Digital Signal Processing

H. Introduction to FIR filters design

Athanassios C. Iossifides
February 2013

2

.1 Discrete time systems and filters
.2 FIR filter design with windows

. Introduction to FIR digital filter design
3


.1 Discrete time systems and
filters
. Introduction to FIR digital filter design
4
H.1 Discrete time systems and filters
LSI system as frequency filters
The frequency response H(e
j
) of an LSI system, leads to the modification
of the input (e
j
) in the frequency domain, and creates the output (e
j
)
according to the formula

Therefore, every LSI system can be considered as a frequency filter, in
the sense that it passes, amplifies, attenuates or cuts frequencies of the
input. The terms LSI system and filter are used interchangeably.
The characteristics of a filter in the frequency domain depend on
The positions of the zeros and the poles of the transfer (system)
function
The positions of zeros and poles are determined
by the coefficients of the difference equation that describes the
system



( ) ( ) ( )
j j j
Y e H e X e =
5
H.1 Discrete time systems and filters
Distortionless response
The response y(n) of an LSI system to an input x(n) has no distortion (in
general) when it is of the form

so that the input signal is only uniformly attenuated with a constant C and
delayed by n
0
samples. Using the properties of DTFT, we have

so that
Therefore, a system does not introduce distortion when
The magnitude of the frequency response is constant
The phase is a linear function of frequency
The group delay is defined as

and provides the delay that undergoes each frequency passing through
the system. In the case of distortionless response systems, the group
delay is constant.


0
( ) ( ) y n Cx n n =

0
0
( ) ( )
jn j
x n n e X e
F

= =
0 0
( ) ( ) or ( )
j jn j j jn
Y e Ce X e H e Ce
( )
( ) , ( ) ( )
j
g
d
H e
d
= =
6
H.1 Discrete time systems and filters
Ideal filters
The frequency response of an ideal filter has unit amplitude (gain) at the
passband, zero amplitude at the stopband and linear phase.





0
| ( )|
j
H e
c

c


B
0
| ( )|
j
H e
c

c


0
| ( )|
j
H e
0

0


B
1

0
| ( )|
j
H e
0

0
| ( )|
j
H e
Lowpass filter (LPF) Highpass filter (HPF)
Bandpass filter (BPF)
Bandstop filter (BSF)
Allpass filter
7
H.1 Discrete time systems and filters
deal filters
he ideal Lowpass Filter (LPF) may be expressed by the frequency
response


The impulse response is calculated by the inverse DTFT as follows




This filter is not causal and therefore it is not realizable.





1, | |
( )
0, | |
c
j
c

H e

s

=

< s

1 1 1
( ) ( ) ( )
2 2 2
sin( )
,
c
c c
c

j jn jn j n j n

c
h n X e e d e d e e
jn
n
n
n

= = =
= < <
} }
8
H.1 Discrete time systems and filters
Transfer functions of real (non-ideal) filters
The principle of zeros and poles positioning for the production of the
desirable frequency response is to put
the poles near to the points of the unit circle that correspond to
the frequencies that we want to amplify (or not attenuate)
the zeros close (or exactly on) the frequencies that we want to
attenuate (make zero)
Additionally, the following restrictions should apply:
All the poles must be in the unto circle so that the filter is stable.
All the complex zeros and poles must be complex conjugate pairs
so that the coefficients of the filter in the diefference equation are
real.





1
0 1
0
1
1 1
(1 )
( )
1 (1 )
M M
k
k k
k k
N N
k
k k
k k
z z b z
H z b
a z p z

= =

= =

= =
+
[
[
9
H.1 Discrete time systems and filters
Examples of real (non-ideal) filters
Lowpass


Highpass


Bandpass

1
1
1
( )
1
z
H z
az

+
=

1
1
1
( )
1
z
H z
az

=
+
2
2
1
( )
1
z
H z
az

=
+
10
H.1 Discrete time systems and filters
FIR filter realization
An FIR filter is described by the difference equation


and a transfer function (including only zeros) of the form


This can be realized in a direct form as drawn below





This requires + 1 multiplications, additions and memory places.


0
( )
M
k
k
k
H z b z

=
=

0 0
( ) ( ) ( ) ( )
M M
k
k k
y n b x n k h k x n k
= =
= =

1
z
0
b
1
b
1
z
1
z
1
z
2
b
3
b
1 M
b
M
b
( ) x n
( ) y n
11
H.1 Discrete time systems and filters
FIR direct form realization
Simplified form









1
z
0
b
1
b
1
z
1
z
1
z
2
b
3
b
1 M
b
M
b
( ) x n
( ) y n
1
z
1
z
1
z
1
z
0
b
1
b
2
b
1 M
b
M
b
( ) y n
( ) x n
12
H.1 Discrete time systems and filters
FIR filters of ljnear phase
A filter has a linear phase when

where = 0 or /2 and constant. For an FIR filter in the interval [0,],
the linear phase condition leads to the following symmetry conditions of
the impulse response:




( ) ,
j
H e = < <
( ) ( ), / 2, 0
( ) ( ), / 2, / 2
h n h M n N
h n h M n N
= = =
= = =

V
13
H.1 Discrete time systems and filters
IIR filter realization
An IR filter is described by the difference equation


and a transfer function of the form






0
1
( )
1
M
k
k
k
N
k
k
k
b z
H z
a z

=
=
+

1 0
( ) ( ) ( )
N M
k k
k k
y n a y n k b x n k
= =
= +

14
H.1 Discrete time systems and filters
IIR filter realization
Direct Form of type I Direct Form of type II









Number of mult.: + + 1 Number of mult.: + + 1
Number of add.: + Number of add.: +
Memory places: + Memory places: max{,}


1
z
1
z
1
z
1
z
0
b
1
b
1 M
b

M
b
( ) y n ( ) x n
1
z
1
z
1
a
1 N
a

N
a
1
z
1
z
0
b
1
b
1 M
b

M
b
( ) y n ( ) x n
1
z
1
a
1 N
a

N
a
15


.2 FIR filter design with windows
. Introduction to FIR digital filter design
16
.2 FIR filter design with windows
The basic idea
The basic idea of the window method (windowing) is based on:
The selection of a non-causal ideal filter of infinite impulse
response duration.
The modification of the impulse response with a proper function
(window) so that to produce a causal and linear phase filter.
The modification of the impulse response h(n) is applied with
multiplication with a proper function w(n) which is called window.
An ideal Lowpass Filter (LPF) with cuttof frequency
c
has a frequency
response of the form


where a is a delay that does not affect the phase linearity of the filter and
is mandatory in order to convert the non-causal filter to a causal one.
1 , | |
( )
0, | |
ja
c j
d
c
e
H e

s
=

< s

17
.2 FIR filter design with windows
Impulse response of ideal filter
The impulse response of an ideal LPF may be found with an IDFT of the
previous formula, which leads to


and is symmetrical with respect to delay a.
sin[ ( )]
( )
( )
c
d
n a
h n
n a

18
.2 FIR filter design with windows
Conversion to an FIR filter
In order to produce a FIR filter, causal and with linear phase, we restrict
the impulse response with a proper window function w(n), that is
symmetrical with respect to a, so that to conserve the symmetry of the
final impulse response.







The greater the delay a is, the greater the impulse response duration is
and the better the ideal filter is approximated (this results in smaller
transition band)
( ) ( ) ( )
d
h n h n w n =
19
.2 FIR filter design with windows
Filter characteristics
The characteristics of the filters that are of interest are:
The order (length) of the filter.
The transition band width.
The attenuation in the stopband.
With the window method it is not possible to control independently the
passband and the stopband and the ripples are not uniform (equal).

=
+
=
+
10
10
1
20log
1
20log
1
p
p
p
s
s
p

20
.2 FIR filter design with windows

Classic windows (h(n) = 0, n e [0, N-1]):

( )
( )
( ) ( )
2
1
2
1
2 4
1 1
Rectangular : ( ) 1, 0 1
Hanning : ( ) 0.5 0.5cos , 0 1
Hamming : ( ) 0.54 0.46cos , 0 1
Blackman: ( ) 0.42 0.5cos 0.08cos , 0 1
n
N
n
N
n n
N N
w n n N
w n n N
w n n N
w n n N

= s s
= s s
= s s
= + s s
window
Transition
bandwidth f
Minimum stopband
attenuation
Rectangular 0.9/ 21dB
Hanning 3.1/ 44dB
Hamming 3.3/ 53dB
Blackman 5.5/ 74dB
21
.2 FIR filter design with windows
Kaiser parametric window
The parametric window Kaiser is defined as




Given
p
,
s
, R
p
, and A
s
, the length of the filter and the parameter are
calculated as:

= +
>

=

+ < <

0.4
Transition bandwidth:
2
7.95
Order (length) : 1
14.36
0.1102( 8.7), 50
Parameter
0.5842( 21) 0.07886( 21), 21 50
s p
s
s s
s s s

f

A
N
f
A A

A A A
( )
2
2
0
1
0
1 1
, 0 1
( )
( )
0,
n
N
I
n N
w n
I

(

s s
=

22
.2 FIR filter design with windows
( ) ( ) ( )
d
h n h n w n =
Windows comparison
Rectangular





Hanning






23
.2 FIR filter design with windows
Windows comparison
Hamming





Blackman






24
.2 FIR filter design with windows
Windows comparison
Kaiser = 2





Kaiser = 8






25
.2 FIR filter design with windows
Example (LPF with cutoff frequency 0.2)
Rectangular Hanning






Hamming Blackman
26
.2 FIR filter design with windows
Example (LPF with cutoff frequency 0.2)
Rectangular Hanning






Hamming Blackman
27
.2 FIR filter design with windows
Example (LPF with cutoff frequency 0.2)
Hamming, = 31 Hamming, = 63






Hamming, = 127 Hamming, = 255
28
.2 FIR filter design with windows
Example of FIR filter design with windows
A. Design a filter with the following characteristics


------------------------------------------------
Attenuation more than 50 dB is given by the Hamming window. We
calculate the transition bandwidth


The cutoff frequency is given by
The order of the filter is

The ripple of the passband is 0.039 dB (calculated only with a computer).
0.2 , 0.25dB
0.3 , 50 dB
p p
s s
R
A
= =
= =
0.3 0.2
0.05
2 2
s p


f

= = =
3.3
1 67

N N
f
(
= + =
(
(
0.25
2
s p
c


+
= =
29
.2 FIR filter design with windows
Example of FIR filter design with windows
( )
1
2
33 1
2
sin[ ( )]
sin[0.25 ( 33)]
( ) ( ) ( ) ( ) 0.54 0.46cos
( 33)
( )
N
c
n
d
N
n
n
h n h n w n w n
n
n

= = = (

30
.2 FIR filter design with windows
Example of FIR filter design with windows
B. Design the following filter with the Kaiser window


------------------------------------------------
We calculate the transition bandwidth


The cutoff frequency is


The ripple in the passband is 0.044 dB (calculated with computer).
0.2 , 0.25dB
0.3 , 50 dB
p p
s s
R
A
= =
= =
0.3 0.2
0.05
2 2
s p


f

= = =
0.25
2
s p
c


+
= =

(
= + + =
(
(
= =
7.95
Order (length) : 1 1 61
14.36
Parameter 0.1102( 8.7) 4.5512
s
s
A
N
f
A
31
.2 FIR filter design with windows
Example of FIR filter design with windows
( )
2
1
0
30 2
1
0
2
4.5512 1 1 sin[ ( )]
sin[0.25 ( 30)]
( ) ( ) ( ) ( )
( 30) (4.5512)
( )
n N
c
d
N
I n
n
h n h n w n w n
n I
n

(



= = =

32
.2 FIR filter design with windows
Example of FIR filter design with windows
C. Design the following bandpass filter


------------------------------------------------
Attenuation over 60 dB can be achieved by the Blackman or the Kaiser
window. We will design the Blackman.
The impulse response of the bandpass filter has the form



The cutoff frequencies are
1 1
2 2
0.4 , 60 dB, 0.5 , 1dB
0.7 , 1dB, 0.8 , 60 dB
s s p p
p p s s
A R
R A
= = = =
= = = =
1 1 2 2
1 2
0.45 , 0.75
2 2
s p s p
c c


+ +
= = = =
1 1
2 1
2 2
1 1
2 2
( ) ( ) ( )
sin[ ( )] sin[ ( )]
( )
( ) ( )
d
N N
c c
d
N N
h n h n w n
n n
h n
n n


=

=

33
.2 FIR filter design with windows
Example of FIR filter design with windows
The transition bandwidth is


The order of the filter is calculated as


The ripple in the passband is equal to R
p
= 0.0033 dB (calculated by a
computer).

5.5
1 111

N N
f
(
= + =
(
(
1 1 2 2
0.05
2 2
p s s p

f


= = =
34
.2 FIR filter design with windows
Example of FIR filter design with windows
( ) ( )
2 4
110 110
sin[0.75 ( 55)] sin[0.45 ( 55)]
( ) 0.42 0.5cos 0.08cos
( 55) ( 55)
n n
n n
h n
n n

(
= + (

(