一种变步长LMS自适应滤波算法研究

为了解决最小均方(LMS)算法的稳定性以及收敛速度和稳态误差之间的矛盾,提出了一种复数型自适应变步长的LMS算法,它的权系数的调整取决于误差曲面在新权值点上的梯度。分析了新算法的收敛特性以及参数选择对算法性能的影响。算法具有较快的收敛速度、鲁棒稳定性且运算小易于实现的特点。计算机仿真的结果与理论分析相一致,证实了算法的收敛性能优于标准的LMS算法并且具有较好的实用性。

Study of a Variable Step Size LMS Adaptive Filtering Algorithm

To solve the problem of stability and the tradeoff between rate of convergence (speed of adaptation) and steady-state excess MSE (quality of adaptation, or accuracy of the adaptive filter) for standard LMS algorithm, this paper presents an adaptive variable step size LMS algorithm (least mean square algorithm) of complex form, where the innovation of the weight-vector depends on the gradient of error surface at the point of new weight. Then the convergence properties and the effects of parameters choice for the algorithm are analyzed. The adaptive variable step size LMS algorithm has fast convergence, robust stability, less computational complexity and easy to implement. Computer simulations confirm the theoretical analysis and show the algorithm performance is practical and superior to the usual LMS algorithm.