Crosscorrelation Functions

ECE 3075A
Random Signals
Lecture 27

Crosscorrelation Functions

School of Electrical and Computer Engineering

Georgia Institute of Technology
Fall, 2003

Fall 2003

ECE 3075A B. H. Juang

Copyright 2003

Lecture #27, Slide #1

Jointly W-S Stationary Processes

A process X(t) is wide sense stationary if

Two processes X (t ) and Y (t ) are jointly wide sense

stationary if

X (t ) and Y (t ) are jointly stationary in wide sense.

X 1 and Y2 are two random variables,
X 1 = X (t1 and Y2 = Y (t1 + )

The crosscorrelation function is defined as

RXY (t1 , t2 ) = E [ X (t1 )Y (t2 )] = dx1 x1 y2 f X 1Y2 ( x1 , y2 ) dy2

where we have used the notation X 1 = X (t1 ), Y2 = Y (t2 )
Fall 2003

ECE 3075A B. H. Juang

Copyright 2003

Lecture #27, Slide #3

RYX ( ) = E[Y1 X 2 ] = dy1 y1 x2 fY1 X 2 ( y1 , x2 ) dx2

)

RXY ( ) = E[ X 1Y2 ] = dx1 x1 y2 f X 1Y2 ( x1 , y2 ) dy2

a function of only the time difference.

Lecture #27, Slide #2

The temporal order is of significance:

2. RXY (t , t + ) = E [ X (t )Y (t + )] = RXY ( ),

Copyright 2003

Crosscorrelation of W-S Stationary Processes

1. Both X (t ) and Y (t ) are wide sense stationary, and

Recall the original definition:

ECE 3075A B. H. Juang

Fall 2003

E [ X (t )] = X = constant (independent of time), and

E [ X (t ) X (t + )] = RXX ( ), a function of only the time difference.

The correlation function between two different

random processes is called a crosscorrelation
function.
The crosscorrelation function measures how
coherently two processes behave together, at
various points in time.
Crosscorrelation functions are very much of interest
in system analysis; for example, in studying the
relationship between the input and the output of an
electronic system, the relationship between the
interest rate and various market indices, etc.

In the above, Y1 = Y (t1 and X 2 = X (t1 + ) .

Let t1 = t2 , it then follows that t2 = t1 + and RXY ( ) = RYX ( ).
Fall 2003

ECE 3075A B. H. Juang

Copyright 2003

Lecture #27, Slide #4

Time Crosscorrelation Functions

1 T
x (t ) y (t + ) dt
T 2T T
1 T
RYX ( ) = lim
y (t ) x (t + ) dt
T 2T T
If the processes are jointly ergodic,

Define

R XY ( ) = lim

Properties of Crosscorrelation Functions

RXY (0) and RYX (0) only measures the correlation of
the two processes at synchronous points.

RXY (0) = RYX (0); RXY ( ) = RYX ( )

1/ 2
| RXY ( ) | [ RX (0) RY (0)]

The maximum of a crosscorrelation function can

occur anywhere, not necessarily at = 0 .

R XY ( ) = RXY ( ), and RYX ( ) = RYX ( )

Example:
Two jointly random processes are of the form

If the two random processes are independent,

X (t ) = 2 cos(5t + ) and Y (t ) = 10 sin(5t + )

RXY ( ) = E[ X 1Y2 ] = E[ X 1 ]E[Y2 ] = X 1Y2 = RYX ( )

where is a random variable uniformly distributed in (0, 2). Find

the crosscorrelation of the two processes

And if any of the process has zero mean, the

crosscorrelation function vanishes everywhere.

RXY ( ) = E [2 cos(5t + ) 10 sin(5t + 5 + )]

= 10 sin(5 ) + E[10 sin(10t + 5 + 2)] = 10 sin(5 )
Fall 2003

ECE 3075A B. H. Juang

Copyright 2003

Lecture #27, Slide #5

Fall 2003

Applications of Crosscorrelation Functions

ECE 3075A B. H. Juang

Copyright 2003

Lecture #27, Slide #6

Sinusoid Plus Noise

Z (t ) = X (t ) Y (t )
Z1 = Z (t1 ) = X (t1 ) Y (t1 ) = X 1 Y1

Let X (t ) = A cos(t + ) where is a random variable uniformly

1
distributed over (0,2 ).
RX ( ) = A2 cos

Z 2 = Z (t1 + ) = X (t1 + ) Y (t1 + ) = X 2 Y2

Let V (t ) be a zero mean noise process, statistically independent

RZ ( ) = E[ Z1Z 2 ] = E[( X 1 Y1 )( X 2 Y2 )]
= E[ X 1 X 2 + Y1Y2 X 1Y2 Y1 X 2 ]
= RX ( ) + RY ( ) RXY ( ) RYX ( )

The autocorrelation function

of the sum is the sum of all
the autocorrelation functions
plus the sum of all the
crosscorrelation functions.

If the two random processes are statistically independent and

one of them has zero mean, the cross terms varnish, resulting in
the autocorrelation function of the sum is the sum of the
autocorrelation functions.

RXY ( ) = E[ X 1Y2 ] = E[ X 1 ]E[Y2 ] = 0 if X 1 = 0 or Y2 = 0.

Fall 2003

ECE 3075A B. H. Juang

Copyright 2003

Lecture #27, Slide #7

of X (t ), the signal, with autocorrelation function :

RV ( ) = B 2 e | |
The observed process is Z (t ) = A cos(t + ) + V (t ) which has

an autocorrelation function :
1
RZ ( ) = A2 cos + B 2e | |
2
2 | |
0, as .
Note that RV ( ) = B e

It is thus possible to recover a sinusoid from noise contamination

as long as we measure the autocorrelation at sufficiently long
time lags.
Fall 2003

ECE 3075A B. H. Juang

Copyright 2003

Lecture #27, Slide #8

Use of Crosscorrelation - Example

A radar sends out a signal X(t) which is returned from a target,
after a round-trip delay of 1 . The receiver receives Y(t),
which includes the delayed and attenuated signal X (t 1 )
and the noise V (t ) :

Y (t ) = aX (t 1 ) + V (t ) where a < 1
We use the crosscorrelation function between X and Y to detect
the presence of the returned signal in Y.

Example 6-8.1
A random process has the form X(t)=A in which A is a random variable
with mean value 5 and variance 10. This random process can be
observed only in the presence of independent noise V(t) having an auto
correlation function of

RV ( ) = 10 exp( 2 | |)

1.
2.

Y (t ) = X (t ) + V (t )
RY ( ) = E[Y (t )Y (t + )]

RXY ( ) = E [ X (t )Y (t + )]
= E [aX (t ) X (t + 1 ) + X (t )V (t + )]

= E[ X (t ) X (t + )] + E[ X (t )V (t + )] + E[V (t ) X (t + )] + E[V (t )V (t + )]

= aRX ( 1 ) + RXV ( ) = aRX ( 1 )

= E[ A2 ] + RV ( ) = A2 + A 2 + RV ( ) = 35 + 10 exp(2 | |)

because X and V are statistically independent and V has zero mean.

RY ( ) = 35 + 10 exp( 2 | |),

RXY ( ) = aRX ( 1 ) attains maximum at = 1 , from which the

ECE 3075A B. H. Juang

Copyright 2003

Lecture #27, Slide #9

Example 6-8.2
A random binary process (see Section 6-2) has amplitude of 12
and ta = 0.01 . It is applied to the half-wave rectifier circuit as
shown on the right.
R
Find the autocorrelation of the output, RY ( )
Find the crosscorrelation function RXY ( )
+
_ X (t ) R Y (t )
Find the crosscorrelation function RYX ( ) .
The amplitude of Y (t ) is either 0 (when X (t ) is -12) or half of X (t ) when it is +12,
i.e. +6. The mean of Y (t ) is thus 3.

Y (t ) = Y + W (t ) where W (t ) is a zero - mean binary process with amplitude 3.

| |

9 + 91
, | | 0.01
RY ( ) = Y 2 + RW ( ) =
0.01
9,
| | 0.01
For crosscorrelation:

If | |> ta , X (t ) and Y (t ) are independent and X (t ) has zero mean, thus RXY ( ) = 0.
If | | t a , Y (t ) is half (= +6) of X (t ) (= +12) half of the time; the other half of time,

1
| |
| |

X (t ) is negative and Y (t ) is 0. Thus RXY ( ) = 12 6 1

= 361
.
2
0.01
0.01
Fall 2003

ECE 3075A B. H. Juang

Copyright 2003

RY ( ) = 35

RY ( ) = 35 + 10 exp( 2 | |) = 35.35 exp( 2 | |) = 0.035

Therefore, at = ( 0.5 ln 0.035) = 1.6762 RY ( ) is within 1% of RY ( ).

distance to the target can be calculated.

Fall 2003

Find the autocorrelation function of the sum of these two processes;

If the autocorrelation function of the sum is observed, find the value of
at which the autocorrelation function is within 1% of its value at = .

Lecture #27, Slide #11

Fall 2003

ECE 3075A B. H. Juang

Copyright 2003

Lecture #27, Slide #10