一种改进的变步长LMS算法

An Improved Variable Step Size LMS Algorithm

LMS算法由于简单而获得了广泛的应用,大量的深入研究不断地改善了它的性能。LMS算法存在收敛速度和稳态失调之间的固有冲突,变步长因子可以获得二者之间的有效平衡。对已有的一些变步长LMS自适应滤波算法进行了分析,在此基础上提出一种改进的变步长LMS算法,步长因子同时考虑了指数为预测误差的一次和二次幂的2项。算法在保持较快收敛速度的同时,获得更优的稳态预测误差。对比仿真实验证明了算法的优越性。

The LMS （Least Mean Square）algorithm has many applications due to its simpleness, and a mass of intensive study on it has greatly improved its performance. There is an inherent conflict between the convergence rate and steady-state misadjustment, which can be overcome by means of a variable size factor. Some variable step size LMS algorithms in literature are analyzed, based on which an improved one is presented. Its step size factor consists of two terms whose exponents are the first and second power of the predictive error. The algorithm can obtain better steady state predictive error while keeping much quicker convergence speed. Comparison and simulation experiments verify its superiority.