作　　者： ; ;

机构地区： 国防科学技术大学机电工程与自动化学院

出　　处： 《计算机工程与应用》 2005年第4期23-25,174,共4页

摘　　要： 源信号之间统计独立是经典独立分量分析模型的基本要求。对实际信号而言,严格的统计独立是很难满足的,统计独立通常解释为尽可能的独立或者物理独立。在探讨了源信号之间存在弱线性相关后,对源信号的构成依次做出了三种假设,分析了独立分量分析对相关信号源的辨识能力。理论研究和实验表明,即使信号源之间存在弱相关性,独立分量分析方法仍然反映信号源的波形特征。 Statistical independence is an important and basic restriction in the classical model of Independent Component Analysis(ICA).However,the assumption cannot be severely satisfied in real environment and practically the requirement often can only be interpreted as independent as possible or physical independent.This paper addresses the blind identification ability of ICA when sources are not independent but somewhat correlated.Three assumptions are made in series and the corresponding analyses are given on the correlation of independent components and correlated sources.The results show the idealistic blind identification ability of ICA,and the robust restoration ability of wave-performing.The performance is also demonstrated in the experiments.

关 键 词： 独立分量分析 盲辨识 盲源分离

领　　域： [自动化与计算机技术] [自动化与计算机技术]