基于EASI的分阶段变步长盲源分离算法

A Variable Step Length in Stages Based on the EASI Blind Source Separation Algorithm is Proposed

固定步长和自适应步长是EASI在线盲源分离算法的两个应用方式。固定步长的最大缺点在于收敛速度与稳态误差间存在着矛盾,即大步长收敛速度快,但稳定性随之降低;反之,小步长稳定性较好,但收敛速度将会大幅下降。为了解决这一矛盾,提出了一种分阶段学习的盲源分离算法,即将整个信号的分离迭代过程根据迭代次数和信号分离度划分为初级阶段、伪初级阶段、捕捉阶段和跟踪阶段四个阶段,根据每个阶段信号的分离特点采取不同的学习速率。仿真实验证实,提出的自适应EASI算法可以有效提高盲源分离算法的稳态性能和收敛速度,该算法不仅在平稳环境中具有良好的性能,而且在非平稳环境中也具有更快的调整能力和更优的算法性能。

Online busy blind source separation algorithm based on the EASI usually have a fixed step length two algorithm and adaptive variable step length.Step in choosing a fixed step algorithm to nod exist the contradiction between convergence speed and steady-state error.Choose large step,initial stages in the iteration convergence speed faster,but as the iteration,the steady state performance is poorer;Conversely,choose small step for a long,slow convergence speed and steady state performance is better On-line blind source separation algorithm,the contradiction between the convergence speed and steady-state error,aiming at the above contradictions,this paper proposes a blind source separation algorithm of learning by stages,namely the whole signal separation iteration process according to the number of iterations and the signal separation degree is divided into four stages:primary stage,pseudo primary stage,capture and tracking phase,each phase based on the characteristics of signal separation in different learning rate.Simulation results confirm that the proposed adaptive EASI algorithm can effectively improve the steady-state performance of blind source separation algorithm and the convergence speed,the algorithm has good performance not only in the stable environment,but also has faster in non-stationary environments adjusting ability and better algorithm performance.