Digital Signal Processing Reference
In-Depth Information
where we have substituted for P ( e ) using Eq. (10.44). The second equality
is obtained by using the identity (G.6) from Appendix G. Since the frequency
response function from s ( n )to
s ( n )isgivenby
[ F ( ) H ( ) G ( )] ↓T
it therefore follows that this transfer function is unchanged when G ( )
replaces G ( ) . Now consider the noise power spectrum which is given by
|
| 2
G ( )
at the output of G ( ) . At the output of the C/D converter this
becomes
| 2
|
G ( )
.
↓T
With G ( ) replaced by G ( )wehave
| 2
↓T
| G ( )
=
| 2
P ( e jωT ) F ( ) H ( )
|
↓T
=
| 2
P ( e )
| 2 (from Eq. (G.6), Appendix G)
|
F ( ) H ( )
↓T |
[ F ( ) H ( ) G ( )] T
2
=
| 2
↓T
.
|
F ( ) H ( )
Observe now that the last numerator can be expanded as follows:
F ( j ( ω +2 πk ) /T ) H ( j ( ω +2 πk ) /T ) G ( j ( ω +2 πk ) /T )
T
k
1
2
F ( j ( ω +2 πk ) /T ) H ( j ( ω +2 πk ) /T )
G ( j ( ω +2 πk ) /T )
T
k
T
k
1
2 1
2
=
| 2
| 2
↓T
|
F ( ) H ( )
|
G ( )
,
↓T
where T =2 π/ω s . Here the second line follows from the Cauchy-Schwartz
inequality. This proves that
| G ( )
| 2
| 2
↓T
|
G ( )
.
↓T
That is, the noise power spectrum (at the output of the C/D converter)
cannot increase at any frequency ω, when we replace the receiver filter G ( )
with G ( ).
10.4.1.A More general noise power spectrum
Lemma 10.2 assumed that S qq ( )=1 . Now imagine that S qq ( ) is arbitrary
but nonzero for all ω. Suppose we rearrange Fig. 10.6 as in Fig. 10.7(a), where
the noise
q ( t ) has power spectrum equal to unity as in Lemma 10.2.
 
Search WWH ::




Custom Search