基于噪音拟合的优化变步长滤波最小均方算法
2021年电子技术应用第11期
钱 拴1,2,高健珍1,2,代永平1,2
1.南开大学 光电子薄膜器件与技术研究所,天津300350; 2.天津市光电子薄膜器件与技术重点实验室,天津300350
摘要: 为了更快地实现主动降噪,设计了噪音多项式拟合模型,提出了改进的变步长滤波最小均方算法(Improved Filtered-x Least Mean Square,IFxLMS)。该算法在统计噪音信号的同时,对噪音信号进行拟合与预测,随后结合误差信号与预测信号对步长进行调节,达到快速调节的目的。为了验证该算法的性能,将该算法与传统变步长滤波最小均方算法对比试验,仿真结果显示,在相同噪音条件下,新算法将噪音信号降到10 dB、20 dB、30 dB、35 dB等信噪比时,所需的迭代次数减少了4次~60次不等,在同时新算法的鲁棒性也优于普通的滤波变步长最小均方算法。
中图分类号: TN911.72
文献标识码: A
DOI:10.16157/j.issn.0258-7998.201091
中文引用格式: 钱拴,高健珍,代永平. 基于噪音拟合的优化变步长滤波最小均方算法[J].电子技术应用,2021,47(11):81-84,89.
英文引用格式: Qian Shuan,Gao Jianzhen,Dai Yongping. Optimal variable step filtered-x least mean square algorithm based on noise fitting[J]. Application of Electronic Technique,2021,47(11):81-84,89.
文献标识码: A
DOI:10.16157/j.issn.0258-7998.201091
中文引用格式: 钱拴,高健珍,代永平. 基于噪音拟合的优化变步长滤波最小均方算法[J].电子技术应用,2021,47(11):81-84,89.
英文引用格式: Qian Shuan,Gao Jianzhen,Dai Yongping. Optimal variable step filtered-x least mean square algorithm based on noise fitting[J]. Application of Electronic Technique,2021,47(11):81-84,89.
Optimal variable step filtered-x least mean square algorithm based on noise fitting
Qian Shuan1,2,Gao Jianzhen1,2,Dai Yongping1,2
1.Institute of Optoelectronic Thin Film Devices and Technology,Nankai University,Tianjin 300350,China; 2.Key Laboratory for Photoelectronic Thin Film Devices and Technology of Tianjing,Tianjin 300350,China
Abstract: In order to achieve active noise reduction faster, a noise polynomial fitting model is designed, and an improved variable step size filtering least mean square algorithm(improved filtered-x least mean square, IFxLMS) is proposed. The algorithm performs fitting and prediction to the noise signal while counting the noise signal, and then adjusts the step length by combining the error signal and the predicted signal to achieve the purpose of rapid adjustment. In order to verify the performance of the algorithm, the algorithm is compared with the traditional variable step filter-x least mean square algorithm. The simulation results show that under the same noise conditions, when the new algorithm reduces the noise signal to 10 dB, 20 dB, 30 dB, 35 dB, etc. The number of iterations required has been reduced from 4 to 60. At the same time, the robustness of the new algorithm is better than that of the ordinary variable step size filtered-x least mean square algorithm.
Key words : filtered-x least mean square algorithm;noise fitting;variable step size;active noise reduction
0 引言
随着城市化进程,环境的噪音问题日益突出[1],降噪的设备及相关算法逐渐成为了研究的热点问题[2],滤波最小均方算法(Filtered-x Least Mean Square,FxLMS)由于其计算量相对较小被大量应用于主动降噪设备[3]。最小均方算法的降噪步长决定了系统的降噪速度以及降噪精度,步长的迭代公式也决定了算法的运算量,进而影响设备降噪的速度[4]。FxLMS可用于主动降噪设备以降低设备局部噪音,包含的降噪场景有电梯[5]、高铁、汽车[6]、耳机[7]以及潜艇等方面,在社会应用中有极大应用价值。
算法迭代步长是FxLMS研究重要方向之一[8],较大的迭代步长可以使得FxLMS算法收敛速度较快,但是系统的稳态性不高;较小的迭代步长可以提供较稳态的结果,但是系统的迭代次数过多,收敛速度较慢。针对以上问题,文献[9]提出归一化泄露FxLMS算法,收敛步长受到误差信号的影响,同时也避免了因误差信号过小而导致的步长过大问题;马英博[5]改善变步长因子更新的方式是计算出误差信号与输入信号之间的相关性,再根据相关性更改步长的迭代;文献[10]使得步长以指数函数变化;文献[11]更改了步长因子的计算公式,使得算法在收敛初期步长小以实现算法的收敛,中期步长变大快速收敛,后期降低收敛因子提高收敛精度。
本文详细内容请下载:http://www.chinaaet.com/resource/share/2000003832。
作者信息:
钱 拴1,2,高健珍1,2,代永平1,2
(1.南开大学 光电子薄膜器件与技术研究所,天津300350;
2.天津市光电子薄膜器件与技术重点实验室,天津300350)
此内容为AET网站原创,未经授权禁止转载。