DSP Lect 13 16

2006 Copyright 2006 Andreas Spanias 13-16-1
Design of FIR Digital Filters;

LINEAR PHASE
Lecture 13-15
Andreas Spanias
spanias@asu.edu
FIR Digital Filters
Advantages:
Linear Phase Design
Quite Efficient for designing notch filters
Always Stable
Disadvantages:
Requires High Order for Narrowband Design
Applications:
Speech Processing, Telecommunications
Data Processing, Noise Suppression, Radar
Adaptive Signal Processing, Noise Cancellation, Echo
Cancellation, Multipath channels
FIR Digital Filters
_
=
=
L
i
i
i n x b n y
0
) ( ) (
( ) n y
( ) n x
T T
0
b
L
b
...
.
.
.
.
.
+
+
+
+
_
1
b
.
L
L
z z X b z z X b z X b z Y

+ + + = ) ( ... ) ( ) ( ) (
1
1 0
L
L
z b z b b
z X
z Y
z H

+ + + = = ...
) (
) (
) (
1
1 0
FIR Filter Frequency Response
O O O
+ + + =
jL
L
j j
e b e b b e H ... ) (
1 0
s
f
f
t 2 = O
o
o
foldover
FIR Filter Design
1. LINEAR PHASE DESIGN
2. FOURIER SERIES DESIGN
3. ZERO PLACEMENT
4. FREQUENCY SAMPLING
5. LEAST SQUARES
6. IMPLEMENTATIONS
LINEAR PHASE DESIGN
Linear Phase (constant time delay) FIR filter design is important
in pulse transmission applications where pulse dispersion must
be avoided. The frequency response function of the FIR filter is
written as:
O O O O
+ + + + =
jL
L
j j j
e b e b e b b e H ... ) (
2
2 1 0
where
)) ( arg( ) ( , ) ( ) (
O O
= O u = O
j j
e H e H M
) (
) ( ) (
O u O
O =
j j
e M e H
GROUP DELAY
The time delay or group delay of a filter is defined as
O
O u
=
d
d ) (
t
therefore if is a linear function of then t is a
constant.
t is given in terms of samples
( ) O u
O
LINEAR PHASE AND IMPULSE RESPONSE
SYMMETRIES
It can be shown that linear phase is achieved if
) ( ) ( n L h n h =
where h(n) is the impulse response of the filter. For L = odd
_
=

+ =
2
1
0
) (
) )( ( ) (
L
n
n L n
z z n h z H
_
=
O
O
|
|
.
|
\
|
|
.
|
\
|
O =
2
1
0
2
2
cos ) ( 2 ) (
L
n
L
j
j
n
L
n h e e H
LINEAR PHASE DESIGN
if we define the pseudomagnitude
_
=
O
|
|
.
|
\
|
|
.
|
\
|
O =
2
1
0
1
2
cos ) ( 2 ) (
L
n
j
n
L
n h e H
Then
< H +
O
>
O
= O u
O
O
0 ) ( ,
2
0 ) ( ,
2
) (
1
1
j
j
e H
L
e H
L
hence the phase response is piecewise linear.
SYMMETRIC AND ANTI-SYMMETRIC
LINEAR PHASE FILTERS
Two Anti-symmetries for L=even or L=odd for
) ( ) ( n L h n h =
) ( ) ( n L h n h =
Two Symmetries for L=even or L=odd for
EXAMPLES OF SYMMETRIES
( ) n h
4 = L
0
( ) n h
0
( ) n h
0
( ) n h
0
4 = L
3 = L
3 = L
EXAMPLES OF PHASE AND SYMMETRY IN h(n)
Normalized frequency (Nyquist == 1)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-200
-150
-100
-50
0
50
P
h
a
s
e

(
d
e
g
r
e
e
s
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-200
-100
0
100
P
h
a
s
e
(
d
e
g
r
e
e
s
)
( ) n h
0
( ) n h
0
4 = L
4 = L
Note that for
) ( ) ( n L h n h =
then
) ( ) (
1
= z H z z H
L
Hence for L=odd and z = -1 then
) 1 ( ) 1 ( ) 1 ( = H H
L
Thus the filter must have a zero at and is therefore not
adequate for high-pass filtering.
t = O
HPF USING CERTAIN LINEAR PHASE FILTERS
Note that for
) ( ) ( n L h n h =
then
) ( ) (
1
= z H z z H
L
Hence any L and z = 1 then
) 1 ( ) 1 ( H H =
Thus the filter must have a zero at and is therefore not
adequate for low-pass filtering.
0 = O
LPF USING CERTAIN LINEAR PHASE FILTERS
Design of FIR Digital Filters
Lecture 14 - FIR DESIGN USING
THE FOURIER SERIES
Andreas Spanias
spanias@asu.edu
Design Using the Fourier Series
In filter design, there is an ideal transfer function H
d
(z) that is
approximated by H(z). For example for a low-pass filter, an
ideal frequency response is given below:
We like to minimize the mean square error, i.e.,
)
H
H
O O
O
H
e= d e H e H
j j
d
2
) ( ) (
2
1
( ) z H
d
c
O O
c
O
Design Using the Fourier Series (Cont.)
Recalling that in general
_
=
O O
=
2
1
) ( ) (
n
n i
ji j
e i h e H
then
)
_
H
H
=
O O
O
H
e= d e i h e H
n
n i
ji j
d
2
2
1
) ( ) (
2
1
Minimization of the integral above leads to an h(n) sequence that is
precisely equal to the sequence of Fourier series coefficients
characterizing H
d
(z). The resultant impulse response sequence is a
sampled truncated sinc function.
Fourier Series Design Example
For the ideal low pass filter the impulse response sequence is an
infinite length sampled sinc function. Lets say the sampling
frequency is 8KHz and we wish to have a cutoff frequency at
2KHz. This results in
2 8000
2000
2 2
H
= = H = O t
s
c
c
f
f
That is
( ) z H
d
2
n
2
n
O
Fourier Series Design Example (Cont.)
The ideal impulse response h
d
(n) is given by
,... 2 , 1 , 0 ,
2
sinc
2
1
) ( =
|
.
|
\
|
= n
n
n h
d
t
For an FIR filter of 11 coefficients
)
`
= ,... 0 , 0 ,
5
1
, 0 ,
3
1
, 0 ,
1
,
2
1
,
1
, 0 ,
3
1
, 0 ,
5
1
, 0 , 0 ... ) (
t t t t t t
n h
This impulse response is not causal, however a shift operator of 5
delays (z
-5
) will convert it into a causal one.
REALIZATION
. .
+
t 5 / 1 t 3 / 1
t / 1
2 1/
_
+ +
.
.
.
.
1
z
1
z
1
z
1
z
1
z
1
z
1
z
1
z
1
z
1
z
( ) n y
( ) n x
Fourier Series Design Example (Cont.)
t t t t t t 5
1
,
3
1
,
1
,
2
1
,
1
,
3
1
,
5
1
10 8 6 5 4 2 0
= = = = = = = b b b b b b b
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-600
-400
-200
0
P
h
a
s
e

(
d
e
g
r
e
e
s
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-80
-60
-40
-20
0
20
M
a
g
n
i
t
u
d
e

R
e
s
p
o
n
s
e

(
d
B
)
Fourier Series Design Example L=32
16 , 15 ,.... 0 ,...., 15 , 16 ,
2
) 16 (
sinc
2
1
=
|
.
|
\
|

= n
n
b
n
t
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-2000
-1500
-1000
-500
0
P
h
a
s
e

(
d
e
g
r
e
e
s
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-80
-60
-40
-20
0
20
M
a
g
n
i
t
u
d
e

R
e
s
p
o
n
s
e

(
d
B
)
Fourier Series Design Example L=64
32 , 31 ,.... 0 ,...., 31 , 32 ,
2
) 32 (
sinc
2
1
=
|
.
|
\
|

= n
n
b
n
t
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-3000
-2000
-1000
0
P
h
a
s
e

(
d
e
g
r
e
e
s
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-100
-50
0
50
M
a
g
n
i
t
u
d
e

R
e
s
p
o
n
s
e

(
d
B
)
Truncating with Hamming Window L=64
Hamming(L) *
2
) 32 (
sinc
2
1
|
.
|
\
|

=
t n
b
n
0 0. 1 0. 2 0. 3 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 1
-4000
-3000
-2000
-1000
0
Normal iz ed frequenc y (Ny quis t = = 1)
P
h
a
s
e

(
d
e
g
r
e
e
s
)
0 0. 1 0. 2 0. 3 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 1
-150
-100
-50
0
50
Normal iz ed frequenc y (Ny quis t = = 1)
M
a
g
n
i
t
u
d
e

R
e
s
p
o
n
s
e

(
d
B
)
F.S. Design Rectangular vs Hamming Window - L=64
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-150
-100
-50
0
50
M
a
g
n
i
t
u
d
e

R
e
s
p
o
n
s
e

(
d
B
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-100
-50
0
50
M
a
g
n
i
t
u
d
e

R
e
s
p
o
n
s
e

(
d
B
)
Hamming
Rectangular
~45 dB
~13 dB
Truncating in time- frequency convolution and F.S. design
The main-lobe width determines transition characteristics
The sidelobe level determines rejection characteristics
Ideal LPF
Narrow mainlobe=
Narrower transition
Wide mainlobe =
Wide transition
Notes On Fourier Series Design
The design performed in the previous example involved truncation
of an ideal symmetric impulse response.A symmetric impulse
response produces a linear phase response.
Truncation involves the use of a window function which is
multiplied with the impulse response. Multiplication in the time
domain maps into frequency-domain convolution and the spectral
characteristics of the window function affect the design.
The main-lobe width determines transition characteristics
The sidelobe level determines rejection characteristics
Design of FIR Digital Filters
Lecture 15 - FIR DESIGN USING
THE KAISER WINDOW
Andreas Spanias
spanias@asu.edu
( )
2
f H
dp + 1
1
dp 1
ds
LPF Ideal
f
fp
fc
Passband band Transition
Stopband
Regions Tolerance
DEFINING DESIGN SPECIFICATIONS
fs
DESIGN USING THE KAISER WINDOW
The Kaiser window is parametric and its bandwidth as well as its sidelobe
energy can be designed. Mainlobe bandwidth controls the transition
characteristics and sidelobe energy affects the ripple characteristics.
1 0 ,
) (
1
) (
0
2 / 1
2
0
s s
|
|
|
.
|
\
|
(
(
|
.
|
\
|
= L n
I
n
I
n w
|
o
o
|
o = L/2 ; associated with the order of the filter
| is a design parameter that controls the shape of the window
I
0
(.) is a zeroth Bessel function of the first kind
DESIGN USING THE KAISER WINDOW (Cont.)
I x
k
x
k
k
0
2
1
1
1
2
( )
!
= +
|
\
|
.
|
(
=
_
25 terms from the Bessel function are sufficient
EXAMPLES OF KAISER WINDOW
0 20 40 60 80 100 120 140
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
|
0
1
KAISER WINDOW DESIGN EQUATIONS
Given f
p
, f
s
, T and dp, ds determine the FIR filter coefficients.
T f f
A
dp ds
p s
) ( 2
log 20
) , min(
10
= AO
=
=
t
o
o
The filter order is
) 2 (
285 . 2
8
AO
=
A
L
| =
>
+ s s
<
0 1102 8 7 50
0 5842 21 0 07886 21 21 50
0 21
0 4
. ( . ),
. ( ) . ( ),
,
.
A A
A A A
A
and the kaiser parameter is given by |
DESIGN PROCEDURE
1. Determine the cutoff frequency for the ideal Fourier Series
method.
f
f f
c
s p
=

2
2. Design the ideal LPF using the Fourier Series.
3. Design the Kaiser window
4. Shift and truncate the ideal impulse response
L n
L
n h n w n h
d LPF
s s
|
.
|
\
|
= 0 ,
2
) ( ) (
Note that this procedure can be generalized for the design of
BPF, HPF, and BSF.
Design by Zero-Placement
As zeros are placed towards the unit circle the frequency
response magnitude decreases at and in the vicinity of the
frequency of the zeros.
.
o
o
foldover
Design by Zero-Placement
Example: Design a linear phase FIR filter for 60Hz interference
cancellation, that will pass a 10Hz signal of interest without
attenuation. The sampling frequency is 500Hz.
25
6
500
60
2
25 500
10
2
60
10
t
t
t
t
= = O
= = O
A second order filter is sufficient, since only a zero pair on the
unit circle is required for 60Hz cancellation. The 10Hz response
is adjusted with a gain factor.
Design by Zero-Placement (Cont.)
A linear phase (steady state) design means symmetric impulse
response:
0 ) (
1 ) (
60 60 60
10 10 10
2
0 1 0
2
0 1 0
= + + =
= + + =
O O O
O O O
j j j
j j j
e b e b b e H
e b e b b e H
and
77 . 2 , 9 . 1
1 0
~ ~ b b
Design by Zero-Placement (Cont.)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-50
0
50
100
150
P
h
a
s
e

(
d
e
g
r
e
e
s
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-60
-40
-20
0
20
M
a
g
n
i
t
u
d
e

R
e
s
p
o
n
s
e

(
d
B
)
-1 -0.5 0 0.5 1
-1
-0.5
0
0.5
1
Real part
Im
a
g
in
a
r
y

p
a
r
t
O
O
O O O
+ =
j j j
e e e H
2
9 . 1 77 . 2 9 . 1 ) (
Frequency Sampling Methods for FIR Filter design
The Frequency Sampling Method (FSM) involves: a) uniform
sampling of a desired continuous frequency response function at N
points, b) applying the N-point inverse Discrete Fourier Transform
to obtain an N-point impulse response.
The FSM guarantees that the FIR frequency response matches that
of the desired filter at the sampled points, however the response
between the sampled points is different. If the desired filter
function is due to an infinite impulse response then the FSM
suffers also from time-domain aliasing.
Min-Max and Parks-McClellan Optimum FIR Design
The Parks-McClellan design is based on Min-Max
Equiripple and linear phase design is possible
This class of methods involve minimizing the maximum error
between the designed FIR filter frequency response and a prototype
{ }
{ } ) ( max min
,..., 1 , 0 ), (
u j
L i i h
e E
=
where
E e W e H e H e
j j
d
j j
( ) ( )( ( ) ( ))
u u u u
=
FIR Filter Design Using MATLAB
+
IN THE MATLAB SP TOOLBOX
cremez - Complex and nonlinear phase equiripple FIR filter design.
fir1 - Window based FIR filter design - low, high, band, stop, multi.
fir2 - Window based FIR filter design - arbitrary response.
fircls - Constrained Least Squares filter design - arbitrary response.
fircls1 - Constrained Least Squares FIR filter design - low and highpass
firls - FIR filter design - arbitrary response with transition bands.
firrcos - Raised cosine FIR filter design.
intfilt - Interpolation FIR filter design.
kaiserord - Window based filter order selection using Kaiser window.
remez - Parks-McClellan optimal FIR filter design.
remezord - Parks-McClellan filter order selection.
+ MATLAB is registered trade mark of the MathWorks
FIR Filter Realizations
Direct Realizations
_
=
=
L
i
i
i
z b z H
0
) (
...
...
( ) n x
( ) n y
1
z
1
z
1
z
0
b
1
b 2
b
1 L
b
L
b
_
_ _ _
Require multiply accumulate instructions
FIR Filter Cascade Realizations
Cascade Realizations
[
=

+ + =
q
i
i i i
z b z b b z H
1
2
2
1
1 0
) ( ) (
...
1
z
1
z
1
z
1
z
_
_
_
_
10
b
11
b
12
b
20
b
21
b
22
b
( ) n x
Reduced Effects from Coefficient Quantization and round-off
In Fixed-Point implementation signal scaling must be
done carefully at each stage
Transform-Domain FIR Filter Realizations
A Transform domain realization is possible using the overlap
and save and the FFT. This yields computational savings for high
order implementations. Input data is organized in 2N-point blocks
and blocks are shifted N points at a time. The data blocks and
N zero-padded coefficients are transformed and multiplied and the
results is inverse transformed. The last N-points are selected as the
result. The blocks are updated and the process is repeated.
F F T I F F T
x y
-
X k ( 0 )
X k ( 1 )
X k ( L )
B k ( 0 )
B k ( 1 )
B k ( L )
Select Last N points
x(n)
y(n)
Transform-Domain FIR Filter Realizations (2)
Implements an N-th order filter with two 2-N point FFTs
For processing N points complexity is reduced from
O(N
2
) to O(Nlog
2
N)
FF T IF F T
x y
-
X k( 0 )
X k( 1 )
X k( L )
B k( 0)
B k (1 )
B k ( L)
Select Last N points
x(n)
y(n)
Simple and Smart FIR Filter Realizations
_
=
=
L
i
i
z
L
z H
0
1
) (
A simple L-tap filter can be built such that no multiplies are not used
This is a LPF with linear phase if L is radix 2 the division can be
implemented with shifts - below an example with L=16
0 0 . 1 0 . 2 0 . 3 0 . 4 0 . 5 0 . 6 0 . 7 0 . 8 0 . 9 1
- 2 0 0
- 1 0 0
0
1 0 0
N o r m a l i z e d f r e q u e n c y ( N y q u i s t = = 1 )
P
h
a
s
e

(
d
e
g
r
e
e
s
)
0 0 . 1 0 . 2 0 . 3 0 . 4 0 . 5 0 . 6 0 . 7 0 . 8 0 . 9 1
- 6 0
- 4 0
- 2 0
0
N o r m a l i z e d f r e q u e n c y ( N y q u i s t = = 1 )
M
a
g
n
it
u
d
e

R
e
s
p
o
n
s
e

(
d
B
)
Reducing Complexity by realizing FIR as IIR
) 1 (
1 1
) (
1
) 1 (
0
+
=
= =
_
z L
z
z
L
z H
L L
i
i
An FIR filter with all its coefficients equal can be realized by
looking at geometric series convergence and pole-zero cancellation
The IIR filter can be implemented using a simple difference equation
with long delay. Precision may become a problem as the pole zero
cancellation is on the unit circle.
1
( ) ( ( ) ( 1)) ( 1) y n x n x n L y n
L
= +
Pole-zero cancellation in FIR to IIR transformation
) 1 (
1 1
) (
1
) 1 (
0
+
=
= =
_
z L
z
z
L
z H
L L
i
i
-1 -0.5 0 0.5 1
-1
-0.5
0
0.5
1
Real part
I
m
a
g
i
n
a
r
y

p
a
r
t
Pole zero
cancellation at z=1
Modifying Filter Response by Subtractive Operations
) ( ) ( 1
O O
=
j
HPF
j
LPF
e H e H
__
=
t
HPF LPF ALL-PASS
t
c
O
c
O
c
O
c
O
Using simple operations like this we can transform prototype
LPF to HPF or BPF, etc.
Implementing Efficiently Digital Cross-Over Using
Subtractive Operations
LPF E to tweeter
to woofer Linear Phase LPF
With delay L/2 samples
-
+
could also implement delay compensation z
L/2
Frequency Sampling
/2

0 /4
Ideal frequency response
N=16 samples
/4

0
Sampled ideal frequency response
/4

0
Interpolated frequency response
(a)
(b)
(c)
1
2 /
0
( ) ( ) ( )
k
N
j j kn N
d d d
n
H k H e h n e
t
O
=
= =
_
( ) ( ) 0 1
d
k
h n h n kN n N
=
= + s s
_
; aliasing

DSP Lect 13 16

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

DSP Lect 13 16

Hochgeladen von

Copyright:

Verfügbare Formate

2006 Copyright 2006 Andreas Spanias 13-16-1

Design of FIR Digital Filters;

Das könnte Ihnen auch gefallen