两种聚类算法在欠定稀疏盲分离中的比较  

Comparison of two cluster algorithms for underdetermined blind sparse source separation

作　　者：谢再晋[1] 林用满[2] 林土胜[2] 

机构地区：[1]华南理工大学理学院,广东广州510641 [2]华南理工大学电子信息学院,广东广州510641

出　　处：《实验技术与管理》2009年第8期40-44,47,共6页Experimental Technology and Management

基　　金：国家自然科学基金项目(604720067);广东省自然科学基金团队项目(04205783)

摘　　要：欠定稀疏盲分离算法主要是采用“两步法”：第一步用混叠信号估计混叠矩阵；第二步根据估计的混叠矩阵求解源信号。在两步法中，C-均值聚类算法和模糊C-均值聚类算法常用来估计混叠矩阵，这两种聚类的研究理论都较成熟，故它们得到很大的应用。该文在欠定稀疏盲分离中，比较了这两种算法。试验结果表明，模糊C-均值聚类算法比C-均值聚类算法估计混叠矩阵更加精确，恢复源信号精度更高，但算法复杂，分离的时间长。For underdetermined blind sparse source separation, a two step strategy using a linear mixing model is used as follows： （1） Firstly, use the sparse mixture model to estimate the mixing matrix for the observed data, and （2） Secondly, solve a linear programming problem to obtain the sources using the estimated mixing matrix. In the method, C-means clustering and fuzzy C-means clustering algorithms often are used for estimated the mixing matrix due to the maturity of the both clustering methods. Both clustering methods have been applied extensively due to the intensive studies of them. The comparison of simulation results indicates fuzzy C-means cluster algorithm outperforms C means cluster algorithm in the estimation of mixing matrix. The algorithm therefore can recover the source signals more accurately. But the drawbacks of this algorithm are its implementation complexity and excessive computation time.

关 键 词：欠定稀疏盲分离 混叠矩阵 C-均值聚类算法 模糊C-均值聚类算法 

分 类 号：TN911.7[电子电信—通信与信息系统]