蚁群聚类的欠定盲源分离方法

利用欠定盲源分离情况下稀疏源信号具有直线聚类的特点，提出了一种估计混叠矩阵的新方法。通过对混叠信号进行标准化处理，使混叠信号形成球形簇，将线性聚类转变成致密聚类；利用蚁群聚类算法对其进行搜索得到聚类中心，从而获得对混叠矩阵的精确估计。该方法能实现源信号数目未知情况下的欠定盲源分离，且能推广到三路或更多路观测信号的情况。对语音信号的仿真结果证明，该方法能精确地分离和恢复原始信号。

Underdetermined blind separation based on ant colony clustering

Taking advantage of the straight line clustering of the sparse source signals in underdetermined blind separation, a method of the mixing matrix estimation is proposed. The aliasing signals are standardized and the aliasing signals are formed spherical cluster, so the linear cluster is turned into density cluster. And then the clustering center is searched and obtained by using the ant clustering algorithm. The aliasing matrix and the source signals are accurately evaluated. The proposed algorithm can separate the source signals in which the number is unknown and it is also effective to separate three or more observed signals. The simulation results of speech signals show that this method can precisely separate and restore the original signals.