Research On Diffusion LMS Algorithms In Wireless Sensor Networks Under Input Noise

The wireless sensor networks(WSN)is a kind of multi-hop self-organizing sensor network with communication ability among nodes.Due to the advantages such as robustness,low power consumption,and scalability,the WSN is widely used in many fields such as military,environmental monitoring,and medical care The distributed algorithm can not only effectively solve the problem of limited communication range and power energy of nodes,but also improve the efficiency and accuracy of data collection in the network.Therefore,the study on distributed algorithms in WSN has great practical significance.The linear regression model is a commonly used mathematical model in distributed algorithms.In the linear regression model,the system input is regarded as the regression vector and has a linear relationship with the measurement output.In many real-life systems,the regression vector may be contaminated with various noises such as additive noise,multiplicative noise,and impulse noise(IN).In this thesis,I firstly propose the distributed algorithms for parameter estimation in the environment of multiplicative regression noise and impulsive regression noise existing in the WSN respectively.Then,I propose the distributed algorithm for parameter estimation when both kinds of regression noises exist in the WSN together.The main contents in this thesis are described in the following:1.I firstly establish the mathematical model in the environment of multiplicative regression noise and analyze the estimation bias when the standard LMS algorithm is used for minimizing the mean-square-error(MSE)cost function.Then,I modify the standard MSE cost function according to the estimation bias to obtain a new cost function with offset compensation.Finally,I use the gradient descent method to minimize the new cost function,and obtain the distributed diffusion Least mean square(DLMS)algorithm with offset compensation;2.I firstly establish the mathematical model and cost function under in the environment of impulsive regression noise.Then,I use the gradient descent method to minimize the costfunction,and obtain distributed DLMS algorithm with threshold.3.Combining the work of(1)and(2),directly propose a threshold setting-offset compensation distributed parameter estimation algorithm;4.The conditions of convergence in both mean and mean-square senses for the proposed algorithm with both offset compensation and threshold are analyzed.In addition,the mean-square steady-state behavior is also analyzed.5.Numerical simulations for the proposed algorithms are provided by comparing my algorithms with the non-cooperative parameter estimation algorithm,the non-threshold distributed DLMS algorithm,and the non-offset-compensated distributed DLMS algorithm to verify the superiority of our algorithm in the problem of distributed parameter estimation.6.Finally,I apply the proposed algorithm to the prediction of fire hazard in the monitoring system of forest fires.Satisfying performances are obtained by using our proposed algorithm.