我们考虑一个extended-Saleh-Valenzuela信道模型,uniform linear array (ULA),在理想假设条件下,
H
\boldsymbol H
H由UE端无近场损伤的
L
L
L簇(clusters)/路径散射确定,表示为:
H
=
N
r
N
t
L
⋅
∑
l
=
1
L
α
l
u
l
v
l
H
(1)
\boldsymbol H = \sqrt{ \frac{N_r N_t}{L} } \cdot \sum_{l=1}^L \alpha_l \boldsymbol u_l \boldsymbol v_l^H \tag{1}
H=LNrNt⋅l=1∑LαlulvlH(1)
其中 α l \alpha_l αl是信道第 l l l条径的复增益, u l ∈ C N r × 1 \boldsymbol u_l \in \mathbb C^{N_r \times 1} ul∈CNr×1是第 l l l条径在接收端的导向向量(receiver array steering vector), v l ∈ C N t × 1 \boldsymbol v_l \in \mathbb C^{N_t \times 1} vl∈CNt×1是第 l l l条径在发送端的导向向量(transmit array steering vector), N t , N r N_t, N_r Nt,Nr分别为发送天线数和接收天线数。
我们定义
u
l
=
[
1
,
e
j
k
d
R
cos
(
ϕ
R
,
l
)
,
⋯
,
e
j
(
N
r
−
1
)
k
d
R
cos
(
ϕ
R
,
l
)
]
T
∈
C
N
r
×
1
(2)
\boldsymbol u_l = [1, e^{jkd_R \cos( \phi_{R,l})}, \cdots, e^{j(N_r - 1)kd_R \cos( \phi_{R,l})}]^T \in \mathbb C^{N_r \times 1} \tag{2}
ul=[1,ejkdRcos(ϕR,l),⋯,ej(Nr−1)kdRcos(ϕR,l)]T∈CNr×1(2)
v l = [ 1 , e j k d T cos ( ϕ T , l ) , ⋯ , e j ( N t − 1 ) k d T cos ( ϕ T , l ) ] T ∈ C N t × 1 (3) \boldsymbol v_l = [1, e^{jkd_T \cos( \phi_{T,l})}, \cdots, e^{j(N_t - 1)kd_T \cos( \phi_{T,l})}]^T \in \mathbb C^{N_t \times 1} \tag{3} vl=[1,ejkdTcos(ϕT,l),⋯,ej(Nt−1)kdTcos(ϕT,l)]T∈CNt×1(3)
其中, k = 2 π λ k=\frac{2\pi}{\lambda} k=λ2π是波数(wave number), d T , d R d_T, d_R dT,dR分别表示发送和接收端阵列天线间的距离, ϕ R , l \phi_{R,l} ϕR,l理解为第 l l l条径的到达角(AOA: Angle of Arrival), ϕ T , l \phi_{T,l} ϕT,l理解为第 l l l条径的出发角(AOD: Angle of Departure)。
定理:若 H ∈ C N r × N t \boldsymbol H \in \mathbb C^{N_r \times N_t} H∈CNr×Nt为式(1)中的信道矩阵,那么 H H H ∈ C N t × N t \boldsymbol {H}^H \boldsymbol H \in \mathbb C^{N_t \times N_t} HHH∈CNt×Nt的特征向量(eigenvector)可以被表征为 { v 1 , v 2 , ⋯ , v L } \{ \boldsymbol v_1, \boldsymbol v_2, \cdots, \boldsymbol v_L \} {v1,v2,⋯,vL}(发送端的array steering vector)的线性组合。
证明:事实上,
H
H
H
∈
C
N
t
×
N
t
\boldsymbol {H}^H \boldsymbol H \in \mathbb C^{N_t \times N_t}
HHH∈CNt×Nt的特征向量/矩阵等价于矩阵
H
\boldsymbol H
H的右奇异向量/矩阵
L
N
t
N
r
H
H
H
=
∑
i
∑
j
α
i
∗
α
j
⋅
(
u
i
H
u
j
)
⋅
v
i
v
j
H
=
V
ˉ
A
V
ˉ
H
(4)
\frac{L}{N_t N_r} \boldsymbol {H}^H \boldsymbol H = \sum_{i} \sum_{j} \alpha_i^{*} \alpha_j \cdot (\boldsymbol u^H_i \boldsymbol u_j) \cdot \boldsymbol v_i \boldsymbol v_j^H = \bar {\boldsymbol V} \boldsymbol A \bar {\boldsymbol V}^H \tag{4}
NtNrLHHH=i∑j∑αi∗αj⋅(uiHuj)⋅vivjH=VˉAVˉH(4)
其中
V
ˉ
=
[
α
1
∗
v
1
,
α
2
∗
v
2
,
⋯
,
α
L
∗
v
L
]
∈
C
N
t
×
L
A
(
i
,
j
)
=
u
i
H
u
j
i
,
j
=
1
,
⋯
,
L
(5)
\begin{aligned} \bar {\boldsymbol V} &= [\alpha_1^{*} \boldsymbol v_1, \alpha_2^{*} \boldsymbol v_2, \cdots, \alpha_L^{*} \boldsymbol v_L] \in \mathbb C^{N_t \times L} \\ A(i,j) &= \boldsymbol u_i^H \boldsymbol u_j \ \ \ \ i,j = 1, \cdots, L \end{aligned} \tag{5}
VˉA(i,j)=[α1∗v1,α2∗v2,⋯,αL∗vL]∈CNt×L=uiHuj i,j=1,⋯,L(5)
令
X
∈
C
L
×
L
\boldsymbol X \in \mathbb C^{L \times L}
X∈CL×L为矩阵
A
V
ˉ
H
V
ˉ
∈
C
L
×
L
\boldsymbol A \bar {\boldsymbol V}^H \bar {\boldsymbol V} \in \mathbb C^{L \times L}
AVˉHVˉ∈CL×L的特征矩阵,即
(
A
V
ˉ
H
V
ˉ
)
X
=
X
D
(6)
(\boldsymbol A \bar {\boldsymbol V}^H \bar {\boldsymbol V}) \boldsymbol X = \boldsymbol {XD} \tag{6}
(AVˉHVˉ)X=XD(6)
其中
D
\boldsymbol D
D是由
A
V
ˉ
H
V
ˉ
\boldsymbol A \bar {\boldsymbol V}^H \bar {\boldsymbol V}
AVˉHVˉ的特征值构成的对角阵。我们在式(6)两端同时乘
V
ˉ
\bar {\boldsymbol V}
Vˉ,得到
V
ˉ
X
⋅
D
=
(
V
A
V
ˉ
H
V
ˉ
)
⋅
X
=
L
N
t
N
r
H
H
H
⋅
V
ˉ
X
(7)
\bar {\boldsymbol V} \boldsymbol {X} \cdot \boldsymbol {D}= (\boldsymbol V \boldsymbol A \bar {\boldsymbol V}^H \bar {\boldsymbol V}) \cdot \boldsymbol X = \frac{L}{N_t N_r} \boldsymbol {H}^H \boldsymbol H \cdot \bar {\boldsymbol V} \boldsymbol {X} \tag{7}
VˉX⋅D=(VAVˉHVˉ)⋅X=NtNrLHHH⋅VˉX(7)
从式(7)可以看出, V ˉ X \bar {\boldsymbol V} \boldsymbol {X} VˉX是矩阵 H H H \boldsymbol {H}^H \boldsymbol H HHH的特征矩阵,特征矩阵中的每一个特征向量处在 V ˉ \bar {\boldsymbol V} Vˉ的列空间中,所以可以被表征为发送端导向向量的线性组合。
[1] V. Raghavan, J. Cezanne, S. Subramanian, A. Sampath and O. Koymen, “Beamforming Tradeoffs for Initial UE Discovery in Millimeter-Wave MIMO Systems,” in IEEE Journal of Selected Topics in Signal Processing, vol. 10, no. 3, pp. 543-559, April 2016, doi: 10.1109/JSTSP.2016.2523442.