Blind source separation using clustering-based multivariate density estimation algorithm

Zhenya He, Luxi Yang, Ju Liu, Ziyi Lu, Chen He, Yuhui Shi

Abstract: A learning algorithm is developed for blind separation of the independent source signals from their linear mixtures. The algorithm is based on minimizing a contrast function defined in terms of the Kullback-Leibler distance. We use a clustering-based multivariate density estimation approach to reduce the number of the parameters to be updated. Simulations illustrate the validity of the algorithm.