Digital Signal Processing Reference
In-Depth Information
The covariance matrix
v
of the observation noise is not diagonal. Consequently,
it is difficult to evaluate
−
n
in closed form in (
5.9
) for a general
v
. One answer
to this problem is the tridiagonal approximation of
v
for sufficiently small
ρ
,this
corresponds to the case where only adjacent node observations are correlated.
Following a similar procedure as in [
27
], we present the upper bound on the
probability of error at the fusion center using the tridiagonal approximation matrix
as well as Bergstrom's inequality [
1
] and finally considering the case where the
correlation coefficients are sufficiently small.
According to Bergstrom's inequality, for any positive definite matrices
P
and
Q
:
(
e
T
(
P
Q
)
−
1
e
)(
e
T
Q
−
1
e
)
+
e
T
P
−
1
e
≥
Q
)
−
1
e
.
(5.21)
e
T
Q
−
1
e
e
T
(
P
−
+
From Eqs. (
5.4
) and (
5.9
),
m
2
e
T
A
−
n
Ae
m
2
e
T
(Σ
v
+
σ
w
A
−
2
)
−
1
e
.Let
P
=
=
σ
w
A
−
2
)
and consider the matrix
Q
given by:
(Σ
v
+
⎛
⎝
⎞
⎠
−
ρ
d
...
−
ρ
d(L
−
2
)
−
ρ
d(L
−
1
)
1
ρ
d
ρ
d(L
−
3
)
ρ
d(L
−
2
)
−
1
...
−
−
σ
v
Q
=
.
(5.22)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
ρ
d(L
−
1
)
ρ
d(L
−
2
)
ρ
d
−
−
...
−
1
From (
5.21
) it can be shown that
e
T
v
+
−
1
σ
w
A
−
2
−
1
e
1
1
D
≥
−
(5.23)
L
=
1
H
2
G
2
2
σ
v
H
2
G
2
+
σ
w
e
T
Q
−
1
e
. Therefore, from (
5.9
) and (
5.23
), the fusion error probability
can be bounded from above by:
where
D
=
m
2
−
2
.
1
1
D
P
e
≤
Q
−
(5.24)
L
=
1
H
2
G
2
2
σ
v
H
2
G
2
+
σ
w
L/σ
v
, we get
When
ρ
=
0,
D
=
1
1
D
L
σ
v
lim
G
−
=
.
(5.25)
L
=
1
H
2
G
2
2
σ
v
H
2
G
2
+
→∞
σ
w
For the correlated observations, the optimization problem (
5.11
) can be reformu-
lated to the following equivalent statement:
⎧
⎨
min
L
=
1
G
2
subject to
β
2
e
T
AΣ
−
n
Ae
(5.26)
−
≤
0
,
⎩
G
≥
0
,
=
1
,
2
,...,L.
Since it is difficult to obtain an analytical closed form of the power allocation
problem, as it was explained in Sect.
5.3.1
, it is useful to have an analytical approx-
imation of the problem that minimizes the fusion error probability bound (
5.24
).
Search WWH ::
Custom Search