Research On Variable Step-size Distributed LMS Algorithm Based On Wireless Sensor Networks

Wireless sensor network(WSN),as distributed network that senses and monitors the physical world,is now maturely applied to military,agricultural,industrial,intelligent transportation and other fields due to its large scale,self-organization,and high reliability.Parameter estimation in distributed way means that each node in the WSN collaboratively estimates the parameter of interest with its neighboring nodes.Due to the inherent limitations of wireless sensor nodes,such as limited power supply,limited computing power and limited storage space,distributed estimation algorithms only need to store a small amount of real-time data on the nodes in the network.By now,the distributed adaptive algorithms have been extensively studied.The study in this thesis focuses on the variable step-size distributed Least Mean Square(LMS)algorithm.Traditional variable step-size distributed LMS algorithms have proved to be able to trade off the two objectives,i.e.achieving low steady-state misadjustment and high convergence rate.However,in some specific cases,such as the great differences among all components of the true parameter,the true parameter being sparse and the background noise being corrupted by impulse noises,the estimation performance of the traditional variable step-size distributed LMS algorithm will be severely deteriorated.To solve these problems,the variable step-size diffusion LMS algorithm is further studied in this paper,and the main work are summarized as follows:1.In order to improve the performance of the traditional variable step-size diffusion LMS algorithm,a novel component-wise variable step-size diffusion LMS algorithm is proposed.Then the numerical simulation of the component-wise variable step-size diffusion LMS algorithm is carried out and compared with the non-component variable step-size diffusion LMS algorithm to verify the advantages of the proposed algorithm;2.In order to suppress the influence of impulse noise on distributed parameter estimation,a robust variable step-size diffusion LMS algorithm is proposed.The robust variable step-size diffusion LMS algorithm is numerically simulated in the environment of impulsive noise and compared with several other algorithms.The ability of the proposed algorithm to suppress impulse noise and the superiority of the proposed algorithm over other algorithms are verified;3.The mean stability analysis is performed on the proposed two algorithms,and the mean-square performance analysis is performed on the component-wise variable step size diffusion LMS algorithm.To solve the distributed estimation problem in some specific cases,a component-wise variable step-size diffusion LMS algorithm and a robust variable step-size diffusion LMS algorithm are proposed in this paper.The effectiveness and superiority of the proposed algorithms are verified by numerical simulationsd.In my future work,I will consider security issues,such as dealing with malicious attacks in distributed networks.Also,I will consider the application of the proposed algorithms,that is,the algorithms are transplanted to the real distributed network.