本論文提出一個透過少量藍牙測向天線結合機器學習強化室內定位性能的架構與方法，包含整合原有的室內定位與燈具控制以及縮短定位所需時間，使本論文開發的藍牙AoA-Dongle能與原有的Dongle兼容於同一Mesh網路並具有藍牙測向功能且提供更快速的即時定位。Mesh網路蒐集到的數據會經由藍牙Gateway以MQTT傳輸至主機端進行定位計算。為了完成以上功能，我們以nRF5340與u-blox AoA天線板為基礎，將兩者結合開發成用於接收藍牙Dongle並實裝於實驗場域中。場域中我們以網格型式布置了12個定位點，並於各點蒐集數據組成資料集，這些資料集以使用的藍牙測向AoA天線數量及是否進行資料預處理做分類，並用於訓練以下四種機器學習模型：K-近鄰演算法（K-Nearest Neighbors, KNN）、支持向量機（Support Vector Machine, SVM）、隨機森林（Random Forest, RF）和多層感知器(Multilayer Perceptron, MLP)，並通過網格搜尋法找到四種模型的最佳參數組合，最後再使用混淆矩陣、F1-Score等指標驗證各模型的性能。
從實驗結果可得知，採用一個及兩個AoA天線特徵時的模型的準確率較原系統未採用時來的高，比起使用原始資料集(即原系統)訓練模型時分別提升11.11%~30.51%不等；在預處理資料集中則分別提升1.17%~6.32%不等。將四種模型均放入定位系統中實時預測，其中以支持向量機表現最佳，取得高達96.58%的準確度。

This thesis proposes a framework and method to enhance indoor positioning performance by utilizing a small number of Bluetooth direction-finding antennas combined with machine learning. The framework integrates existing indoor positioning and lighting control, allowing the developed Bluetooth AoA-Dongle to be compatible with existing dongles within the same mesh network while providing Bluetooth direction-finding functionality.
To achieve this, we used the nRF5340 and u-blox AoA antenna boards as the basis, combining them to develop a system that can receive Bluetooth dongles and implement it in an experimental field. We arranged 12 positioning points in a grid pattern and collected data at each point to form a dataset. These datasets were categorized based on the number of Bluetooth AoA antennas used and whether data preprocessing was performed, and were used to train the following four machine learning models: K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest (RF), and Multilayer Perceptron (MLP). The optimal parameter combinations for the four models were found using the grid search method. Finally, the performance of each model was validated using confusion matrices and F1-Score metrics.
Experimental results indicate that the models' accuracy, when using one or two AoA antenna features, is higher than the original system without the feature, showing an improvement ranging from 11.11% to 30.51% compared to using the original dataset (i.e., the original system). In the preprocessed datasets, the improvement ranges from 1.17% to 6.32%. All four models were then incorporated into the positioning system for real-time prediction, with the Support Vector Machine performing the best, achieving an accuracy of up to 96.58%.

摘要...i
Abstract...ii
誌謝...iii
目錄...iv
表目錄...vii
圖目錄...ix
第一章 緒論...1
1.1 研究背景與動機...1
1.2 研究目的...1
1.3 論文內容大綱...2
第二章 相關技術探討...3
2.1 藍牙通訊技術(Bluetooth)...3
2.1.1 低功耗藍牙(Bluetooth Low Energy)...3
2.1.2 藍牙網狀網路(Bluetooth Mesh Network)...4
2.2 藍牙測向(Bluetooth Direction Finding)...5
2.2.1 定頻擴展訊號(Constant Tone Extension, CTE)...6
2.2.2 IQ採樣(In-Phase and Quadrature Components, IQ Sampling)...7
2.2.3 藍牙到達角(AoA)技術說明...7
2.3 MQTT通訊協定(Message Queuing Telemetry Transport)...10
2.4 室內定位方法...11
2.4.1 指紋定位法(Fingerprinting)...11
2.4.2 三角測量法(Triangulation)...11
2.4.3 概似法(Proximity)...11
2.5 機器學習演算法...11
2.5.1 K-近鄰演算法 (K Nearest Neighbor, KNN)...11
2.5.2 支持向量機(Support Vector Machine, SVM)...13
2.5.3 隨機森林(Random Forest, RF)...15
2.5.4 多層感知器(Multilayer Perceptron, MLP)...17
2.5.5 機器學習模型比較...21
2.5.6 網格搜尋(Gird Search)...21
2.6 相關研究...22
第三章 系統架構...24
3.1 系統架構...24
3.2 定位功能架構...25
3.3 定位功能流程...25
第四章 系統設計與實作...26
4.1 實驗場域...26
4.2 實驗設備...27
4.3 藍牙網狀網路建置...30
4.3.1 信標(C209-Beacon)開發...30
4.3.2 BLE Dongle開發...31
4.3.3 藍牙網狀網路環境...33
4.4 RSSI數據蒐集與流程...39
4.5 訓練資料集說明...41
4.6 資料預處理...42
4.7 機器學習模型建立流程...45
4.8 模型驗證與評估方法...48
4.8.1 混淆矩陣...48
4.8.2 K折交叉驗證...49
4.9 實測流程...49
第五章 實驗結果...50
5.1 原始資料集之定位模型訓練結果...50
5.1.1 混淆矩陣訓練結果(無AoA天線資料集)...51
5.1.2 混淆矩陣訓練結果(一個AoA天線資料集)...53
5.1.3 混淆矩陣訓練結果(兩個AoA天線資料集)...55
5.2 預處理資料集之定位模型訓練結果...57
5.2.1 混淆矩陣訓練結果(無AoA天線資料集)...60
5.2.2 混淆矩陣訓練結果(一個AoA天線資料集)...62
5.2.3 混淆矩陣訓練結果(兩個AoA天線資料集)...64
5.3 原始資料集之實測模型定位分析...66
5.3.1 模型實測結果(無AoA天線)...66
5.3.2 模型實測結果(一個AoA天線)...68
5.3.3 模型實測結果(兩個AoA天線)...70
5.4 預處理資料集之實測模型定位分析...73
5.4.1 模型實測結果(無AoA天線)...74
5.4.2 模型實測結果(一個AoA天線)...76
5.4.3 模型實測結果(兩個AoA天線)...78
5.5 比較各定位模型實測結果...80
5.6 模型預測耗時比較...81
第六章 結論與未來展望...83
6.1 結論...83
6.2 未來展望...83
參考文獻...84

[1]Joe,“BLE 簡介：Bluetooth low energy,”Blogger, http://shyuanliang.blogspot.tw/2014/10/v-behaviorurldefaultvmlo.html, Oct. 2014(accessed Mar. 2024)
[2]J. Wang,“傳統BT與BLE架構有什麼不一樣呢？,”Blogger, https://wjungle.blogspot.com/2012/01/btble.html, Jan. 2012(accessed Mar. 2024)
[3]S.-T. Tu.“藍牙筆記,”Medium, https://medium.com/@yuchunlin/003-%E8%97%8D%E8%8A%BD%E7%AD%86%E8%A8%98-975b4f44ccc8, Mar. 2016 (accessed Mar. 2024)
[4]S.-T. Tu,“藍芽低功耗介紹,”hackmd, https://hackmd.io/@ShenTengTu/H1aUE5ltV?type=view, Apr.2019(accessed Mar. 2024)
[5]K. Townsend,“Introduction to Bluetooth Low Energy,”Adafruit, https://learn.adafruit.com/introduction-to-bluetooth-low-energy/introduction, Mar. 2014(accessed Mar. 2024)
[6]M. Afaneh,“Bluetooth GATT: How to Design Custom Services & Characteristics [MIDI device use case],”novelbits, https://novelbits.io/bluetooth-gatt-services-characteristics/, Oct. 2023(accessed Apr. 2024)
[7]H. C. C. Yoon, J. Cho and Y. W. Kim,“CoAP over BLE-GATT for OCF,”presented at the 2017 International Conference on Information and Communication Technology Convergence (ICTC), Jeju, Korea (South), 2017.
[8]賴冠廷,“藍牙低功耗網狀網路路由設計與應用,”國立雲林科技大學, 2022.
[9]M. Woolley,“The Bluetooth® Mesh Primer,”Bluetooth, https://www.bluetooth.com/bluetooth-mesh-primer/, Jan. 2019(accessed May. 2024)
[10]M. Woolley,“Bluetooth Direction Finding,”Bluetooth, https://www.bluetooth.com/wpcontent/uploads/Files/developer/RDF_technical_Overview.pdf,Oct. 2021 (accessed May. 2024)
[11]E. Essa, B. A. Abdullah and A. Wahba,“Improve Performance of Indoor Positioning System using BLE,”presented at the 14th International Conference on Computer Engineering and Systems (ICCES), Cairo, Egyp, 2019.
[12]D. G. and P. P. M. Malajner,“Indoor AoA Estimation Using Received Signal Strength Parameter and a Support Vector Machine,”presented at the 2019 International Conference on Systems, Signals and Image Processing (IWSSIP), Osijek, Croatia, 2019.
[13]F. Mavilia, P. Barsocchi,F. Furfari and M. Girolami,“Evaluating the Impact of Anchors Deployment for an AoA-based Indoor Localization System,”presented at the 18th Wireless On-Demand Network Systems and Services Conference (WONS), Italy, 2023.
[14]M. S. Y. Zheng, J. Liu and J. Li,“Exploiting AoA Estimation Accuracy for Indoor Localization: A Weighted AoA-Based Approach,”IEEE Wireless Communications Letters, vol. 8, no. 1, pp. 65-68, 2019, doi: 10.1109/LWC.2018.2853745.
[15]Y. Shuai, Y. Zheng and H. Huang,“Hybrid Software Obsolescence Evaluation Model Based on PCA-SVM-GridSearchCV,”2018 IEEE 9th International Conference on Software Engineering and Service Science, pp. 449-453, 2018.
[16]陳晏羚,“基於藍牙網狀網路技術智慧型燈具控制物聯網系統之設計與實作,”國立虎尾科技大學, 雲林縣, 2021.
[17]魏嘉儀,“以藍牙測向為基礎室內定位系統之設計與實作,”國立虎尾科技大學, 雲林縣, 2022.
[18]李育瑋,“於藍牙網狀網路以機器學習改善室內定位穩定性方法之設計與分析,” 國立虎尾科技大學, 雲林縣, 2022.
[19]T. Rithanasophon and C. Wannaboon,“Integration of Machine Learning and Kalman Filter Approach for Fingerprint Indoor Positioning,”2023 20th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology (ECTI-CON), Nakhon Phanom, Thailand, 2023, pp. 1-4.
[20]D. Zhu and J. Yan,“A Deep Learning Based Bluetooth Indoor Localization Algorithm by RSSI and AOA Feature Fusion,”2022 International Conference on Computer, Information and Telecommunication Systems (CITS), Piraeus, Greece, 2022, pp. 1-6.
[21]J. Wang, J. G. Hwang, J. Peng, J. Park and J. G. Park, "Gaussian Filtered RSSI-based Indoor Localization in WLAN using Bootstrap Filter," 2021 International Conference on Electronics, Information, and Communication (ICEIC), Jeju, Korea (South), 2021, pp. 1-4.