Independent Component Analysis Based on Parametric Parzen Window Estimation
Cheng Xiao-gang · Signal Processing · 2009
When ICA is used for analyzing the blind sources,the probability density function(PDF)is always needed.However, it's impossible to prior know the pdf of the sources.In general,the non-parametric Parzen window is adopted for estimating the pdf.In the present paper,we present a strategy of variate window width of Parzen window for the pdf estimation.Based on it,the minimal criterion of the mutual information entropy is designed for ICA.The computer simulation demonstrates that the signal-to-interface ration is improved.