针对FastICA计算终点判断的算法改进

An Algorithm Improvement for Judging Computing End-point in FastICA

独立分量分析不能分离高斯分布信号,导致对含高斯噪声系统计算不收敛;FastICA可以从系统中逐个计算出独立分量,通过计算系统残余信息的自相关函数值,判断残余信息属性,找出独立分量分析的计算终点,对FastICA算法进行改进,可以避免无效计算,节省计算时间。改进后的算法可以自动判断含噪声的线性系统的独立分量数目,与预先定义分量数目的独立分量分析相比,具有更好的降噪效果。

The independent component analysis（ICA） can not separate the mixed Gaussian signal,and ICA may be nonconvergent in system which contains Gaussian noise.FastICA is one current ICA algorithm.After one independent component （IC）is figured out in FastICA,the next IC is figured out from the system remaining information.Computing autocorrelation value of remaining information and judging its attribute,the computation end point of ICA is found out and improvement is made in FastICA algorithm.The improvement may avoid useless computing and save computing time.The improved algorithm can judge automatically ICs number of linear system contained noise,Compared with ICA predefining ICs number,the improved algorithm has better effect in noise reduction.