請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/79086
標題: | 次世代蜂巢網路下基於無線訊號場域圖之效能診斷系統 A REM-enabled Diagnostic Framework for Next Generation Cellular Networks |
作者: | Hsiu-Wen Yen 顏修溫 |
指導教授: | 逄愛君 |
關鍵字: | 蜂巢式網路,群眾外包,效能分析,無線訊號場域圖,機器學習,空間性統計, Cellular Networks,Crowdsourcing,Performance Profiling,Radio Environment Map,Machine Learning,Spatial Statistics, |
出版年 : | 2018 |
學位: | 碩士 |
摘要: | 近年來行動裝置全面普及、手機應用程式的市場持續擴大,使得我們對於行動數據的需求量呈現指數性上升。為了能夠填補服務上的短缺,小型基地台被視為一種有效提升覆蓋範圍與網路容量的解決方案。由於大部分的行動數據流量產生於室內區域,所以現今多數的小型基地台被部署在室內中,稱之為家庭基站(Femtocell)。密集地建設小型基地台確實提升網路系統的服務能力,但卻也伴隨著一些負面隱憂。家庭基站之部建缺乏預先規劃、室內訊號傳播較為複雜等等因素可能導致室內環境存在嚴重的覆蓋與容量問題。因此電信營運商需要妥善地處理這些負面問題,才能夠真正發揮密集建設的功效。於本研究中,我們提出一個數據驅動之效能診斷系統。其運用機器學習技法去分析用戶所回傳之資訊。經妥善訓練之預測模型能建立目標環境的無線訊號場域圖(Radio Environment Map),用以作為後續之診斷與管理用途。我們從辦公室場域中蒐集真實數據以驗證室內診斷之可行性。接著執行一系列的實驗去衡量7種資料分析方法的效能,分別從精確度、時間成本與資料需求量之角度進行討論。結果顯示隨機森林(Random Forest)是較適合用於本診斷系統的演算法。 With the inevitable growth of mobile application markets, wireless traffic demands have been exponentially increasing in recent years. Small cells are viewed as a promising solution to deal with the resource shortages over next generation cellular networks. Since most data traffic is generated indoors by users, a majority of small cells nowadays are installed indoors, called femtocells. However, unplanned femtocell deployment and complex interior layouts of buildings may lead to severe coverage and capacity problems in indoor environment. Therefore, it is crucial to address such negative impacts to truly benefit from the dense deployment. In this work, we aim to design a data-driven diagnostic framework for fault detections. The framework utilizes various popular machine learning techniques to analyze crowdsourced measurements which are uploaded from network subscribers. The well-trained prediction models can be used to construct a global look of a radio environment, namely radio environment map (REM), for diagnosis and management purposes. Moreover, REMs enable operators to detect existing coverage and capacity problems at any chosen location. We have collected real traces from an indoor testbed to verify the feasibility of indoor diagnoses. Then, a series of experiments is conducted to evaluate the performance of 7 statistical algorithms in terms of the accuracy, the time cost and the sensitivity to data volumes. Experimental results show that random forest algorithm is the most suitable method for the diagnostic procedure in all aspects. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/79086 |
DOI: | 10.6342/NTU201801539 |
全文授權: | 有償授權 |
電子全文公開日期: | 2023-08-23 |
顯示於系所單位: | 資訊工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-107-R05922034-1.pdf 目前未授權公開取用 | 8.33 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。