Digital Signal Processing Reference
In-Depth Information
where
μ
k
are the eigenvalues of (
U
f
H
†
HU
f
)
M
.Deno ingthe
M
dominant
singular values of the channel as
σ
0
,h
≥ σ
1
,h
≥ ...≥ σ
h,M−
1
(19
.
84)
we can show, as we did in Sec. 12.4.2, that
σ
h,
0
≥
σ
h,
1
≥
σ
h,M−
1
≥
μ
0
,
μ
1
,
μ
M−
1
.
(19
.
85)
Thus,
φ
=det
σ
q
Σ
f
(
U
f
H
†
HU
f
)
M
M−
1
I
+
σ
s
(1 +
σ
s
σ
q
σ
f,k
σ
h,k
)
≤
(19
.
86)
k
=0
with equality achieved whenever
U
f
is chosen such that
U
f
H
†
HU
f
is diago-
nalized, with the top
M
diagonal elements equal to the
M
dominant values of
σ
h,k
.
This is readily achieved by choosing
U
f
=
V
h
,
(19
.
87)
where
V
h
is the unitary matrix appearing in the channel SVD (Eq. (19.31)).
Summarizing, we have proved that
1+
σ
s
σ
q
σ
f,k
σ
h,k
−
1
/M
M−
1
Mσ
s
E
mse
≥
.
(19
.
88)
k
=0
The bound in Eq. (19.88) depends on the elements
σ
f,k
≥
0 of the precoder
matrix. These can be optimized subject to the power constraint to minimize
the right-hand side of Eq. (19.88) as shown next. After this we will show how
the minimized bound can actually be achieved by choosing the remaining free
parameters (feedback matrix
B
,
and the unitary part
V
f
of the precoder).
19.4.4 Minimizing the bound (19.88)
For simplicity of notation let
k
=
σ
s
σ
q
α
k
=
σ
f,k
≥
σ
h,k
.
0
,
(19
.
89)
But the power constraint can be rewritten as Eq. (19.38), that is,
M−
1
p
σ
s
σ
f,k
=
k
=0
To minimize the right-hand side of Eq. (19.88) we have to maximize
M−
1
(1 +
β
k
α
k
)
(19
.
90)
k
=0
Search WWH ::
Custom Search