digital signal processing lecture 10
TRANSCRIPT
Remote Sensing LaboratoryDept. of Information Engineering and Computer Science
University of TrentoVia Sommarive, 14, I-38123 Povo, Trento, Italy
Remote Sensing LaboratoryDept. of Information Engineering and Computer Science
University of TrentoVia Sommarive, 14, I-38123 Povo, Trento, Italy
Digital Signal ProcessingLecture 10
Digital Signal ProcessingLecture 10
E-mail: [email protected] page: http://rslab.disi.unitn.it
Quote of the Day
When you look at yourself from a universal standpoint, something inside always reminds or informs you that there are
bigger and better things to worry about.
Albert Einstein
University of Trento, Italy
University of Trento, Italy B. Demir
Reminder
Pole-zero plot and region of convergence of a causal exponential sequence x[n]=anu[n] with a) decaying exponential, b) unit step sequence and c) growing sequence
University of Trento, Italy B. Demir
Reminder
Pole-zero plot and region of convergence for the a) causal, b) anticausal and c) two sidedexponential sequences.
University of Trento, Italy B. Demir
Example
Pole-zero plot and region of possible convergence for the z transform in
1
1 1
11 (1 0.5 )
zX zz z
University of Trento, Italy B. Demir
System function of LTI systems
Procedure for the analytical computation of the output of an LTI system using the convolution theorem of the z-transform:
University of Trento, Italy B. Demir
System function of LTI systems
Equivalent system function of linear time-invariant systems combined in a) parallel connection and b) cascade connection
University of Trento, Italy B. Demir
Stability-Causality
University of Trento, Italy B. Demir
Stability-Causality
University of Trento, Italy B. Demir
Linear constant-coefficient difference equations
0 0[ ] [ ]
N M
k kk k
a y n k b x n k
0
0
Mk
kkN
kk
k
b zH z
a z
A class of LTI systems whose input and output sequences satisfy a linear constant-coefficient difference equation is written as:
Using the properties of linearity and shifting we obtain:
Then the system function (transfer function) is obtained as:
0 0( ) ( )
N Mk k
k kk k
a z Y z b z X z
University of Trento, Italy
Example
University of Trento, Italy
Example
B. Demir
University of Trento, Italy B. Demir
Example-Cont
University of Trento, Italy B. Demir
Linear constant-coefficient difference equations
1
0 0 1 0 1
10 0
0 1 1
1
1
M MMk
k kkN Mk k k
N N Nk
k k kk k k
z z z zb zb bH z za aa z p z z p
H(z) is a rational function, that is, the ratio of two polynomials in z-1.
We recall that any rational function of z-1 can be expressed in the form of:
Where the first summation is included only if M≥N.
10 1 1
M N Nk k
kk k k
AH z C zp z
University of Trento, Italy B. Demir
Linear constant-coefficient difference equations
0 1 0[ ]
M N Nn
k k kn k k n
h n C A p
If we assume that the system is causal, then the ROC is the exterior of a circle starting at the outermost pole, and the impulse response is:
For a causal system to be stable, the impulse response must be absolutely summable.
This is possible if and only if |pk|<1. Hence a causal LTI system with a rational system function is stable if and only if all poles of H(z) are inside the unit circle in the z plane. The zeros can be anywhere.
0 1[ ] [ ] ( ) [ ]
M N Nn
k k kk k
h n C n k A p u n
University of Trento, Italy B. Demir
Example
Pole-zero plot for the system function given by:
This causal system is stable because its poles are inside the unit circle.
2
1 2 3
11 0.9 0.6 0.05
zH zz z z
University of Trento, Italy
Example
B. Demir
University of Trento, Italy B. Demir
One-sided z-Transform
University of Trento, Italy B. Demir
Linear constant-coefficient difference equations
University of Trento, Italy B. Demir
Example
University of Trento, Italy B. Demir
Example-Cont
University of Trento, Italy
Systems that are designed to pass some frequency components withoutsignificant distortion while severely or completely eliminating others areknown as frequency selective filters.
By definition, an ideal frequency selective filter satisfies the requirementsfor distortionless response over one or more frequency bands.
Ideal filters are used in the early stages of a design process to specifymodules in a signal processing systems. However they are not realizablein practice, they must be approximated by practical and non-ideal filters.
This is usually done by minimizing some approximation error between thenon-ideal filter and a prototype ideal filter.
Ideal and Practical filters
B. Demir
University of Trento, Italy
Filters-Reminder
B. Demir
Ideal frequency selective filters either pass or eliminate a region of inputspectrum perfectly and they have abrupt transitions between passbands andstopbands.
However ideal filters can not be implemented in practice; therefore they have tobe approximated by practically realizable filters. Practical filters differ from idealfilters in several respect:
The bandpass responses are not perfectly flat; The stopband responses can not completely eliminate bands of
frequencies; The transition between passband and stopband regions takes place over
a finite transition band.
University of Trento, Italy
Filters-Reminder
University of Trento, Italy B. Demir
Frequency responses of four types of ideal filter
a) Lowpass filter; b) bandpass filter; c) highpass filter; bandstop filter.
Ideal and Practical filters
University of Trento, Italy
As the impulse response corresponding to each of these ideal filters isnoncausal and of infinite length, these filters are not realizable.
In practice, the magnitude response specifications of a digital filter in thepassband and in the stopband are given with some acceptabletolerances.
In addition, a transition band is specified between the passband andstopband.
Ideal and Practical filters
B. Demir
University of Trento, Italy
Filter design
B. Demir
Ideal frequency selective filters either pass or eliminate a region of inputspectrum perfectly and they have abrupt transitions between passbands andstopbands.
However ideal filters can not be implemented in practice; therefore they have tobe approximated by practically realizable filters. Practical filters differ from idealfilters in several respect:
The bandpass responses are not perfectly flat; The stopband responses can not completely eliminate bands of
frequencies; The transition between passband and stopband regions takes place over
a finite transition band.
University of Trento, Italy
Filters-Reminder
University of Trento, Italy
Filters-Reminder
University of Trento, Italy
1 , 0 ,
kj
k
H e
?][ nh
Ex Example from DTFT:
B. Demir
Ideal Low Pass Filter:
University of Trento, Italy
1 1[ ]2 2
kk k
k
j n j nj ne e eh njn jn
1 1 2 sin sin[ ] ,2 2
k kj n j nk ke e j n nh n n
jn jn n
sin
j j nk
n
nH e en
Example-Cont:
B. Demir
1 1 1[ ]2 2 2
kk
k k
j nj j n j n eh n H e e d e d
jn
University of Trento, Italy
sin
Mj j nk
Mn M
nH e en
Example-Cont:
B. Demir
University of Trento, Italy
Filters-Reminder
University of Trento, Italy B. Demir
Ideal and Practical filters
Typical characteristics of a practical bandpass filter
A good filter should have only a small ripple in the passband, highattenuation in the stopband, and very narrow transition bands.
University of Trento, Italy
Frequency Response of Rational System Functions
• All LTI systems of practical interest are described by a difference equation of the form:
and have a rational system function:
• For a stable system the system function convergences on the unit circle. Therefore, we obtain:
which expresses H(ejω) as a ratio of two polynomials in the variable ejω
0 0[ ] [ ]
N M
k kk k
a y n k b x n k
0
0
Mk
kkN
kk
k
b zH z
a z
0
0
Mj k
kj k
Nj k
kk
b eH e
a e
B. Demir
University of Trento, Italy
Frequency Response of Rational System Functions
• DTFT of a stable and LTI rational system function:
• Magnitude Response:
• Phase response:
0 1
0
1
1
1
Mj
kj k
Nj
kk
z ebH ea p e
0 0 1
0
0 1
1
1
MMjj k
kkj k k
N Nj k j
k kk k
z eb ebH eaa e p e
0
1 10
( ) (1 ) (1 )M N
j j jk k
k k
bH e z e p ea
B. Demir
University of Trento, Italy
Dependence of frequency response on poles and zeros
• This equation consists of factors of the form and
• The factor is a complex number represented by a vector drawn from the point zk (zero) to the point z=ejω in the complex plane and can be written as:
• is the distance from the zk to the point z=ejω , and is the angle of the vector with the positive real axis.
( )0 1 0 1
0 0
1 1
1
1
M Mj j
k kj j N Mk k
N Nj j
k kk k
z e e zb bH e ea ap e e p
jke z j
ke p
jke z
kjjk ke z Q e
kQ k
University of Trento, Italy B. Demir
Dependence of frequency response on poles and zeros
University of Trento, Italy
Dependence of frequency response on poles and zeros
The factor is a complex number represented by a vector drawn from the point pk (pole) to the point z=ejω in the complex plane.
is the distance from the pole to the point, and is the angle of the vector with the positive real axis.
Then, we can write:
jke p
kjjk ke p R e
kRk
0 1 0
1 10 0
1
exp ( ) ( ) ( )
M
k M Mj k
i kNk k
kk
Qb bH e N Ma aR
B. Demir
University of Trento, Italy
Dependence of frequency response on poles and zeros
• We note that if because , and
if because
0 1 0
1 10 0
1
exp ( ) ( ) ( )
M
k M Mj k
i kNk k
kk
Qb b
H e N Ma aR
0
0
ba
0
0
0ba
0
0
0ba
0
0
0ba
1je
0 1je
0 1
0
1
M
kj k
N
kk
Qb
H ea R
0
1 10
( ) ( ) ( )M M
ji k
k k
bH e N M
a
B. Demir
University of Trento, Italy
Dependence of frequency response on poles and zeros
0 1
0
1
M
kj k
N
kk
Qb
H ea R
0
1 10
( ) ( ) ( )M M
ji k
k k
bH e N M
a
distance of the - zero from
distance of the - pole from
angle of the - zero with the real axis
angle of the - pole with the real axis
jk
jk
k
k
Q k th e
R k th e
k th
k th
where ω is the angle of the point z=ejω with the positive real axis and
Therefore the magnitude response at a certain frequency ω is given by theproduct of the lengths of the vectors from the zeros to z=ejω divided by theproduct of the lengths of vectors drawn from the poles to z=ejω.
Similarly the phase response is obtained by subtracting from the sum ofangles of zeros the sum of angles of poles. All angles are determined withrespect to the positive real axis. We can determine H(ejω ) for each value ofω or equivalently any location of the point ejω on the unit circle.
B. Demir
University of Trento, Italy
Dependence of frequency response on poles and zeros
To suppress a frequency component at ω=ω0, we should place a zero atangle ω0 on the unit circle.
To enhance or amplify a frequency component at ω=ω0 we should place apole at angle ω0 close but inside the unit circle.
Complex poles or zeros should appear in complex conjugate pairs to assurethat the system has real coefficients. This stems from the fact that thefrequency ω is defined as the angle with respect to positive real axis.Therefore all symmetries should be defined with respect to the real axis.
Poles and zeros at the origin do not influence the magnitude responsebecause their distance from any point on the unit circle is unity. However apole at the origin adds (or subtracts) a linear phase of ω rads to the phaseresponse. We often introduce poles and zeros at z=0 to assure that N=M
B. Demir
University of Trento, Italy B. Demir
Significance of poles and zeros
University of Trento, Italy B. Demir
Significance of poles and zeros
University of Trento, Italy B. Demir
Significance of poles and zeros
University of Trento, Italy
Significance of poles and zeros
B. Demir
University of Trento, Italy
Allpass systems
The frequency response of an allpass system has constant magnitude at all frequencies.
If we assume that G=1, allpass systems preserve that the power andenergy of their input signals.
In this sense allpass systems are said to be lossless systems. Thesimplest allpass system simply scale and delay the input signal:
A more interesting all pass system is:
jH e G
kapH z Gz
1 *
11k
apk
z pH zp z
B. Demir
University of Trento, Italy
Allpass systems
We note that each pole pk of an allpass system should be accompanied by a complex reciprocal zero 1/pk
*.
Since causal and stable allpass systems must have all poles inside the unit circle, all zeros are outside the unit circle.
B. Demir
University of Trento, Italy
Invertibility and minumum-phase systems
The zeros of H(z) become the poles of its inverse system and vice versa.A causal and stable H(z) should have its poles inside the unit circle; itszeros can be anywhere.
In inverse filtering applications the system Hinv(z) should be causal andstable as well.
A causal and stable LTI system with a causal and stable inverse isknown as a minumum-phase system.
A LTI system is stable and causal and also has a stable and causalinverse if and only if both the poles and the zeros of H(z) are inside theunit circle.
invh n h n n
1invH z H z
B. Demir
1
invH zH z
University of Trento, Italy
Minimum phase and allpass decomposition
Any system with a rational system function can be decomposed into a minimum phase system and an allpass system.
Suppose that H(z) has one zero z=1/a* where |a|<1 outside the unit circle, and all the other poles and zeros are inside the unit circle. Then we can factor H(z) as:
If we repeat this process for every zero outside the unit circle, we obtain:
B. Demir
1 *1( ) ( )( )H z H z z a
1 *1
1 1
( )( ) ( )(1 )(1 )z aH z H z az
az
min( ) ( ) ( )apH z H z H z
minimum phase system allpass system
University of Trento, Italy
Relationship between time, frequency and z-transform domains
B. Demir