基于KM-PCA稀疏信号的盲源分离算法

Blind source separation algorithm based on KM-PCA sparse signal

欠定盲信道估计是欠定盲源分离的关键组成部分,其估计精度直接影响到源信号的估计精度.基于充分稀疏假设,在K均值聚类的基础上,提出一种新的欠定盲信道估计算法——K均值与主成分分析方法(KM-PCA算法).该算法首先对观测数据进行K均值聚类,然后对聚类分析结果分别进行主成分分析,修正其聚类中心,从而提高混叠矩阵的估计精度.采用语音信号进行的仿真实验表明,KM-PCA算法简单有效,估计精度优于传统的欠定盲信道估计算法.

Underdetermined blind channel estimation is the key element of underdetermined blind source separation and the estimation accuracy will directly affect the estimation accuracy of the source signal. Un- der the assumption that the sources were fully sparse, a new underdetermined blind channel estimation algorithm--analysis method of K-means and principal component algorithm （KM-PCA） was proposed. This algorithm was firstly used to perform K-means clustering of the observed data. Then the principal compo- nent analysis was conducted for the clustering results and the cluster centers were revised, so that the esti- mation accuracy of mixing matrix was improved. The phonetic signals were simulated and its result showed that KM-PCA algorithm was simple and effective, whose estimation accuracy was superior to that of tradi- tional underdetermined blind channel estimation algorithms.