| DC 欄位 |
值 |
語言 |
| DC.contributor | 通訊工程學系 | zh_TW |
| DC.creator | 賴韋州 | zh_TW |
| DC.creator | Wei-Chou Lai | en_US |
| dc.date.accessioned | 2021-1-6T07:39:07Z | |
| dc.date.available | 2021-1-6T07:39:07Z | |
| dc.date.issued | 2021 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=107523022 | |
| dc.contributor.department | 通訊工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 近年來物聯網設備在無線通訊使用率節節攀升，其所造成的能量消耗以及電力需求逐漸增加，能量獵取技術被提出以解決有限電池的電量問題。為了達到最大化的系統通道容量，傳統的凸優化方法受到電池容量和能量獵取時間因果性的限制必須假設系統知道未來所有時刻的能量獵取及通道增益等狀態資訊才能實現傳輸功率控制，這在實際情況下難以達成且具有高計算複雜度。因此本論文提出深度神經網路(Deep Neural Network, DNN)以及捲積神經網路(Convolutional Neural Network , CNN)學習過去能量獵取的傳輸模式來預測未來功率控制值，本文所提之方法在不需要未來能量獵取及通道增益等狀態資訊的情況下，能有效降低複雜度且達到即時應用。
本篇論文中，提出了兩種基於深度神經網路的功率控制方法，用於多用戶上鏈綠能無線通訊系統之功率控制以最大化系統的通道容量。在多用戶環境中，吾人使用凸優化(Convex Optimization)及歷年能量獵取資料產生之最佳功率控制值以及最佳電池狀態作為輸入，再讓多層感知器(Multilayer perceptron, MLP)以及捲積神經網路(Convolutional Neural Network , CNN)學習其輸入/輸出的關係。為了讓兩種神經網路進行學習，皆採用監督式學習，使用加權和最小均方誤差(Weighted-sum MMSE, WMMSE)作為監督式學習的標籤(Label)，並且比較兩種方法於最大通道容量效能之差異。模擬結果顯示本論文提出的方法不僅符合實際應用還能大幅降低計算複雜度並且在多用戶應用上都能近似於參考功率控制值的系統通道容量。 | zh_TW |
| dc.description.abstract | In recent years, the utilization rate of wireless communication of IoT devices has been steadily increasing, and the energy consumption and power demand caused by it have gradually increased. Energy harvesting technology has been proposed to solve the problem of limited battery power. In order to achieve the maximum system channel capacity, the traditional convex optimization method is limited by the causality of battery capacity and energy harvesting time. It must be assumed that the system knows the state information of energy harvesting and channel gain at all times in the future to achieve transmission power control. Under the circumstances, it is difficult to achieve and has high computational complexity. Therefore, this paper proposes Deep Neural Network (DNN) and Convolutional Neural Network (CNN) to learn the transmission mode of energy harvesting in the past to predict the future power control value. The method proposed in this paper does not require the future state information, such as energy harvesting and channel gain, it can effectively reduce complexity and achieve real-time applications. In this paper, two power control methods based on neural networks are proposed for power control of multi-user uplink green energy wireless communication systems to maximize the channel capacity. In a multi-user scenario, we use convex optimization and the best power control value generated by energy harvesting data over the years and the best battery state as input, and then let the Multilayer Perceptron (MLP) and convolutional neural network (CNN) learn its input/output relationship. In order to allow the two neural networks to learn, both use supervised learning, using weighted-sum MMSE (WMMSE) as the label, and compare the maximum capacity difference between DNN and CNN. The simulation results show that the method proposed in this paper not only conforms to the actual application, but also greatly reduces the computational complexity and can approximate the system channel capacity of the reference power control value in multi-user applications. | en_US |
| DC.subject | 能量獵取 | zh_TW |
| DC.subject | 凸優化 | zh_TW |
| DC.subject | 深度神經網路 | zh_TW |
| DC.subject | 多層感知器 | zh_TW |
| DC.subject | 捲積神經網路 | zh_TW |
| DC.subject | 加權和最小均方誤差 | zh_TW |
| DC.subject | Energy harvesting | en_US |
| DC.subject | Convex optimization | en_US |
| DC.subject | Deep Neural Networks | en_US |
| DC.subject | Multilayer perceptron | en_US |
| DC.subject | Convolutional Neural Network | en_US |
| DC.subject | Weighted-sum MMSE | en_US |
| DC.title | 基於神經網路之多用戶綠能上鏈無線通訊功率控制研究 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Neural Network-based Power Control for Multiusers Uplink Green Wireless Communications | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |