基于NPCA的盲源分离算法

Blind Source Separation with Nonlinear PCA

主要讨论了基于非线性主分量分析(NPCA)的盲源分离,从理论与实验2个方面详细分析了算法的特性与效果。针对算法中的非线性函数选择的问题,采用了在线统计的方法,即根据不同的输入信号选择不同的非线性函数。从实验结果可以看出,该方法不仅可以很好地解决源信号为亚高斯信号混合的盲源分离问题,而且对源信号为亚高斯和超高斯信号混合的盲源分离问题也取得了很好的效果。

This paper discusses Blind Source Separation （BSS） with Nonlinear PCA both theoretically and experimentally. The main problem of Nonlinear PCA is how to choose proper nonlinear function. A novel approach called on-line selection is proposed.That is, different nonlinear functions are selected according to different inputs. The experiments show the effectiveness of this approach not only with sub-sub Gaussian sources but also with sub-super Gaussian sources, which is more difficult to perform.