Abstract: The non-parametric density estimation—generalized k-nearest neighbor(GKNN) estimation based novel independent component analysis(ICA) algorithm which is fully blind to the sources is proposed using a linear ICA neural network. The proposed GKNN density estimation is directly evaluated from the original data samples, so it solves the important problem in ICA and blind source separation(BSS): how to choose nonlinear functions as the probability density function(PDF) estimation of the sources. Moreover, the GKNN-ICA algorithm can separate the hybrid mixtures of source signals which include Gaussian, super-Gaussian, sub-Gaussian, and symmetric distribution ones using only a flexible model and it is completely blind to the sources. The algorithm presented in this paper provides the way for wider applications of ICA methods to real world signal processing. Simulations confirm the effectiveness of the proposed algorithm.

Key words: blind source separation, independent component analysis, nonparametric estimation, generalized k-nearest neighbor