新的变步长LMS算法在系统辨识中的应用

在分析最小均方误差(LMS)自适应滤波算法和变步长LMS算法的基础上,提出了一种新的变步长算法,该算法用误差的平均值来控制步长的变化,进一步的解决了收敛速度和稳态误差的矛盾。讲述了新算法的具体改进方式,并将该算法和变步长G-SVSLMS算法以及固定步长算法分别应用到系统辨识中,通过MATLAB进行仿真,结果证实文中提出的算法在明显提高收敛速度的同时,并拥有好的稳态误差。

Application of Novel Variable Step Size LMS Algorithm in System Identification

Based on analysis of LMS (Least Mean Square Error) adaptive filtering algorithm and variable step size LMS algorithm, a novel variable step algorithm is proposed. The error mean of this algorithm is used to control the step variation, and further to mitigate the contradiction of between convergence rate and steady-state error. The specific improved method to modify the novel algorithm is described, and the proposed algorithm, variable step size G-SVSLMS algorithms and fixed-step algorithm are respectively applied to the system identification. Simulation with MATLAB indicates that this algorithm could significantly improve convergence rate while having a fair steady-state error.