ee225d spring,1999 digital filters
TRANSCRIPT
![Page 1: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/1.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.1
PE Spring,1999
25D
ORGAN / B.GOLD LECTURE 7
University of CaliforniaBerkeley
College of EngineeringDepartment of Electrical Engineering
and Computer Sciences
rofessors : N.Morgan / B.GoldE225D
Digital Filters
Lecture 7
![Page 2: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/2.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.2
n)
n
1– dz) 1 az 1––( )-------------------------
an 1+
a–------------ n 0≥
25D
ORGAN / B.GOLD LECTURE 7
1. Example of inverse z-transform use.
- Let input be and filter be u n( ) y n( ) ay n 1–( ) x(+=
X z( ) x n( )z 1–
n 0=
∞
∑=
Y z( ) 11 az 1––-----------------=
x n( ) u n( )=
so y n( ) 12πj-------- zn
1 z 1––(-----------------∫°=
Basic theorem1
2πj-------- zn 1– dz
1 az 1––-----------------∫° an for n 0≥=
0 for n 0<=
This allows computation of the integral to be
This result can be proved by iteration.
y n( ) 1 –1
-----=
![Page 3: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/3.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.3
)
h m( )e j– wn
n 1+
nsform evaluated
nsmit h m( ) 0⇒
m ∞→ thus sum 0→,as n ∞→
25D
ORGAN / B.GOLD LECTURE 7
2. Steady state respond to a complex exponential ejwnu n(
y n( ) h m( )x n m–( )m 0=
n
∑ x m( )h n m–( )m 0=
n
∑= =
If x n( ) ejwnu n( ) , then y n( ) from above is =
y n( ) h m( )ejw n m–( )
m 0=
n
∑ ejwn h m( )e j– wm
m 0=
n
∑= =
m 0=
n
∑ m n 1+=
∞
∑–m 0=
∞
∑= , so y n( ) ejwn h m( )ejwn
m 0=
∞
∑m =
∞
∑–=
Steady state value of y n( ) ejwn H z( )[ ]z ejw==
So the Frequency response is the value of the z-tra
on the unit circle.
steady state Tra
![Page 4: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/4.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.4
esponse z
z a–-----------=
x z-plane,
is the product of
ectors to the poles.
25D
ORGAN / B.GOLD LECTURE 7
3. Geometric Interpretation of Steady State Frequency R
for simple first order diff equation : H z( ) 11 az 1––-----------------=
ω
ef
Unit circle
H z( ) at z ejw is ef--=
General Rule
Given a collection of poles and zeros in the comple
the Frequency response at any is where
all vectors to the zeros and is the product of all v
[special rules apply for multiple pole and zeros.]
ND---- NW
D
![Page 5: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/5.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.5
n
25D
ORGAN / B.GOLD LECTURE 7
Preview of the Rest of the Material
1. Filtering concepts. - approximate problem
2. Sampling and Impulse Invariance
3. Bilinear Transformation
4. The DFT
5. Circular Convolution and Linear Convolutio
6. Basic FFT Concept.
7. DFT’s and Filter Banks
![Page 6: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/6.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.6
Ideal low pass
Ideal band pass
Ideal band step
Ideal band pass differentiator
ω
ω
ω
ω
25D
ORGAN / B.GOLD LECTURE 7
5. Approximation Problem
Example of Ideal Filters
Important Point
Linear Analog filters of R, L, C
must have frequency responses
that are rational functions in ω.
Similary, linear digital filter
must have rational functions in
ejω
A ω( )
ωC– ωC
![Page 7: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/7.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.7
y specifying
ifferent n.
ω
25D
ORGAN / B.GOLD LECTURE 7
Analog designers tackle the approximation problem b
a REAL function on the jω axis.
Example : H jω( ) 2 1
1ωωc
-----
2n
+
-----------------------=
Figure 7.1 : Butterworth Frequency Response for D
H ω( )
![Page 8: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/8.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.8
e for n=4.
ω
25D
ORGAN / B.GOLD LECTURE 7
Figure 7.2 : Chebyshev Frequency Respons
T ω( )
![Page 9: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/9.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.9
tion in the complex
ehe s-plane.
.
25D
ORGAN / B.GOLD LECTURE 7
If a suitable is chosen, it can lead to a specifica
s-plane of and this function holds true Everywher in t
Let’s normalize, so that and then let
H jω( ) 2
H s( )
γ ωωC
------= s jγ=
So H s( )H * s( ) 11 s2–( )n+----------------------=
![Page 10: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/10.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.10
n 1=
n 2=
n 3=
H s–( )
25D
ORGAN / B.GOLD LECTURE 7
H s( )H s–( ) 11 s2–------------=
H s( )H s–( ) 11 s4+-------------=
H s( )H s–( ) 11 s6–------------=
H s( )
![Page 11: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/11.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.11
s, A = Butterworth,
r a low pass filter
25D
ORGAN / B.GOLD LECTURE 7
Figure 7.4 : Comparison of Group Delay for Four Type
B = Chebyshev, C = Elliptic, D = Bessel, fo
with a 500Hz corner frequency.
![Page 12: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/12.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.12
bank yielded
sponse and
quency responses?
n impulse response.
me
25D
ORGAN / B.GOLD LECTURE 7
Raders channel vocoder experiment - Butterworth filter
better results than Chebyshev.
Note : Bessel and Lenner filters have good phase re
were used in Vocoders.
Question :
How do we construct digital filters that give good fre
* Impulse Invariance - Linear analog filters have a give
h t( )
tiT
![Page 13: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/13.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.13
’s.
25D
ORGAN / B.GOLD LECTURE 7
Figure 7.16 : Aliasing Effects of Hopped FFT
![Page 14: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/14.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.14
e
es1T––
eal
jω
H s( )
s1– α
25D
ORGAN / B.GOLD LECTURE 7
Construct a Digital Filter that has an impulse respons
that are the samples of . h t( )
h n( ) h nT( )=
h n( ) L 1– A1
S S1+-------------
A1es1t–= =
h n( ) A1es1nT–=
H z( ) h n( )z n–
n 0=
∞
∑ A1
1 es1T– z 1––
----------------------= =
Start with a simple analog filter.
and
s1 is r
![Page 15: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/15.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.15
.
.
25D
ORGAN / B.GOLD LECTURE 7
Procedure
- Find impulse response of suitable analog filter
- Sample it to find .
- Take z-transform to find transfer function
h n( )
H z( )
![Page 16: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/16.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.16
es1T––
Case.
and
y n 1–( ) x n( )+
s1
1RC--------–=
25D
ORGAN / B.GOLD LECTURE 7
Example of Impulse Invariant Design for a Very Simple
Filter is
Digital Filter
+
RC
z 1–
e–s1T
H z( ) 11 z 1– e
s1T–
---------------------=
y n( ) es1T=
1 sC( )⁄
R 1sC------+
------------------ 11 sRC+--------------------=
![Page 17: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/17.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.17
to find the digital filter.
is maps into
of z-plane unit circle
z-plane unit circle.
ircle.
25D
ORGAN / B.GOLD LECTURE 7
Aliasing is prevented by using the bilinear transfom
When ax
Stated without proof - Left half s-plane imterior
Right half s-plane exterior of
sz 1–z 1+----------- z,→ 1 s+
1 s–-----------=
s jω= z 1 jω+1 jω–---------------= z 1= jω
unit c
⇒
⇒
![Page 18: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/18.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.18
11– )RC
-------------------- H z( )=
1 RC–1 RC+-----------------
appear at .z 1–=
25D
ORGAN / B.GOLD LECTURE 7
As
11 sRC+-------------------- 1
1z 1–z 1+-----------RC+
-----------------------------⇒ z+z 1 z(+ +----------------------=Simple example
H z( ) z 1+1 RC–( ) z 1 RC+( )+
-----------------------------------------------------=
ω Π H ejω( ) 0⇒,→No Folding.
For more complex filter designs, multiple zeros
![Page 19: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/19.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.19
o z-transformo Fourier transformo Laplace transformo Fourier Series
25D
ORGAN / B.GOLD LECTURE 7
Discrete Fourier Transform
Related tRelated tRelated tRelated t
Consider a finite duration sequence.
Inverse
Important parameters
Size of DFT
Size of data.
Window
How often the DFT is done.
Sampling rates.
Xk x n( )Wnk
n 0=
n 1–
∑= W ej
2πN------
–
=
x n( ) 1N---- XkW
nk–
k 0=
N 1–
∑=
N
![Page 20: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/20.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.20
recall that an “odd”
equence is symmetric.
25D
ORGAN / B.GOLD LECTURE 7
Table 7.1 : Relations between Sequence and its DFT ;
sequence is antisymmetric, and an “even” s
![Page 21: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/21.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.21
cular convolution
can be made equivalent
25D
ORGAN / B.GOLD LECTURE 7
* The DFT can implement an FIR filter exactly.
a) The product of two DFT’s corresponds to the cir
of two signals
to linear convolution.
b) By augmenting with zeros, circular convolution
and
Because
![Page 22: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/22.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.22
y3
y2
y1
x0
x2
x3
y0
y2
y1
x0
x2
x3
25D
ORGAN / B.GOLD LECTURE 7
The DFT can Implement Linear Convolution.
Circular
y0 x1
Xk x0W0 x1W
k x2W2k x3W
3k+ + +=
Yk y0W0 y1W
k y2W2k y3W
3k+ + +=
y3 x1XkYk x0y0 x1y3 x2y2 x3y1+ + +=
x0y1 x1y0 x2y3 x3y2+ + + +
Convolution
![Page 23: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/23.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.23
y1x3
y1x3
25D
ORGAN / B.GOLD LECTURE 7
y0
y3
y2
x0
x1
x2
y0
y3
y2
x0
x1
x2
x0y2 x1y1 x2y0 x3y3+ + + +
x0y3 x1y2 x2y1 x3y0+ + + +
CircularConvolution
![Page 24: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/24.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.24
0
0
0 0
y0
25D
ORGAN / B.GOLD LECTURE 7
0 0 0 x0
x0y0
x1 x2 x3 0 0
y3 y2 y1 y0 0 0 0
0y3 y2 y1 y00 0 0
y3 y2 y1 y00 0 00
y3 y2 y1 y00 00
0y3 y2 y1 y0000
y3 y2 y1 y0000
y3 y2 y1
x0y1 x1y0+
x0y2 x1y1 x2y4+ +
x3y3
Linear Convolution
![Page 25: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/25.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.25
ces.rcles Carry the Same
25D
ORGAN / B.GOLD LECTURE 7
Figure 7.10 : Circular Convolution of Two 8 Point SequenOnly y(0), y(2), y(4) and y(7) are shown. All Outer CiSequence as the Upper Left Circle.
![Page 26: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/26.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.26
th Sequences by DFT.
25D
ORGAN / B.GOLD LECTURE 7
Figure 7.11 : Linear Convolution of Two Finite Leng
![Page 27: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/27.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.27
25D
ORGAN / B.GOLD LECTURE 7
M
L
![Page 28: EE225D Spring,1999 Digital Filters](https://reader031.vdocuments.site/reader031/viewer/2022012105/61dc12c6e4fd6f1f484da096/html5/thumbnails/28.jpg)
EE 2 LECTURE ON DIGITAL FILTERS
N.M 7.28
avings of ~ 3:1
12 L, 5=
1+ 18=
60
1
25D
ORGAN / B.GOLD LECTURE 7
DFT of each Row - LM2 Operations
Twiddle the Resulting Matrix - LM Operation
DFT of each Column - ML2 Operation
Total
DFT of
Complete Array
BUT e.g.
s
ML M L 1+ +( )
ML( )2 ML ML( )=
M 1000 L, 20= =
M L 1+ + 1021=
ML 20 000,=
M =
M L+
ML =
savings of ~ 20: