摘要
为显著提高对稀疏系统的辨识性能,提出了一种自适应算法。该算法将与稀疏性有重要关系的l1范数引入LMS算法的代价函数中,并导出新的滤波器权系数更新公式。该公式在迭代过程中向权系数不断添加一个指向零矢量的修正量,使得在稀疏系统中占主要地位的零系数加速收敛,从而显著提高自适应算法的收敛速度和跟踪速度。理论分析并推导了算法的均值收敛过程。仿真结果表明:该算法无论对一般稀疏系统还是分簇稀疏系统,都能明显改善收敛性能,并且表现出良好的稳健性和通用性。
This paper presents an adaptive algorithm using the unknown system sparsity to improve the system identification performance.A gradient descent recursion of the filter coefficient vector was deduced through introducing l1 norm,which has vital relationship with sparsity,to the cost function of LMS algorithm.Adding a zero-vector-approaching correction to the tap coefficient of the filter enables faster convergence of zero coefficients（which play a main role in sparse systems）,thus notably increases the convergence speed and tracking speed in system identification.Simulation results show that the algorithm evidently improves the convergence performance of both general sparse systems and clustering sparse systems,and exhibits universal applicability and excellent robustness.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2010年第10期1656-1659,共4页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金资助项目(60872087)
