改进的复值快速独立分量分析算法

摘要

An improved CFastICA algorithm was proposed to solve the problem of initial value sensibility and to improve convergence speed.First,Newton decline factor was used to optimize Newton iteration convergence direction to make the separated matrix be close to the optimal value to a certain extent,then remove the conver-gence factor and use the original Newton iteration realize fast convergence.Simulation results showed that the proposed algorithm had the same convergence precision with Newton decline CFastICA,and its convergence time was 52.85% less than Newton decline CFastICA.The combination property of proposed algorithm was sig-nificantly better than both of CFastICA and Newton decline CFastICA under the low SNR.%针对复值快速独立分量分析算法(CFastICA)对初始权值敏感且收敛速度较慢的问题，提出了改进的CFastICA 算法。该算法首先利用牛顿下降因子优化牛顿迭代的收敛方向，使分离矩阵在一定程度上接近最优值，然后去除牛顿收敛因子，利用普通牛顿迭代实现分离矩阵快速收敛。仿真实验表明：提出的算法拥有和牛顿下降 CFastICA 同样的收敛精度，收敛时间比牛顿下降 CFastICA 减少了近53%，且在低 SNR 下，提出算法的综合收敛性能明显优于 CFastICA 和牛顿下降 CFastICA 算法。