A FOCUSS-Based Algorithm for Nonlinear Overcomplete Independent Component Analysis

A new neural network approach is proposed to solve the blind signal separation problem under both nonlinear and overcomplete conditions. To our knowledge, most previous algorithms including FOCal Underdetermined System Solver (FOCUSS) focus on linear distortion which may not accord with practical applications. We base our work on the FOCUSS algorithm but develop it further to apply for nonlinear situations. The corresponding parameter learning algorithm for the proposed neural network is also presented through formal derivation. Simulation has been carried out to show that the proposed method can separate the independent sparse signals successfully under nonlinear and overcomplete conditions. Key-Words: FOCUSS, Nonlinear Independent Component Analysis (ICA), Neural Network, Blind Signal Separation, Bayesian Framework.