摘要
该文针对SVSLMS算法步长函数在误差e(n)接近零处不具有缓慢变化的缺点和MLMS算法由于采用固定步长使得在稳态阶段权值更新到期望值速度过慢的不足进行了讨论。通过更新SVSLMS算法步长函数和在权值调整式中增加动量项,该文提出了一种改进算法—SVS-MLMS算法。该算法具有步长函数在误差e(n)接近零处能够缓慢变化的优点,使得在自适应稳态阶段的步长稳定在最优值,进而使权值收敛到最佳。仿真结果证明该算法在学习曲线收敛速度加快和稳态误差减小方面取得了较好的效果。该文还讨论了算法中三个参数a,b,r的取值对算法收敛性能的影响,确定了它们的最优值。
This paper discusses that the step size function of the SVSLMS algorithm has a shortcoming of slight change for e（n） close to zero and the MLMS algorithm has a defect of updating weights too slow to expectation values because of adopting invariable step size. An improved algorithm called SVS - MLMS is presented by updating the step size function of the e（n） algorithm and adding a momentum to the weight adjustment expression. There is an advatage in the improved algorithm that the step size function changes slightly for e（n） close to zero, which results in that the step size stabilizes at an optimum value during the stage of adaptive stationary state and the weight converges optimally. Simulation results prove that this improved algorithm achieves an effect in accelerating the converging rate and reducing the stationary error. In addition, this algorithm discusses how the parameters a, b, r affect the convergence performance and determines their optimum values.
出处
《计算机仿真》
CSCD
2006年第5期74-76,97,共4页
Computer Simulation
关键词
动量项
变步长
自适应滤波
Momentum term
Variable step size
Adaptive filtering
