基于比率矩阵聚类的欠定盲分离算法

利用稀疏分量的直线聚类性,提出了欠定盲源分离中估计混合矩阵的一种方法。该方法通过构造比率矩阵对观测信号进行分选,剔除了源信号频谱重叠的部分,然后利用鲁棒竞争的聚类学习算法获得对混合矩阵的精确估计,解决了源信号在频域不充分稀疏的条件下准确估计混合矩阵的问题。在估计出混合矩阵的基础上,利用最短路径法分离出源信号。由仿真结果可以看出,与传统的K均值估计混合矩阵的方法相比,方法具有更好的鲁棒性。

Underdetermined Blind Separation Based on Ratio Matrix Clustering

A method of the mixing matrix estimation in underdetermined source separation is proposed,which is based on the linear clustering of sparse component.Removing the overlapped source signals spectrum through choosing observation signals and constructing ratio matrix,the mixing matrix is estimated precisely by using Robust Competitive Agglomeration.Then,the source signals can be recovered by the shortest path method.The experiment proves that the method has better accuracy than K-means in estimating the mixing matrix.