Improved Blind Source Separation Based on Non-Holonomic Natural Gradient Algorithm with Variable Step Size

摘要

The traditional natural gradient algorithm works badly when the source signal amplitude changes rapidly or becomes zero at a certain time. In addition, it cannot resolve very well the contradiction between the convergence speed and the error in steady state because the step-size is fixed. In order to solve the above problems, this paper proposes an improved blind source separation algorithm based on non-holonomic natural gradient by choosing an adaptive step-size and a suitable nonlinear activation function. Simulation result demonstrates that the new algorithm performance is superior to the traditional natural gradient algorithm.