請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/83907
標題: | 應用機器學習於高速訊號穩定度之智慧檢測系統 Smart Detection System for High-Speed Signal Stability Using Machine Learning |
作者: | 吳冠廷 Kuan-Ting Wu |
指導教授: | 蕭浩明 Hao-Ming Hsiao |
關鍵字: | 訊號補償,訊號量測,電路板驗證,機器學習, signal equalization,signal measurement,validation of printed circuit board,machine learning, |
出版年 : | 2022 |
學位: | 碩士 |
摘要: | 隨著微處理器的效能提升,高速訊號的資料傳輸速度要求也越來越高,然而高速訊號速度的提升往往也會增大符碼間干擾的問題,造成訊號傳遞錯誤的情況產生。目前最主要的解決辦法便是在訊號傳送通道的兩端設置訊號補償器,為訊號提供補償以降低訊號失真的情況,針對每個通道的特性,訊號補償器的參數設定會大不相同,參數的好壞也就因此大幅的影響通訊系統的好壞,故需要設計相關的演算法搜索最佳的參數值,以確保穩定的通訊系統做後續功能性的測試,然而近年來訊號線的數量增加以及越趨複雜的補償器設計都增加了最佳參數的搜索時間。 近年來機器學習的相關技術發展相當快速,透過大數據的幫助,在各個領域都能夠協助研發人員解決以前所面臨的各式難題,故本研究提出一透過機器學習模型結合電路板電氣特性達到快速預測訊號表現的方法。 本研究所提出的預測模型共可以分為兩個部分,系統資訊之蒐集以及迴歸預測,在系統資訊蒐集的階段,會透過向量網路分析儀以及時域反射量測法兩種量測手法測量電路板各訊號線的阻抗、頻域響應等資料並做特徵擷取,另外也設計了訊號餘裕實驗測量每條訊號線在不同的訊號補償器設置下的訊號表現,將上述的兩種資料整合後,在迴歸預測的部分,會做為訓練資料用以訓練各種機器學習模型,使得模型能夠根據訊號線的電氣特性預測出不同訊號補償器參數下的訊號餘裕值。經過本研究的相關實驗與分析,成功地訓練出一能在短時間內預測訊號餘裕的模型,透過此方法便能夠快速地降低訊號補償器最佳參數地搜索時間,降低電路板開發、驗證的時間。 With the improvement of the performance of microprocessors, the data transmission speed of high-speed signal is also getting higher and higher. However, the improvement of high-speed signal speed often aggravate the problem of inter-symbol interference, resulting in the occurrence of signal transmission errors. The main solution is to set up signal equalizers at both ends of the signal transmission channel to provide compensation for the signal to reduce signal distortion. According to the characteristics of each channel, the parameter settings of the signal equalizers are very different. The parameter settings therefore greatly affect the quality of the communication system. Thus, it is necessary to design algorithms to search for the best parameter to ensure a stable communication system for subsequent functional validation tests. However, recently, the number of signal lines has increased and more complex signal equalizer designs lead to greater search time for optimal parameters. In recent years, the technologies of machine learning have developed rapidly. With the help of big data, several difficult problems in various fields could be solved. Therefore, this study proposes a machine learning model combined with the electrical characteristics of the circuit board to predict signal performance. The prediction model proposed in this study can be divided into two parts, the collection of system information and the regression prediction. First, the impedance, and frequency response of each signal line on the circuit board are measured by Vector Network Analyzer (VNA) and Time Domain Reflectometry (TDR). The feature vectors are then extracted for following training. In addition, a signal margin validation test is designed to measure the signal margin of each signal line under different signal equalizer settings. After integrating the data, they are used to train various machine learning models. The models can predict the signal margin under different signal equalizer parameters according to the electrical characteristics of the signal line. After the relevant experiments and analysis, a model that can predict the signal margin in a short time has been successfully trained. This proposed methodology can reduce the search time for the optimal parameters of the signal compensator, and thus reduce the development and validation time of printed circuit board. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/83907 |
DOI: | 10.6342/NTU202201407 |
全文授權: | 未授權 |
顯示於系所單位: | 機械工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-110-2.pdf 目前未授權公開取用 | 3.69 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。