Digital Signal Processing

Ch5: Design of Digital Filters

Lecturer: Nguyen-Son Vo (Ph.D), email: vonguyenson@gmail.com

1/37
 Contents
5.1 General considerations

5.2 Design of FIR filter

5.3 Design of IIR filter from analog filters

5.4 Freq. transformations

5.5 Summary

5.6 Problems

2/37
 5.1 General considerations
 What we have gained?
• FIR & IIR
• DFT/IDFT & FFT/IFFT
• Desired filter characteristics, i.e., magnitude & phase res.
 What we are going to do?
• Further study on FIR & IIR in terms of filter designs, e.g.,
sidelobes in the stopband, complexity,…
• Freq. transformation in both analog & digital sig. for
lowpass, bandpass, bandstop, or high pass filters
• Various necessary design of digital filters
3/37
 5.1 General considerations
 Causality & its implications
• Let us examining the impulse res. h(n) of an ideal lowpass
filter with freq. res. given by

1,   c
H ( )  
0, c |  | 

• The impulse res. of this filter is

 c
  , n  0
h( n)  
 c sin c n , n  0
  c n
4/37
 5.1 General considerations
 Causality & its implications
• But non-causal and thus cannot be realized in practice

Matlab: Ch_5_Ideal_Lowpass_Filter.m 5/37

 5.1 General considerations
 Causality & its implications
• What are the necessary & sufficient conditions that H()
must satisfy to ensure the resulting filter to be causal?
 Paley-Wiener theorem
• Paley-Wiener theorem
If h(n) has finite energy & h(n) = 0 for n < 0, then

 ln H ( ) d   


Conversely, if |H()| is square-integrable the above integral is

finite, the resulting filter with H() = |H()|ej() is causal
6/37
 5.1 General considerations
 Causality & its implications
• Limit to LTI sys. specified by difference equation in practice
N M 1
y (n)    ak y (n  k )   bk x(n  k )
k 1 k 0
M 1
 bk e j k
k 0
H ( )  N
1   ak e j k
k 1
 Causal & physically realizable
 Approximate any of the ideal freq. response characteristics
by properly selecting ak, bk, M, & N
7/37
 5.1 General considerations
 Characteristics of practical freq.-selective filters

8/37
 5.2 Design of FIR filters
 Symmetric & anti-symmetric FIR filters
• An linear-phase FIR filter of length M described by dif. equ.
M 1 M 1
y ( n)   h(k ) x(n  k )   bk x(n  k )
k 0 k 0

h(k )  bk
M 1
H ( z)   h( k ) z  k
k 0

 It is a polynomial of degree M – 1 in the variable z-1

 The roots of this polynomial yield the zeros of the filter
9/37
 5.2 Design of FIR filters
 Symmetric & anti-symmetric FIR filters
• An FIR filter has linear phase if its unit sample res. satisfies
the condition h(n)   h( M  1  n), n  9,1,.., M  1

10/37
 5.2 Design of FIR filters
 Symmetric & anti-symmetric FIR filters
• An FIR filter has linear phase if its unit sample res. Satisfies
the condition h(n)   h( M  1  n), n  9,1,.., M  1

11/37
 5.2 Design of FIR filters
 Symmetric & anti-symmetric FIR filters
• Utilizing the sym & anti-sym, we have

H ( z )  h(0)  h(1) z 1  h(2) z 2  ...  h( M  1) z ( M 1)

 ( M 3)/2 
  M  1 
   h(n)  z
 ( M 1)/2        , M odd
z h 
( M 1 2 n )/2
z ( M 1 2 n )/2

  2  n 0 
( M /2) 1
 z ( M 1)/2  h(n)  z ( M 1 2 n)/2  z ( M 1 2 n)/2 , M even
 
n 0

• We also have
M 1 M 1
z  z 1
H ( z)   h( k ) z k
 1
 H (z )   h( k ) z k
k 0 k 0 12/37
 5.2 Design of FIR filters
 Symmetric & anti-symmetric FIR filters
M 1 M 1
H (z ) 1
 h( k ) z  z
k  ( M 1) 1
H (z )   h(k ) z k z ( M 1)
k 0 k 0
M 1
z  ( M 1) 1
H (z )   h(k ) z ( M 1 k )   H ( z ), h(n)   h( M  1  n)
k 0

 H ( z )   H ( z 1 ) z ( M 1)

• If z1 is a root or a zero of H(z), the 1/ z1 is also a root

• If z1 is a complex-valued root, z1* is also a root, H(z) also has
a zero at 1/ z1*
13/37
 5.2 Design of FIR filters
 Symmetric & anti-symmetric FIR filters
• When h(n) = h(M – 1 – n)

H ( )  H r ( )e  j ( M 1)/2
  M  1  ( M 3)/2  M 1 
 
h   2  h ( n ) cos    n , M odd
  2  n 0  2 
H r ( )  
( M /2) 1
  M 1 
 
2 h ( n ) cos    n , M even
 n 0  2 
  M 1 
  2  , if H r ( )  0
  
( )  
  M  1    , if H ( )  0
  2  r
14/37
 5.2 Design of FIR filters
 Symmetric & anti-symmetric FIR filters
• When h(n) = –h(M – 1 – n) page 683/1104
H ( )  jH r ( )e  j ( M 1)/2  H r ( )e j[  ( M 1)/2  /2]
 ( M 3)/2  M 1 
 
2 h ( n ) sin    n , M odd
 n 0  2 
H r ( )  
( M /2) 1
  M 1 
 
2 h ( n ) sin    n , M even
 n 0  2 
  M 1 
 2    2  , if H r ( )  0
  
( )  
 3    M  1  , if H ( )  0
 2   r
 2  15/37
 5.2 Design of FIR filters
 Symmetric & anti-symmetric FIR filters
• h(n) = h(M – 1 – n):
(M + 1)/2 filter coefficients when M is odd
M/2 filter coefficients when M is even
None-zeros res. at  = 0
• h(n) = –h(M – 1 – n):
(M – 1)/2 coefficients when M is odd, due to h(M – 1)/2 = 0
M/2 coefficients when M is even
Zero res. at  = 0 &  = 
We now design the desired freq. response Hd() of the FIR filter
16/37
 5.2 Design of FIR filters
 Design of linear-phase FIR filters using windows
• Fourier transform relations of Hd() & hd(n)

H d ( )   d
h ( n ) e  j n

n 0

1 j n
hd (n)  
2 
H d ( )e d

• hd(n), which is infinite, must be truncated at n = M – 1, to

yield an FIR filter of length M
• To do so, we multiply hd(n) by a “”rectangular window
1, n  0,1,..., M  1
w(n)  
0, otherwise
17/37
 5.2 Design of FIR filters
 Design of linear-phase FIR filters using windows
• So, the unit response of the FIR filter becomes
hd (n), n  0,1,..., M  1
h(n)  hd (n) w(n)  
0, otherwise
• It is easy to derive the Fourier transform of w(n)
M 1
1  e j M sin( M / 2)  j ( M 1)/2
W ( )  e  j n

1  e j

sin( / 2)
e
n 0

sin( M / 2)  ( M  1) / 2,sin( M / 2)  0

W ( )  and ( )  
sin( / 2)  ( M  1) / 2   ,sin( M / 2)  0
18/37
 5.2 Design of FIR filters
 Design of linear-phase FIR filters using windows
• Freq. res. of rectangular window

19/37
 5.2 Design of FIR filters
 Design of linear-phase FIR filters using windows
Ex. 5.1: Assume that we want to design a sym. lowpass linear-
phase FIR filter having a desired freq. res. Below
e j (M 1)/2 , 0    c
H d ( )  
0, otherwise
A delay of (M – 1)/2 units is incorporated into Hd() in
anticipation of forcing the filter to be of length M
c
1 sin c [n  ( M  1) / 2]
hd (n)   e j[ n ( M 1)/2] d   , n  ( M  1) / 2
2 c
 [n  ( M  1) / 2]

Clearly, hd(n) is noncausal & infinite in duration

20/37
 5.2 Design of FIR filters
 Design of linear-phase FIR filters using windows
Limit hd(n) by w(n) we have
sin c [n  ( M  1) / 2]
h( n)  , 0  n  M  1, n  ( M  1) / 2
 [n  ( M  1) / 2]
 M  1  c
If M is odd, the value of h(n) at n = (M – 1)/2 is h  
 2  

21/37
 5.2 Design of FIR filters
 Design of linear-phase FIR filters by freq.-sampling
Hd() is equally sampled by
 M 1
 k  0,1,..., , M odd
2  2 1
k  (k   ),  ,   0,
M k  0,1,..., M  1, M even 2
 2
 2 
Define H (k   )  H  (k   ) 
M 
M 1
Because H ( )   h(n)e  j n
n 0
M 1
We have H (k   )   h(n)e  j 2 ( k  ) n / M , k  0,1,..., M  1
n 0
22/37
 5.2 Design of FIR filters
 Design of linear-phase FIR filters by freq.-sampling
Inversely we obtain
M 1
1
h( n) 
M
 H (k   )e j 2 ( k  ) n / M , n  0,1,..., M  1
k 0

When  = 0, they reduce to DFT & IDFT

Now, if we sample H() by  k = 2(k + )/M in slides 14 & 15,
we have
 2 
H (k   )  H r  (k   )  e j[  /2 2 ( k  )(M 1)/2 M ]
M 

where  = 0 when h(n) is sym. &  = 1 when h(n) is anti-sym.

23/37
 5.2 Design of FIR filters
 Design of linear-phase FIR filters by freq.-sampling
If we further define
 2 
G (k   )  (1) H r 
k
(k   )  , k  0,1,..., M  1
M 
Then
 2  j[  /2 2 ( k  )( M 1)/2 M ]
H (k   )  H r  (k   )  e
M 
 G (k   )e j k e j[  /2 2 ( k  )( M 1)/2 M ]
Finally, h(n) can be obtained in 4 cases, i.e.,  = 0,  = 1/2,  =
0, &  = 1, by substituting H(k + ) to the eq. in slide 23
Detail results can be referred to Table 10.3
24/37
 5.2 Design of FIR filters
 Design of linear-phase FIR filters by freq.-sampling

Can
we
design
a passband
Filter?

Matlab: Ch_5_Ex_5_2.m 25/37

 5.2 Design of FIR filters
 Design of …

Read at home… 26/37

5.3 Design of IIR filters from analog filters
 Techniques
• There are several methods used to design digital filter (DF)
having an infinite-duration unit sample response
• AF design is a mature & well-developed field
• Design a AF and then convert the design into the DF
• An analog filter can be described by its system function

B( s)
 k
 s k

k 0
H a (s)   N
A( s )
 k
 s k

k 0
27/37
5.3 Design of IIR filters from analog filters
 Techniques
• Its impulse res. is related to Ha(s) by Laplace transform

H a (s)   h(t )e  st dt

• It can be described by linear constant-coefficient dif. equ.
N
d k y (t ) M
d k x(t )
 k dt k
  k
dt k
k 0 k 0
Note:
1) An analog LTI sys. with the sys. function is stable if its poles
lie in the left half of the s-plane (LHP)
2) Not a linear-phase filter
28/37
5.3 Design of IIR filters from analog filters
 IIR filter design by approximation of derivatives
• An equivalent dif. equ. for the first derivative is given by
dy (t ) y (nT )  y (nT  T ) y (n)  y (n  1)
 
dt t  nT T T

29/37
5.3 Design of IIR filters from analog filters
 IIR filter design by approximation of derivatives
• An equivalent dif. equ. for the second derivative is given by
d 2 y (t ) d  dy (t ) 
 
dt 2 dt  dt  t  nT
t  nT
[ y (nT )  y (nT  T )] / T  [ y (nT  T )  y (nT  2T )] / T

T
y (n)  2 y (n  1)  y (n  2)

T2

1 2 1 z 1  2
1  2z z
s 
2
 
2  T
T  
30/37
5.3 Design of IIR filters from analog filters
 IIR filter design by approximation of derivatives
• It is easily follows to obtain the kth derivative
1 z 1  k
s 
k
 T 
 
H ( z )  H a ( s ) s (1 z1 )/T
• An Lth-order dif. is proposed in the form of
dy (t ) 1 L y (nT  kT )  y (nT  kT )
  k
dt t  nT T k 1 T

where {k} are a set of parameters that can be selected to

optimize the approximation
31/37
5.3 Design of IIR filters from analog filters
 IIR filter design by approximation of derivatives
• It is easily follows to obtain the kth derivative
1 z 1  k
s 
k
 T 
 
H ( z )  H a ( s ) s (1 z1 )/T
• An Lth-order dif. is proposed in the form of
dy (t ) 1 L y (nT  kT )  y (nT  kT )
  k
dt t  nT T k 1 T

where {k} are a set of parameters that can be selected to

optimize the approximation
32/37
5.3 Design of IIR filters from analog filters
 IIR filter design by approximation of derivatives
• The resulting mapping between s-plane & z-plane is now

1 L
s    k ( z k  z k )
T k 1
• When z = ej, we have
2 L
s  j   k sin  k
T k 1

• Given an AF, we can convert it to a DF by the relationship

between s & z

33/37
5.3 Design of IIR filters from analog filters
 IIR filter designs by …

Read at home… 34/37

 5.4 Freq. transformations
 To convert LPF to X-PF

Read at home… 35/37

 5.5 Summary

Filter using window

FIR filter

IIR (s2z)

Others (further read at home)

Using Matlab to design FIR filter

36/37
 5.6 Problems
 Problems
• All the problems related the lecture in Chapter 10 [4]
 Matlab to compute, plot, and check
• Filter using window
• FIR filter
• IIR
 Problems are solved mostly by students in the class
 Group leaders have to submit Matlab files to my Zalo before the next
lecture (more details higher scores)

DO NOT COPY FROM EACH OTHER!

37/37