利用Wi-Fi CSI作精細室內定位

Abstract

最近, 利用Wi-Fi 做室內定位的技術變得更加引人關注, 因為利用廣泛被佈置的Wi-Fi系統可以減少系統在硬體建置的負荷, 此外, 有先前的研究驗證使用Wi-Fi在正交分頻調變(Orthogonal frequency-division multiplexing, OFDM) 運作下的精密評估資訊, 也就是通道狀態訊息(Channel State Information, CSI), 用以作為位置指紋比傳統以接收訊號強度(RSSI)更具代表性。這篇論文中, 將會分享關於利用學校已建置的Wi-Fi系統去實作一個以通道狀態訊息為基礎的定位系統。 我們的系統包含了「指紋資料庫」和「位置評估系統」兩個部分。因為在我們的測試環境中可以明顯觀察到多個不同的通道狀態訊群，所以我們利用K-means 演算法分群，並保留每個探勘點的多個指紋在「指紋資料庫」中。 因為精密的通道狀態訊息是以高維度的向量呈現，所以我們利用一種統計模型R平方數 (R-square value) 來做指紋比對。除了單一通道狀態訊息封包的比對測試方法之外，文中也提供一個可以達到更高定位精準度的多通道狀態訊息封包的比對測試方法。同時為了避免受到易受位置影響的通道狀態訊息而導致的誤判, 亦提供了一個可行的權重投票估計機制。最後我們顯示系統的評估結果, 也可以看到我們的系統表現比傳統利用接收訊號強度實作的系統突出。

Nowadays, Wi-Fi-based indoor localization techniques have become attractive, because widely deployed Wi-Fi system could reduce the overhead of infrastructure. Moreover, some prior works argue that Wi-Fi OFDM-based fine-grained estimation data, Channel State Information (CSI), is more representative than traditional RSSI as location fingerprints. In this paper, we shared the experience of utilizing the school built-in Wi-Fi system to build a CSI-based localization system. Our system includes “Fingerprint Database” and “Localization System.” Due to multiple obvious CSI clusters could be observed in our testbed, we utilize K-means algorithm to retain multiple fingerprints for each survey point in our “Fingerprint Database”. Because fine-grained CSI are high-dimension vectors, a statistical module (i.e., R-square value) is proposed for fingerprint comparison. Not only Single-CSI comparison, but also a Multiple-CSI comparison testing method is also proposed, which reaches higher accuracy. To reduce the location misjudgment caused by location sensitive CSI, a feasible weighted voting estimation process is also proposed. Finally, we evaluate our system in our testbed and show our system outperforms traditional RSSI-based localization system.