本論文提出命名為FPFE (Fingerprint Feature Extraction)及FPFE-C(Fingerprint Feature Extraction with Calibration)的室內定位 (indoor localization)方法。這二個方法利用參考點(reference point)的低耗能藍芽(Bluetooth Low Energy, BLE) 信標指紋(beacon fingerprint)進行定位，會在室內環境中佈置4個以上的信標節點(beacon node)，而這些信標節點會週期性持續送出廣告(advertisement)封包。我們可以從不同已知位置的參考點接收這些廣告封包並記錄其接收訊號強度(Received Signal Strength Indicator, RSSI)，將這些接收訊號強度當作每個參考點的個別信標指紋，做為定位資料之用。FPFE定位法先使用自動編碼器或主成份分析進行信標指紋特徵擷取，然後再進行參考點與位置未知之目標點(target point)的特徵相似程度比較，挑選出特徵相似度較高的參考點之後以其位置平均權重計算出目標點的位置。每次進行定位時，可能受到溫度、濕度、場地不同與設備不同等各種環境因素影響，因此FPFE-C定位法另外挑選若干參考點為校正點(calibration point)，透過每次重新測量校正點的信標指紋，並計算其與相對應參考點原始信標指紋的比值之後再對定位資料進行校正，達到更好的定位精準度。若其他使用者欲使用本論文所提的定位方法時，便可以不必再收集所有參考點的信標指紋，只需透過FPFE-C方法所使用的校正方式，微調定位資料即可。本研究以長為8公尺，寬為5公尺的範圍作為實驗區域，在區域四周佈置信標節點，透過大量實驗資料評估所提方法的定位精準度。實驗結果顯示，FPFE定位法的平均定位誤差為0.68公尺；而在具有環境變化影響之校正定位實驗中，FPFE定位法的平均定位誤差為2.13公尺，當加上校正點後之FPFE-C定位法之平均定位誤差為1.63公尺，下降了0.4公尺。

The study proposes two indoor localization methods named FPFE (Fingerprint Feature Extraction) and FPFE-C (Fingerprint Feature Extraction with Calibration). These two methods use Bluetooth Low Energy (BLE) beacon fingerprints of reference points for the purpose of localization. Four or more beacon nodes are deployed in an indoor environment which periodically and continuously broadcast advertisement packets. We can measure Received Signal Strength Indicator (RSSI) values of packets of different beacon nodes for every reference point and take the combination of the RSSI values as the beacon fingerprint of the reference point with a known position. Similarly, we can obtain the beacon fingerprint of a target point with an unknown position. The FPFE method first uses the autoencoder or the principal component analysis (PCA) to extract features of beacon fingerprints, and then calculate the Minkowski distances between the feature of the target point and the features of all reference points. The FPFE method then selects k reference points with the k smallest Minkowski distances and use their positions to estimate the target point position. The FPFE-C method also uses the similar concept adopted by the FPFE method for positioning. However, the FPFE-C method additionally considers dynamically changing environmental factors. It chooses few reference points as calibration points to re-measure their beacon fingerprints periodically. The re-measurement is also conducted when there are drastic environmental changes, such as the change of packet-receiving devices, significant changes of surrounding temperature and humidity, or and even when the whole localization system is moved and applied in a brand-new place. The average ratios of beacon fingerprints of calibration points and corresponding reference points are used to adjust all beacon fingerprints. This study takes an 8 m by 5 m region with four beacon nodes deployed at four corners for evaluating the performance of the proposed methods. With 187 reference points, the FPFE method achieves the average positioning error of 0.68 m. For the scenario with different packet receiving devices, the FPFE-C method reduces the positioning error from 2.13 m to 1.63 m with 12 calibration points.

中文摘要 VI
Abstract VIII
誌謝 X
目錄 XI
圖目錄 XIV
表目錄 XVI
一、 緒論 1
1.1 研究背景與動機 1
1.2 研究目的 2
1.3 論文結構 2
二、 背景知識 3
2.1 Bluetooth 3
2-1-1 Bluetooth 4.0 3
2-1-2 Bluetooth 4.0後版本 6
2-1-3藍牙廣播傳送方式 6
2.2 Beacon 7
2-2-1 概述 7
2-2-2 技術原理 8
2.3 訊號衰減 9
2-3-1 弗里斯傳輸方程式 10
2-3-2 環境擾動因素 11
2-3-3 RSSI 訊號衰減與距離關係 14
2.4 訊號濾波器 14
2-4-1 均值濾波器 14
2-4-2 中值濾波器 15
2-4-3 卡爾曼濾波器 15
2.5 深度學習 16
2-5-1 類神經網路 16
2-5-2 深度學習介紹 20
2-5-2-1 監督式學習 22
2-5-2-2 非監督式學習 23
2-5-2-3 半監督式學習 23
2-5-3 多層感知器 23
2-5-4 自動編碼器 24
2.6 特徵擷取 26
2-6-1 主成分分析 26
2-6-2 自動編碼器做特徵擷取 28
2.7 相似性度量 29
2-7-1 歐氏距離 29
2-7-2 曼哈頓距離 29
2-7-3 閔可夫斯基距離 30
2.8 定位技術 31
2-5-1 接受訊號強度測距法(RSSI) 32
2-5-2 到達時間測距法（Time Of Arrival，TOA） 32
2-5-3 到達時間差測距法（Time Difference Of Arrival，TDOA） 33
2-5-4 角度到達測距法（Angle Of Arrival ，AOA） 34
2-5-5 質心定位法 35
2-5-5 指紋定位法(Fingerprinting localization) 35
2.9 相關文獻研究 36
三、 研究方法 37
3.1 系統流程 37
3.2 定位方法 38
3-2-1 布置方式 38
3-2-2 收集與建立指紋資料庫 40
3-2-3 指紋資料之前處理 40
3-2-4 特徵擷取之模型訓練 41
3-2-5 收集校正資料 42
3-2-4 特徵擷取且校正指紋定位法(FPFE-C localization method) 43
四、 實驗及結果 45
4.1 實驗介紹 45
4-1-1 實作硬體設備 45
4-1-2 實作軟體設備 47
4-1-3 實驗場景介紹 48
4.2 FPFE定位法實驗結果 50
4-2-1 不同候選參考點數f之筆較 50
4-2-2 特徵擷取維度不同之比較 52
4-2-3 不同相似性度量之比較 52
4.3 FPFE-C定位法實驗結果 52
4-3-1 無校正經FPFE定位法實驗結果 53
4-3-2 經校正後FPFE-C定位法之實驗結果 53
五、 結論和未來展望 56
參考文獻 57

[1] Bluetooth Technology Website, “Bluetooth”, July 28, 2016 https://www.bluetooth.com/
[2] Shyuanliang blogspot. “Bluetooth 4.0”, July 28, 2016 http://shyuanliang.blogspot.tw/2014_10_01_archive.html
[3] Apple, “iBeacon” , July 28, 2016 https://developer.apple.com/ibeacon/
[4] iBeacon work method , http://d10pb0rwjcag47.cloudfront.net/wp-content/uploads/2014/04/Rover-LabsiBeacon.jpg
[5] Antenna-theory, “Friis Transmission Equation”, July 28, 2016 http://www.antenna-theory.com/basics/friis.php
[6] S.Viswanathan and S.Srinivasan, " Improved path loss prediction model for short range indoor positioning using bluetooth low energy.", in IEEES ENSORS, 2015
[7] M.Varsamou and T.Antonakopoulos. "A bluetooth smart analyzer in iBeacon networks", in IEEE Consumer Electronics??? Berlin (ICCE-Berlin), 2014
[8] K.Benkic, et al, "Using RSSI value for distance estimation in wireless sensor networks based on ZigBee, " in Systems, signals and image processing, 15th international conference on. IEEE, 2008
[9] 神經元介紹 http://www.hkpe.net/hkdsepe/human_body/neuron.htm
[10] Yang, Y. C., & Jiang, J. R. (2019, October). Web-based Machine Learning Modeling in a Cyber-Physical System Construction Assistant. In 2019 IEEE Eurasia Conference on IOT, Communication and Engineering (ECICE) (pp. 478-481). IEEE.
[11] Wiki ,Sigmoid https://zh-yue.wikipedia.org/wiki/Sigmoid_%E5%87%BD%E6%95%B8
[12] Tanh function, http://c.biancheng.net/cpp/html/194.html
[13] Wiki, ReLU https://zh.wikipedia.org/wiki/%E7%BA%BF%E6%80%A7%E6%95%B4%E6%B5%81%E5%87%BD%E6%95%B0
[14] Tensorflow’s MLP https://www.itread01.com/p/447163.html
[15] Autoencoders https://ithelp.ithome.com.tw/articles/10195480
[16] 淺談降為方法中的PCA與t-SNE https://medium.com/d-d-mag/%E6%B7%BA%E8%AB%87%E5%85%A9%E7%A8%AE%E9%99%8D%E7%B6%AD%E6%96%B9%E6%B3%95-pca-%E8%88%87-t-sne-d4254916925b
[17] 莫煩PYTHON 自編碼(Autoencoder), https://morvanzhou.github.io/tutorials/machine-learning/ML-intro/2-5-autoencoder/
[18] 百科知識, 曼哈頓距離 https://www.easyatm.com.tw/wiki/%E6%9B%BC%E5%93%88%E9%A0%93%E8%B7%9D%E9%9B%A2
[19] K.Benkic, et al, "Using RSSI value for distance estimation in wireless sensor networks based on ZigBee, " in Systems, signals and image processing, 15th international conference on. IEEE, 2008
[20] Y.Jingjing, W.Zhihui and Z.Xiao, “RSSI Based Bluetooth Low Energy Indoor
Positioning”, in IEEE International Journal of Smart Home 2015
[21] M.S.Gast, “Building applications with IBeacon: proximity and location services with bluetooth low energy”, in O'Reilly Media, Inc., 2014.
[22] R.Mehra and A.Singh, " Review of Angle of Arrival (AOA) Estimations Through Received Signal Strength Indication (RSSI) for Wireless Sensors Network (WSN)", in IEEE Advance Computing Conference (IACC), 2013
[23] M.I.Jais, P.Ehkan, R.B.Ahmad and I.Ismail, “Review of Angle of Arrival (AOA) Estimations Through Received Signal Strength Indication (RSSI) for Wireless Sensors Network (WSN) ” in IEEE International Conference on Computer, Communication, and Control Technology ,2015
[24] D.Quande and X.Xu, “A Novel Weighted Centroid Localization Algorithm Based on RSSI for an Outdoor Environment”, in IEEE Journal of Communications Vol. 9, No. 3, March 2014
[25] D.J.Sretenovic, S.M.Kostić, M.I.Simić, “Experimental Analysis of Weight Compensated Weighted Centroid Localization Algorithm Based on RSSI”, in IEEE TELSIKS 2015
[26] Ding, H., Zheng, Z., & Zhang, Y. (2016, November). AP weighted multiple matching nearest neighbors approach for fingerprint-based indoor localization. In 2016 Fourth International Conference on Ubiquitous Positioning, Indoor Navigation and Location Based Services (UPINLBS) (pp. 218-222). IEEE.
[27] Yang, L., Chen, H., Cui, Q., Fu, X., & Zhang, Y. (2015, May). Probabilistic-KNN: A novel algorithm for passive indoor-localization scenario. In 2015 IEEE 81st Vehicular Technology Conference (VTC Spring) (pp. 1-5). IEEE.
[28] Kuxdorf-Alkirata, N., Maus, G., & Brückmann, D. (2019, August). Efficient calibration for robust indoor localization based on low-cost BLE sensors. In 2019 IEEE 62nd International Midwest Symposium on Circuits and Systems (MWSCAS) (pp. 702-705). IEEE.
[29] Seeckoy, “iBeacon”, July 28, 2016 http://www.seekcy.com/