Source Recovery in Underdetermined Blind Source Separation Based on Artificial Neural Network

We propose a novel source recoveryalgorithm for underdetermined blind sourceseparation, which can result in better accuracyand lower computational cost. On the basisof the model of underdetermined blind sourceseparation, the artificial neural network withsingle-layer perceptron is introduced intothe proposed algorithm. Source signals areregarded as the weight vector of single-layerperceptron, and approximate gonorm is tak-en into account for output error decision ruleof the perceptron, which leads to the sparserecovery. Then the procedure of source re-covery is adjusting the weight vector of theperceptron. What＇s more, the optimal learningfactor is calculated and a descent sequence ofsmoothed parameter is used during iteration,which improves the performance and signifi-cantly decreases computational complexity ofthe proposed algorithm. The simulation resultsreveal that the algorithm proposed can recoverthe source signal with high precision, while itrequires lower computational cost.