Research On Methods For Reducing Communication Cost Of Diffusion LMS Algorithm

Distributed estimation is that the node utilizes the local noised observation data and exchanges information with the neighboring nodes to estimate the parameters of interest by collaboration.Distributed estimation has high reliability,strong robustness,and is playing an increasingly important role in the military,environment,medical treatment,urban transportation,etc.It is often used in wireless sensor networks where nodes are powered by batteries and exchange information through wireless communication.The core of distributed estimation lies in the distributed algorithm.Diffusion Least Mean Square(DLMS)is a classical distributed estimation method with strong robustness and high accuracy.However,frequent data exchange between nodes leads to the large communication overhead of the DLMS algorithm,and the data processing capability and battery energy of the sensor nodes are limited.Therefore,it is necessary to reduce the communication overhead and make a tradeoff between estimation performance and communication overhead.DLMS algorithm that adopts multi-hop communication can improve the performance of estimation.However,it also results in an increased network communication overhead.Hence,this thesis firstly propose the Multi-hop Low Communication LMS(MLCLMS)algorithm.The proposed algorithm develops the Mean Square Deviation(MSD)optimization criterion.The node determines whether to exchange intermediate estimation information with the neighboring nodes by the MSD optimization criterion.Meanwhile,some nodes are set as relay nodes,which only forward the intermediate estimation information of the optimal neighboring nodes to the remaining nodes.In consequence,the estimation performance is guaranteed.The simulation results show that compared with DLMS algorithm,in which nodes exchange intermediate estimation information continuously,the MLCLMS algorithm reduces the amount of transmitted data,and reduces the communication overhead.Secondly,for the existing dimensionality reduction transmissions that affect the estimation performance of DLMS algorithm,this thesis designs a Selective Sending Least Mean Square(SSLMS)algorithm based on similarity judgment.In the proposed algorithm,by comparing the similarity of the local estimation information generated at adjacent time indices,each node selects a part of parameters to be estimated that have more information,broadcasts them to the neighboring node,and receives those selected parameters from neighboring nodes.The local estimation information is used to compensate the unexchanged parameters to be estimated of the neighboring nodes.The simulation results show that the SSLMS algorithm reduces the communication overhead of the network and achieves good estimation performance,which realizes the tradeoff between communication overhead and estimation performance when reducing the dimensionality of information sent by nodes.Two distributed least mean square algorithms based on diffusion strategy proposed in this thesis can reduce the communication overhead of nodes and the energy consumption of sensor,which has broad application prospects in resource-limited wireless network.