Research On Distributed Sparse Parameter Estimation Method Based On The Multi-agent Network

With the rapid development of modern communications and sensor technology,people are increasingly demanding for high channel spectrum efficiency and information transfer rate,the problem of sparse channel estimation and distributed multi-agent network's collaborative estimation are particularly prominent.This paper studies the application of sparse identification in distributed multi-agent network,i.e.,based on the identification algorithm for sparse system,utilize distributed collaborative strategy to achieve synergy estimate of the sparse system parameters in the network.Sparse system parameter identification means adding sparsity constraint items in the cost function,apply zero attraction of different weights for zero coefficients and non-zero coefficients to improve the convergence rate.Distributed collaborative estimate refers to each agent in the distributed network architecture exchange network data sharing policy through various information and process these data synergistically,realize synergy estimates for the whole network.First introduced identification algorithm based on sparse LP norm constraint,L0-LMS algorithm and L1-LMS algorithm,both the two are improve algorithms based on the standard least mean square(LMS)algorithm using the sparse transcendental characteristics of system.On the bases of the single agent sparse identification algorithm and existing results of distributed collaborative estimation methods research,we studied the problem of every agent collaborative estimate sparse system parameters in the distributed network.By applied distributed collaboration strategy optimize sparse identification algorithm,at the same time of achieving to collaborative estimate the sparse parameters,we enhance the distributed network's ability of identifying.In the actual engineering environment,the input and output signals of agents are usually accompanied by noise interference,resulting in adaptive estimation deviation.Bias compensation principle is analyzing the bias in the estimate process and rectifying the cost function by adaptive estimating the noise variance,obtaining the updated formula of unbiased estimation,to achieve an unbiased estimate to the parameter of unknown target.In this paper,based on forward prediction noise variance estimation method we proposed bias compensation algorithm which is suitable for cluster sparse system,and combined different distributed strategies,we also proposed distributed bias compensation RZA-LMS algorithms for multi-agent network.The simulation results shows that the proposed bias compensation scheme is able to improve the accuracy of sparse system identification effectively,and distributed bias compensation RZA-LMS algorithms possess good convergence in the multi-agent network.