| 摘要: | 以無線感測器網路(WSN, Wireless Sensor Network) 節點的全向天線接收指向天線的訊號時，其接收訊號強度階度(簡稱接收階度: Received Signal Strength Indication, RSSI) 會隨收訊角度(Angle of Arrival, AoA) 的不同而變化。依據實驗的結果，我們觀察到以下二個特性: (1) 在AoA 介於0 度到90 度之間時，RSSI 相對於AoA 呈現近似拋物線函數的變化；(2) 在AoA 介於0 度到90 度之間時，全向天線針對二支相同位置且相互垂直的指向天線的RSSI 差值則呈現近似線性函數的變化。運用以上兩個特性，我們使用相互垂直指向天線RSSI 差值來估算AoA 以設計新的定位機制。在我們的定位機制中，運用RSSI 差值進行定位，可以降低環境因素對定位精準度的影響；而我們利用儲存RSSI 相對於AoA 的關係函數降低對儲存空間的需求，以符合感測器對儲存空間的限制。我們的定位機制區分兩個階段: 訓練階段與定位階段。訓練階段為在系統部署前實際量測AoA 對指向天線RSSI 的影響，並利用二次線性迴歸分析取得在不同距離下角度與接收階度的關係函數，同時 利用量測所得的資料，以線性迴歸分析計算相互垂直天線的RSSI 差值與AoA 的關係函數，並將所得的關係函數儲存於目標節點(target node，也就是位置待測的感測器節點)。在定位階段，我們在環境中部署二組已知位置的錨節點(anchor node)，每組錨節點上裝配著兩支同型且相互垂直的指向天線。目標節點在接收到同一組錨節點的二支天線發出的訊號後，先利用RSSI 與AoA 的關係函數粗估其與錨節點的距離，再運用該距離所相對應的RSSI 差值與角度的關係函數，估算錨節點與目標節點的角度。而藉由取得對兩組已知位置錨節點的角度，目標節點即可計算出自己的位置。我們在10 公尺見方的室內測試環境中一邊的二個端點部署二組錨節點，以驗證所提出方法的定位精準度，實驗的結果的平均定位誤差為124 公分。為了更進一步提高定位精準度，我們利用二個去除離群值(outlier) 的修正方法，可以進一步降低29% 的定位誤差，修正後的平均定位誤差為89 公分。 As a node of a wireless sensor network (WSN) with an embedded omni-directional antenna receives signals emitted from a directional antenna, the received signal strength indication(RSSI) varies with the angle of arrival (AoA) of the received signal. From experiments,we have the following two observations about the RSSI values of signals which a sensor node receives from a directional antenna. (1) If the distance of the node and the antenna is fixed, RSSI varies like a parabola function of AoA between −90◦ and 90◦ with a symmetry axis at AoA=0◦. (2) If we put two same-type directional antennas with perpendicular orientation at the same position, then the difference of the signal RSSI values which the node receives from the two antennas varies like a linear function of AoA between 0◦ and 90◦. Based on the above observations, we design and implement a novel localization scheme, called ALRD, for a sensor node to estimate AoA and then its position by RSSI value difference of two perpendicular directional antennas. The proposed localization scheme consists of two phases: the learning phase and the localizing phase. In the learning phase, we measure the RSSI values of signals received from a directional antenna at different distances and angles. For a fixed distance d, we perform regression analysis on the measured RSSI values to obtain two approximation functions: a quadratic function Rd = f( ) and a linear functions Dd = g( ), where is AoA, Rd is RSSI, and Dd is the RSSI difference of two signals received from two perpendicular directional antennas at the same position. These approximation functions, rather than all measured RSSI values, are then loaded into the limited storage of the sensor nodes for them to calculate AoA values to locate themselves. In the localizing phase, two location-known beacon nodes, either of which is equipped with two same-type perpendicular directional antennas, are deployed to transmit beacon signals periodically. By the Rd functions and the RSSI values received from two antennas of one beacon node, a sensor node can roughly estimate the distance d′ to the beacon node. By the Dd′ functions and d′, the sensor node can estimate AoA. With estimated AoA values of two distinct beacon nodes, a sensor node can then calculate its position. We have implemented ALRD and apply it to a WSN in a 10 by 10 meters indoor square area with two beacon nodes being installed at two ends of an area edge. Our experiments demonstrate that the average localization error is 124 centimeters. We further propose two methods, namely maximum-point-minimumdiameter(MPMD) and maximum-point-minimum-rectangle (MPMR), to improve localization accuracy by removing outliers from positioning results. The experiment results demonstrate the two methods can reduce the average localization error to be 89 centimeters, i.e., about 29% improvement in accuracy. |
