1

Random Signal Analysis

• Random Variables and Random Processes

• Signal Transmission through a Linear System
 2

Discrete Random Variables

• A discrete random variable takes on a countable number

of possible values.
Suppose that a discrete random variable X takes on one of the values
x1,…, xn.

 Distribution functions: n

Probability Mass Function: p ( xi )  Pr{ X  xi }

 p( x )  1
i 1
i

Cumulative Distribution Function: F (a )  Pr{ X  a}   p ( xi )

xi  a
 Moments:
n
Expected Value, or Mean:  X  E[ X ]   xi p ( xi )
n i 1
The m-th Moment: E[ X ]   x p( xi ), m  1, 2,...
m m
i
i 1
 3

Continuous Random Variables

• A continuous random variable has an uncountable set of

possible values.
X is a continuous random variable if there exists a nonnegative
function f, defined for all real x  (, ), having the property that for
any set B of real numbers, Pr{ X  B}   f ( x)dx.
B

 f is called the probability density function (pdf) 

of X, denoted as: f X ( x) 

f X ( x)dx  1
a
 Cumulative Distribution Function (cdf): FX (a)   f X ( x)dx


 Expected Value, or Mean:  X  E[ X ]   xf X ( x)dx


 The m-the Moment: E[ X m ]   x m f X ( x)dx, m  1, 2,...

 4

Variance

• Define variance of a random variable X as:

Var[ X ]  E ( X  E[ X ]) 2 

 Var[X] describes how far apart X is from its mean on the average.
 Var[X] can be also obtained as: Var[ X ]  E[ X 2 ]   E[ X ]
2

 Var[X] is usually denoted as  X2 .

 The square root of Var[X],  X , is called the standard deviation
of X.
 5

Example 1. Uniform Distribution

• X is a uniform random variable on the interval (, ) if its pdf is given by

 1 f X ( x)
  x
f X ( x)     
1
 
0
 otherwise
  x

 cdf:  0 a 

a  a 
FX (a )   f X ( x)dx    a
 


 1 a
  
 Mean:  X   xf X ( x)dx   Variance:
 2
 The second moment:  X2  E[ X 2 ]   X2
  2     2 (   )2
E[ X 2 ]   x 2 f X ( x)dx  
 3 12
 6

Example 2. Gaussian (Normal) Distribution

• X is a Gaussian random variable with parameters 0 and  02 if its pdf is

given by
fX(x)
1  ( x  0 ) 2 
f X ( x)  exp   
2 0  2 02 
0 x
• X is denoted as X   ( 0 ,  02 )

 Mean:  X  0  Variance:  X2   02

a 1  ( x  0 ) 2 
 cdf: FX (a )   f X ( x)dx  1   exp   dx
 a
2 0  2 2
0 
x  0
z 
 1  z2  a   x2 
 1  a0
0
exp   dz  1  Q   Q( )  
 1
exp   dx
0 2  2  
 0  
2  2
 7

More about Q Function

 1  x2 
Q( )   exp   dx

2  2
• Q() is a decreasing function of .

• For X   ( 0 ,  0 ) ,
2

  1  ( x  0 ) 2   a  0 
Pr{ X  a}   f X ( x)dx   exp   dx  Q  
a a
2 0  2 2
0    0 

fX(x)
Pr{ X  a}

0 a x

Read the textbook for more discussion about Q function.

 8

Random Processes

• Sample values of a random process at time t1, t2, …, are a

collection of random variables {X(t1), X(t2), …}.

– Continuous-time random process: t   (set of real numbers)

– Discrete-time random process: t   (set of integers)

• Statistical description of random process X(t)

– A complete statistical description of a random process X(t) is known

if for any integer n and any choice of (t1 ,..., tn )   , the joint PDF of
n

( X (t1 ),..., X (tn )) is given.

Difficult to be obtained!
 9

Statistical Averages

• The mean of the random process X(t):


 X (tk )  E[ X (tk )]   xf X (t ) ( x)dx
 k

X(tk) is the random variable obtained by observing the random process X(t)
at time tk, with the pdf f X (t ) ( x).
k

• The autocorrelation function of the random process X(t):

 
RX (t1 , t2 )  E[ X (t1 ) X (t2 )]    x x f X (t1 ), X (t2 ) ( x1 , x2 )dx1dx2
  1 2

f X ( t1 ), X ( t2 ) ( x1 , x2 ) is the joint pdf of X(t1) and X(t2).

 10

Power and Power Spectrum of Random Signal

Time Domain
1 T /2
 T T / 2
Deterministic signal s(t): Ps  Tlim
2
| s (t ) | dt

Random signal (described as a random process X(t)):

 1 T /2 2  1 T /2
PX  E  lim 
T  T T / 2
X (t )dt   Tlim
  T

T / 2
RX (t , t )dt

Frequency Domain
Deterministic signal:


1 1 T /2
Ps  Gs ( f )df , Gs ( f )  lim | ST ( f ) |  lim  s (t   ) s* (t )dt
2
 T  T T  T T / 2

Random signal:
 1 1 T /2
 
PX  GX ( f )df , GX ( f )  lim E | X T ( f ) |  lim  RX (t   , t )dt
T  T
2
T  T T / 2
 11

Example 1. Wide-Sense Stationary (WSS) Processes

• A random process X(t) is wide-sense stationary (WSS) if the following

conditions are satisfied:

–  X (t )  E[ X (t )] is independent of t;
– RX (t1 , t2 ) depends only on the time difference =t1-t2, and not on t1 and t2
individually.

1 T /2
 Power: PX  lim
T  T 
T / 2
RX (t , t )dt  RX (0)

 Power spectrum: GX ( f )  lim 1 T / 2 RX (t   , t )dt  RX ( )

 T  T T / 2
 12

Example 2. Cyclostationary Processes

• A random process X(t) with mean X(t) and autocorrelation function

RX(t+, t) is called cyclostationary, if both the mean and the
autocorrelation are periodic in t with some period T0, i.e., if
–  X (t  T0 )   X (t )
– RX (t    T0 , t  T0 )  RX (t   , t )

1 T /2 1

T0 / 2
 Power: PX  lim
T  T T / 2
RX (t , t )dt 
T0 
T0 / 2
RX (t , t )dt

1 T /2 1
T T / 2
T0 / 2
 Power spectrum: GX ( f )  Tlim

R X (t   , t ) dt 
T0 
T0 / 2
RX (t   , t )dt
 13

Signal Transmission through a

Linear System
 14

Linear Time Invariant (LTI) System

Random signal 
Impulse Response Y (t )  X (t )  h(t )   X ( )h(t   )d

X(t) h(t)

• If a WSS random process X(t) passes through an LTI system with

impulse response h(t), the output process Y(t) will be also WSS
with mean 
Y   X  h(t )dt   X H (0)


autocorrelation
RY ( )  RX ( )  h( )  h( )

and power spectrum

GY ( f )  GX ( f ) | H ( f ) |2
 15

Example 3. Gaussian Processes

• A random process X(t) is a Gaussian process if for all n and all (t1, …, tn),
n
the random variables { X (ti )}i 1 have a jointly Gaussian pdf.

 For Gaussian processes, knowledge of the mean and auto-

correlation gives a complete statistical description of the process.

 If a Gaussian process X(t) is passed through an LTI system, the

output process Y(t) will also be a Gaussian process.
 16

Example 4. White Processes

• A random process X(t) is called a white process if it has a flat spectral

density, i.e., if GX(f) is a constant for all f.
GX ( f )
N0

 Power: PX   GX ( f )df  
2



0 f
 Autocorrelation:
N0
N0 : two-sided power spectral density
GX ( f )  RX ( )   ( ) 2
2

 If a white process X(t) passes through an LTI system with impulse

response h(t), the output process Y(t) will not be white any more.

Power spectrum of Y(t): GY ( f ) 

N0
2 | H ( f ) |2
 
Power of Y(t): PY   H ( f ) df  
N0 2 N0
2 2 h 2 (t )dt
 