Research On Distributed Collaborative Estimation

We study the problem of distributed collaborative estimation over distributed networks.Distributed networks are a collection of agents which are distributed over a geographic areaand collect data from their environments, and distributed signal processing can deal withextraction of useful information from data collected at agents.Distributed collaborativeestimation is one of branches of distributed signal processing, i.e. the agents in distributednetwork have a common objective to estimate parameter vector. The target parameters ofdistributed collaborative estimation is always unknow, we choose adaptive algorithm as thedistributed algorithm, and using different distributed strategies to estimate.This paper described two basic adaptive estimation algorithm, the least-mean-square(LMS) algorithm and recursive least-squares (RLS) algorithm, and comparison was madebetween the above two algorithms. The implementation of LMS algorithm is simple, butthe convergence process is too slow. The estimation of RLS algorithm converges fast withhigh accuracy. On the basis of adaptive algorithms, we study the distributed stratergy ofdistributed network, i.e. incremental strategy and diffusion strategy. By analyzing thecharacteristics of the two strategies, we established the collaborative estimation model overdistributed network. The distributed LMS algorithm and distributed RLS algorithm wereproposed, and comparison was made to the performance of their estimates. Simulationresults show that the distributed algorithms have higher accuracy than single agentalgorithms, and distributed RLS algorithm has faster convergence rate with good estimationaccuracy compared to the distributed LMS algorithm.In the real environment, the collected data of each agent are inevitable influence ofnoise in distributed network. Based on the study of the distributed adaptive algorithms, weimproved the original model of distributed network to study the distributed collaborativeestimated with regressor noise. Because of the presence of regression noise, resulting theoriginal estimation results are biased. We choose the RLS algorithm as the research object,a real-time estimation algorithm of noise variance was proposed. On this basis, based ondistributed strategies, we propose distributed bias compensated recursive least-squares(BCRLS) algorithm. Simulation results show that the distributed BCRLS algorithm caneffectively achieve bias compensation, and with the higher SNR, the distributed BCRLSalgorithms have better effects. Furthermore, we adjust the diffusion strategy base on the real-time SNR, to make the diffusion BCRLS algorithms have better unbiased estimations.Simulation results show that the modified diffusion BCRLS algorithms (BCRLS-M)according to the real-time SNR, have lower mean square error (MSE) compare to thediffusion BCRLS algorithms..