一個智慧型WiFi無線區域網路定位系統之設計

Abstract

隨著網際網路和無線區域網路的蓬勃發展，與位置感知相關的服務和系統變成近年來相當熱門的議題。雖然GPS系統已經廣泛地使用在旅遊和駕駛導航上，但它並不適合應用於室內的環境或高樓的周邊，而且其精確度無法滿足某些需要面對面的服務或應用。此外目前的GPS系統主要是設計用來做純粹的定位或導航，如果要擁有網路傳輸的能力，必須另外結合具有網路傳輸功能的系統。基於以上GPS的不足，有很多定位的方法或技術在過去幾年相繼被提出來，但是其中大部分不是缺乏足夠的精確度，就是架設或維修的過程複雜，導致成本過高而不切實用。 在這篇論文中，我們針對室內的環境，在網路通訊的應用層上架構一個智慧型的無線區域網路定位系統，由於其架構在應用層上，這個系統並不需要修改或更新現有的無線網路設備。 而且，這個系統具有機器學習的能力，所以在延伸定位範圍和系統維護時將會非常實用。其本概念就是，當某個移動裝置被定位出來之後，它的相關資訊可以加入系統的位置表(Location Tables)中，作為往後定位的參考樣本。但是這樣的方式在應用上會面臨一些問題，我們將會在論文中討論並且提出我們的解決方法。 我們也提出了幾種方法來改進我們系統的精確度、效率和彈性，當系統經過長時間的使用和訓練後，其準確度會相應地提升，但是位置表內相對的會有龐大的樣本資料。如果每次定位時都去搜尋並找出有用的樣本，對系統而言將會是相當大的負擔。在我們設計的系統裡，只有在一開始樣本數量較少時才需要一一比對位置表裡的樣本，一旦樣本成長到相當數量時，並不需要針對每次的定位去做樣本搜尋比對，只有在某些情況下才必須去執行完全搜尋比對動作，這將有助於系統效能的提升。

Due to the fast growing in the user population of Internet and Wireless LAN, location-aware services and systems become a popular topic. Although GPS systems are widely deployed for traveling and driving guidance, it is not suitable for indoor environment. Further, it does not provide sufficient accuracy for face to face applications. Besides, current GPS system was solely designed for positioning purpose, so it needs to be combined with wireless communication service to implement location-aware function for mobile computing purposes. Because of these GPS disadvantages, quite a few indoor positioning schemes were proposed in the past years, but most of them are either expensive or featuring low accuracy. In this thesis, we proposed an Intelligent WiFi LAN Positioning System which was implemented on the application layer to position the mobile station in indoor environment. Since it was implemented on application layer, it does not need any change of wireless equipments. Further, it is intelligent because it has the machine learning capability, which is quite useful for extending the positioned area and system maintenance. The basic idea of this learning capability is that the positioned locations can be reused as new samples for future positioning. But there are some potential problems for using all of these positioned locations as future samples. We will discuss the problem and provide our solution in this work. We also introduce schemes to improve the accuracy, performance and flexibility for our system. After the system has been trained for a certain period, the accuracy will be improved, but there will be a large number of samples in the Location Tables. Searching through the table one by one will be a time consuming task on Location Server, in our system, searching all samples is required only in the beginning or in few special cases.