Research On Distributed Multi-task Adaptive Diffusion Estimation Algorithm

With the rapid development of sensing and communication technology,as well as signal processing technology,the integration and miniaturization of signal processing chips,distributed information processing on the network has attracted more and more attention.In practice,distributed processing on the network depends on the local computation of each node and the cooperation with the adjacent nodes,so as to realize the completely decentralized distributed information processing.In practice,the completely decentralized distributed information processing can be achieved depends on the distributed processing on the network,which relying on the local computation of each node and the cooperation with the adjacent nodes.Cooperation is beneficial when nodes pursue a single goal,that is,a single task estimation problem.However,in many applications,nodes may pursue different goals,namely the multi-task estimation problem,and blind cooperation will lead to unsatisfactory estimation results.In this paper,the problem of adaptive estimation algorithm for distributed multitask on network is considered.Aiming at this issue,this paper considers not only the similarity between different tasks that exists possibly,but also the change of tasks in possible situation.In the current research of distributed estimation algorithm,information exchange strategies in network are mainly divided into three types: incremental strategy,diffusion strategy and diffusion strategy.Compared with the other two strategies,diffusion strategy has a better flexibility,stronger robustness,and can realize real-time adaptive updates on large-scale networks.Thus,in this paper,the distributed multi-task diffusion estimation algorithm and its application in adaptive clustering,as well as and information security are studied.We take into account the condition that the tasks are independent from each other and the task may change or be abnormal due to interference,which mainly for the distributed adaptive clustering problem,specifically in this paper.And aiming at these points,we propose a distributed adaptive diffusion estimation algorithm based on maximum entropy criterion and a distributed adaptive diffusion estimation algorithm based on event trigger.In addition,a distributed adaptive clustering framework is derived based on normal task and same task adaptive clustering strategies,so that network nodes can identify their clusters and make effective cooperation strategies.The simulation results show that the proposed algorithm performs well under the mixed interference of Gaussian noise and impulse noise.This paper also optimizes the intra-cluster collaboration weight selection scheme,which allows network nodes to locally optimize their intra-cluster collaboration weight.The simulation results show that the proposed method can obtain a lower mean steady-state network mean square error than the weight selected by other methods that commonly used in the literature.In this paper,we also consider the possible similarity between tasks and the hostile environment of the network,and a distributed multi-task adaptive diffusion estimation algorithm based on security information sharing is proposed.In the algorithm,a distributed adaptive detector is designed,that is,a threshold test is constructed by similarity between tasks to detect the trust neighbor of each node.Then,based on the detected trusted neighbors,the multi-task diffusion least mean square(SM-DLMS)algorithm is derived.Meanwhile,we theoretically analyze the stable performance of the proposed SM-DLMS algorithm,and simulation results are also provided to show the robustness and effectiveness of the algorithm under network attack.