3.

DISCRETE-TIME RANDOM PROCESSES

Outline

Random variables

Random processes

Filtering random processes

Spectral factorization

Special types of random processes

• Autoregressive moving average processes

• Autoregressive processes

• Moving average processes

1 3 Random processes September 12, 2016

Random variables

Definitions

A random variable is a function that assigns a number to each outcome of a

random experiment.

Probability distribution function:

Pr

 


 






Probability density function:


 









Mean or expected value:
Z 
 

 





 






Variance:
Z


Var
 


 
 


 
















2 3 Random processes September 12, 2016

Random variables

Definitions

Joint probability distribution function:

Pr

 

 











Joint probability density function:

 
 


 


 








 


Correlation:











Covariance:

Cov









 


 















Correlation coefficient


























3 3 Random processes September 12, 2016

Random variables

and uncorrelated and strongly correlated




(small )



Linearly dependent

4 3 Random processes September 12, 2016

Random variables

Definitions

Two random variables and are independent if

 



 










Two random variables and are uncorrelated if

 or or






 




Two random variables and are orthogonal if






Orthogonal random variables are not necessarily uncorrelated

Zero-mean uncorrelated random variables are orthogonal

5 3 Random processes September 12, 2016

Random processes

Definitions

A random process is an indexed sequence of random variables (a “signal”)




Mean and variance:


 



 


 







Autocorrelation and autocovariance:








 









 







 

 




 
















Cross-correlation and cross-covariance
 






 












 




 

 




 

















Uncorrelated and orthogonal processes are defined as for variables but now





6 3 Random processes September 12, 2016
Random processes

Stationarity

First-order stationarity if . Implies

 


 



 


 












Second-order stationarity if .

 



 
 


 

























Implies


 
























Stationarity in the strict sense, if the process is stationary for all orders

Wide-sense stationarity, if i) ; ii) , and iii)


 


 



















Two processes and jointly wide-sense stationary if i) both and







are wide-sense stationary and ii)


 

























Properties of WSS processes:
symmetry:


 









mean-square value:

 










maximum value:










mean-square periodicity: periodic with period

 








 









7 3 Random processes September 12, 2016

Random processes

Autocorrelation and autocovariance matrices

We consider a WSS process and collect samples in a vector


















 
Autocorrelation matrix:

















 
 






 








 .. .. ..
 
 

 



 . . . 
 













Autocovariance matrix:




 
 

 







where

 













The autocorrelation matrix of a WSS process is Toeplitz, Hermitian, and



nonnegative definite; hence the eigenvalues of are nonnegative




8 3 Random processes September 12, 2016

Random processes

Sample mean:


X
Realization 1






Realization 2

Realization 3

Realization 4

Realization 5
Ensemble mean:





When is the sample mean equal to the ensemble mean (expectation)?

9 3 Random processes September 12, 2016

Random processes

Ergodicity

Sample mean:





X


 












A WSS process is ergodic in the mean if (mean-square convergence)

or




 





















(unbiased) and Var
 

 













Necessary and sufficient condition:




X


 












Sufficient condition: and





 








10 3 Random processes September 12, 2016

Random processes

White noise

White noise is a discrete-time random process with autocovariance:




 

 











i.e. for .
 






All variables are uncorrelated with variance (probability density not important)

 


The power spectrum of zero-mean white noise is constant:


X








 








 







11 3 Random processes September 12, 2016

Random processes
Power spectrum

The power spectrum of a WSS process is the DTFT of the autocorrelation:


X X
Also:

 





 


 











 


 

















Since the autocorrelation is conjugate symmetric, the power spectrum is real:








 

 






 












If the stochastic process is real, the power spectrum is even:











 


 


 









 












The power spectrum is nonnegative:








The total power is proportional to the area under the power spectrum:
Z


(use inverse DTFT, take )





 

 






















12 3 Random processes September 12, 2016

Random processes

Power spectrum

The eigenvalues of the autocorrelation matrix are upper and lower



bounded by the maximum and minimum value, respectively, of the power spec-
trum:











 













The power spectrum is related to the mean of as





 





  



 X   
 







 






 



  








If has a nonzero mean or a periodicity, the power spectrum contains im-




pulses

13 3 Random processes September 12, 2016

Filtering random processes

Suppose is a WSS process with mean and correlation that is filtered








by a stable LSI filter with unit sample response ; then the output is also





WSS with






 









 












 










where is the “(deterministic) autocorrelation” of :








 X




 


 














The power of is given by





X X





 
 

 

 




 


















where we assume zero outside and






















14 3 Random processes September 12, 2016
Filtering random processes

In terms of the power spectrum, this means that











 

 


 




















So assuming no pole/zero cancelations between and , if has a








pole (zero) at , then also has a pole (zero) at and another at










the conjugate reciprocal location





If is a narrow-band bandpass filter with center frequency , bandwidth








, and magnitude 1, then the output power is



Z















 




















 






so the power spectrum describes how the power is distributed over frequency


15 3 Random processes September 12, 2016
Spectral factorization

If the power spectrum of a WSS process is a continuous function of ,




 






then may be factored as






X

 



 

 
























Proof:

If is analytic in then we can write













X X
and
 




















 























so is the IDTFT of , and since is real,







 






















16 3 Random processes September 12, 2016
Spectral factorization

Proof (continued):

Now we can write

( ) ( )




X X

 

 
 














If we now define the second exponential as
( )


X

 
 














then we can express the third exponential as
( ) ( )





X X
 

 






























and so we obtain

with


 


 

 

 











17 3 Random processes September 12, 2016

Spectral factorization

The filter is causal, stable, and minimum phase; moreover it is monic:



 





 







A process that can be factorized as described earlier is a regular process

Properties of a regular process

• A regular process can be realized as the output of a filter driven by




white noise with variance
 


• If the process is filtered by the inverse filter , then the output is white









noise with variance (whitening)
 


• The process and the white noise contain the same information (compres-
sion)

18 3 Random processes September 12, 2016

Spectral factorization

Suppose the power spectrum is a rational function




 








then the spectral factorization tells us we can factor this as
  




 





 


 

 

 


















 



where
















 

 









whose roots are all inside the unit circle

Since is real, we have ; so the poles and zeros occur in




 











conjugate reciprocal pairs and we simply relate the zeros inside the unit circle to
the zeros of and the poles inside the unit circle to the zeros of







19 3 Random processes September 12, 2016
Special types of random processes

Autoregressive moving average processes

Suppose we filter white noise of variance with the filter

 




P

 








 





 
P





 
















The power spectrum of the output can then be written as












 









 


 





 

 

















 









Such a process is known as an autoregressive moving average process of order

, or ARMA





The power spectrum of an ARMA process has poles and zeros with






conjugate reciprocal symmetry

20 3 Random processes September 12, 2016

Special types of random processes

Autoregressive moving average processes

From the LCCDE between and :







X X


 

 

 

















we can multiply both sides with and take the expectation:










X X X





 


 




 





 

 



























The crosscorrelation between

and can further be expressed as






X






  

  

 

 




 












For , this leads to the Yule-Walker equations







 X

 


 

  




 
X
 












 




 
























21 3 Random processes September 12, 2016

Special types of random processes

Autoregressive moving average processes

The Yule-Walker equations can be stacked for :

 



   

 

 










..
   
   
.

 
    





   
.. .. .. ..


   

 . . . 




 . 

    




 

..
    












    








 . 













   







.. .. ..
  .. 



. . . .
   
   
   














Given the filter coefficients and , it gives a recursion for the autocor-












relation

Given the autocorrelation, we may compute the filter coefficients and













22 3 Random processes September 12, 2016
Special types of random processes

Autoregressive processes

An ARMA process is an autoregressive process, or AR :












 


 

 

 



 












 







The Yule-Walker equations are given by


X

 





 




 

















Stacking the Yule-Walker equations for :

 



    
 

 










..
    
.
    
 

    









 



 .. .. .. ..  .. 
    









 
 .
 . . 
 . 

.
 















Estimating from the Yule-Walker equations is easy (linear)







23 3 Random processes September 12, 2016

Special types of random processes

Moving average processes

An ARMA process is a moving average process, or MA :







 


 






 

 








 












The Yule-Walker equations are given by


X 


 


 




 


 


 





 



















The autocorrelation function is zero outside







Estimating from the Yule-Walker equations is not easy (nonlinear)







24 3 Random processes September 12, 2016