基于自适应核估计的ICA算法

ICA Algorithm Based on Adaptive Kernel Estimation

在自然梯度算法的框架下,本文利用随机变量概率密度函数非参数估计的自适应核函数法,给出了一种能够对任意混合信号(超高斯和亚高斯信号,对称和非对称分布信号)进行盲分离的算法。本算法无需选择非线性函数,而是根据信号的统计特性自动地直接估计评价函数。通过仿真实验,验证了本文算法的有效性。

Under the frame of natural gradient algorithm, an ICA algorithm based on adaptive kernel estimation is proposed, which can separate arbitrary mixed signals （such as super-Gaussian and sub-Gaussian, symmetric and asymmetric signals）. This algorithm does not need to choose non-linear function artificially, but automatically estimates score function according to the statistic identity of signals. Simulations show that this algorithm can successfully separate mixed signals.