Multi-user pdf estimation based criteria for adaptive blind separation of discrete sources

This paper deals with criteria for adaptive blind separation of discrete sources. The criteria are based on the estimation of the probability density function (pdf) of the recovered signal using a parametric model and the divergence of Kullback-Leibler to measure the similarities between the involved signals. Two strategies that guarantee the recovering of all sources are employed: the first one introduces a penalty when the sources are correlated and the second one constrains the filtering to an orthogonal global system response. Simulations are carried out to evaluate the performance of the criteria compared with existing blind methods in typical multi-user environments such as spatial and space-time processing.