Distributed bias-compensated normalized least-mean squares algorithms with noisy input

作     者：Lu FAN Lijuan JIA Ran TAO Yue WANG 

作者机构：School of Information and Electronics Beijing Institute of Technology 

出 版 物：《Science China(Information Sciences)》 (中国科学:信息科学(英文版))

年 卷 期：2018年第61卷第11期

页      面：193-207页

核心收录：

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 081104[工学-模式识别与智能系统] 08[工学] 0835[工学-软件工程] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：supported by National Natural Science Foundation of China (Grant No. 61421001) 

主　　题：distributed parameter estimation normalized least-mean squares bias-compensated algorithms 

摘      要：In this paper, we study the problem of distributed normalized least-mean squares(NLMS) estimation over multi-agent networks, where all nodes collaborate to estimate a common parameter of *** consider the situations that all nodes in the network are corrupted by both input and output noise. This yields into biased estimates by the distributed NLMS algorithms. In our analysis, we take all the noise into consideration and prove that the bias is dependent on the input noise variance. Therefore, we propose a bias compensation method to remove the noise-induced bias from the estimated results. In our development, we first assume that the variances of the input noise are known a priori and develop a series of distributed-based bias-compensated NLMS(BCNLMS) methods. Under various practical scenarios, the input noise variance is usually unknown a priori, therefore it is necessary to first estimate for its value before bias removal. Thus, we develop a real-time estimation method for the input noise variance, which overcomes the unknown property of this noise. Moreover, we perform some main analysis results of the proposed distributed BCNLMS algorithms. Furthermore, we illustrate the performance of the proposed distributed bias compensation method via graphical simulation results.