Digital Signal Processing Reference
In-Depth Information
Proof of Theorem 15.2.
As explained in Sec. 12.4.2 we have
σ
h,k
≥ μ
k
for
0
≤
k
≤
M
−
1
.
Using this we see that
M−
1
M−
1
1
μ
k
≥
1
σ
h,k
E
mse
=
σ
q
σ
q
(15
.
42)
k
=0
k
=0
The lower bound is achieved by choosing
U
f
=
V
h
so that
U
f
H
†
HU
f
=
Σ
h
Σ
h
.
Thus the optimal precoder
F
canbewrittenintheform(15.40)
indeed. The equalizer can be computed from the zero-forcing condition:
GHF
=
I
. Taking
G
to be the minimum-norm left inverse of
HF
we get
G
=
(
HF
)
†
HF
−
1
(
HF
)
†
.
(15
.
43)
Since
I
M
=
U
h
Σ
h
I
M
0
=
U
h
(
Σ
h
)
M
0
HF
=
U
h
Σ
h
V
h
V
h
(15
.
44)
0
we have
⎡
⎤
Σ
h
)
M
0
]
U
h
U
h
⎣
⎦
=(
Σ
h
)
2
M
,
(
HF
)
†
HF
=[(
Σ
h
)
M
0
where (
Σ
h
)
M
is the
M
M
diagonal matrix containing the
M
dominant
singular values of the channel. Substituting into Eq. (15.43), the equalizer
G
takes the final form in Eq. (15.40). The MSE expression then achieves
the lower bound in Eq. (15.42), which proves Eq. (15.41).
×
15.3.2 Pure-MMSE transceivers
We now extend the results of Sec. 15.3.1 to the case where the zero-forcing
condition is not imposed. From Sec. 13.3 we know that given an arbtirary
precoder
F
, the MMSE equalizer has the closed form expression
G
=
σ
s
F
†
H
†
σ
s
HFF
†
H
†
+
σ
q
I
−
1
,
(15
.
45)
and the corresponding reconstruction error is
J
)+
σ
s
Tr
σ
q
HFF
†
H
†
−
1
.
I
J
+
σ
s
E
mse
=
σ
s
(
M
−
(15
.
46)
Search WWH ::
Custom Search