摘要
为了降低非高斯噪声对系统性能的影响,核风险敏感损失函数(Kernel Risk-Sensitive Loss,KRSL)因其较高的凸性而被广泛应用为自适应滤波器的代价函数.基于此,为了提高非高斯情况下系统的滤波精度,本文采用广义高斯密度(Generalized Gaussian Density,GGD)函数作为KRSL的核函数,进而提出了一种广义核风险敏感损失函数(Generalized Kernel Risk-Sensitive Loss,GKRSL),并给出了GKRSL的重要性质.为了进一步识别稀疏系统,结合GKRSL的优点,采用递归更新方式提出了一种基于稀疏惩罚约束的广义递归核风险敏感损失(Generalized Recursive Kernel Risk-Sensitive Loss with Sparse Penalty Constraint,GRKRSL-SPC)算法.仿真结果表明,GRKRSL-SPC算法能够显著提高非高斯噪声下系统的滤波精度和鲁棒性.
The kernel risk-sensitive loss(KRSL)is widely used as the cost function of adaptive filters to reduce the influence of non-Gaussian noises on system performance owing to its high convexity.In this paper,a generalized kernel risk-sensitive loss(GKRSL)is proposed by using a generalized Gaussian density(GGD)function as the kernel of KRSL to improve the filtering accuracy of system in non-Gaussian noises.The important properties of GKRSL are presented for optimization.Furthermore,combined with the advantages of the GKRSL under the sparse penalty constrain,the recursive updating method is used to generate a novel generalized recursive kernel risk-sensitive loss with the sparse penalty constrain(GRKRSL-SPC)algorithm for identification of sparse systems.The superiorities of GRKRSL-SPC from the aspects of accuracy and robustness are verified by Monte Carlo simulations.
作者
王代丽
王世元
张涛
齐乐天
WANG Daili;WANG Shiyuan;ZHANG Tao;QI Letian(College of Electronic Information Engineering,Southwest University/Chongqing Key Laboratory of Nonlinear Circuits and Intelligent Information Processing,Chongqing 400715,China)
出处
《西南大学学报（自然科学版）》
CAS
CSCD
北大核心
2022年第4期196-205,共10页
Journal of Southwest University(Natural Science Edition)
基金
国家自然科学基金项目(62071391)
重庆市自然科学基金项目(cstc2020jcyj-msxmX0234)
中央高校基本科研业务费项目(2020jd001).
关键词
广义相关熵
核风险敏感损失函数
稀疏系统
辨识
自适应滤波
generalized correntropy
kernel risk-sensitive loss
sparse system
identification
adaptive filtering
