Study On Network-Constrained Detection Algorithm Based On Graphical Model

The vision of"wireless sensor networks"is a confluence of emerging technology in both miniaturized devices and wireless communications. It is of growing interest in a variety of scientific fields and engineering applications e.g., geology, biology, surveillance, fault-monitoring and so on. For the reason of each node's finite power, the analysis and research on the energy constraints have important theoretical meaning and high practical value.Graphical Model (GM) is a marriage between probability theory and graphical theory. It provides some effective methods to solve for probability inference problem. The thesis aims to improve the performance of detection algorithm after importing the GM.Therefore, this thesis will be carried out as follows:(1) Derive the new team decision methods based on Person-By-Person optimality rule. After analysis, we can see that the traditional detection algorithm does not consider completely about the relevance among the nodes, and original team decision method has its limitations. Given some model assumptions, to minimize the Bayesian risk function, we derive the algorithm in detail according to the Bayesian detection paradigm and some GMs'algorithms. Lastly, analyze the influence to detection algorithm from parameters (such as correlation, signal-to-noise ratio) according to a simple simulation example, and the results also show its efficiency.(2) Develop the more complexity team decision method based on the channel models. For the complexity of environment, the signal may occur to attenuation or produce kinds of biases. In this situation, after modeling the channel, the algorithm will be applied into more general scene, and given the analysis for the maneuvering target tracking problem.(3) A new tracking algorithm called N-KF is developed in problem of the maneuvering target tracking. It is designed based on the minimization of the global risk, and aims to improve the performance of the tracking. After computing the conditional probability distribution and the threshold value of the innovation, we can choose the model according to the decision result. Then track the target by the Kalman Filtering (KF). Lastly, the N-KF is compared with the KF through the two classical models——maneuvering model and mutation model, the results shows the efficiency and superiority of the N-KF.