Digital Signal Processing Reference
In-Depth Information
q
(
n
)
s
(
n
)
s
(
n
)
W
M
W
M
Λ
a
Λ
b
H
(a)
diagonal
normalized IDFT
circulant channel
diagonal
normalized DFT
q
(
n
)
x
(
n
)
s
(
n
)
x
(
n
)
s
(
n
)
W
M
W
M
Λ
a
Λ
b
U
H
U
(b)
circulant
unitary
diagonal
diagonal
unitary
Figure 17.7
. (a) The MMSE cyclic-prefix transceiver, and (b) the modified version
with unitary matrices
U
and
U
†
inserted.
Lemma 17.1.
Diagonal structure of error covariance
. For the MMSE cyclic-
prefix system shown in Fig. 17.7(a) (with or without zero forcing), the error
covariance matrix is diagonal.
♠
♦
Proof.
The reconstructed signal in Fig. 17.7(a) is given by
s
(
n
)=
Λ
b
WHW
†
Λ
a
M
s
(
n
)+
Λ
b
W
√
M
q
(
n
)
.
Since
H
is circulant,
WHW
†
is diagonal (Appendix D). So
s
(
n
)=
Λ
0
s
(
n
)+
Λ
1
Wq
(
n
)
for some diagonal matrices
Λ
0
and
Λ
1
.
The reconstruction error is therefore
e
(
n
)=
s
(
n
)
−
s
(
n
)=(
Λ
0
−
I
)
s
(
n
)+
Λ
1
Wq
(
n
)=
Λ
2
s
(
n
)+
Λ
1
Wq
(
n
)
,
where
Λ
2
is also diagonal. Since
s
(
n
)and
q
(
n
) are zero-mean uncorrelated
processes with covariances
σ
s
I
and
σ
q
I
,
respectively, the covariance of the
error is
R
ee
=
σ
s
Λ
2
Λ
2
+
σ
q
Λ
1
WW
†
Λ
1
=
σ
s
Λ
2
Λ
2
+
Mσ
q
Λ
1
Λ
1
,
which is diagonal indeed.
As explained in detail in Sec. 16.3, the MMSE system without zero forcing
creates a bias in the reconstruction error. This
bias should be removed
before the
Search WWH ::
Custom Search