用于稀疏系统辨识的改进惩罚LMS算法研究  

The Improvement of LMS Algorithm for Sparse System Identification

作　　者：万涛[1] 刘遵雄[1] 王树成[1] 

机构地区：[1]华东交通大学信息工程学院,江西南昌330013

出　　处：《华东交通大学学报》2013年第6期62-66,共5页Journal of East China Jiaotong University

基　　金：国家自然科学基金项目(61065003);教育部人文社会科学研究规划基金项目(11YJCZH160)

摘　　要：基于加权零吸引因子最小均方算法(RZA-LMS),提出了一种应用于系统辨识的新型自适应滤波算法(ARZA-LMS)。RZA-LMS通过在标准LMS算法迭代过程中添加零吸引因子,促进了滤波器小权系数的收敛,从而在辨识稀疏系统时,加快了算法的整体收敛速度。但是RZA-LMS算法中的零吸引因子,选择了固定的e,过于武断,降低了算法的鲁棒性。通过在参数e与误差信号e之间建立非线性关系,使零吸引因子在最小化MSE更具有灵活性,提出了一种改进的RZA-LMS,提高了对系统辨识的收敛速度和稳定性。最后,计算机仿真验证了新算法的性能明显优于原算法和若干现有稀疏系统辨识的方法。Based on RZA-LMS,a novel adaptive algorithm is presented for sparse system identification. The RZA-LMS algorithm generates a zero attractor in the LMS iteration due to the penalty item on coefficients,and the zero attractor promotes sparsity in taps during the filtering process,therefore convergence can be acceler-ated when identifying sparse systems. For the parameter e of the e-law compression in the zero attractor is con-stant,the algorithm is not robust. The proposed approach adaptively establishes nonlinear relationship between the parameter e and the error signal e,which makes the algorithm more flexible in an attempt to minimize the MSE. Simulation results demonstrate the advantages of the proposed filter in both convergence rate and steady-state behaviors under sparsity assumptions on the true coefficient vector.

关 键 词：自适应滤波器 最小均方算法 压缩传感 稀疏信道 零吸引因子 L1范数 

分 类 号：TN911.7[电子电信—通信与信息系统]