一种基于矢量加速的变步长频域最小均方算法  

A Variable Step-size Frequency-domain Least Mean Square Algorithm Based on Vector Acceleration

作　　者：易清明[1] 曾杰麟 石敏[1] 

机构地区：[1]暨南大学信息科学技术学院,广州510632

出　　处：《计算机工程》2015年第7期285-288,293,共5页Computer Engineering

基　　金：广东省工程技术研究中心基金资助项目(2012gczx A003)

摘　　要：针对现有变步长频域批处理最小均方(VSFBLMS)算法收敛速度慢的问题,提出一种基于矢量加速的VSFBLM S算法。利用VSFBLM S计算得到的基本步长参数对当前收敛阶段进行判断,并根据收敛阶段分别在前期和后期选择较大和较小迭代次数的权值更新公式进行系数更新,从而加快算法前期的收敛速度并保证后期失调量较小。采用基于自适应滤波器噪声抵消模型进行算法性能测试,结果表明,相比VSFBLMS算法,该算法的收敛速度有较大提高,且在后期具有与VSFBLMS算法趋于一致的失调量。To enhance the convergence speed of existed Variable Step-size Frequency-domain Block Least Mean Square （VSFBLMS） algorithm, an Accelerating Vector based Variable Step-size Frequency-domain Block Least Mean Square （AV-VSFBLMS） algorithm is proposed. The algorithm judges the current convergence stage of basic step parameters obtained by using the algorithm of VSFBLMS computing, selects a larger iteration number of coefficient updating formulas in early process stage while a smaller iteration number of coefficient updating formulas in later process stage according to the convergence statue. It helps to improve the convergence rate in early stage and also ensures a low misalignment in later stage. Experimental result based on adaptive noise cancelling model shows that the proposed algorithm outperforms other VSFBLMS algorithms in convergence speed with a lower misalignment.

关 键 词：自适应滤波 矢量加速 变步长 频域批处理最小均方算法 重叠保留法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]