摘要
研究算法的优化问题,对比于传统的正则化最小均方算法(NLMS),成系数比例自适应算法(PNLMS)拥有较快的初始收敛速度,但是PNLMS并不是一种最优化的算法。改进了采用L0范数的IPNLMS算法以提高对稀疏系统进行辨识的性能。分析了近年来的几种系数比例算法的性能及其局限性,通过建立步长因子μ与误差信号e之间的非线性关系,提出了一种结合Sigmoid函数和L0范数的变步长系数成比例的规则化的LMS滤波算法。并对其与文中提到的算法进行了比较和分析,拥有更好的收敛性和稳态误差。
It is known that the proportionate normalized least mean square （PNLMS） algorithm achieves a better performance than traditional normalized least mean square （ NLMS ） algorithm, in terms of fast initial convergence rate. However, the PNLMS has been widely observed not to be optimal. In order to improve the performance of sparse system identification, the variable step L0 norm constraint IPNLMS algorithm was studied and improved in this paper. Firstly, the performance and limitation of the several proportionate LMS algorithm in recently were analysised. Sec- ondly, based on the theories of the adaptive-fiher step, the Sigmoid IPNLMS-L0 algorithm was proposed. Finally, the simulations demonstrate that the proposed algorithm significantly can achieve convergence and steady-state mis- alignment.
出处
《计算机仿真》
CSCD
北大核心
2012年第11期166-169,共4页
Computer Simulation
基金
国家自然科学基金项目(61065003)
江西省自然科学基金项目(2010GZS0034)
