Multi-input single-output neural network blind separation algorithm based on penalty function

A real-time linear mixed signals blind separation algorithm based on multi-input and single-output neural network is proposed in this paper. The algorithm embodies an idea of blind separation by extracting source signals one by one. Firstly mixed signals are whitened so that mixture matrix is changed into an orthogonal matrix, then the method of courant penalty function is adopted to solve the problem of blind separation restricted by an equation. The paper also easily discusses how to choose the nonlinear function. Simulation results show the algorithm has both of a great ability to separate and a good separation effect.