0% found this document useful (0 votes)
62 views

Discrete-Time Linear Shift Invariant System With WSS Random Inputs

1) Discrete-time linear shift-invariant (LTI) systems are characterized by their unit sample response h[n]. 2) The discrete-time Fourier transform (DTFT) of h[n] is the transfer function H(w) of the system. 3) If the input to a discrete-time LTI system is a wide-sense stationary (WSS) random process, then the output will also be a WSS random process. The output's mean and autocorrelation can be determined from the input's mean, autocorrelation, and the system's impulse response.

Uploaded by

Akhil Kathi
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
62 views

Discrete-Time Linear Shift Invariant System With WSS Random Inputs

1) Discrete-time linear shift-invariant (LTI) systems are characterized by their unit sample response h[n]. 2) The discrete-time Fourier transform (DTFT) of h[n] is the transfer function H(w) of the system. 3) If the input to a discrete-time LTI system is a wide-sense stationary (WSS) random process, then the output will also be a WSS random process. The output's mean and autocorrelation can be determined from the input's mean, autocorrelation, and the system's impulse response.

Uploaded by

Akhil Kathi
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
You are on page 1/ 11

Discrete-time Linear Shift Invariant System with WSS Random Inputs

Discrete-time Linear Shift Invariant System with Deterministic Inputs

We have seen that the Dirac delta function d (t ) plays a very important role in the analysis
of the response of the continuous-time LTI systems to deterministic and random inputs.
Similar role in the case of the discrete-time LTI system is played by the unit sample
sequence d [n] defined by
1 for n = 0

d [ n] = �
0 otherwise

Any discrete-time signal x[ n] can be expressed in terms of d [n] as follows:

x[n] = � x[k ]d [n - k ]
k =-�

A discrete-time linear shift-invariant system is characterized by the unit sample response


h[ n] which is the output of the system to the unit sample sequence d [n].

d [n] x h[n]
(n) LTI
system

The DTFT of the unit sample response is the transfer function of the system and given by

H (w ) = � h[n]e - jw n
n =-�

The transfer function in terms of z - transform is given by



H ( z ) = � h[ n] z - n
n =-�

where H ( z ) is a function of the complex variable z. It is defined on a region of

convergence (ROC) on the z - plane.

An analysis similar to that for the continuous-time LTI system can be applied to the
discrete-time LTI system. Such an analysis shows that the response y[n] of a the linear

time-invariant system with impulse response h[n] to a deterministic input x[ n] is


y[n] = � x[k ]h[n - k ] = x[n]* h[ n]
k =-�
By taking the DTFT of both sides, we get
Y (w ) = X (w ) H (w )
More generally, we can take the z-transform of the input and the response and show that
Y ( z) = X ( z) H ( z)
Remark
 If the LTI system is causal, then
h[ n] = 0 for n < 0

In this case, the ROC of H ( z ) is a region in the z - plane given by r < z < �. For
example, suppose
n
�1 �
h[ n] = � �u[ n]
�2 �
1 1
H ( z) = < z <�
Then, 1 2
1 - z -1
2
 Similarly, if the LTI system is anti-causal, then
h[ n] = 0 for n > 0

In this case, the ROC of H ( z ) is a region in the z - plane given by z �r .

 The contour z = 1 is called the unit circle. Thus H (w ) represents H ( z )

evaluated on the unit circle.


 H ( z ) can be expressed as the ratio of two polynomials in z
N ( z)
H ( z) =
D( z )
The polynomials N ( z ) and D( z ) helps us in analyzing the properties of a linear

system in terms of the zeros and poles of H ( z ) defined by


Zero- the point in the z - plane where N ( z ) = 0. H ( z ) = 0 at such a point.
Pole- the point in the z - plane where D( z ) = 0. H ( z ) � � at such a point. The

ROC of H ( z ) does not contain any pole.


Zero
Im( z)
z =1 Pole

0 �

0 � Re( z )

 For the stability of the LTI system, the unit-sample response h[ n] should
decay to zero as n � �. A necessary and sufficient condition for the stability
of a discrete-time LTI system is that all its poles lie strictly inside the unit
circle.
 A discrete-time LTI system is called a minimum- phase system if all its poles
and zeros lie inside the unit circle. A minimum-phase system is always stable
as its poles lie inside the unit circle. Because the zeros of the system lie

1
inside the unit circle, the inverse system with a transfer function will
H ( z)
have all its poles inside the unit circle and be stable.
 A discrete-time LTI system is called a maximum- phase system if all its poles
and zeros lie outside the unit circle.

Response of a discrete-time LTI system to WSS input


Consider a discrete-time linear time-invariant system with impulse response h[n] and

input { X [ n]} as shown in Fig. Assume { X [n]} to be a WSS process with mean m X and

autocorrelation function RX [m].

h[n]
X [n] Y [n]
x(n) y(n)
The output random process {Y [n]} is given by

Y [n] = X [ n] * h[ n] = � h[ k ] X [ n - k ]
k =-�

Given the WSS input { X [n]}, the output process {Y [n]} is also WSS. We establish this result
in the following section.
Mean and Autocorrelation of the output
The mean of the output is given by

E Y [n] = E � h[k ]X [n - k ]
k =-�

= � h[k ]EX [ n - k ]
k =-�

=m X � h[ k ]
k =-�

= m X H (0)

where H (0) is the dc gain of the system given by


H (0) = H (w ) w =0

= � h[ n]e - jwn
n =-� w =0

= � h[ n]
n =-�

Thus the mean of the output process {Y [ n]} is constant. We write


mY = EY [n] = m X H (0)

The cross-correlation between the output and the input random processes is given by
RYX [n, n - m] = E Y [n] X [n - m]
= E ( X [n]* h[n]) X [ n - m]

= E � h[k ] X [n - k ]X [n - m]
k =-�
� � .
= � h[k ] � EX [n - k ] X [n - m]
k =-� l =-�

= � h[k ]RX [m - k ]
k =-�

= h[m]* RX [m]

Thus RYX [n, n - m] does not depend on n , but on lag m and we can write

RYX [m] = h[ m]* RX [ m]

The autocorrelation function of the out put is


RY [n, n - m] = E Y [n]Y [ n - m]

= E (Y [n] �h[k ] X [n - m - k ]
k -�

= E � h[k ]RYX [m + k ] .
k =-�

= h[-m]* RYX [m]


= h[-m]* h[m]* RYX [m]

RY [n, n - m] is a function of lag m only and we write


RY [m]=h[- m]* RYX [m]
=h[-m]* h[m]* RX [m]
� �
= � � h[ k ]h[l ]RX [ m + k - l ]
k=-�l =-�

The mean-square value of the output process is


EY 2 [ n] = RY [0]
� �
= � � h[ k ]h[l ]RX [k - l ]
k=-�l =-�

Thus if { X [ n]} is WSS then {Y [n]} is also WSS.


Taking the DTFT of RY [ m], we get

w ) = | S(w ) |2 ) X (w
SY (Η

S Y (w )
S X (w )
2
Hw 

In terms of z - transform, we get


SY ( z ) = S X ( z ) H ( z ) H ( z -1 )

Notice that if H ( z ) is causal, then H ( z -1 ) is anti-causal and vice versa.


Similarly if H ( z ) is minimum-phase then H ( z -1 ) is maximum-phase.

H (z ) H ( z -1 )
S X ( z)
Remark
Finding the probability density function of the output process {Y [n]} is a difficult task.

However, if { X [n]} is a WSS Gaussian random process, then the output process {Y [n]} is
also Gaussian with the probability density function determined by its mean and the
autocorrelation function.

Example
1
Suppose H ( z) = and {x[n]} is a zero-mean unity-variance white-noise
1 - a z -1
sequence. Then

EX 2 [n] = s X2
s X2
\ S X (w ) =
2p
\ SY (w ) = H (w ) H ( -w ) S X (w )
� 1 � 1
� �s X2
=� �
� �
1 - a e- jw �
� �1 - a e jw �2p
1 s X2
=
1 - 2a cos w + a 2 2p
Similarly
SY ( z ) = H ( z ) H ( z -1 ) S X ( z )
� 1 � 1 �
� s X2
=� �

1 - a z -1 �
� �1-a z �
�2p
By partial fraction expansion and inverse z - transform, we get
1
RY [m] = a |m|
1-a 2

Remark
 Though the input is an uncorrelated process, the output is a correlated process.
 For the same white noise input, we can generate random processes with
different autocorrelation functions or power spectral densities.

Spectral factorization theorem


Consider a discrete-time LTI system with the transfer function H ( z ) and the white noise

sequence {W [ n]} as the input random process as shown in the Fig. below.

H ( z)
{W [n]} x( { X [n]} y(n
n) )
Then
s W2
S X ( z ) = SW ( z ) H ( z ) H ( z ) =
-1
H ( z ) H ( z -1 )
2p
We have seen that S X ( z ) is the product of a constant and two transfer functions

H ( z ) and H ( z -1 ). This result is of fundamental importance in modeling a WSS process


because of the spectral factorization theorem stated below:

p
If S X (w ) is an analytic function of w , and �| ln S X (w ) | dw < �, then
-p

s W2
S X ( z ) = s W2 H c ( z ) H a ( z ) = H C ( z ) H C ( z -1 )
2p
where
H c ( z ) is the causal minimum phase transfer function

H a ( z ) = H c ( z -1 ) is the anti-causal maximum phase transfer function

and s W2 a constant and interpreted as the variance of a white-noise sequence.

Thus a WSS random signal { X [ n]} with continuous spectrum S X (w ) that satisfies the

p
Paley Wiener condition �| ln S X (w ) | dw < � can be considered as an output of a linear
-p

filter fed by a white noise sequence.

Innovation sequence
W [n] X [n ]
H c (z )
Figure Innovation Filter
Minimum phase filter => the corresponding inverse filter exists.
1
X [n ] H c ( z)
V [n]
Figure whitening filter
Proof of the spectral factorization theorem
1
Since ln S X ( z ) is analytic in an annular region r < z < that includes the unit circle
r

z = 1, ,

ln S X ( z ) = � c[ k ] z - k
k =-�

1 p iw n
where c[k ] = -p ln S X (w )e
� dw is the kth order cepstral coefficient.
2p
For a real signal c[ k ] = c[- k ]
1 p
and c[0] = � ln S XX ( w)dw
2p -p

� c[ k ] z - k
S X ( z) = ek =-�
� -1
c[ k ] z - k � c[ k ] z - k
c[0] k�
=e e
=1 e k =-�


�c[ k ] z - k
Let H C ( z) = ek =1 z >r
= 1 + hc (1)z -1 + hc (2) z -2 + ......
( Q hc [0] = Limz��H C ( z ) = 1
H C ( z ) and ln H C ( z ) are both analytic

Therefore, H C ( z ) is a minimum phase filter.


Similarly let
-1
� c(k ) z- k
H a ( z) = e k =-�


�c ( k ) z k 1
= ek =1 = H C ( z -1 ) z<
r
Therefore,
s v2
S XX ( z ) = H C ( z ) H C ( z -1 )
2p

s v2
where = ec (0)
2p
Pole of the
Zero of the maximum-phase

Im( z)
minimum-phase filter
filter

0 z =1 �
Pole of the

� minimum-phase
filter
0
0 � Re( z )

0

Remarks

e jw + e - jw
 Note that S X (w ) of a real process is a function of cos w = .
2

1
Therefore, S X ( z) is a function of z + . Consider rational spectrum S X ( z )
z
N ( z)
so that S X ( z) = , where N ( z) and D( z ) are polynomials in z. If
D( z )

1
zi is a root of S X ( z ), so is . Thus the roots of S X ( z) are symmetrical
zi

about the unit circle z = 1. H c ( z ) groups the poles and zeros inside the unit

circle and H a ( z ) groups the poles and zeros outside the unit circle.

 S X ( z ) can be factorized into a minimum-phase and a maximum-phase factors

i.e. H C ( z ) and H C ( z -1 ).
 In general spectral factorization is difficult, however for a signal with rational
power spectrum, spectral factorization can be easily done.
1
 Since is a minimum phase filter, the inverse filter HC ( z) exists and stable.

1
Therefore we can have a filter to filter the given signal to get the
HC ( z)

innovation sequence.
 X [n ] and v[ n] are related through an invertible transform; so they contain the

same information.
Example
Suppose the power spectral density S XX (w) of a discrete random sequence { X [ n ]} is given by

2.18 - 1.12 cos w


S X (w ) = - p �w �p
1.25 - cos w
Then
2.18 - 0.6( z + z -1 )
S X ( z) =
1.25 - 0.5( z + z -1 )
2�1.09 - 0.3( z + z -1 ) �
= � �
1.25 - 0.5( z + z -1 )
2(1 - 0.3 z )(1 - 0.3 z -1 )
=
(1 - 0.5 z )(1 - 0.5 z -1 )
(1 - 0.3 z -1 )
\ H c ( z) = ,
(1 - 0.5 z -1 )
(1 - 0.3 z )
H a ( z) = and
(1 - 0.5 z )
s v2
=2
2p
Wold’s Decomposition
Any WSS signal X [n ] can be decomposed as a sum of two mutually orthogonal
processes
 a regular process X r [n] and a predictable process X p [n] , X [n ] = X r [n ] + X p [n ]

 X r [n] can be expressed as the output of linear filter using a white noise

sequence as input.
 X p [n] is a predictable process, that is, the process can be predicted from its own

past with zero prediction error.

You might also like