一种新的自适应步长ICA算法

学习速率的优选问题是自适应ICA算法中的一个重要问题。文中定义了描述信号分离状态的去相关性测度。以此为基础提出一种新的自适应步长选取算法,算法将信号分离过程分为3个阶段,在不同的阶段采用不同的步长选取方法。有效地解决了算法的稳态误差和收敛速度之间的矛盾,其仿真结果证实,该算法不仅在平稳环境中具有较快的收敛速度,而且具有良好的稳态性能和数值稳定性。

A New Adaptive Step-size Algorithm for ICA

An important problem in adaptive ICA algorithm is the opting step-size.This paper defines the measure of non-linear decorrelation which describes the separation states.On the basic of the measure non-linear decorrelation,a new ICA algorithm of adaptive step-size is presented.The algorithm divides the process of separation into three stages,in which different step-size algorithms are adopted.The simulation shows that this algorithm has both higher convergence rate and the lower steady-state error.