降低雜訊干擾之室內無線定位演算法

Abstract

室內無線定位的快速發展可應用於商業與公共安全。但是對於室內無線定位而言，最根本的問題還是在於接收到的信號有著很劇烈的變動，即便是在同一點上所收到的信號亦是如此。在此篇論文中，我們的目標就要發展出一套卓越的定位系統，進而從已接收到含有劇烈變化的信號中，萃取出其強健特性，降低動態雜訊對系統的影響。為了要達到這個目的，在我們的方法中主要有兩個步驟。第一：我們先將受多重路徑干擾的信號轉換到頻譜上，再取log將多重路徑干擾的現象轉換成可加性的隨機變數，進而可以用平均的方式取得受干擾較少的信號資訊。第二：利用PCA的轉換，可以讓我們得到信號特性加強的效果。因此，經過這些步驟之後，我們即可獲得對室內動態環境較不易受影響的信號特性。就我們所知，這些步驟加強了信號特性的強健性，用這些信號特性在一般室內環境下做定位，不易受到動態環境的影響。除此之外，我們所使用的方法不但間單方便、容易實作，還不需額外的硬體或感知網路的架設。 為了評估我們所提出的方式，我們在真實的無線網路環境中收集無線基地台所發出的信號。並從實驗的結果發現，我們的方法較傳統的定位作法在精準度上提升不少。由數據顯示，我們的結果在平均誤差距離上，較傳統的方法減少了52.8%而在變異數上改善了74.9%。此外，實驗的結果也透露出，在建立定位系統上，我們的方法較傳統作法需較少的基地台數目及訓練樣本數。這樣也就可以達到節省資源的目的。

Positioning in wireless is growing rapidly in commercial interest and public safety for context aware applications. The essential challenge in location positioning technology is the severe fluctuation of the receive signal strength (RSS) even for a fixed location. In this article, our work is to develop a novel localization algorithm which would extract the robust feature from received signal strength (RSS). And the algorithm is less sensitive to time varying noise effect. There are two steps for our approach. The first step is to transform the dynamic multipath into an additive random variable in the logarithmic spectrum domain and thus can be averaged out the proposed method. Then, we apply the PCA technique to replace the elements by principal components (PCs) which are generated through a transformation such that the retained information can be maximized. Therefore, the transformed information would be less sensitive to time varying indoor environment. To our knowledge, this work is to enhance the robustness to multipath fading condition, which is common in the indoor environment. This approach is not only simple and easy to be implemented but also required neither new hardware nor extra sensor network installation. In order to evaluate the performance of our algorithm, we collect realistic RSS in an indoor WLAN environment. The results show that the positioning accuracy is significantly improved. The numerical results show that mean and variance of estimated error are reduced by 52.8% and 74.9% on the average. Moreover, the experimental results also show that fewer training samples and access points are required to build the positioning models.