Digital Signal Processing Reference
In-Depth Information
mization problem define the Lagrangian function
+
λ
σ
s
,
M−
1
M−
1
σ
q
k
x
k
|
ψ
=
x
k
−
p
0
(11
.
11)
H
k
|
2
k
=0
k
=0
where
λ
is the Lagrange multiplier. Setting
∂ψ/∂x
k
= 0 we get
−
σ
q
k
+
λσ
s
=0
,
x
k
|
H
k
|
2
which yields
c
2
σ
q
k
x
k
=
,
(11
.
12)
|
H
k
|
σ
s
where the constant
c
2
=1
/
√
λ
should be such that the power constraint (11.9) is
satisfied, that is,
M−
1
−
1
p
σ
s
σ
q
m
c
2
=
.
(11
.
13)
|
H
m
|
m
=0
The optimum multipliers
α
k
therefore can be taken to be the nonnegative num-
bers defined by
α
k,opt
=
c
σ
q
k
|
1
/
2
.
(11
.
14)
H
k
|
σ
s
From Eq. (11.4) we have
1
H
k
α
k,opt
β
k,opt
=
(11
.
15)
which therefore yields
σ
s
σ
q
k
1
/
2
1
=
1
c
1
|
β
k,opt
|
=
×
(11
.
16)
|
H
k
α
k,opt
|
|
H
k
|
1
/
2
The minimized objective function
E
mse
is obtained by substituting the optimal
α
k
into Eq. (11.8). The result is
M−
1
2
E
ZF mmse
=
σ
s
p
0
σ
q
k
|
.
(11
.
17)
H
k
|
k
=0
A few remarks are in order:
1.
Effective noise.
Note that the quantity
σ
q
k
/
H
k
|
2
appears in all the ex-
pressions. This can be regarded as the effective noise variance of the
k
th
channel. This will become more clear when we discuss Fig. 11.4 below.
|
2.
Power allocation.
The signal power at the output of
α
k
is the power input
to the
k
th channel
H
k
.
This is given by
P
k
=
σ
s
α
k,opt
=
γσ
q
k
|
H
k
|
where
γ
is a temporary notation for the constant part. Thus channels with
large effective noise σ
q
k
/
H
k
|
2
get
more power
in the optimal system.
|
Search WWH ::
Custom Search