关键词: 盲信号分离, 自适应, 学习率, 峭度

Abstract: Because of quick convergence rate and good separation performance，natural gradient algorithm occupies importance position in blind source separation.Natural gradient algorithm adopts fix-step，so they cannot resolve the contradiction between convergence speed and the error in the steady state.By building a nonlinear function relationship between the step size factor and the square sum of the kurtosis，the paper proposes an adaptive natural gradient algorithm.Computer simulation result confirms the algorithm’s validity，and shows that the algorithm’s performance is superior to natural gradient algorithm.

Key words: blind source separation, adaptive, learning rate, kurtosis