基于累积张量分解的雷达信号欠定盲分离算法

针对时域和频域不充分稀疏条件下的雷达信号欠定盲分离问题,提出了基于信号不同时延的累积量与三阶张量分解估计混合矩阵的方法,并通过修正子空间投影算法完成对雷达源信号的恢复。首先将混合信号的四阶累积量表示成三阶张量,利用三阶张量分解获得混合矩阵估计值;通过求解雷达源信号任意时频点处对应的估计矩阵的列矢量,得到该时频点处最优超定矩阵的伪逆并恢复源信号。该算法可以解决复杂电磁环境下时频域同时混叠的雷达信号盲分离问题,仿真结果表明与现有算法相比提高了盲分离中混合矩阵估计性能和源信号恢复性能。

Underdetermined blind separation of radar signals based on tensor decomposition

Considering the underdetermined blind separation of radar signals which is non-disjoint in time frequency domain, a method based on observed signals and the cumulant is proposed for estimating matrix, and then the modified subspace projection is used for recovering radar signal. Firstly, the fourth-order cumulant is constructed based on observed signals and the cumulant is expressed as the third-order tensor, the mixed matrix is estimated by tensor decomposition with enhanced line search alternating least square. Finally, the over-determined matrix, which is calculated by estimating the column vector corresponds to the active original signal at any time-frequency point, is used to complete the estimation of signal by Moore-Penrose. The proposed method can solve the blind separation of non-disjoint radar signals in the time-frequency domain under complex electromagnetic environment. Simulation results show that the proposed method outperforms the existing methods in mixed matrix estimation and source recovery.