基于自然梯度的非平稳信号自适应盲分离算法

An Adaptive Blind Source Separation Approach of Non-stationary Signals by Natural Gradient Rule

基于自然梯度原则并利用信号的时间相关属性对一类代价函数进行推导,获得一种新的非平稳信号自适应盲分离算法.算法利用样本的多时延解相关方法以及迭代计算的形式获得盲混合信号的分离矩阵,无需对观测样本进行分块处理,计算工作量低.仿真结果表明,算法分离精度高,迭代过程平稳,对多个信号源的盲分离可实现良好的分离性能.

A new adaptive blind source separation algorithm of non-stationary signals was presented by using natural gradient rule and time-correlation property of the source signals acting on a cost function. The algorithm uses the multiple time-delayed de-correlation method and iterative calculation mode to get the separation matrix and no block separation for the samples is needed, so the computing cost is low. The simulation shows that the algorithm can get high separation performance and stationary separation process even for multiple blind mixed signals.