Nonlinear single channel source separation

A new model of nonlinear single channel source separation is proposed in this paper. The proposed model is a linear mixture of the independent sources followed by an element-wise post-nonlinear distortion function. In addition, the paper develops a novel solution that efficiently compensates for the nonlinear distortion and performs source separation. The proposed solution is a two-stage process that consists of a Gaussianization transform and a maximum likelihood estimator for the sources. The paper also discusses the theory behind the proposed solution. Simulations have been carried out to verify the theory and evaluate the performance of the proposed algorithm. Results obtained have shown the effectiveness of the algorithm even in presence of the strong nonlinearity.