关键词: 最小均方算法, 变步长, 自适应滤波, 系统辨识

Abstract: By building a nonlinear function relationship between μ and the error signal e（n），this paper proposes an improved variable step size LMS（Least Mean Square） adaptive filtering algorithm.This variable step size algorithm avoids the shortcoming of changing step size of SVSLMS（Variable Step Size LMS based on Sigmoid Function）.Also in the stage of adaptive steady state，it has the virtue of e（n） slightly changing at point close to zero.Meanwhile the algorithm efficiently overcomes the discrepancy between the convergence rate and the steady error.When applied to system identification，this algorithm constitutes a significant improvement in the identification speed with very small steady error in stationary environment and is of better tracking capability，as compared with the traditional algorithms with step size.

Key words: LMS algorithm, variable step size, adaptive filtration, system identification