TDOA Localization Technique Based On Data Mining

With the prevalence of mobile terminals and diversification of wireless service, the wireless circumstances are becoming more and more complicated. Streets of cities and countryside are filled with mobile terminals of all kinds, bringing about big challenges to wireless supervision department. Among the regulatory techniques in civil fields as well as military fields, source localization plays a significant role. Confronted with the more complex network, modern localization has to deal with the scenario with multi-source and multi-path. The solution is urgently needed in source localization to precisely localize multiple sources in transportation environment with complex superposition of multipath. In the meanwhile, the growth of national wireless supervision network brings about new technique vision for the upcoming problem. Armed with the improvement of the data sampling, huge dividend introduced by the big data may instill innovation for various scenarios. Upon the things, the research center of this article is to work out how to combine the big data and source localization technique better for magnitude gain.Traditional time difference localization methods are always constrained to get the maximum likelihood estimates of the source with a few adjacent receivers. This brings about the weakness of low resistance to multi-path and multi-source scenarios since the matrix manipulation is easy to get sick results for singularity. However, with high-density data and the stochastic property of received information, data mining tends to produce good result with high robustness. The thesis gets down to the following two parts in detail.First, the grouped TDOA decision model is brought in. Different from targeting at the optimization of a single algorithm or introducing extension vertically in old ways, the thesis gets to the horizontal generalization of traditional TDOA in the first place. That is commit localization with grouped receivers in traditional ways but merge the obtained results of groups after that. Via analysis of the stochastic property of the process, we reach the conclusion of specified property of grouped localization result with regards to results of NLOS (Non Line of Sight) receivers. Then it becomes possible to convert the old problem into a data-driven localization decision model.Second, with the introduction of data mining and the clustering algorithm high correlated with the scenario, the model gets diverse solutions. In the presence of computer simulation different methods are put into practical with given multi-source multi-path localization scenario. Through comparison, image mining algorithm applicable is deduced and the high robustness gets proved. Localization precision of TDOA Localization algorithm base on Least Square combined with clustering method as DBSCAN matches up with such ideal scenario as single-source line-of-sight positioning. Afterwards, data structure of segment tree is introduced to lower down the computation complexity of DBSCAN searching process. In this way, a stable algorithm with high efficiency is built with test verification.Eventually, conclusion for whole thesis is made. Also, in accordance with present situation and development of localization fields, future outlook gets arrived at.