考虑一个多用户多输入单输出MU-MISO
下行通信场景,基站端配置有
N
N
N根天线,其服务该小区下
K
K
K个单天线用户。假定信道为平坦瑞利衰落信道,记为
H
∈
C
N
×
K
\mathbf{H}\in\mathcal{C}^{N\times K}
H∈CN×K,基站最大发射功率为
P
P
P。则最小均方误差准则下的预编码矩阵为下面优化问题的最优解
min
W
E
n
,
s
[
∣
∣
H
H
W
s
+
n
−
s
∣
∣
2
]
s.t.
Tr
(
W
H
W
)
≤
P
\min_{\mathbf{W}}\mathbb{E}_{\mathbf{n},\mathbf{s}}\left[||\mathbf{H}^H\mathbf{Ws}+\mathbf{n}-\mathbf{s}||^2\right]\\ \text{s.t.}\quad \text{Tr}(\mathbf{W}^H\mathbf{W})\leq P
WminEn,s[∣∣HHWs+n−s∣∣2]s.t.Tr(WHW)≤P
拉格朗日函数为
L
(
W
,
λ
)
=
E
n
,
s
[
∣
∣
H
H
W
s
+
n
−
s
∣
∣
2
]
+
λ
[
Tr
(
W
H
W
)
−
P
]
=
E
n
,
s
[
(
(
H
H
W
−
I
K
)
s
+
n
)
H
(
(
H
H
W
−
I
K
)
s
+
n
)
]
+
λ
[
Tr
(
W
H
W
)
−
P
]
=
Tr
(
(
H
H
W
−
I
K
)
H
(
H
H
W
−
I
K
)
)
+
K
+
λ
[
Tr
(
W
H
W
)
−
P
]
=
Tr
(
W
H
H
H
H
W
)
−
Tr
(
W
H
H
)
−
Tr
(
H
H
W
)
+
2
K
+
λ
[
Tr
(
W
H
W
)
−
P
]
\begin{aligned} L(\mathbf{W},\lambda)&=\mathbb{E}_{\mathbf{n},\mathbf{s}}\left[||\mathbf{H}^H\mathbf{Ws}+\mathbf{n}-\mathbf{s}||^2\right]+\lambda\left[\text{Tr}(\mathbf{W}^H\mathbf{W})- P\right]\\ &=\mathbb{E}_{\mathbf{n},\mathbf{s}}\left[\left(\left(\mathbf{H}^H\mathbf{W}-\mathbf{I}_K\right)\mathbf{s}+\mathbf{n}\right)^H\left(\left(\mathbf{H}^H\mathbf{W}-\mathbf{I}_K\right)\mathbf{s}+\mathbf{n}\right)\right]+\lambda\left[\text{Tr}(\mathbf{W}^H\mathbf{W})- P\right]\\ &=\text{Tr}\left(\left(\mathbf{H}^H\mathbf{W}-\mathbf{I}_K\right)^H\left(\mathbf{H}^H\mathbf{W}-\mathbf{I}_K\right)\right)+K+\lambda\left[\text{Tr}(\mathbf{W}^H\mathbf{W})- P\right]\\ &=\text{Tr}\left(\mathbf{W}^H\mathbf{H}\mathbf{H}^H\mathbf{W}\right)-\text{Tr}\left(\mathbf{W}^H\mathbf{H}\right)-\text{Tr}\left(\mathbf{H}^H\mathbf{W}\right)+2K+\lambda\left[\text{Tr}(\mathbf{W}^H\mathbf{W})- P\right]\\ \end{aligned}
L(W,λ)=En,s[∣∣HHWs+n−s∣∣2]+λ[Tr(WHW)−P]=En,s[((HHW−IK)s+n)H((HHW−IK)s+n)]+λ[Tr(WHW)−P]=Tr((HHW−IK)H(HHW−IK))+K+λ[Tr(WHW)−P]=Tr(WHHHHW)−Tr(WHH)−Tr(HHW)+2K+λ[Tr(WHW)−P]
令
∂
L
∂
W
∗
=
H
H
H
W
−
H
+
λ
W
=
0
\frac{\partial L}{\partial \mathbf{W}^*}=\mathbf{H}\mathbf{H}^H\mathbf{W}-\mathbf{H}+\lambda \mathbf{W}=\mathbf{0}
∂W∗∂L=HHHW−H+λW=0
变量为复矩阵,应令关于其共轭的导数为0
可得
W
=
(
H
H
H
+
λ
I
N
)
−
1
H
=
H
(
H
H
H
+
λ
I
K
)
−
1
\begin{aligned} \mathbf{W}&=\left(\mathbf{H}\mathbf{H}^H+\lambda\mathbf{I}_N\right)^{-1}\mathbf{H}\\ &=\mathbf{H}\left(\mathbf{H}^H\mathbf{H}+\lambda\mathbf{I}_K\right)^{-1} \end{aligned}
W=(HHH+λIN)−1H=H(HHH+λIK)−1
上面的第2个等号,是由于
A
+
A
B
A
=
A
(
I
+
B
A
)
=
(
I
+
A
B
)
A
A+ABA=A(I+BA)=(I+AB)A
A+ABA=A(I+BA)=(I+AB)A
所以
(
I
+
B
A
)
−
1
A
−
1
=
A
−
1
(
I
+
A
B
)
−
1
(I+BA)^{-1}A^{-1}=A^{-1}(I+AB)^{-1}
(I+BA)−1A−1=A−1(I+AB)−1
两边同时左乘右乘
A
A
A,得到
A
(
I
+
B
A
)
−
1
=
(
I
+
A
B
)
−
1
A
A(I+BA)^{-1}=(I+AB)^{-1}A
A(I+BA)−1=(I+AB)−1A
相关文献指出
λ
o
p
t
=
P
/
K
\lambda_{opt}=P/K
λopt=P/K