未知数量稀疏源的盲分离方法

提出了一种新的用于未知数量稀疏源的盲分离的统一方法.为了改善聚类分离的精度,首先选取混合空间中半径给定、中心位于原点的超球面以外的数据点,并将这些数据点映射到中心位于原点的单位超球面上.由此,原来的超平面线性聚类变为致密聚类,各聚类互相重叠的现象消失.然后,通过对这些映射到单位超球面上的数据点进行聚类分离来估计混合矩阵,再利用混合矩阵估计源,其中最佳不相似阈值和相应的聚类数量是自动生成的.仿真实验验证了该方法对实际音频信号包括语音信号的有效性.仿真结果表明该方法精确、简便,稳定性好,且计算量较小.

A new method for blind separation of sparse sources with unknown source number

In this paper,a new unified method for blind separation of sparse sources with unknown source number was presented.In order to improve the accuracy of clustering separation,all data-points outside the hypersphere of a given radius centered at origin,in mixture space were selected and then projected onto the unit hypersphere centered at origin.Thereafter the clusters became compact clusters,and the phenomenon for clusters to overlap each other disappeard.The mixing matrix was firstly estimated by clustering separation for the data points on the unit hypersphere,and then the estimation of sources was made by use of the mixing matrix.In the method,the optimal threshold of dissimilarity and the corresponding cluster number were produced automatically.The validity of the method for voice signals including speech signal was verified by computer simulation.The simulation results show that the method was accurate and convenient,meanwhile save calculation time in comparison with other methods.