基于拟牛顿优化方法的自适应FIR滤波算法

An Algorithm for FIR Adaptive Filters Based on Quasi-Newton Optimization

基于拟牛顿优化方法,提出了一种稳健的自适应FIR滤波算法。新算法用最小二乘误差(LSE)代替了均方误差(MSE)作为代价函数,它具有和常规递归最小二乘(CRLS)算法相近似的追踪能力,且不存在数值计算不稳定性的问题。在收敛速度以及稳态效果方面也要优于DeCampos的拟牛顿(QN)算法。由计算机仿真比较了有关算法的性能。

A robust algorithm for FIR adaptive filters based on the quasi-Newton class of optimization algorithm is described. In particular, a least squares error （LSE） based cost function is minimized instead of the commonly used mean square error （MSE）. The proposed algorithm has almost as good a tracking ability as that of the conventional recursive least squares （CRLS） algorithm and is numerically stable. Its convergence speed and performance during the stable period are also superior to those of De Campos＇s quasi-Newton （QN） algorithm. The performance of the algorithms is evaluated via simulations.