欠定盲分离时变混合矩阵的估计  

Time-Varying Mixing Matrix Estimation for Underdetermined Blind Source Separation

作　　者：付卫红[1] 李爱丽 黄坤[1] 马丽芬[1] 严新[1] 杨博[2] 

机构地区：[1]西安电子科技大学综合业务网理论及关键技术国家重点实验室,西安710071 [2]航天恒星科技有限公司,北京海淀区100086

出　　处：《电子科技大学学报》2015年第4期508-512,共5页Journal of University of Electronic Science and Technology of China

基　　金：国家自然科学基金(61201134;61201135);中央高校基本科研业务费专项资金(72124669);高等学校学科创新引智计划(B08038);重大专项(2012ZX03001027-001);中国航天科技集团公司卫星应用研究院创新基金(2014_CXJJ-TX_06)

摘　　要：针对欠定盲分离中时变混合矩阵的估计问题,在稀疏域二维最小偏差角算法的基础上,提出了一种改进的欠定盲分离时变混合矩阵估计算法。该算法通过判断原始阵各列上是否都有观测点聚集和聚集在原始阵上的观测点以外的点的聚集方向,来检测变化时刻;并利用基于点密度大区域检测算法估计混合矩阵。改进算法对于混合矩阵发生某些列增加、消失和变化时均能检测出变化,并且在大幅提高变化时刻检测概率和混合矩阵估计精度的同时,降低了复杂度。实验仿真结果表明,在20 d B信噪比时,混合矩阵估计精度提高了60%以上。Aiming at the problem of time-varying mixing matrix estimation in underdetermined blind source separation, an improved time-varying mixing matrix estimation algorithm is presented based on the method of planar minimum offset angle of sparse domain. By determining whether the original columns has observed the signals and the direction of the observed signals that are not in the direction of the original matrix the change time instant can be detected. The mixing matrix can be estimated by the method called the large point density area detecting. The improved algorithm can detect the changes of the columns increasing, disappearing and changing. Moreover, the detection probability of the changed time instant and the estimation accuracy of the mixing matrix are substantially increased, while the complexity of the algorithm is decreased. Experimental results show that when SNR is 20 d B, the estimation accuracy of the mixing matrix is improved by more than 60%.

关 键 词：盲源分离 混合矩阵估计 信号处理 时变 欠定 

分 类 号：TP911.7[自动化与计算机技术]