一种改进的变步长NLMS算法  

An improved variable-step NLMS algorithm

作　　者：刘志骋 杨博 

机构地区：[1]武警工程大学信息工程系,陕西西安710086

出　　处：《现代电子技术》2015年第22期12-13,16,共3页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(60940007)

摘　　要：针对传统NLMS使用固定步长而出现的收敛速度和稳态误差的矛盾,提出一种改进的变步长NLMS算法。该算法建立了步长与误差的函数关系,使步长随着输出误差和噪声误差的变化而动态更新,从而降低稳态误差。理论分析和仿真结果表明,与现有NLMS算法相比,改进的算法具有更快的收敛速度和更低的稳态误差。Aiming at the contradiction between convergence speed and steady state error of the traditional NLMS, an improved variable-step NLMS algorithm is proposed, in which the functional relationship between step-length and error is established, which makes dynamic update of step length changed with variation of output error and noise error to reduce steady-state error. The theoretical analysis and simulation results show that, compared with the existing NLMS algorithm, the improved algorithm has faster convergence speed and lower steady-state error.

关 键 词：自适应滤波 归一化最小均方误差算法 变步长 稳态误差 

分 类 号：TN710-34[电子电信—电路与系统] TN911.7