基于最小均方误差和稀疏特征的欠定盲源分离  

An algorithm for under-determined blind source separation based on the least-mean-square error and sparse features

作　　者：白树忠[1] 刘琚[2] 孙国霞[2] 

机构地区：[1]山东大学电气工程学院,山东济南250061 [2]山东大学信息科学与工程学院,山东济南250100

出　　处：《山东大学学报（工学版）》2008年第4期97-101,共5页Journal of Shandong University（Engineering Science）

基　　金：教育部留学回国人员科研启动基金资助项目([2005]55);教育部新世纪优秀人才支持计划基金资助项目(NECT-05-0582);高等学校博士学科点专项科研基金资助项目(20050422017);北京大学视觉与听觉信息处理国家重点实验室开放基金资助项目(0403)

摘　　要：针对欠定条件下的盲源分离问题,即观测信号个数小于信源个数的情况,提出了一种基于最小均方误差和稀疏特征的算法.首先,利用变换后信源的稀疏特征,采用一新的势函数通过聚类算法估计混叠矩阵.然后利用混叠矩阵和信源自身的相关性,通过寻找信源在聚类方向时间点上的精确值,以均方误差最小为准则寻找最佳分离矩阵实现信源的分离,克服了传统的分离算法在寻找最佳分离子矩阵方面的缺点.仿真结果显示使用该方法分离的信号具有更高的信噪比,和其他同类方法相比具有更优越的分离性能.An algorithm was presented based on the least-mean-square error and sparse features for under-determined blind source separation, i.e. , observed signal numbers are less than sources numbers. Based on the clustering method, the mixing matrix was first estimated by a new potential function using the sparseness of sources. By using the estimated mixing matrix and the self-cor- relation of sources and searching the accurate values at the source clustering directions, the optimal sub-matrix for separation was obtained according to the least-mean-square error criterion. This can overcome the disadvantages of traditional algorithm in searching the optimal sub-matrix. Simulation results show that the separated signals have higher SNR, and the proposed approach has better separation performance compared with other similar methods.

关 键 词：稀疏性 欠定分离 最小均方误差 

分 类 号：TN911.7[电子电信—通信与信息系统]