一种基于拟牛顿法的递归最小二乘方法

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De Campos的拟牛顿(QN)算法。通过计算机仿真比较了有关算法的性能。

A Recursive Least Squares Algorithm Based on Quasi-Newton Optimization

A robust algorithm for FIR adaptive filters based on the quasi-Newton class of optimization al- gorithm is described.A least squares error(LSE)based cost function is minimized instead of the commonly used mean square error(MSE).The proposed algorithm has a tracking ability comparable to that of convention- al 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 per- formance of the algorithms is evaluated via simulations.