Research and Simulation of EASI Blind Source Separation Algorithm Based on Gradient Method

摘要

This paper mainly analyzes the blind source separation (BSS) model, studies some BSS algorithm of randomized gradient algorithmatural gradient algorithm, and EASI algorithm, etc. EASI algorithm insteads the relatively gradient of the randomized gradient and together the albino algorithm with the blind separation algorithm, which avoiding the complex inverse matrix calculation process. Through the simulation of three mixed signal based on EASI algorithm, the algorithm can realize convergence with the step of 0.004 in iterative 500 times or so, the biggest string sound error under steady state is less than 1. EASI algorithm can realize convergence with the step of 0.002 in iterative 1000 times or so, the biggest string sound error under steady state is less than 0.5. This result shows that the gradient algorithm is very sensitive to the step length selection, tiny difference will bring different separation results.