Theory Study of a Novel ICA Algorithm Based on Wavelet Density Estimation
Cheng Wu, Hongwei Li, Guoqing Wang · 2006
The purpose of this paper is to analyze the feasibility of applying wavelet density estimation to ICA algorithm in theory. We propose a novel ICA algorithm based on wavelet density estimation in this paper. The density estimation of the separated signals is directly evaluated by truncated wavelet to approximate the arbitrary density function of sources so that nonlinear function can be adaptively estimated in the proposed algorithm. Hence, this algorithm is able to separate arbitrary distribution source signals in theory. In this paper, we analyze the principles and effectiveness of the proposed algorithm in theory and discusses the local stability of the algorithm