d ( )
g  
d

3/74
– Condition of linear phase response
 ( )   (3)

 ( )     (4)

Where  and  is constant

Constant group delay and phase delay response

4/74
• If a filter satisfies the condition given in equation (3)
– From equation (1) and (2)
 ( )  
N
 h(n) sin  nT
= tan 1 n0
N

 h(n) cos  nT
n0

thus
N

 h(n) sin  nT
tan = n 0
N

 h(n) cos  nT
n 0

5/74
 N

 h(n)  cos nT sin   sin  nT cos    0

n0
N

 h(n) sin(   nT )  0
n0

h(n)  h( N  n  1), 0n N

  NT / 2

– It is represented in Fig 7.1 (a),(b)

6/74
• When the condition given in equation (4) only
– The filter will have a constant group delay only
– It is represented in Fig 7.1 (c),(d)

h ( n )   h( N  n )

  NT / 2

  /2

7/74
Center of symmetry

Fig. 7-1.

8/74
Table 7.1 A summary of the key point about the four types of linear phase FIR filters

9/74
 Example 7-1
(1) Symmetric impulse response for linear phase response.
No phase distortion
h(n)  h( N  n) or h(n)  h( N  n)

(2) h(n)  h( N  n), 0  n  N , N  10

h(0)  h(10)
h(1)  h(9)
h(2)  h(8)
h(3)  h(7)
h(4)  h(6)

10/74
 • Frequency response H ( )

H ( )  H (e jT )
10
=  h(k )e  jkT
k 0

=h(0)  h(1)e jT  h(2)e  j 2T  h(3)e  j 3T  h(4)e  j 4T  h(5)e  j 5T
+h(6)e j 6T  h(7)e  j 7T  h(8)e  j 8T  h(9)e  j 9T  h(10)e  j10T
=e  j 5T [h(0)e j 5T  h(1)e j 4T  h(2)e j 3T  h(3)e j 2T  h(4)e jT  h(5)
+h(6)e jT  h(7)e j 2T  h(8)e  j 3T  h(9)e  j 4T  h(10)e  j 5T ]

H ( )=e  j 5T [h(0)(e j 5T  e  j 5T )  h(1)(e j 4T  e  j 4T )  h(2)(e j 3T  e  j 3T )
 h(3)(e j 2T  e  j 2T )  h(4)(e jT  e  jT )  h(5)]
 e  j 5T [2h(0) cos(5T )  2h(1) cos(4T )  2 h(2) cos(3T )
 2h(3) cos(2T )  2h(4) cos(T )  h(5)]
5
H ( )=  a(k ) cos(kT )e j 5T  H ( ) e j ( )
k 0

5
where H ( ) =  a( k ) cos( kT )
k 0

 ( )  5T
11/74
(3) N  9
h(0)  h(9)
h(1)  h(8)
h(2)  h(7)
h(3)  h(6)
h(4)  h(5)

H ( )=e  j 9T /2 [h(0)(e j 9T  e  j 9T )  h(1)(e j 7T  e  j 7T )  h(2)(e j 5T  e  j 3T )
 h(3)(e j 3T  e  j 3T )  h(4)(e jT  e  jT )]
 e  j 9T /2 [2h(0) cos(9T / 2)  2h(1) cos(7T / 2)  2 h(2) cos(5T / 2)
 2h(3) cos(3T / 2)  2h(4) cos(T / 2)]
= H ( ) e j ( )
5
where H ( )= b(k ) cos[ (k  1/ 2)T ]
k 1
 ( )  (9 / 2)T
N 1 N 1
b( k )  2h(  k ), k  1, 2, ,
2 2

12/74
 3. Zero distribution of FIR filters
 Transfer function for FIR filter
N
H ( z )   h( k ) z  k
k 0

N
H ( z )   h( N  k ) z  k
k 0
0
=  h( k ) z k z  N
k  N

=z  N H ( z 1 )

13/74
 Four types of linear phase FIR filters
– H ( z ) have zero at z  z0
z0  re j

z0 1  r 1e  j

h(n) is real and z0 is imaginary

z0*  re  j

( z0* ) 1  r 1e j

(1  re j z 1 )(1  re  j z 1 )(1  r 1e j z 1 )(1  r 1e j z 1 )

14/74
– If zero on unit circle
r  1, 1/ r  1

z0  e j

z0 1  e j  z0*

(1  e j z 1 )(1  e  j z 1 )

– If zero not exist on the unit circle

(1  rz 1 )(1  r 1 z 1 )

– If zeros on z  1
(1  z 1 )
15/74
 Necessary
zero

Necessary Necessary Necessary

zero zero zero

Fig. 7-2.
16/74
 4. FIR filter specifications

 Filter specifications
 p peak passband deviation (or ripples)
 r stopband deviation

 p passband edge frequency

r stopband edge frequency

s sampling frequency

17/74
 ILPF

Satisfies spec’s

Fig. 7-3.

18/74
 Characterization of FIR filter
N
y ( n )   h( k ) x ( n  k )
k 0

N
H ( z )   h( k ) z  k
k 0

– Most commonly methods for obtaining h(k )

• Window, optimal and frequency sampling methods

19/74
 5. Window method

 FIR filter
– Frequency response of filter H I ( )
– Corresponding impulse response hI (n)
1 
hI (n) 
2 H
 I ( )e jn d 

– Ideal lowpass response

1  1 c j n
2  2 c
j n
hI (n)  1 e d   e d

 2 f c sin(nc )
 , n  0, -  n  
= nc
 2 fc , n0


20/74
Fig. 7-4.

21/74
 Truncation to FIR
h(n)  hI (n) w(n) "windowing"

H(e j )  1
2 H I (e j ) W(e j )

 1
2 

H I (e j )W(e j (  ) )d

– Rectangular Window

1 , n  0,1,..., N
w(n)  
0 , elsewhere

22/74
Fig. 7-5.

23/74
Fig. 7-6.

24/74
Fig. 7-7.

25/74
 Table 7.2 summary of ideal impulse responses for standard frequency selective filters

hI (n)

hI (n) hI (n)

f c , f1 and f 2 are the normalized passband or stopband edge frequencies; N is the length of
filter

26/74
 Common window function types
– Hamming window

0.54  0.46 cos  2 n / N  , 0n N

w n  
 0 , elsewhere

F  3.32 / N

where N is filter length and

F is normalized transition width

27/74
– Characteristics of common window functions

Fig. 7-8.
28/74
Table 7.3 summary of important features of common window functions

29/74
– Kaiser window

  
2 

 I0   1   2 n N  

   N  
w n     , 0n N
 I0   

 0 , elsewhere

where I 0 ( x) is the zero-order modified Bessel function of the first kind

2
L
 ( x / 2) k 
I 0 ( x)  1   
k 1  k ! 
where typically L  25

30/74
• Kaiser Formulas – for LPF design

A  20log10 ( )
  min( p ,  s )
0, if A  21dB

  0.5842( A  21)0.4  0.07886( A  21), if 21dB  A  50dB
0.1102( A  8.7) if A  50dB

A  7.95
N
14.36F

31/74
 Example 7-2
– Obtain coefficients of FIR lowpass using hamming
window
Passband cutoff frequency f p : 1.5kHz
Transition width f : 0.5kHz
Stopband attenuation  50dB
Sampling frequency f s : 8kHz

• Lowpass filter
 sin  nc 
2 f c , n0
hI  n    nc
 2 fc , n0

F  f / f s  0.5 / 8  0.0625

N  3.32 / F  3.32 / 0.0625  53.12

N  54
32/74
• Using Hamming window
h(n)  hI (n) w(n) 0  n  54

w(n)  0.54  0.46cos(2 n / 54) 0  n  54

f c'  f p  f / 2  (1.5  0.25) 1.75[ kHz ]
f c  f c' / f s  1.75 / 8  0.21875

  N 
 c2 f sin  ( n  )c
2 , N
 n , 0 n N
 N 2
hI (n)   (n  )c
 2
 N
 2 fc , n
2

33/74
 2  0.21875
n  0 : hI (0)  sin(27  2  0.21875)
27  2  0.21875
 0.00655
w(0)  0.54  0.46 cos(0)  0.08
h(0)  hI (0) w(0)   0.00052398

2  0.21875
n 1 : hI (1)  sin(26  2  0.21875)
26  2  0.21875
 0.011311
w(1)  0.54  0.46 cos(2 / 54)
 0.08311
h(1)  hI (1) w(1)   0.00094054

34/74
 2  0.21875
n  2 : hI (2)  sin(25  2  0.21875)
25  2  0.21875
 0.00248397
w(2)  0.54  0.46 cos(2  2 / 54)
 0.092399
h(2)  hI (2) w(2)  0.000229516
  

2  0.21875
n  26 : hI (26)  sin( 1 2  0.21875)
1 2  0.21875
 0.312936

35/74
 w(26)  0.54  0.46 cos(2  26 / 54)
 0.9968896
h(26)  hI (26) w(26)  0.3112226

n  27 : hI (27)  2 f c  2  0.21875
 0.4375
w(27)  0.54  0.46 cos(2  27 / 54)
1
h(27)  hI (27) w(27)  0.4375

36/74
Fig. 7-9.

37/74
 Example 7-3
– Obtain coefficients using Kaiser or Blackman window
Stopband attenuation : 40dB
passband attenuation : 0.01dB
Transition region f : 500 Hz
Sampling frequency f s : 10kHz
Passband cutoff frequency f p : 1200 Hz

20 log(1   p )  0.01dB,  p  0.00115

20 log( r )  40dB,  r  0.01

   r   p  0.00115

38/74
• Using Kaiser window

A  7.95 58.8  7.95

N   70.82
14.36F 14.36(500 /10000)
N  71
  0.1102(58.8  8.7)  5.52
f c  f c' / f s  1, 450 /10, 000  0.145

 N 
2 f c sin  n   c 
hI (n)   2 
, 0n N
 N
 n   c
 2

39/74
 h(n)  hI (n) w(n)

2  0.145
n  0 : hI (0)  sin(35.5  2  0.145)
35.5  2  0.145

 0.00717

  2n  N  
2

I0   1    
  N   I (0)
w(0)    0  0.023
I0 ( ) I 0 (5.52)

h(0)  hI (0) w(0)  0.000164935

40/74
 2  0.145
n  1: hI (1)  sin(34.5  2  0.145)
34.5  2  0.145

 0.0001449

  69 
2 

I 0  5.52 1    
  71   I (1.3)
w(1)    0  0.0337975
I 0 (5.52) I 0 (5.52)

h(1)  h( 1)  hI (1) w(1)  0.000004897

41/74
 2  0.145
n  2 : hI (2)  sin( 34.5  2  0.145)
33.5  2  0.145

 0.007415484

  69 
2 

I 0  5.52 1    
  71   I (1.8266)
w(2)    0  0.04657999
I 0 (5.52) I 0 (5.52)

h(2)  h(2)  hI (2) w(2)  0.000345413

   

42/74
 2  0.145
n  35 : hI (35)  sin(0.5  2  0.145)
0.5  2  0.145

 0.280073974

  1  
2

I 0  5.52 1    
  71   I (5.51945)
w(35)    0  0.999503146
I 0 (5.52) I 0 (5.52)

h(35)  h(35)  hI (35) w(35)  0.279934818

43/74
Fig. 7-10.

44/74
 Summary of window method
1. Specify the ‘ideal’ or desired frequency response of filter, H I ( )
2. Obtain the impulse response, hI (n), of the desired filter by
evaluating the inverse Fourier transform
3. Select a window function that satisfies the passband or
attenuation specifications and then determine the number of filter
coefficients
4. Obtain values of w(n) for the chosen window function and the
values of the actual FIR coefficients, h(n), by multiplying
hI (n) by w( n)
h(n)  hI (n) w(n)

45/74
 Advantages and disadvantages
– Simplicity
– Lack of flexibility
– The passband and stopband edge frequencies cannot be
precisely specified
– For a given window(except the Kaiser), the maximum
ripple amplitude in filter response is fixed regardless of
how large we make N

46/74
 6. The optimal method

 Basic concepts
– Equiripple passband and stopband
E ( )  W ( )[ H I ( )  H ( )]

Weighted Weighting Ideal desired Practical

Approx. error function response response

min[max E ( ) ]

• For linear phase lowpass filters

– m+1 or m+2 extrema(minima and maxima)
where m=(N+1)/2 (for type1 filters) or m =N/2 (for type2 filters)

47/74
 Practical response
Ideal response

Fig. 7-11.

48/74
Fig. 7-12.

49/74
– Optimal method involves the following steps
• Use the Remez exchange algorithm to find the optimum set of
extremal frequencies
• Determine the frequency response using the extremal
frequencies
• Obtain the impulse response coefficients

50/74
 Optimal FIR filer design
N
H ( z )   h( k ) z  k
k 0

where h( n)  h(  n)
N /2 N /2
H ( )  h(0)   2h( k ) cos kT   a( k ) cos kT
k 1 k 0

where a (0)  h(0) and a ( k )  2h( k ), k  1, 2, , N / 2

Let f   / 2 f s  T / 2

1 , 0  f  fp
H1  f   
0 , f r  f  0.5

1
 , 0  f  fp This weighting function permits different peak
W  f   k
 1 error in the two band
, f r  f  0.5
51/74
 N /2
H ( f )  H (e j 2 f
)   a(k ) cos k 2 f
k 0

E ( f )  W ( f )[ H ( f )  H I ( f )]
where f are 0, f p  and  f r , 0.5

min[max E ( f ) ]

Find a (k )

52/74
– Alternation theorem
Let F  0, f p    f r , 0.5
 

If E ( f ) has equiripple inside bands F and more than m+2 extremal point

then H ( f )  HI ( f )

E ( fi )   E ( f i 1 )  e, i  0,1, , l  1

f 0  f1    f l (l  m  1)

where e  max E( f )
f  0, f p  &  f r ,0.5 

53/74
• From equation (7-33) and (7-34)

W ( f i )  H ( f i )  H I ( f i )   ( 1) i e, i  0,1, 2,  , m  1

• Equation (7-35) is substituted equation (7-32)

m
e
 a(k ) cos 2 kf
k 0
i  H I ( f i )  (1) i
W ( fi )
, i  0,1, 2,  , m  1

• Matrix form

1 cos 2 f 0 cos 4 f 0  cos 2 mf 0 1/ W ( f 0 )   a (0)   H I ( f 0 ) 

1 cos 2 f cos 4 f1  cos 2 mf1 1/ W ( f1 )   a (1)   H I ( f1 ) 
 1  
            
    
1 cos 2 f m cos 4 f m  cos 2 mf m ( 1) m / W ( f m )   a (m)   H I ( f m ) 
1 cos 2 f m1 cos 4 f m 1  cos 2 mf m1 ( 1) m! / W ( f m 1 )   e   H I ( f m1 ) 

54/74
– Summary
Step 1. Select filter length as 2m+1
Step 2. Select m+2 f i point in F
Step 3. Calculate a(k ) and e using equation (7)
Step 4. Calculate E ( f ) using equation (5). If E( f )  e in some
of f , go to step 5, otherwise go to step 6
Step 5. Determine m local minma or maxma points
Step 6. Calculate h(0)  a(0), h(k )  a(k ) / 2 when k  1, 2, , m
N
where H ( z )   h(k ) z
k 0
k

55/74
– Example 7-4
• Specification of desired filter
– Ideal low pass filter
– Filter length : 3
–  pT  1[rad], rT  1.2[rad]
– k 2

• Normalized frequency
f 0  0.5 / 2 , f1  1/ 2 , f 2  1.2 / 2

56/74
• From H I ( f 0 )  H I ( f1 )  1, H I ( f 2 )  0, W ( f 0 )  W ( f1 )  1/ 2 and W ( f 2 )  1

1 0.8776 2   a(0)  1 
1 0.5403 2  a(1)   1 
    
1 0.3624 1   e  0 

a (0)  0.645, a(1)  2.32, e  0.196

H ( f )  0.645  2.32 cos 2 f : not the optimal filter

• Cutoff frequency

1 1.2
f 0  f1  f 2  0.5
2 2

H I ( f 0)  1, H I ( f1)  H I ( f 2)  0, W ( f 0)  0.5, and W ( f1)  W ( f 2)  1

57/74
1 0.5403 2   a(0)  1 
1 0.3624 1  a(1)   0 
    
1 1 1   e   0

a(0)  0.144, a(1)  0.45, e  0.306

H ( f )  0.144  0.45cos 2 f : has the minimum max E ( f )

(N=3)

58/74
Fig. 7-13.

59/74
 Optimization using MATLAB
– Park-McClellan
– Remez
b  remez(N,F, M )

b  remez(N,F, M, WT )

where N is the filter order (N+1 is the filter length)

F is the normalized frequency of border of pass band
M is the magnitude of frequency response
WT is the weight between ripples

60/74
– Example 7-5
• Specification of desired filter
– Band pass region : 0 – 1000Hz
– Transition region : 500Hz
– Filter length : 45
– Sampling frequency : 10,000Hz

• Normalized frequency of border of passband

F  [0, 0.2, 0.3, 1]

• Magnitude of frequency response

M  [1 1 0 0]
61/74
Table 7-4.

62/74
Fig.7-14.

63/74
– Example 7-6
• Specification of desired filter
– Band pass region : 3kHz – 4kHz
– Transition region : 500Hz
– Pass band ripple : 1dB
– Rejection region : 25dB
– Sampling frequency : 20kHz

• Frequency of border of passband

F  [2500, 3000, 4000, 4500]

M  [0 1 0]

64/74
• Transform dB to normal value

Ap

10 20  1  Ar
p  Ap  r  10 20

10 20
1

where A p and Ar ripple value(dB) of pass band and rejection

band
• Filter length
– Remezord (MATLAB command)

65/74
Table 7-5.

66/74
Fig. 7-15.

67/74
 7. Frequency sampling method

 Frequency sampling filters

– Taking N samples of the frequency response at intervals
of kFs / N k  0,1, , N  1
– Filter coefficients h(n)
N 1
1
h( n) 
N
 H
k 0
( k ) e j (2 / N ) nk

where H (k ), k  0,1, , N  1 are samples of the ideal or target frequency response

68/74
– For linear phase filters (for N even)
 N2 1 
1 
h( n )   2 H (k ) cos[2 k (n   ) / N ]  H (0) 
N  k 1 
 

where   ( N  1) / 2

– For N odd
• Upper limit in summation is ( N  1) / 2

69/74
Fig. 7-16.

70/74
 – Example 7-7
1  N /21 
(1) Show the h ( n )  
N  k 1
2 H ( k ) cos[2 k ( n   ) / N ]  H (0) 


• Expanding the equation

1 N 1
h(n)   H (k ) e j 2 nk / N
N k 0
1 N 1
  H (k ) e  j 2 k / N e j 2 kn / N
N k 0
1 N 1
  H (k ) e j 2 ( n  ) k / N
N k 0
1 N 1
  H (k ) cos[2 k ( n   ) / N ]  j sin[2 k ( n   ) / N ]
N k 0
h(n) is real value
1 N 1
  H (k ) cos[2 k ( n   ) / N ]
N k 0 71/74
(2) Design of FIR filter
– Band pass region : 0 – 5kHz
– Sampling frequency : 18kHz
– Filter length : 9

Fig. 7-17.

72/74
• Samples of magnitude in frequency
 1, k  0,1, 2
H (k )  
 0, k  3, 4

Table 7-6.

73/74
8. Comparison of most commonly
method
– Window method
• The easiest, but lacks flexibility especially when passband and
stopband ripples are different
– Frequency sampling method
• Well suited to recursive implementation of FIR filters
– Optimal method
• Most powerful and flexible

74/74