Beruflich Dokumente
Kultur Dokumente
We can do the correlation and power spectral analysis of such signals in the usual
manner. However, for analysis of nonlinear operations like the multiplication with a
random process, the following trigonometric representation is useful.
Fig. Power spectrum of a band-pass random process
and
RX C X S ( ) RX ( ) cos 0 (t ) cos 0t RY ( )sin 0 (t ) sin 0t
RXY ( ) cos 0 (t ) sin 0t RYX ( ) sin 0 (t ) cos 0t
RX ( )[cos 0 (t ) cos 0t sin 0 (t )sin 0t ]
RXY ( )[cos 0 (t )sin 0t si n 0 (t ) cos 0t ]
RX ( ) cos 0 RXY ( ) sin( 0 )
RX ( ) cos 0 RYX ( )sin 0
{Y (t )}
{ X (t )}h(t ) 1t
j if >0
H j if <0
0 if =0
H ( ) j sgn( )
2
and H ( ) 1
SY H S X ( ) S X
2
and
jS XX ( ) for 0
S XY ( ) H S XX ( )
jS XX ( ) for 0
jS XX for >0
SYX ( ) H * S XX ( )
jS XX for <0
SY S X
and
S XY ( ) SYX
RX ( ) RY ( )
and
RXY ( ) RYX ( )
The Hilbert transform of X (t ) is denoted as X (t ). Therefore, from (2) and (3) we
establish
X C (t ) X (t ) cos 0t X (t )sin 0t
and X S (t ) X (t ) cos 0t X (t ) sin 0t
and
X (t ) X c (t ) cos 0t X S (t ) sin 0t
The realization for the in phase and the quadrature phase components is shown in the
figure below.
From the above analysis, we can summarise the following expressions for the
autocorrelation functions
RX ( ) RX ( ) RX ( ) cos 0 RYX ( )sin 0
C S
where
R X ( ) Hilbert transform of RX ( )
= h( s) RX ( s)ds
X2 c X2 s RX (0).
RX ( )h( ) 1 R X ( )
Similarly,
RX C X S ( ) =RX ( )sin 0 RYX ( ) cos 0
=RX ( ) sin 0 R X ( ) cos 0
and
j[ S X ( 0 ) S X ( 0 )] B
S X C X S ( )
0 otherwise
Notice that the cross power spectral density S X C X S ( ) is purely imaginary. Particularly, if
S X ( ) is locally symmetric about 0 ,
S X C X S ( ) 0
implying that
RX C X S ( ) 0
Consequently, the zero-mean processes { X c (t )} and { X s (t )} are also uncorrelated.
Example
Suppose the band-limited white-noise process { X (t )} has the PSD S X ( ) as shown in
Fig below.
N0
2
B
c
S Xc ( ) S X s ( )
N0
2
B B
2
2
B
N 0 B sin 2
RX ( ) cos 0
2 B
2
The plot of S X c ( ) S X s ( ) is as shown in the Fig. Therefore,
B
sin
N B 2
RX c ( ) RX s ( ) 0
2 B
2
Remark
(1) The representation of the band-pass process { X (t )} in terms of the in-phase and
the quadrature phase components is not unique. By selecting different c , we can
have different representations.
(2) The band-pass process { X (t )} can be represented as
X (t ) A(t ) cos( 0 t (t ))
where
A(t ) X c2 (t ) X s2 (t )
and
X s (t )
(t ) tan 1 .
X c (t )
A(t ) and (t ) are respectively called the envelope and the phase of the process
{ X (t )}.
(3) If { X (t )} is a Gaussian process, then { X (t )} ( being linear transform of { X (t )} )
is also Gaussian. Consequently, the processes { X c (t )} and { X s (t )} are also
Gaussian.
(4) Under the condition of local symmetry of S X ( ) about 0 , { X c (t )} and { X s (t )}
are uncorrelated. If { X c (t )} and { X s (t )} are also Gaussian. processes, then
{ X c (t )} and { X s (t )} will be independent . Using the results on the PDF of
functions of RVs, we get following.
The envelope A(t ) X c2 (t ) X s2 (t ) will be Rayleigh-distributed. Thus
x2
x 2 RX c {0)
e , for x 0
f A( t ) ( x) RX {0)
c
0 otherwise
1 X s (t )
The phase (t ) tan will be U [0 2 ] distributed
X c (t )