基于约束NMF的盲源分离算法  

An Blind Source Separation Algorithm Based on Constrained NMF

作　　者：赵知劲[1,2] 卢宏[1] 徐春云[1] 

机构地区：[1]杭州电子科技大学通信工程学院,浙江杭州310018 [2]中国电子科技集团第36研究所通信系统信息控制技术国家级重点实验室,浙江嘉兴314001

出　　处：《压电与声光》2010年第6期1049-1052,共4页Piezoelectrics & Acoustooptics

基　　金：国防科技重点实验室基金资助项目

摘　　要：非负矩阵分解(NMF)是解决非独立源信号混合的盲分离的另一条新途径。该文提出一种基于约束NMF的盲源分离算法,在对NMF估计得到的源信号施加最小相关约束的基础上,对混合矩阵估计施加行列式约束,实现NMF的唯一分解。与已有算法相比,本算法放宽了对混合矩阵的稀疏性要求,大幅提高了信号分离质量。该算法仍适用于独立源信号分离问题。The nonnegative matrix factorization（NMF） is another new approach for resolving the problem of the blind source separation（BSS） with the non-independent source signals mixed.In this paper,an algorithm for BSS based on the constrained NMF is proposed.Through imposing the least correlated component constraints on the source signals estimated by NMF,the determinant criterion is applied to constrain the estimated mixing matrix,so that the unique factorization of NMF has been realized.Comparing with the present algorithm,the sparsity of the mixing matrix is relaxed and the quality of the recovered source signals has been improved significantly.The proposed algorithm is also suitable for separating the independent source signals.

关 键 词：盲源分离(BSS) 非负矩阵分解(NMF) 行列式准则 最小相关约束 

分 类 号：TN911[电子电信—通信与信息系统]