Adama Science and Technology

University

Probability and Random Processes(ECE 3103 )

Chapter 5
Random Processes
 Contents

1. Random Process and Ensembles

2. Autocorrelation Function and Its Properties
3. Cross-correlation Function and Its Properties
4. Stationary Random Processes
5. Ergodic Random Processes
6. Power Spectral Density of Random Processes
 Determinism Vs Randomness
There are two main kinds of processes in Nature, distinguished by their
time evolution.
 In a deterministic process, the future state of the system is
completely determined by the present state.
 Physical systems whose time evolution is described by differential
equations are deterministic; e.g:
• classical mechanics (Newton’s equation)
• quantum mechanics (Schrödinger’s equation)
• the weather (chaotic but deterministic!)
 It is a process where given the starting point, we can know or
predict the behavior of the process. Consider the following
𝑥(𝑡) = [𝑥(𝑡 − 1)]2 𝑎𝑛𝑑 𝑥(0) = 𝑎 (𝑠𝑜𝑚𝑒 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡) … . 𝑖𝑛𝑖𝑡𝑖𝑎𝑙 𝑜𝑟
𝑠𝑡𝑎𝑟𝑡𝑖𝑛𝑔 𝑝𝑜𝑖𝑛𝑡 (0) = 𝑎 (𝑔𝑖𝑣𝑒𝑛): 𝑥(1) = [𝑥(1 − 1)]2= 𝑥(0)2 = 𝑎2
𝑥(2) = [𝑥(2 − 1)]2 = 𝑥(1)2 = (𝑎2)2 = 𝑎4
𝑥(3) = [𝑥(3 − 1)]2 = 𝑥(2)2 = (𝑎4)2 = 𝑎8
So, the process goes as follows: 𝑎, 𝑎2, 𝑎4, 𝑎8, … … … ….
If we know starting point, we will know the rest of the sequence
without any ambiguity (doubt). This is the deterministic process. In a
deterministic process, each subsequent step is said to be known with
probability 1(complete certainty). Another example of deterministic
process is a sinusoidal process. No uncertainty about its time-
dependent behavior at any instant of time. But many real-world
4
problems are random in nature & hence random processes.
 Random Process = Stochastic Process
Random process is a family of random variables or random signals. It
is a collection (or ensemble) of random functions representing the
evolution of some system of random values over time. Its another
name: Stochastic is a Greek term which means random or guessing.
It is a probabilistic counterpart of deterministic process (means
Stochastic processes are non-deterministic) and the time evolution of
it is subject to a probability distribution.
 Examples of stochastic processes are
• Random walks
• Markov chains
• Birth-death processes
• Queues 5
 Random process cont.…
The value of a random signal can’t be predicted ahead of time. There
are many examples of random signals.
 speech & music signals, output voltage of a SOLAR DETECTOR
which depends on weather conditions, the marks you score in
university examination etc.
A stochastic process is simply a collection of random variables i.e.,
random signals indexed by time (t). if the index (t) is continuous, it is
regarded as continuous-time process. If ‘t’ is discrete, we call
discrete-time process.
Models are mathematical representation of real systems &
fundamentals tools in science and Engineering.
Random processes have 2 fundamental properties:
 They are functions of time
 They are random i.e., before conducting the experiment, it is not
possible to define the waveforms that will be observed in the future.
6
 In general, the function x(t) is a member of an ensemble (family,
set, collection) of functions.
 If one scans all possible outcomes of the underlying random
experiment, we shall get an ensemble of signals.
 Random Process cont.…
• A general Random or Stochastic Process can be described as:
Collection of time functions (signals) corresponding to various
outcomes of random experiments.
• Collection of random variables observed at different times.
• A random process is a process (i.e., variation in time or one
dimensional space) whose behavior is not completely predictable
and can be characterized by statistical laws.
 Daily stream flow
 Hourly rainfall of storm events
 Stock index are examples of random processes 16
 Mean and Variance of Random Process
Stochastic process X ={X(t), t є T}. Let X(t) denote state of X at time t; X(t) is the
random variable obtained by observing the stochastic process at time t. The mean
of a random process, mX(t), is thus a function of time specified by:

m X (t )  E[ X (t )]  E[ X t ]   xt f ( xt , t )dxt


=}

Where f(xt , t) is the probability density function for X(t):

P{X(t) = Xi} is the probability distribution of X(t).

• For the case where the mean of X(t) does not depend on t, we have
mX (t ) E[ X (t )] mX (a constant).
• The variance of a random process, also a function of time, is defined by

 X2 (t )  E [ X (t )  m X (t )]2   E[ X t2 ]  [m X (t )]2 17
 Average of random process

• Random process X  {X (t), t  0}  {X (A, t), t

 0}
is a function with two variable , the time t and
the random event A.

• Ensemble average
The Ensemble average of X (t) : E{X
The Ensemble average of the autocorrelation
(t)} :
RX (t, t   )  E{X (t)  X (t
  )}
Time average
Operator : T
1
Tlim
 T 
{}dt 2T
The time average of X (t)
:  X (t)  lim 1 T
T  T
x(t)dt
2T
where x(t) is any specific sample function.
The time average of the autocorrelation :
T
 X (t) X (t   )  lim1  x(t)x(t 
T  T
 )dt
2T
 Example:
Consider process: X (t)  a cos(t 
a, are constant. RV )
is uniform in the interval (-π, π).
(1)Find the Ensemble average of X (t) and its autocorrelation function.
(2)Find the time average of X (t) and its autocorrelation function.
 Observe that 𝑅𝑋𝑋 (𝜏 ) is
Solution: (1) Ensemble average X(t) independent of variable t (time).
So, for the given problem
autocorrelation will not change
after 1 hour or 1 day or 1 year.

i. Mean 𝑿 ̅ = 𝟎
Now conclude that

ii. Autocorrelation is not

dependent on time. Hence the
given Random Process is WSS
(Wide Sense Stationary)
E{X (t)} 
a2
0
RX (t, t   )  E{X (t) X (t   )}

Solution :
(2) The time average of X(t)

T
 X (t)  lim 1 
T 
x(t)dt T

1
T 2T
Tlim a cos(t 
T 

 )dt 2T
1 a
 lim  [sin(T   )  sin(T  
T  2T
)] | sin(T   )  sin(T   ) |
(2
2) sinA−sinB=2cos()sin()

0 sin(ωT+φ)−sin(−ωT+φ)=2cos(φ)sin(ωT)
∣2cos(φ)sin(ωT)∣≤2
• Random process {X (t), t  T } is Ergodic if it satisfies following
condition.
E{X (t)}   X (t) 
RX (t, t   )  E{X (t) X (t   )}
= X
Example: (t)X (t   ) 
Consider process X (t)  a cos(t,  ) a, are constant. RV  is uniform in
the interval (-π, π)
X (t) is Ergodic ? X (t)  a cos(t  )  For ergodic process: Ensemble
E{X (t)}  0   X (t)  averages = time averages
RX (t, t   )  E{X (t) X (t 
 )}
 cos( )
=

 X (t) X (t   ) 
 Definition 2: we define the variance function of the Stochastic
process X as :
Dx(t) = D{X(t)} = E{|X(t) - mx(t)|2} = E{|X(t)|2} - |mx(t)|2
 It is a measure of the “randomness” of the Stochastic process X at
time t (designated by X(t)).
 The smaller is Dx(t), the more the samples of X(t) concatenate.
 It is equal to the difference between the mean-square value and the
square of the mean.
Correlation: is a measure of similarity between two signals (two
random process).
➢ If two random signals come from same random process, auto-
correlation function is defined.
➢ If two random signals come from different random processes, it
will be called the cross-correlation function.
Autocorrelation
Correlation of a function with itself is known as autocorrelation. Autocorrelation is
used to compare a signal with a time-delayed 𝝉 version of itself. If the signal is
periodic, then the signal will be perfectly correlated with a version of itself (if the
time delay is an integer number of periods).
Autocorrelation function can be used to detect non-randomness in data. It describes
the general dependency of x(t) with its value at a short time latter 𝑖. 𝑒. , 𝑥( 𝑡 + 𝝉),
where 𝝉 = 𝑡𝑖𝑚𝑒 𝑑𝑒𝑙𝑎𝑦.

Definition 3: Define the autocorrelation function of the Stochastic process X(t) as:

Rx(t1, t2) = E{(X(t1)∙ X(t2)}, where X(t1) and X(t2) are random variables
obtained by observing X(t) at times t1 and t2 respectively.

 It is a function of two random variables t1and t2; 24

 It is the measure of the degree to which two times samples of the same
Stochastic process are related.
 The Autocorrelation function provides a measure of similarity between to
observations of the random process x(t) at different points in time t1 and t2.

R(t1 , t2 )  E X (t1 ) X (t2 )  R(t2 , t1 )

Properties of ACF (Auto Correlation Function)
ACF and their properties play a useful role in the analysis of systems with random
inputs.
1. ACF of a random process X(t) is defined as 𝑅𝑋𝑋 (𝜏 ) = 𝐸[𝑋(𝑡)𝑋(𝑡 + 𝜏 )]

2. ACF is an even function of 𝝉 i. e. , 𝑅𝑋𝑋()𝑅= 𝑅 𝑋𝑋 (−𝜏) i.e, mirror image.

25
3. The mean-square value of the random process can always be obtained by setting
𝝉=𝟎
𝑖. 𝑒. , 𝑅𝑋𝑋 (0) = 𝐸[𝑋2]
4. The largest value of auto-correlation function always occurs at 𝝉 = 𝟎
𝑖. 𝑒. , 𝑅𝑋𝑋 (0) ≥ 𝑅𝑋𝑋 (𝜏)
NOTE: Auto-correlation is a characteristic of data in which the correlation between
the values of the same signal or waveform is measured.

Note that correlation between 2 random variables X and Y is the expected value of
their product i.e., E[XY] = correlation. Correlation depends on how rapidly the
value of the random variable can change with respect to time (t).

26
This is auto-covariance function of the random process {X(t)}
Property 4
NOTE: Two random processes X(t) and Y(t) are said to be uncorrelated if their cross-
correlation is equal to the product of their mean functions i.e., E[XY] = E[X]. E[Y].
Two random processes X(t) and Y(t) are said to be orthogonal if their cross-
correlation is zero.
cosαsinβ=​[sin(α+β)−sin(α−β)]
Stationary Random Process

• Strict sense stationary process

(SSS)
• Wide sense stationary process
(WSS)
STATIONARY PROCESS
Random processes whose statistics do not change with respect to time are called
stationary. Statistics are mean, variance etc. In other words, process in which
probabilistic rules do not change with time.
A stochastic process is called stationary if
Its mean is constant over time
Its variance is constant over time
Its covariance between two time periods depends only on the distance or lag
between two time periods but not the actual value of the time at which
covariance is computed
Its autocorrelation structure does not change over time.
Note that if the statistical properties like mean, variance etc., depends on time
then the random process is said to be non-stationary.
47
Ergodic process
A stationary process is said to be ergodic its ensemble averages & time averages
are equal.
Ensemble = collection of similar functions
Note that random process is a collection of waveforms (functions). The entire
collection of these functions is called ensemble. Any particular member of
ensemble is called sample function.
Some stationary random processes possess the property that almost every
member of the ensemble exhibits the same statistical behavior as the whole
ensemble. Thus, it is possible to determine this statistical behavior by examining
only one typical sample function (waveform). Such processes are said to be
ergodic.
Ensemble averages: example:
1. Tossing 500 coins at a time (multiple coins) 48
Time averages: example:
1. Tossing single coin 500 times
2. Measuring voltage across 1 resistor at different times
For ergodic process: Ensemble averages = time averages
Ensemble averages: properties of the process are obtained over a collection
(ensemble) of waveforms.
Time averages: properties of the process are obtained by statistics of singe
waveform at different times.
Generally, the statistics (mean, variance, covariance etc) are estimated from time
averages.
Note that an ergodic process is always stationary, but reverse may not be true.

WSS (Wide Sense Stationarity)

Let X(t) represent a Random Process. The conditions for a WS process are:
(𝑖). 𝐸[𝑋(𝑡)] = = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 49
(𝑖𝑖). 𝑅𝑋𝑋(𝜏) = 𝐸[𝑋(𝑡)𝑋(𝑡 + 𝜏)] 𝑖𝑠 𝑖𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑜𝑓 𝑡𝑖𝑚𝑒 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡
Jointly WSS processes:
Let X(t) and Y(t) are two random processes (RPs),The conditions for a process to

(𝑖). 𝐸[𝑋(𝑡)] = = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡

be a WSS are:

(𝑖𝑖). 𝐸[𝑌(𝑡)] = = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡

(𝑖𝑖𝑖). 𝑅𝑋𝑌(𝜏) = 𝐸[𝑋(𝑡)𝑌(𝑡 + 𝜏)]𝑖𝑠 𝑖𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑜𝑓 𝑡𝑖𝑚𝑒 𝑡𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡
Note: If 2 RPs X(t) and Y(t) are statistically independent and are at least WSS,
then 𝑅𝑋𝑌(𝜏) =
Independent RPs are uncorrelated. But converse is not always TRUE. So, the
condition for 2 RPs to be statistically independent:
(𝑖𝑖). 𝑅𝑋𝑌 (𝜏) = 𝐸[𝑋(𝑡)] 𝐸[𝑌(𝑡 + 𝜏)]

If mean value is not constant and 𝑅𝑋𝑋 (𝜏) is a function of time (t), then X(t) is not
50
 If mean value is constant and 𝑅𝑋𝑋 (𝜏) autocorrelation is independent of time (t),
then X(t) is a WSS.
Let X(t) represent a random process (RP). Conditions for non-stationarity are:
 The mean of X(t) is not constant
 The autocorrelation function depends on time, t
 The auto-covariance function depends on time t

Some properties of AC are:

(𝑖) 𝑀𝑒𝑎𝑛 𝑆𝑞𝑢𝑎𝑟𝑒 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑋(𝑡) = 𝐸[𝑋2(𝑡)] = 𝑅𝑋𝑋 (0) = 𝑃𝑜𝑤𝑒𝑟 𝑜𝑓𝑡ℎ𝑒
𝑝𝑟𝑜𝑐𝑒𝑠𝑠 𝑋(𝑡).
Note that [𝑋2(𝑡)] is also known as 2nd moment. And the 1st moment is E[X(t)].
(𝑖𝑖). 𝑅𝑋𝑋 (𝜏) 𝑖𝑠 𝑎𝑛 𝑒𝑣𝑒𝑛 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 𝑜𝑓 𝜏 𝑖. 𝑒. , 𝑅𝑋𝑋 (-)𝑅= 𝑅 (𝜏)
(𝑖𝑖𝑖). 𝐼𝑓 𝑡ℎ𝑒 𝑅𝑃 𝑍(𝑡)𝑖𝑠 𝑎 𝑠𝑢𝑚 𝑜𝑓 2 𝑅𝑃𝑠 𝑋(𝑡)𝑎𝑛𝑑 𝑌(𝑡)𝑖. 𝑒. , 𝑍(𝑡) = 𝑋(𝑡)
𝑋𝑋

+ 𝑌(𝑡) 51
Mean =
 Power spectral density for random
processes
• Let {X (t), t  T } be a wide-sense stationary
processes (WSS) ,
 1
T
 1
T

PXav  E lim
2T | X (t) | dt  T  2T
2
E  X (t) 2

dt
T  T  lim T

is the total average power of X

PXav (t)
can be denoted as
PXav 
1
PXav   S X
2
S () is the power spectral density of X(t)
()d
X
 Spectral Property
If X (t) is WSS ,
(1) PXav   RX (  0)

(2) RX ( ) and S X () a re Fourier transform pair. If

X (t) is a continuous WSS:

S X ()  dτ

RX ( ) = d


If X (t) is a discrete WSS:
and RX ( ) = d
S X ()  R X ( )  jk

 k
(3) S X () is e
a even function and more than zero.
S X ()  S X (); S X ()  0
 
1
(1) Showing :
PXav   S X ()d  RX ( 

2
X (t) G(
0)) transform
Fourier

According to Parseval’s theorem , the total energy of X (t)

 
2 dt  1   G(2)
QX  -  X d 

2
Let X(t)(t) X
X T (t)  
t 
T defined (t),0, T
t 

as: T
X T (t) Fourier
  G
transform T ()
 T

GT ()  T
X (t)e  jt
dt  X (t)e  jt
dt
  T
the total energy of X T (t)
T 
∞
dt  dt  1   GT (2)
 X
T 2 2
QX 

(t) d 

limX (t)
QX T  2
T
T  ∞
QX
The time average of QX
T

T 
QX 1 1 1
2 2

2T 
T

 X
T 
(t) dt  2T 2

GT ( ) d
2T  of QX
Then the ensemble average T

2T
T 
  1   1 
Q X
T
 X (t) dt  E 1 G 2
( ) 2
d  
E  E T
2T
 
T

 2T 2  
    
2T 
• Continuing…
the total average power X
of Q  X  1 T (t:) 1
T
2
PXav  lim  
T
 lim   X (t)  2

 
2T  
E
T 
 2T E T 
 2T T
dt  T lim
 T E X (t) 
dt
T T
1 1
 lim  RX (t, t)dt  R X (  0)dt  R( X  
T  2T T  2T
 T lim T 0)

PXav  RX ( 
0)  

 Q  1 2
PXav  lim E X T
1 GT () d
 lim E
 2T 2  
T  T 
 2T 

1   1  
   lim E GT ()  2  d
2 2T   
  
T 
 1  2
Let SX ()  E GT ()  , then

lim T  2T 

1
PXav   S X
(2) Showing:  S
R X ( )Fourier
transform X
1  
() 2
S X ()  E GT ()

lim T  2T 
T

GT () X (t)e  jtdt

 T

1  2 1  2 1
E GT ()   E GT ()   E  GT ()  G 

2T   2T   2T
T
()
1 T 
T 
 2T E   X dt T X (s)e  jsds
jt

T 
(t)e T
1  
T
 2T E T X jt
dt T X (s)e 
(t)e1 
 T T
js
ds
  j (t s )
dtds 
2T E  TT  X (t)X
(s)e  
T T
1  j (t s )
dtds

2T TT  E  X (t) X
 
T T
(s)1 e  j (t s )
dtds
2T   
 RX (t  s) 
 T T
e
• Continuing…
let u  t  s ,    t  u   , s  u -
t-s 2
t t 1 1
J    
1 2
1  1 2
u s s
2 2
u  2 2
T T 
S X ( ) 1   j (t s )
dtds
   R (t  s)  
 2T  T T X

e
1  J  RX  )  ej dud  u u
 2T  , -T  2 T -T  2 
(
,   T
 2T 
2T | |
 j
 1
   J  RX ( )  e du d 
2T 2T (2T | |) 
2T
 | |   j

  1   R X (  )  e
2T 
2T 
 d    RX (t  s)  e -j (t-s)dtds    Rx( )  e d
T T 
S X ( )  1   
 j

lim T  2T
 T T
(3) showing: S X ()  S X ()

RX ( )  RX ( )

Sx(ω) = dτ let 1  

Sx(-ω) = dτ

= d(-τ)

= dτ1

Sx(ω) = Sx(-ω)
Example: X (t )is a WSS process , the mean | |
RX ( )  4e 
m X (t) = 0, the autocorrelation c,os( )
(1)Find the power spectral density S X (ω);
(2)Find the total average power PXav
Solution: (1)
RX ( )  Sx()
Fourier transform

Sx(ω) = dτ

1 1 
4  2 
  1  (  
2
1  (  
(2)  ) 1 +  )
PXav   S X ()d
2
 RX (  0) 
 Example: X (tis) a WSS process , the mean m X (t) ,= the
0 power spectral
density:
SX () =

(1) Find the autocorrelation RX (τ )

(2) Find the total average power PXav
Solution: (1). 1  S ()  e j 1  22 
 X d 2 41
 e j

2   52  d
 
    4
1  2k1 4k2 
 2  e j
2    1   4
2

 
 k1 e- 2
d (2
1

 k e
2

) PXav    S X
1 7 2
k =  6 , k 12
1 2  R (  0)
()dX
5
 k1  k2 12