一种新的变步长NLMS自适应算法

A Novel Variable Step Size NLMS Algorithm

文章根据最优步长理论提出一个向量因子来控制步长的变化,其长度及形式与权向量相同,并且由输入数据相关性大的部分决定两个向量中每次应更新的部分,从而使权向量异步更新达到最佳。计算机仿真结果表明,在输入数据相关性很强时,该算法与传统的变步长LMS及NLMS算法相比不仅具有更快的收敛速度,更小的稳态误差及更平稳的收敛过程,同时计算量也有一定的减少。

A vector factor is proposed to control the change of the step size according to optimal step size theory, with its length and form equal to the weight vector. The update part of the two vectors is determined by the larger part of the input correlation, and then the asynchronous update of the weight vector, finally becomes optimal. Computer simulation results shows that the new algorithm has higher speed of convergence, lower steady state error and more stable converge process than other classic variable step size LMS and NLMS algorithms when the correlation of input data is strong. Furthermore, the computing complexity is also cut down.