EE160 Handout #5

January 26, 2010

The Wiener-Khintchine Theorem

Haykins proof(!) of the Wiener-Khintchine theorem on pages 50-52 leaves a lot to be

desired. Here is a reasonably complete proof, courtesy of B.P. Lathi, Modern Digital and
Analog Communication Systems, and Prof. McElieces EE160 course notes from a few years
back.
Let x(t) be a real wide-sense stationary process with autocorrelation function

Rx ( ) = Ex(t)x(t + ).

Assume further that Rx ( ) satisfies the Dirichlet conditions (see e.g., pages 197-200 of
Signals and Systems). This implies that Rx ( ) is absolutely integrable, i.e., that
Z
|Rx ( )|d,

converges and that the Fourier transform

Z
Rx ( )ej2f d,

exists.
Now for each sample function x(t), we can define the truncated Fourier transform
Z T /2

XT (f ) = x(t)ej2f t dt.
T /2

The corresponding truncated power spectral density is T1 |XT (f )|2 . Since x(t) is a random
process, for each f , T1 |XT (f )|2 is a random variable. Let us denote its expectation by

1
ST (f ) = E |XT (f )|2 .
T
A natural definition for the power spectral density of the process x(t) is therefore

Sx (f ) = lim ST (f ). (1)
T

Of course, we may worry about whether the above limit exists. The Wiener-Khintchine
theorem, however, puts this issue to rest by asserting that the limit exists for all f , and by
identifying its value.

Theorem 1 For all f , the limit in (1) exists and

Z
Sx (f ) = Rx ( )ej2f d. (2)

Proof: Note that
Z 2
 T /2 
2 j2f t 
E |XT (f )| = E x(t)e dt

 T /2 
Z T /2 Z T /2
= E x(t)x( )ej2f (t ) dtd
T /2 T /2
Z T /2 Z T /2
= Ex(t)x( )ej2f (t ) dtd
T /2 T /2
Z T /2 Z T /2
= Rx (t )ej2f (t ) dtd.
T /2 T /2
Now a simple exercise in calculus shows that
Z T /2 Z T /2 Z T
f (t )dtd = (T | |)f ( )d.
T /2 T /2 T

(Show this!) Applying the above formula to our problem yields

Z T
E |XT (f )|2 = (T | |)Rx ( )ej2f d,
T
and so
1
Z T | |
Z
E |XT (f )|2 = (1 )Rx ( )ej2f d = Rx,T ( )ej2f d, (3)
T T T
where we have defined
(
| |
(1 T )Rx ( ) | | < T
Rx,T ( ) =
0 | | T
Integrals of the form (3) appear to be tricky to analyze. However, their asymptotic behav-
ior can be inferred from the following Lebesgue-dominated-convergence theorem (see, e.g.,
Bartle, The Elements of Integration).
Theorem 2 Let (fn ) be a sequence of complex-valued measurable functions which converges
almost everywhere to a complex-valued measurable function f . If there exists an integrable
function g such that |fn | g for all n, then f is integrable and, in fact,
Z Z
f d = lim fn d. (4)
n

If we take the fn s as the complex-valued functions Rx,T ( )ej2f , then the corre-
sponding limit is f = Rx ( )ej2f . Moreover, we can take the integrable function to be
g = |Rx ( )|. Then from the definition of Rx,T ( ) it is straightforward to see that
 
Rx,T ( )ej2f  |Rx ( )|,
 

and the Lebesgue-dominated-convergence theorem applies. Therefore

Z
1
Sx (f ) = lim E |XT (f )|2 = lim Rx,T ( )ej2f d
T T T
Z
= lim Rx,T ( )ej2f d
T
Z
= Rx ( )ej2f d.