
非高斯噪声下基于Wilcoxon范数的变步长符号扩散式仿射投影算法
郭 莹 于和芳 赵 璐 李 飞 刘振宇
*
(沈阳工业大学信息科学与工程学院 沈阳 110870)
摘 要:扩散式仿射投影算法(DAPA)是实现分布式网络参数自适应估计的一种重要方法,该算法在输入信号存
在相关性时仍快速收敛,但抑制具有脉冲特性的非高斯噪声能力弱,且固定步长对收敛性有所限制。为此,该文
提出了基于Wilcoxon范数的变步长符号扩散式仿射投影算法(VSS-DWAPA)。首先,引入稳健估计理论中抗异常
值能力强的Wilcoxon范数作为代价函数并根据其取值特点进行了符号量化,推导出了新的迭代方程;其次,针对
固定步长的局限性,采用迭代方式实现了误差信号对步长的控制,在初始阶段和接近收敛阶段选择不同的步长,
使算法具有更好的适应性。仿真结果表明,在非高斯噪声下本文的VSS-DWAPA算法在收敛性、跟踪性等方面均
优于现有一些扩散式自适应滤波算法,同时在高斯噪声环境下也具有较好的性能。
关键词:自适应网络;非高斯噪声;Wilcoxon范数
中图分类号:TN911.7 文献标识码:A 文章编号:1009-5896(2021)02-0303-07
DOI: 10.11999/JEIT200371
Variable Step Size Sign Diffusion Affine Projection Algorithm
Based on Wilcoxon Norm under Non-Gaussian Noise
GUO Ying YU Hefang ZHAO Lu LI Fei LIU Zhenyu
(School of Information Science and Engineering, Shenyang University of Technology, Shenyang 110870, China)
Abstract: Diffusion Affine Projection Algorithm (DAPA) is an important method to realize the adaptive
estimation of distributed network parameters. The algorithm can converge rapidly even when the input signal
has correlation. The disadvantage of DAPA is that the ability to suppress non-Gaussian noise with impulsive
characteristics is weak, and the fixed step size limits the performance of the algorithm. In this paper, a Variable
Step size Sign Diffusion Wilcoxon Affine Projection Algorithm (VSS-DWAPA) is proposed. Firstly, the
Wilcoxon norm which has strong ability to resist outliers is introduced as the cost function, and sign
quantization is carried out according to its value characteristics, and then a new iterative equation is derived.
Secondly, considering the limitation of fixed step size, the control of error signal to step size is realized through
iterative method. That is, in the initial stage and the almost convergent stage, the step size is selected
differently, which effectively makes it have better adaptation. The simulation results show that the proposed
VSS-DWAPA is superior to some existing diffusion adaptive filtering algorithms in convergence, stability and
tracking. It can also work well in Gaussian noise environment.
Key words: Adaptive network; Non-Gaussian noise; Wilcoxon norm
1 引言
分布式自适应估计是一种多节点协作的信息处
理方式,即传感器网络中的各个节点均通过自适应
迭代方式参与计算,并按照特定的协作策略与邻居
节点进行信息交互,从而实现对感兴趣参数的有效
估计。在各种节点协作策略中扩散策略
[1]
更具灵活
性和适应性,适于实现大规模网络参数的自适应估
计。因而,在救灾管理、精确农业、电力系统建设
等众多领域
[2–4]
得到了广泛应用。
根据输入信号的特性不同,扩散式自适应估计
算法可分为白输入信号算法和有色输入信号算法。
最早提出的扩散式自适应估计算法
—
DLMS
(Diffusion Least Mean Square)及其改进算法
[5–7]
就
是在白输入信号的假设之下得到的,对于有色输入
收稿日期:2020-04-15;改回日期:2020-08-20;网络出版:2020-10-28
*通信作者: 刘振宇 liu_zhenyu0419@sina.com
基金项目:国家自然科学基金(61803272)
Foundation Item: The National Natural Science Foundation of
China (61803272)
第43卷第2期 电 子 与 信 息 学 报 Vol. 43No. 2
2021年2月 Journal of Electronics & Information Technology Feb. 2021