Quantitative Information Fusion Kalman Filter

In recent years, WSN(Wireless Sensor Network) systems have attracted muchattention of people in business, based on their many advantages such as low cost,flexibility, reliability and so on. And they have been used in many areas widely, e.g.military surveillance, environmental monitoring, space exploration and intelligenttransmission. For there are the problems of limited energy and limited bandwidth of thecommunication channel in the WSN, the transmitted data can be quantized beforetransmission in order to reduce the number of bits and the energy consumption thattransmission requires. At the same time, the unreliability of communication will causethe phenomenon of losing data during the transmission. Thus, for the WSN systemswith the phenomenon of dropouts, the filtering algorithms based on quantizing thetransmitting data are developed. The studied main contents are as follows:In the WSN systems, the uniform quantization method is used to quantize theobservations of each sensor. Then, the quantized data are transmitted to the fusioncenter to carry on the fusion filtering, during which there are the phenomena of randomdropouts. When the probability of dropouts occurring at each time is known, thecentralized fusion Kalman filter based on the quantized observations with the dropoutsof quantized observations is presented. And when the dropout situation of quantizeddata at each time is known, the weighted measurement fusion Kalman filter based onthe quantized observations with random dropouts is presented.Based on transmitting the innovations of each sensor, the two kinds of quantizationmethods are used to study the WSN systems with the random dropouts for the values ofinnovations are small. The one is that innovations are quantized according to their signssimply, which requires lower bandwidth and energy. The local SOI Kalman filters withrandom dropouts are obtained based on the projection theory, and the step by step fusion SOI Kalman filter with random dropouts is obtained using of the step by step fusionalgorithm. The other quantization method is the uniform quantization, in which thequantization noise is an approximation of its upper bound. The local quantizedinnovation Kalman filter with the random dropouts is proposed using of the standardKalman filtering algorithm, and the step by step fusion quantized innovation Kalmanfilter with random dropouts is gained using of the step by step fusion algorithm.Finally, based on transmitting the local quantized state estimators of each sensor,the fusion filtering of the WSN system with the random dropouts is studied using twokinds of methods respectively, by extending the scalar-uniform quantization method tothe vector quantization. One of the fusion methods is that the fusion center only fusesthe local quantized state estimators received at current time, and gets the suboptimalfusion Kalman filter with quantized state estimators. The other is that the fusion centeruses the state estimator of the previous time to predict to get the approximate local stateestimator of the sensor which is lost at current time, and obtains the improvedsuboptimal fusion Kalman filter with quantized state estimators using the covarianceintersection algorithm.And a large number of simulations are done, and verify the effectiveness of theproposed algorithm.