基於動態傳送頻率調控與調整連線配置之多用戶邊緣AI服務保障QoS機制;A QoS Assurance Mechanism for Multi-User Edge AI Services Based on Dynamic Frequency Adjustment and Connection Configuration

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/98271

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/98271

Title:	基於動態傳送頻率調控與調整連線配置之多用戶邊緣AI服務保障QoS機制;A QoS Assurance Mechanism for Multi-User Edge AI Services Based on Dynamic Frequency Adjustment and Connection Configuration
Authors:	林俞諠;Lin, Yu-Hsuan
Contributors:	資訊工程學系
Keywords:	擴增實境;邊緣運算;邊緣AI;服務品質;多使用者;Augmented Reality;Edge Computing;Edge AI;Quality of Service;Multi-user
Date:	2025-07-18
Issue Date:	2025-10-17 12:34:00 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著 AR 技術在博物館、美術館等展覽場合中的應用越來越熱門，AR 眼鏡成為提供沉浸式導覽與互動體驗的重要工具。現代AR眼鏡不僅需即時捕捉使用者視角中的影像，還需結合人工智慧（Artificial Intelligence, AI）技術進行物件辨識等處理，讓使用者的體驗更加豐富，然而為了維持輕量與可穿戴性，AR 眼鏡本身在計算能力上有所限制，難以執行複雜的深度學習推論任務。因此，邊緣運算（Edge Computing）成為近年來支援即時AR應用的重要架構，使用邊緣伺服器（Edge Server）來處理AR眼鏡所無法處理的計算工作，以滿足AR應用對低延遲的要求。然而由於邊緣伺服器的運算資源有限，在多個使用者同時使用的情況下，工作量超出邊緣伺服器所能處理的上限時，可能無法即時回傳運算結果，造成延遲增加以及使用者收到的每秒回應數量減少，進而導致使用者的體驗下降。為了解決這個問題，本研究提出了根據邊緣伺服器資源狀態與使用者人數動態地調整使用者對服務的傳送頻率與連線配置。此方法能夠讓邊緣伺服器在資源足夠的情況下能夠盡可能提供服務給更多使用者且讓所有使用者都獲得高於最低標準的服務品質。實驗結果顯示，系統根據本演算法調整後，和其他調整方法相比，在不同服務配置下可額外提供1至5名使用者使用服務且每名使用者均能夠獲得最低標準以上之服務品質，提升使用者數量25%至100%。;As augmented reality (AR) technology becomes increasingly popular in museum and exhibition settings, AR glasses have emerged as a key tool for delivering immersive guided experiences and interactive content. Modern AR glasses not only need to capture the user’s perspective in real time but also integrate artificial intelligence (AI) technologies to perform object recognition and enhance the user experience. However, in order to maintain lightweight and wearable designs, AR glasses are limited in computational capability and are often unable to execute complex deep learning inference tasks locally. To address this limitation, edge computing has recently become a crucial architecture for supporting real-time AR applications. By offloading computationally intensive tasks to edge servers located closer to the user, the system can meet the low-latency requirements of AR applications. Nonetheless, edge servers have limited processing resources, and when a large number of users access the system concurrently, the workload may exceed the server′s capacity. This can result in increased inference latency and reduced response frequency per user, ultimately degrading the overall user experience. To solve this issue, this study proposes a dynamic adjustment mechanism that adapts each user′s request frequency and service connection configuration based on the current resource state of the edge server and the number of active users. This approach allows the edge server to provide service to as many users as possible while ensuring that every user receives a level of service quality above a predefined minimum threshold. Experimental results show that, compared with other adjustment strategies, the proposed algorithm enables the system to accommodate an additional 1 to 5 users under different service deployment configurations, corresponding to a 25% to 100% increase in the number of supported users, while ensuring that each user receives a service quality above the minimum acceptable threshold.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	25	View/Open

社群 sharing

Loading...