摘要
最小均方(LMS)自适应滤波算法易于实现,在很多领域得到了广泛地应用,但是存在加快算法收敛和减小稳态误差之间的矛盾,而固定步长LMS算法无法解决矛盾。用反正切函数atan建立了步长因子与误差之间一种新的非线性函数关系,给出了一种新的变步长LMS算法。反正切函数较Sigmoid函数简单且易于控制,并且可以使步长在误差接近为零时变化缓慢,从而可以使算法具有更小的稳态误差。还分析了参数、、对算法性能的影响。计算机仿真结果与理论分析一致,算法的性能优于固定步长LMS算法和SVSLMS算法。
LMS algorithm can be easily realized and applied in many fields,but there is an inconsistency between the fast algorithm convergence and the low steady state error,fixed step size LMS algorithm can not resolve this problem.A novel variable step size LMS algorithm which established a new non-linear functional relationship between μ（n） and e（n）with arc tangent function was proposed in this paper.Arc tangent function is simpler and can be easier controlled than Sigmoid function,and the function can make the step size change more slightly when the error is near to 0,thereby the algorithm can have a lower steady state error.In this paper,how the parametersα、β、γ affected the performance of the novel algorithm was discussed too.Computer simulation results confirmed the theoretical analysis and showed that the algorithm performed better than fixed step size LMS algorithm and SVLMS algorithm.
出处
《计算机仿真》
CSCD
2008年第9期93-95,154,共4页
Computer Simulation
关键词
最小均方算法
自适应滤波
变步长
算法
LMS algorithm
Adaptive filtering
Variable step size
Algorithm
