均方偏差分析的多态可变步长LMS算法

Multi-state variable step size LMS algorithm basedon analysis of mean square deviation

最小均方(least mean square,LMS)算法在时变信道的最小稳态均方偏差(mean square deviation,MSD)由输入功率、噪声功率、随机扰动信号功率以及滤波器长度共同决定。为达到系统中最小的MSD值,传统的LMS算法存在有迭代次数较多和收敛速度慢等问题,提出了一种多态可变步长最小均方(multi-state variable step size least mean square,MVSS-LMS)算法。该算法通过添加暂态递减步长作为过渡,实现以更快的收敛速度达到系统中最小的MSD值。理论分析与仿真结果表明,与目前最新的Prob-LMS算法相比,所提算法在时变信道以及突变信道都具有更快的收敛速度和更低的MSD值,且算法的复杂度更低。

The minimum steady state mean square deviation(MSD)of the least mean square(LMS)algorithm in the time varying channel is determined by input power,noise power,random disturbance signal power and filter length.The traditional LMS algorithm adopts large iterations to achieve the system’s minimum MSD value so that its convergence speed is slow,so we propose a Multistate Variable Step Size Least Mean Square(MVSSLMS)algorithm.By adding the transient decreasing step size as the transition,the system can achieve the minimum MSD value with a faster convergence speed.According to theoretical analysis and simulation results,the proposed algorithm has a faster convergence speed and lower MSD value in time-varying channels and abrupt channels comparing with the latest Prob-LMS algorithm.The complexity of MVSS-LMS algorithm is also comparatively lower than that of Prob-LMS.