摘要: |
当被识别系统是稀疏系统时,传统的遗漏最小均方(LLMS)自适应算法收敛性能较差,特别在非高斯噪声环境中,该算法性能进一步恶化甚至算法不平稳收敛。为了解决因信道的稀疏性使算法收敛变慢的问题,对LLMS算法的代价函数分别利用加权1-norm和加权零吸引两种稀疏惩罚项进行改进;为了优化算法的抗冲激干扰的性能,利用符号函数对已改进的算法迭代式作进一步改进。同时,将提出的两个算法运用于非高斯噪声环境下的稀疏系统识别,仿真结果显示提出的算法性能优于现存的同类稀疏算法。 |
关键词: 稀疏系统识别 自适应算法 冲激干扰 收敛性 |
DOI: |
|
基金项目: |
|
Sparse penalty constraint leaky least mean square algorithms against impulsive interference. |
YAN Guojie,LIN Yun |
() |
Abstract: |
The leaky least mean square(LLMS) adaptive filtering algorithm converges slowly when the identified system is sparse. Especially when the noise is non-Gaussian impulsive interference,the performance of LLMS algorithm deteriorates severely.To solve the problem that the convergent rate becomes slower because the system is sparse,the cost function of the conventional LLMS algorithm is improved by the two penalty functions,the reweighted zero-attracting(the log-sum penalty) and reweighted 1-norm(RL1),respectively. To address the problem of the impulsive interference,the iterative functions are improved by introducing the sign function. Simultaneously,the simulations are made for the proposed algorithms to prove to be better performances compared with existing leaky-style algorithms in the case of impulsive interference. |
Key words: sparse system identification adaptive filtering algorithm impulsive interference convergence |