基於到達角使用深度學習觀察定位效能;Study on Performance of Localization Based on Angle of Arrival with Deep Learning Algorithms

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Communication Engineering > Electronic Thesis & Dissertation > Item 987654321/86452

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/86452

Title:	基於到達角使用深度學習觀察定位效能;Study on Performance of Localization Based on Angle of Arrival with Deep Learning Algorithms
Authors:	蘇博威;Su, Po-Wei
Contributors:	通訊工程學系
Keywords:	到達角;三角測量法;最小平方法;深度神經網路;自動編碼器;均方根誤差;AOA;Triangulation;Least Square method;Deep Neural Network;Auto Encoder;RMSE
Date:	2021-10-06
Issue Date:	2021-12-07 12:51:19 (UTC+8)
Publisher:	國立中央大學
Abstract:	無線定位(Wireless Localization)的技術隨著通訊的發展，受到越來越多人的關注，由於接收到的訊號資訊不同，分析方式也不同，根據基地台傳送訊號到達的角度(Angle of Arrival,\ AOA)，再利用三角測量技術(Triangulation)或是最小平方法(Least Square method)達到估計用戶端位置。本篇論文，我們提出深度學習演算法來處理定位估計。在深度學習演算法中使用的是深度神經網路(Deep Neural Network,\ DNN)，針對二維平面空間進行分類，三角測量技術與最小平方法主要的問題就是當多個基地台(Base Stations, \ BS)接收到的角度是相近的就容易造成線性估計的誤差增加，而深度學習演算法透過學習特徵的方式使用非線性系統分類進而改善定位估計的效能。但在高密度類別的情況下，深度神經網路分類的正確率是有限的，因此透過自動編碼器(Auto Encoder,\ AE)結合深度神經網路改善此問題，自動編碼器的概念像是根據學習資料的結構特徵，先區分相同類型的數據，再將這些數據根據特徵細分，這有助於減輕後續深度神經網路估計用戶端位置的負擔，最後在結果分析與比較中，可以看出此方法在穩定性和精確性都有不錯性能。;Thanks to the development of modern communication technologies, localization of end-user with wireless base stations has drawn more and more attention. Subject to the received signals to be utilized, the employed localization method is different. According to the angle-of-arrival (AOA) of the signal received by the base station equipped with multiple antennas, conventional methods such as triangulation and least squares methods can be used to determine the position of the end-user. In this thesis, we study the approach of deep learning methods. Here, the deep neural network (DNN) is revised to perform spatial classification on a two-dimensional plane. The main problem of triangulation and least square method is that when the angles of arrival at multiple base stations are similar, a large localization error will increase. By contrast, the deep learning methods improve the performance of position estimation at learning features through non-linear system classification. However, in the case of high-density classification categories, the accuracy of the conventional neural network method is limited. Therefore, the autoencoder is combined with the neural network to solve this problem. The autoencoder acts like a group of spatial filters, decomposing the input into multiple smaller spatial sub-regions. The range of the input covered by each spatial sub-region is narrower than that of the original input ,and hence, the distribution of the input is better centralized for classification. From simulation analysis and performance comparison, the autoencoder method achieves better performance than the LS and triangulation methods and lower complexity than the DNN method.
Appears in Collections:	[Graduate Institute of Communication Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	95	View/Open

社群 sharing

Loading...