阵列信号处理中协方差矩阵为什么重要

阵列信号处理中协方差矩阵为什么重要

如下图所示为阵列探测示意图，其中目标到达角定义为信号入射方向与阵列法线之间的夹角，在窄带(点信源)和远场(平面波)的假设下，同一信号到达不同阵元存在波程差，这个波程差导致了各接收阵元的相位差，该相位差组成阵列的导向矢量和阵列流型矩阵。

图1 阵列信号探测示意图

设阵元数为$M$，信源数为$N$，第n个信源的到达角为${\theta }_n$则其对应的导向矢量可以表示为$\mathbf{a}({\theta }_n)=[1,e^{-i2\pi d\sin ({\theta }_n)/\lambda },...,e^{-i2\pi (M-1)d\sin ({\theta }_N)/\lambda }{]}^T$，所有信源组成的阵列流型矩阵可以表示为

$$\begin{array}{c}\begin{array}{rl}\mathbf{A}(\mathbf{\theta })& =[\mathbf{a}({\theta }_0),...,\mathbf{a}({\theta }_p),...,\mathbf{a}({\theta }_{N-1})]\\ & =\left[\begin{array}{ccccc}& 1,& 1,& ...,& 1\\ & e^{-i2\pi d\sin ({\theta }_0)/\lambda },& e^{-i2\pi d\sin ({\theta }_1)/\lambda },& ...,& e^{-i2\pi d\sin ({\theta }_{N-1})/\lambda }\\ & ⋮,& ⋮,& ⋮,& ⋮\\ & e^{-i2\pi (M-1)d\sin ({\theta }_0)/\lambda },& e^{-i2\pi (M-1)d\sin ({\theta }_1)/\lambda },& ...,& e^{-i2\pi (M-1)d\sin ({\theta }_{N-1})/\lambda }\end{array}\right]\end{array}\end{array}\text{\hspace{1em}(1)}$$

从(1)可以看出，$\mathbf{A}(\mathbf{\theta })$是一个范德蒙矩阵，且当$M>=N$，即信源数小于阵列维数时，阵列流型矢量是非奇异的。

所以阵列信号的接收模型可以表示为

$$\mathbf{x}(t)=\mathbf{A}(\mathbf{\theta })\mathbf{s}(t)+\mathbf{n}(t)\text{\hspace{1em}(2)}$$

其中$\mathbf{s}(t)=[s_0(t),...,s_{N-1}(t){]}^T$，$\mathbf{n}(t)=[n_0(t),...,n_{M-1}(t){]}^T$。计算阵列接收信号的协方差矩阵如下

$$\begin{array}{c}\begin{array}{rl}{\mathbf{R}}_x& =\mathrm{E}[\mathbf{x}(t){\mathbf{x}}^H(t)]=\mathbf{A}\mathrm{E}[\mathbf{s}(t){\mathbf{s}}^H(t)]{\mathbf{A}}^H+\mathrm{E}[\mathbf{n}(t){\mathbf{n}}^H(t)]=\mathbf{A}{\mathbf{R}}_s{\mathbf{A}}^H+{\sigma }^2\mathbf{I}\end{array}\end{array}\text{\hspace{1em}(3)}$$

对协方差矩阵进行奇异值分解可以得到

$${\mathbf{R}}_x=\mathbf{A}{\mathbf{R}}_s{\mathbf{A}}^H+{\sigma }^2\mathbf{I}=\mathbf{U}\mathbf{\Sigma }{\mathbf{U}}^H=[{\mathbf{U}}_s,{\mathbf{U}}_n]\left[\begin{array}{c}{\mathbf{\Sigma }}_s,\mathbf{O}\\ \mathbf{O},{\mathbf{\Sigma }}_n\end{array}\right]\left[\begin{array}{c}{\mathbf{U}}_s^H\\ {\mathbf{U}}_n^H\end{array}\right]\text{\hspace{1em}(4)}$$

其中${\mathbf{\Sigma }}_s=\mathrm{diag}({\sigma }_1^2+{\sigma }^2,...,{\sigma }_N^2+{\sigma }^2)=\mathrm{diag}({\lambda }_1,...,{\lambda }_N)$，${\mathbf{\Sigma }}_n={\sigma }^2\mathbf{I}$，令$\mathbf{U}=[{\mathbf{u}}_1,...,{\mathbf{u}}_N,|{\mathbf{u}}_{N+1},...,{\mathbf{u}}_M]$，则${\mathbf{U}}_s=[{\mathbf{u}}_1,...,{\mathbf{u}}_N]$，${\mathbf{U}}_n=[{\mathbf{u}}_{N+1},...,{\mathbf{u}}_M]$。根据奇异值分解的性质可知，$\mathbf{U}$是一个酉矩阵， { u 1 , . . . , u M } \lbrace\boldsymbol{u}_1,...,\boldsymbol{u}_M\rbrace {u1​,...,uM​}构成了一组标准正交基，即${\mathbf{u}}_i^H{\mathbf{u}}_i=1$，${\mathbf{u}}_i^H{\mathbf{u}}_j=0,i\ne j$，基于此，下面来证明一些结果：

(1) ${\mathbf{U}}_s^H{\mathbf{U}}_s=\mathbf{I},{\mathbf{U}}_n^H{\mathbf{U}}_n=\mathbf{I}$

证明：

$$\begin{array}{c}\begin{array}{rl}{\mathbf{U}}_s^H{\mathbf{U}}_s& =\left[\begin{array}{c}{\mathbf{u}}_1^H\\ ⋮\\ {\mathbf{u}}_N^H\end{array}\right][{\mathbf{u}}_1,...,{\mathbf{u}}_N]=\left[\begin{array}{cccc}& {\mathbf{u}}_1^H{\mathbf{u}}_1,& ...,& {\mathbf{u}}_1^H{\mathbf{u}}_N\\ & ⋮,& ...,& ⋮\\ & {\mathbf{u}}_N^H{\mathbf{u}}_1,& ...,& {\mathbf{u}}_N^H{\mathbf{u}}_N\end{array}\right]=\mathbf{I}\\ {\mathbf{U}}_n^H{\mathbf{U}}_n& =\left[\begin{array}{c}{\mathbf{u}}_{N+1}^H\\ ⋮\\ {\mathbf{u}}_M^H\end{array}\right][{\mathbf{u}}_{N+1},...,{\mathbf{u}}_M]=\left[\begin{array}{cccc}& {\mathbf{u}}_{N+1}^H{\mathbf{u}}_{N+1},& ...,& {\mathbf{u}}_{N+1}^H{\mathbf{u}}_M\\ & ⋮,& ...,& ⋮\\ & {\mathbf{u}}_M^H{\mathbf{u}}_{N+1},& ...,& {\mathbf{u}}_M^H{\mathbf{u}}_M\end{array}\right]=\mathbf{I}\end{array}\end{array}$$
(2) ${\mathbf{U}}_s{\mathbf{U}}_s^H+{\mathbf{U}}_n{\mathbf{U}}_n^H=\mathbf{I}$

证明：
$$\begin{array}{c}\begin{array}{rl}{\mathbf{U}}_s{\mathbf{U}}_s^H& =[{\mathbf{u}}_1,...,{\mathbf{u}}_N]\left[\begin{array}{c}{\mathbf{u}}_1^H\\ ⋮\\ {\mathbf{u}}_N^H\end{array}\right]=\sum _{i=1}^N{\mathbf{u}}_i{\mathbf{u}}_i^H\\ {\mathbf{U}}_n{\mathbf{U}}_n^H& =[{\mathbf{u}}_{N+1},...,{\mathbf{u}}_M]\left[\begin{array}{c}{\mathbf{u}}_{N+1}^H\\ ⋮\\ {\mathbf{u}}_M^H\end{array}\right]=\sum _{i=N+1}^M{\mathbf{u}}_i{\mathbf{u}}_i^H\\ & {\mathbf{U}}_s{\mathbf{U}}_s^H+{\mathbf{U}}_n{\mathbf{U}}_n^H=\sum _{i=1}^M{\mathbf{u}}_i{\mathbf{u}}_i^H\end{array}\end{array}$$
而${\mathbf{u}}_i{\mathbf{u}}_i^H=\mathbf{C},C_{i,j}=\{\begin{array}{l}0,i\ne j\\ 1,i=j\end{array}$，所以，显然有${\mathbf{U}}_s{\mathbf{U}}_s^H+{\mathbf{U}}_n{\mathbf{U}}_n^H=\mathbf{I}$

(3)${\mathbf{U}}_s{\mathbf{U}}_s^H={\mathbf{U}}_s({\mathbf{U}}_s^H{\mathbf{U}}_s{)}^{-1}{\mathbf{U}}_s^H$，${\mathbf{U}}_n{\mathbf{U}}_n^H={\mathbf{U}}_n({\mathbf{U}}_n^H{\mathbf{U}}_n{)}^{-1}{\mathbf{U}}_n^H=\mathbf{I}-{\mathbf{U}}_s({\mathbf{U}}_s^H{\mathbf{U}}_s{)}^{-1}{\mathbf{U}}_s^H$

证明：由证明1可得${\mathbf{U}}_s^H{\mathbf{U}}_s=\mathbf{I},{\mathbf{U}}_n^H{\mathbf{U}}_n=\mathbf{I}$，所以${\mathbf{U}}_s{\mathbf{U}}_s^H={\mathbf{U}}_s({\mathbf{U}}_s^H{\mathbf{U}}_s{)}^{-1}{\mathbf{U}}_s^H$和${\mathbf{U}}_n{\mathbf{U}}_n^H={\mathbf{U}}_n({\mathbf{U}}_n^H{\mathbf{U}}_n{)}^{-1}{\mathbf{U}}_n^H$显然是成立的，再由证明结果2，可以得到${\mathbf{U}}_n{\mathbf{U}}_n^H={\mathbf{U}}_n({\mathbf{U}}_n^H{\mathbf{U}}_n{)}^{-1}{\mathbf{U}}_n^H=\mathbf{I}-{\mathbf{U}}_s({\mathbf{U}}_s^H{\mathbf{U}}_s{)}^{-1}{\mathbf{U}}_s^H$也是成立的。

由上面的证明结果(3)可知，${\mathbf{U}}_s{\mathbf{U}}_s^H$和${\mathbf{U}}_n{\mathbf{U}}_n^H$分别表示在s空间和n空间上的投影算子，所以${\mathbf{U}}_s{\mathbf{U}}_s^{H}X$，${\mathbf{U}}_n{\mathbf{U}}_n^{H}X$分别表示将信号$\mathbf{X}$往s空间和n空间上进行投影，基于这个性质可以做很多事情。

(4)可以写成
$${\mathbf{R}}_x=[{\mathbf{U}}_s,{\mathbf{U}}_n]\left[\begin{array}{c}{\mathbf{\Sigma }}_s,\mathbf{O}\\ \mathbf{O},{\mathbf{\Sigma }}_n\end{array}\right]\left[\begin{array}{c}{\mathbf{U}}_s^H\\ {\mathbf{U}}_n^H\end{array}\right]={\mathbf{U}}_s{\mathbf{\Sigma }}_s{\mathbf{U}}_s^H+{\mathbf{U}}_n{\mathbf{\Sigma }}_n{\mathbf{U}}_n^H=\sum _{i=1}^N{\lambda }_i{\mathbf{u}}_i{\mathbf{u}}_i^H+\sum _{i=N+1}^M{\sigma }^2{\mathbf{u}}_i{\mathbf{u}}_i^H\text{\hspace{1em}(5)}$$
对${\mathbf{R}}_x$求逆可以得到
$${\mathbf{R}}_x^{-1}=\sum _{i=1}^N\frac{1}{{\lambda }_1}{\mathbf{u}}_i{\mathbf{u}}_i^H+\sum _{i=N+1}^M\frac{1}{{\sigma }^2}{\mathbf{u}}_i{\mathbf{u}}_i^H\text{\hspace{1em}(6)}$$
现在考虑杂波抑制前的场景，此时信号中的直达波和多径杂波信号远远强于目标信号和噪声信号，所以上述s空间即为杂波空间，n空间即为目标和噪声空间。由于${\lambda }_1,...,{\lambda }_N\gg {\sigma }^2$，所以可以对(6)进行如下近似
$${\mathbf{R}}_x^{-1}\approx \sum _{i=N+1}^M\frac{1}{{\sigma }^2}{\mathbf{u}}_i{\mathbf{u}}_i^H=\frac{1}{{\sigma }^2}{\mathbf{U}}_n{\mathbf{U}}_n^H\text{\hspace{1em}(7)}$$
结合上面证明(3)的结论可知，此时${\mathbf{R}}_x^{-1}$等效于n空间的投影算子。所以利用${\mathbf{R}}_x^{-1}$可以将原始信号往能量较弱的信号成分构成的子空间进行投影，可以较好地实现信号分离，正是有这样的性质，在很多空域处理算法中经常能看到${\mathbf{R}}_x^{-1}$的身影。所以${\mathbf{R}}_x$在空域信号处理中具有举足轻重的地位。

---

参考文献

[1]H. Krim and M. Viberg, “Two decades of array signal processing research: the parametric approach,” IEEE Signal Process. Mag., vol. 13, no. 4, pp. 67–94, 1996.

[2]Wan X, Yi J, Zhao Z, et al. Experimental research for CMMB-based passive radar under a multipath environment[J]. IEEE Transactions on Aerospace and Electronic Systems, 2014, 50(1): 70-85.