一种基于二阶统计的盲源分离方法及性能分析

A SOS-Based Algorithm and Its Performance Surfaces for Blind Source Separation

提出了一个基于信息论原理的目标函数 ,该目标函数可以作为衡量输出分量独立性的标测度。最小化该目标函数并利用信号的非平稳特性和两种网络结构形式的等价性 ,得到一种可以进行非平稳信号的盲分离的训练算法 ;计算机仿真结果表明了该算法的有效性。最后对目标函数的性能进行了分析。

Higher-order statistics(HOS) is always used in present algorithms for blind source separation. In this paper, we propose an informaion theory based on objective function for measuring the statistics independent of source signals, and develop a learning algorithm for blind separation of nonstationary signals through minimizing the objective function, in which the property of nonstationary and direct architecture neural network are applied. The analysis shows the equivalence of two neural network architectures in some special cases. Only second-order statistics(SOS) is used in the proposed algorithms. The computer simulation shows the validity of the proposed algorithm. Finally, the paper gives out the performance surface of the object function