本論文針對藍牙低功耗測向（BlueTooth Low Energy Direction Finding, BLE-DF）於室內定位的應用需求，提出一套兼具低功耗、低硬體成本與高精度的角度與定位估計方法。首先建立符合2M-PHY CTE 天線交替切換特性的接收訊號模型，並針對信號源落入菲涅耳區（近場）時產生的球面波曲率問題，提出「近場 相位差校正」演算法，透過二次展開分離方向與曲率項，並以最小平方法進行校正。角度估計演算法搭配三角觀測法，即可求得空間座標。 本論文另設計兩類卷積自編碼網路，包括DoA-CAE與Position-CAE，其中DoA-CAE 針對ULA-2 與ULA-4 回歸入射角度，而Position-CAE 採用端對端結構，直接回歸目標座標。藉由模擬分析各類近場訊號下的演算法與模型，以角度均方根誤差與距離均方根誤差為判定標準，驗證所提方法的穩健性與優越性。 最後以nRF52833DK 與外部SP8T RF switch 實作多天線基站，實測誤差趨勢。 ;This thesis addresses the application needs of Bluetooth Low Energy Direction Finding (BLE-DF) for indoor positioning and proposes a method that achieves high accuracy while maintaining low power consumption and minimal hardware cost. A receive signal model is first developed to reflect the 2M-PHY CTE characteristics with alternating antenna switching. When the transmitter lies in the Fresnel (near-field) region, spherical wavefront curvature leads to phase distortion. To mitigate this, a near-field phase-difference correction algorithm is proposed, which uses a second-order expansion to decouple the directional and curvature components and applies a least-squares method for correction. The resulting direction-of-arrival (DoA) estimates, combined with triangulation, yield the spatial coordinates of the target. Two types of convolutional autoencoder networks are also developed—DoA-CAE and Position-CAE. The DoA-CAE performs angle regression for ULA-2 and ULA-4 arrays, while the Position-CAE adopts an end-to-end structure to directly regress the target coordinates. Through simulation analysis under various near-field signal conditions, both algorithms and models are evaluated using root-mean-square error (RMSE) of angle and position to validate their robustness and effectiveness. Finally, a multi-antenna base station is implemented using the nRF52833DK and an external SP8T RF switch, and experimental measurements are conducted to observe the resulting error trends.
