文档介绍：非线性ICA
非线性独立分量分析
著名“鸡尾酒会”问题——将若干传声器放置在同一房间的不同位置,每一个传声器接收的是房间内不同人的声音的混合声音,我们对房问内人的声音特征及所处位置一无所知,目的是根据多个传声器观测的声音信号,分离出每一个人的声音。换句话说,不知道源信号和源信号传播通道的参数,在假设源信号之间相互统计独立,仅凭借传感器阵或换能器阵的观测数据,分离或估计最初源信号的波形以及传播通道参数(或称为混合矩阵),这就是盲信号处理。
独立分量分析的含义是把信号分解成若干个互相独立的成分, 它是为了解决盲信号分离的问题而发展起来的。如果信号本来就是由若干独立信源混合而成的,ICA就能恰好把这些信源分解开来。故在一般的文献中通常把ICA等同于BSS 。
盲信号处理(BlindSignal Processing,BSP) 它是指在不知道系统传递函数和源信号的情况下,只利用传感器接收信号,来求解系统传递函数或恢复源信号的方法。“盲”的含义是
1)所有的源信号都是不可观测的;2)没有关于混合信道的信息,即系统的传递函数不可知。盲信号处理的基本思想是利用所有源信号之间,且源信号和噪声之间都是相互统计独立的假设来弥补信道等先验信息的缺乏.
Nonlinear ponent analysis The famous problem is the
" cocktail party problem " -- some microphones placed in the same room, each microphone receives is a mixture of different voice room sound , we asked the sound characteristics of human and the location to be ignorant of the real, objective is to a microphone observations of the sound signal, separating each one's voice. In other words, do not know the parameters of the source signal and the source signal transmission channel, between the assumption of statistical independence between source signals, only by virtue of observation data of sensor array or transducer array, separation or estimated waveform originally signal munication channel parameters ( or mixed matrix ), this is the blind signal processing
ponent analysis ponent analysis is to pose the signal into several
ponents, it is in or