基于矩阵对角化的盲源分离算法研究

The Research of Blind Source Separation Algorithms Based on Matrix Diagonalization

盲源分离是从观测信号中恢复源信号的一种有效方法,目前已成为信号处理领域的研究热点。首先对三种盲源分离的算法进行分析,它们是:四阶盲辨识(FOBI)、特征矩阵的联合近似对角化(JADE)、二阶盲辨识(SOBI)。分析表明这些算法均有各自的不足,而另一方面,它们都是通过矩阵对角化实现盲源分离的。一个很自然的想法是将这些算法结合起来,以提高盲源分离的性能。仿真结果表明,JADE法和SOBI法的结合可以获得不错的盲分离效果。

Blind source separation is an efficient method to recover source signals from observed signals, and it has become an attractive research in the field of signal processing. First this paper analyses three types of blind source separation algorithms, that is, fourth-order blind identification (FOBI), joint approximative diagonalization of eigenmatrix (JADE) and second-order blind identification (SOBI). Analysis shows that all these algorithms have drawbacks. On the other hand, all of them implement blind source separation by matrix diagonalization. A natural idea is to combine these algorithms, for improving performance in blind separation. The result of simulation indicates that the combination of JADE and SOBI does a good job in blind separation.