Distributed Robust Estimation Algorithm Over Sensor Network

With the rise of internet of things and internet technology,sensor network integrated with microelectronics technology,sensor technology,wireless communication technology and distributed information processing technology have gradually entered all aspects of people's lives.Sensor networks can collaboratively monitor,sense and collect the physical information of various environments or detection objects in the network coverage area in real time,and process and transmit the available data.There are widely used in military defense,industrial and agricultural control,environmental monitoring,remote control of dangerous areas,and so on.Parameter estimation is one of the important applications of sensor networks.It establishes certain statistical models and uses various algorithms to acquire an estimate of some parameter from their measurements corrupted by noises,so as to determine the interdependence and interrelationship and between physical quantities.The parameter estimation method over sensor networks can be divided into centralized and distributed methods based on the node cooperation mechanism.The centralized estimation method requires a fusion center in the sensor network,and each node in the network transmits the measurements to the fusion center for centralized processing to obtain the global optimal estimates.However,in this way,if the network fusion center doesn't work properly,the parameter estimation system of the entire network will be paralyzed.In contrast,the distributed-estimation method relies the node's local data exchange and cooperative processing between the neighbor nodes to obtain a global optimal estimates,so that the entire sensor network can perform well even when some nodes of the network are interfered or even paralyzed.However,in practical applications,the measurement environment of sensor network is complex.For example,some nodes may suffer from impact noise or impulse interferences,or maybe face maliciously attack,resulting in some outliers mixed in the measurements.As a result,this adverse effect will be transmitted to the entire network through the node cooperation mechanism,and result in performance degradation of the whole network.In order to solve this problem,this paper proposes two distributed robust adaptive estimation algorithm based on LMS algorithm and RLS algorithm,respectively.The algorithms both introduce an e1-norm in cost function based on the sparsity of outliers,arming at detect and then reject these outliers,and moreover,exploit data exchange and cooperation between neighbors to further improve the estimation performance.In the distributed robust LMS algorithm,we consider online estimation of outliers and unknown interested parameter by using the measurement data,thus transform the estimation problem into a joint optimization problem for outliers and unknown parameters.This algorithm takes advantage of the insensitivity of the Huber function to be outliers,substitutes the errors between expected output and actual output into Huber function to remove outliers from measurements.Then with gradient descent in the traditional LMS algorithm,the gradient value of the Huber function is substituted into the iterative equation of the distributed LMS algorithm,and finally the unknown parameter estimate is obtained and updated at each iterative time.In the distributed robust RLS algorithm,the outliers in the measurements are firstly estimated by the coordinate descent method in one time,which will be used to compensate the deviation of the system output.Then,apply distributed RLS algorithm to obtain parameter estimates.Through a series of computer simulation experiments,the paper proves that the proposed algorithms have good estimation performance.