基于平行因子分析的欠定混合矩阵估计算法  

Underdetermined Mixed Matrix Estimation Based on Parallel Factor Analysis

作　　者：王少波 郭英[1] 眭萍 李红光[1] 杨鑫 WANG Shaobo;GUO Ying;SUI Ping;LI Hongguang;YANG Xin(Institute of Information and Navigation,Air Force Engineering University,Xi an 710077,China)

机构地区：[1]空军工程大学信息与导航学院

出　　处：《探测与控制学报》2019年第6期101-106,共6页Journal of Detection & Control

基　　金：国家自然科学基金项目资助(61601500);全军研究生资助课题(JY2018C169)

摘　　要：针对现有算法在解决非稀疏信号的欠定混合矩阵估计中,存在的计算时间长、初值敏感且容易陷入局部收敛的问题,提出了基于平行因子分析的欠定混合矩阵估计算法。该算法利用信号的协方差矩阵构造三阶张量,采用直接三线性分解确定交替最小二乘(ALS)算法的初始迭代矩阵,然后在迭代过程中采用标准线搜索加速收敛,最终实现张量分解得到混合矩阵。仿真实验表明,该方法不要求信源的稀疏性,较ALS算法估计精度可以提高约3 dB,迭代次数减少约41.4%~84.3%,是一种有效的欠定混合矩阵估计算法。Aiming at the shortcomings of the existing algorithm in solving the under-determined hybrid matrix estimation,and the problems of non-sparse signals,the shortcomings of long calculation time,initial value sensitivity and easy to fall into local convergence are presented,an underdetermined hybrid matrix estimation algorithm based on parallel factor analysis was proposed.The algorithm used the covariance matrix of the signal to construct the third-order tensor,and used the direct trilinear decomposition to determine the initial iterative matrix of the alternating least squares algorithm,and then used the standard line search to accelerate convergence in the iterative process.Simulation results showed that the method did not require the sparseness of the source.The estimation accuracy of the ALS algorithm could be improved by about 3dB,and the number of iterations was reduced by about 41.4%-84.3%.

关 键 词：欠定盲源分离 混合矩阵估计 交替最小二乘 直接三线性分解 标准线搜索 

分 类 号：TN975[电子电信—信号与信息处理]