Quantized Information-Based Target State Estimation And Fusion

The main objective of the current study is to propose state estimate and in-formation fusion algorithm for target tracking in wireless sensor networks (WSN).Through the introduction of measurement information quantification, the study effec-tively solves the wireless sensor network resources constraint problem. Combiningquantitative information with target state estimation and information fusion algorithm,the study offers a new approach for wireless sensor networks target positioning track-ing. The quantitative strategy and information fusion are optimization from the angleof system performance, it can balance the system performance and the resources con-straint, it also can effectively raise energy and communication bandwidth utilizationratio, and improve practical applicability of the system. Specifically, the project hascompleted the wireless sensor network positioning tracking modeling, the quantitativestrategy and the noise probability density estimate of the quantitative measurementsystem. The project has also completed state estimation and information fusion algo-rithm based on quantitative innovation. The performance analysis of the correspondingalgorithm is also given.Chapters 1 introduces the research background, research content and researchstatus of the quantitative information state estimation.Chapter 2 Firstly, the mathematical model of the discussed problems is presented.Secondly, preliminaries about the considered problem are given. The common stateestimation methods, i.e. kalman type filters are introduced. Thirdly, the approximatecorrection of the additive white noise model in quantized Kalman filter is investigated.Finally, the concept of system indeterminacy and robustness are introduced.Chapter 3 investigates the problem of state estimation with quantised measure-ments for general vector state-vector observation model in WSNs, which broadens the scope of sign of innovations Kalman filtering (SOI-KF) and multiple-level quantisedinnovations Kalman filter (MLQIKF). Adhering to the limited power and bandwidthresources WSNs must operate with, this paper introduces a novel decentralised un-scented Kalman filtering (UKF) estimators based on quantised measurement innova-tions. In the quantisation approach, the region of a measurement innovation is parti-tioned into L contiguous, non-overlapping intervals. After quantised, the measurementinformation is broadcasted by using a variable number of bytes coding method. A fil-tering algorithm for general vector state-vector observation case is developed based onthe quantised measurement information. Performance analysis and Monte Carlo simu-lations reveal that under the same bandwidth constraint condition, the performance ofnovel quantised UKF tracker, indeed better than those of SOI-KF and MLQIKF in er-ror covariance matrix (ECM) and root mean-square error (RMSE) and almost identicalto these of an UKF based on analogue-amplitude observations.Chapter 4 provides a performance analysis for Kalman filtering based on quan-tized measurement innovations (QIKF). In this performance analysis framework, thetrue covariance matrix of estimating error is strictly analyzed without Gaussian as-sumption on predicted distribution. By taking the quantization errors as a random per-turbation in observation system, an equivalent state-observation system is obtained.Accordingly, the QIKF for the original system is equivalent to a Kalman-like filter-ing for the equivalent state-observation system. The boundedness of the true errorcovariance matrix of QIKF is obtained under some weak conditions. We also obtaina sufficient condition for the stability of the QIKF in general vector case. Then, therelationship between the standard Kalman filtering and the QIKF for the original sys-tem is discussed. In order to guarantee the stability of QIKF, the design of numberof quantized levels is given. The relationship between filtering performance and thenumber of quantization levels is also discussed. Finally, the validity of these results isdemonstrated by numerical simulations.Chapter 5 gives a modified quantized innovation Kalman filtering (MQIKF) toimprove the stability of quantized innovation Kalman filtering . Two matrices pa-rameters and are introduced into the QIKF to develop the MQIKF. Then, theperformance of the MQIKF for linear time invariant system is investigated. Firstly,the statistic of estimation errors is discussed. Secondly, the stochastic stability of the MQIKF is analysed. Here, an diagonal matrix is introduced to take the quantization er-rors into account. It makes the Lyapunov method can be used to analyze the stochasticstability in a rigorous mathematical way. Thus the connection between a good sta-bility behavior of the MQIKF and the instrumental parameters ??1 and ??2 is revealed.Furthermore, a simple way to design the parameters ??1 and ??2 is given to make theestimation error of the MQIKF remain mean square bounded. Finally, the validity ofthe proposed algorithm is demonstrated by numerical simulations.Chapter 6 provides a networked data-fusion method for the target tracking inWireless Sensor Networks (WSNs). The proposed scheme reduces the energy con-sumption dramatically. Specifically, in the proposed scheme, each sensor collectsone noise-corrupted sample, performs a quantizing operation, and transmits quan-tized message to the fusion center, while the latter combines the received quantizedmessages to produce a final estimate. The novel data-fusion method is based on thequantized measurement innovations and decentralized Kalman filtering (DKF) withfeedback. For the proposed algorithm, the performance analysis of the estimation pre-cision is provided. Finally, Monte Carlo simulations show the effectiveness of theproposed scheme.Chapter 7, based on the research of quantized innovation Kalman filtering, givesthe algorithm process design of multi-target tracking in WSNs. At the same time,the simulations are provided to verify the actual performance of above all kinds ofestimation fusion algorithms.Finally, the results of the dissertation are summarized and further research topicsare pointed out.