Research On Non-uniform Quantization Filter In Sensor Networks

Sensor network fusion systems are usually faced with the communication bandwidth constraints.Quantizing these signals should be one of reasonable ways to satisfy the bandwidth because the information could be quantized to decrease the transmitted bits.In actual sensor networks,measurements must be quantized before reaching the fusion estimator.Current quantization methods are mostly about uniform quantization and less research of non-uniform quantization.Hence,this paper mainly studies non-uniform quantization filtering in sensor networks.The main contents in this paper include the following three aspects:(1)We study the EKF in nonlinear systems where uniform quantization is adopted.Firstly,the theory of uniform quantization and the algorithm of quantitative filtering are analyzed.Then,the performance of filter is analyzed from the aspects of estimation error.Finally,for a series of models with linear observation equation and nonlinear state equation,the EKF with uniform quantization is researched and compared with the EKF without quantization.(2)We present a method of the non-uniform quantization filter in sensor networks.Firstly,a new non-uniform quantization strategy based on the least square sum of quantization error is proposed.Then,we further present a method to compute non-uniform quantizing points and build a recursive updating method of quantizing points which include semi real-time update and real-time update.Eventually,the effectiveness of proposed strategy is verified through computer simulation.(3)We study the allocation of quantitative copies in sensor networks.Firstly,the Kalman Filter based on Quantized Innovations is introduced.Then,we present an optimal function about the allocation of quantitative copies and provide a kind of cross quantitative copies allocation scheme and another fixed quantitative copies allocation scheme.Eventually,we analyze the advantages and shortcomings between different schemes via computer simulation.