基于Parzen核估计的ICA算法及其性能分析  

Independent Component Analysis Algorithm Based on Parzen Kernel Estimator and Performance Analysis

作　　者：王法松[1] 李宏伟[1] 朱培民[2] 

机构地区：[1]中国地质大学数学与物理学院,武汉430074 [2]中国地质大学地球物理与空间信息学院,武汉430074

出　　处：《武汉理工大学学报》2005年第9期93-96,共4页Journal of Wuhan University of Technology

基　　金：国家自然科学基金(60472062);湖北省自然科学基金(2004ABA038).

摘　　要：基于概率密度非参数估计的Parzen核估计,提出了一种新的独立成分分析(ICA)算法,实现了对源信号分布的全“盲”要求。该算法由观测信号样本出发,实现了对分离信号评价函数的直接估计,从而在一定程度上解决了ICA算法中如何选取估计信号评价函数的难题且能对任意的源混合信号(包括:超高斯与亚高斯分布,对称与非对称分布)进行有效盲分离。模拟实验从统计性质和计算时间2个方面说明了所提算法的性能。A novel independent component analysis（1CA） algorithm based on nonparametric density estimation——Parzen kernel estimate was proposed, which was truly blind to the source signals. The nonparametric density estimation was directly evaluated from the original data samples. It .solved an important problem in ICA： how to choose nonlinear functions as the probability density function estimation of the sources. The proposed 1CA algorithm was able to .separate a wide range of source signals, including sub- and super-Gaussian .sources, symmetric and asymmetric .sources. Simulations showed the effectiveness of the proposed algorithm along two sets of criteria： statistical and computational.

关 键 词：独立成分分析 评价函数 自然梯度 Parzen核估计 

分 类 号：TN911.72[电子电信—通信与信息系统]