摘要
针对变步长最小均方(LMS)自适应滤波算法的步长因子在算法收敛程度加深时快速变小的问题,提出在步长因子迭代过程中引入历史误差平方的遗忘加权和补偿项,并加入滑动窗,提高算法的收敛速度,同时减小了稳态失调。利用互功率谱相位(CSP)方法对信号进行时延估计,采用对互功率谱函数进行多帧加权平滑的方式来提高算法的抗噪能力。实验仿真结果表明,改进后的算法在信噪比较低的条件下,也能获取到有效的时延估计,证明新方法具有很好的鲁棒性,是有效的。
According to the issue that step factor of variable step length least mean square（ LMS） adaptive filtering algorithm reduces quickly when the algorithm convergence degree deepen,this paper introduced the sum of forgetting weighted for historical error square and sliding window to improve the convergence rate of algorithm and reduce the steady-state misadjustment in the iterative process of step factor. It used the cross-power spectrum phase（ CSP） method for signal time delay estimation,adopted the method of multi-frame weighted smoothing for cross-power spectrum function to improve algorithm＇s anti-noise ability. Simulation results show that the improved algorithm can also get the effective time delay estimation at low SNR conditions,proved that the new method has good robustness and is effective.
出处
《计算机应用研究》
CSCD
北大核心
2015年第10期2940-2942,2946,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(51274118)
关键词
声信号
变步长LMS算法
遗忘加权
互功率谱相位
时延估计
acoustic signal
variable step size LMS algorithm
forgetting weighted
cross-power spectrum phase
time delay estimation
