Digital Signal Processing Reference
In-Depth Information
Appendix to Chapter 8
8.A Monotonicity of noise gain
With
A
M
denoting the (
M
+
L
)
×
M
banded Toeplitz matrix (8.2), denote
R
M
=
A
†
M
A
M
.
(8
.
25)
Using Eq. (8.8) with
T
=
A
M
,wehave
A
M
2
=Tr
A
M
(
A
M
)
†
=Tr(
A
†
M
A
M
)
−
1
(8
.
26)
so that
M
Tr
R
−
M
.
1
M
A
M
2
=
1
noise gain =
(8
.
27)
Proving that the noise gain is monotonically increasing is therefore equivalent
to proving that
M
+1
Tr
M
Tr
R
−
M
.
1
1
R
−
M
+1
≥
(8
.
28)
It should be recalled here that
R
M
is positive definite and Toeplitz (see Eq.
(8.23)). We begin by proving the following lemma. The results in this appendix
are based on the work of Ohno [2006].
♠
Lemma 8.2.
Let
R
M
+1
be (
M
+1)
×
(
M
+ 1) positive definite. Partition
it as
R
M
+1
=
R
M
,
v
(8
.
29)
v
†
c
where
R
M
is
M
×
M
.Then
[
R
−
M
+1
[
R
−
M
]
kk
]
kk
≥
(8
.
30)
for 0
≤
k
≤
M
−
1.
Similarly, suppose
S
M
is the lower
M
×
M
principal
submatrix of
R
M
+1
,thatis,
R
M
+1
=
d
u
†
uS
M
.
(8
.
31)
Then we have
[
R
−
M
+1
[
S
−
M
]
kk
]
k
+1
,k
+1
≥
(8
.
32)
for 0
≤
k
≤
M
−
1
.
♦
Note that since
R
M
+1
is positive definite,
R
M
and
S
M
are positive definite, and
c, d >
0
.
The Lemma does not assume that the matrices are Toeplitz.
Proof.
Using the formula for inverses of matrices in partitioned form (Sec.
B.4.3, Appendix B) we get
⎡
⎣
⎤
=
I
−
R
−
M
v
0
R
−
M
.
0
I
0
R
−
M
+1
1
c −
v
†
R
−
M
v
⎦
−
v
†
R
−
M
0
1
1
Search WWH ::
Custom Search