在無線感測網路 (Wireless Sensor Network, WSN) 的定位方法中，指紋比對 (fingerprinting) 方法透過事先訓練量測環境中不同位置的無線電訊號樣式，在定位時從訓練的樣式中選出與當前訊號相似者來估測目標位置，可以達到很好的準確度但需要花費較高的訓練成本。指向性天線 (Directional antenna) 因其各個角度的訊號增益程度不同，可透過指向性天線通訊的訊號強度變化情況，量測訊號入射角度 (Angle of Arrival, AoA) 以進行定位。本論文提出一個結合指紋比對與指向性天線的無線感測網路定位方法，稱為FLCDAA (Fingerprinting Localization with Cross Directional Antenna Arrays for WSNs)，我們在各個已知位置的錨節點 (Anchor node) 上配置四隻呈十字型排列的指向性天線，對目標節點 (Target node) 估測位置。各個錨節點能夠獨自對鄰近的目標節點定位。本文方法可分為訓練與定位二個階段：在訓練階段，我們利用事先量測指向性天線的天線訊號樣式 (Antenna radio pattern) 取代傳統指紋比對定位方法的環境樣式以降低訓練成本；在定位階段，錨節點根據四隻天線所收到的接收訊號強度階度 (Received signal strength indicator, RSSI)，比對訓練的樣本進行定位。本論文另外提出了將訓練階段所蒐集到的樣式透過迴歸分析工具簡化成近似關係函數以減少比對樣式數目加快定位的方法。我們在室內體育館實驗檢驗所提出的方法的效果，在半徑為6公尺的圓型區域內定位的平均距離誤差為24 cm，並且能夠迅速完成定位。為了降低量測成本，針對不同的量測距離與量測角度精密度進行實驗，以得知實作時合適的量測取樣精密度；並且透過使用一隻天線的量測樣式取代所有同型號天線的實驗，測試只訓練一隻天線時，因為天線間個體差異對定位結果造成的影響，以降低訓練成本提高可擴大性(scalability)。Fingerprinting method measures the radio signal patterns at certain location in the environment in an offline phase, and it localizes target devices according the similarity between current signal pattern and prior collected patterns. Directional antenna can distinguish the signal strength from different angle of arrival (AoA) based on its individual power gain for different directions. This paper presents a localization method for wireless sensor networks (WSNs), which using fingerprinting and directional antenna, called FLCDAA (Fingerprinting Localization with Cross Directional Antenna Arrays for WSNs). We set four directional antennas be cross arrangement at location-known anchor nodes, which an anchor node can independently localize its neighboring location-unknown target nodes. In offline phase, we measures and records the received signal strength indicator (RSSI) value which an anchor node receives a signal emitted from target node at different distances and angles, we store 4-tuple RSSI value as the radio pattern of antennas. In online phase, anchor node compares the RSSI values of signal emitted from target node with prior collected patterns for finding the best match one to estimate the related location of the sensor node corresponding to the anchor node. Furthermore, FLCDAA presents an accelerating method, performing linear regression to generating approximation functions to fit the AoA and RSSI-Difference value of any two vertical antennas in an anchor, and we use the approximation functions to reduce amount of pattern compared in online phase. We have to implemented an anchor of FLCDAA in a circular area with a radius of 6 meters within a indoor gym, we gather 100 patterns at each measuring position, take half of patterns be training data and the other half be the test data. The experiment average error is 24 centimeters for base method。In order to apply in bigger area, and try to train one antenna to replace other same type antennas to reduce training cost for scalability.
