提升 5G 與 B5G 使用者體驗之AI 策略;AI Strategies for Enhancing QoE in 5G and Beyond 5G

NCU Institutional Repository > 資訊電機學院 > 資訊工程研究所 > 博碩士論文 > Item 987654321/98225

請使用永久網址來引用或連結此文件: https://ir.lib.ncu.edu.tw/handle/987654321/98225

題名:	提升 5G 與 B5G 使用者體驗之AI 策略;AI Strategies for Enhancing QoE in 5G and Beyond 5G
作者:	高曉雯;Kao, Hsiao-Wen
貢獻者:	資訊工程學系
關鍵詞:	第五代行動通訊;使用者體驗;人工智慧/機器學習;資料驅動;超可靠低延遲通訊;行動切換;5G network;quality of experience (QoE);AI/ML;data-driven;Ultra-Reliable and Low Latency Communications (URLLC);handover
日期:	2025-07-12
上傳時間:	2025-10-17 12:30:58 (UTC+8)
出版者:	國立中央大學
摘要:	5G (5th-Generation Mobile Network) 不僅促成了許多即時寬頻應用實現無線化，如虛擬實境與雲端遊戲，更推動了諸如車載即時影音通訊等新型行動服務的興起。5G 及其邁向後 5G (Beyond-5G, B5G) 及 6G (6th-Generation Mobile Network) 的核心願景之一，即是實現「以人為中心」的通訊，不僅支援多元化服務，亦致力於提升使用者的體驗品質（Quality of Experience, QoE）。然而，無線網路於資料傳輸過程所遭遇之通道時變性使得在 5G 系統中對 QoE 的掌控仍面臨顯著挑戰。為因應此問題，本論文提出兩項以人工智慧（Artificial Intelligent, AI）為基礎的策略，旨在於 5G 與其後續網路中提升 QoE 水準。所有策略均基於實際場景下於台灣商用 5G 網路中、以駕駛環境所蒐集之真實資料進行設計與驗證。首項提出的 AI 策略是一個兩層式的創新邏輯回歸（Logistic Regression）演算法模型，用於即時預測影片應用中的 QoE 降低情形。該架構透過監控網路層的無線訊號品質，經由 AI 模型推論使用者的觀看體驗使否出現劣化，並將此模型與作者提出的可持續運作 QoE 架構結合，利用 5G 核網中的網路資料分析功能 (Network Data Analytics Function, NWDAF) 為中心，介接 5G 傳輸通道品質資料收集及 5G 資源調度功能元件，如網路切片 (Network Slicing)，做到即時收集數據、推論、動態調整網路資源、改善 QoE 品質的自主循環運作。該架構在 AI 模型訓練階段實現了跨域資料處理—結合網路層的訊號品質指標與應用層的觀賞品質指標進行訓練，於 AI 模型上線進行推論的階段，依應用服務需求調整網路資源，從而有效支援 QoE 敏感型應用、促進新商業模式的發展，由於依需調用網路資源，也達到了提升能源使用效率的效益。第二項 AI 策略聚焦於提升用戶在 5G 網路移動過程中，進行基地台切換的連線品質穩定，這對於支援 5G 時代的新型態應用—超可靠與低延遲通訊（Ultra-Reliable Low-Latency Communications, URLLC）於行動場景中使用至關重要。傳統利用多條連線重複傳送資料封包是有效提升通訊可靠度的方法，但也造成額外的頻寬與能源消耗。為解決此問題，本研究設計了一組連線品質可靠性預測 AI 模型，於基地台切換過程中，預估來源鏈路 (Source Link) 與目標鏈路 (Target Link) 在接續的一段時間區間內連線品質下降的可能性，並據此做出是否建立資料包複製的決策。本研究同時針對 5G 條件式切換（Conditional Handover, CHO）方法，針對其應用於多無線接取技術的雙重連結（Multi-Radio Dual Connectivity, MR-DC）網路架構提出演進框架，結合上述 AI 模型以提供 URLLC 應用在基地台切換過程中仍可維持高可靠度的通訊。總體而言，本文所提出的 AI 驅動策略係從傳統網路層性能導向轉向以使用者為核心的 QoE 最適化，為未來智慧化行動服務的實現奠定基礎。 ;The advent of 5G (5th-Generation Mobile Network) networks has not only enabled the wireless transformation of numerous real-time broadband internet services—such as virtual reality (VR) and cloud gaming—but has also fostered the emergence of novel, mobility-native applications, including in-vehicle real-time video communications. A key vision of 5G and beyond-5G (B5G) networks is to deliver human-centric communication [1], supporting not only a wide range of service types but also enhancing the Quality of Experience (QoE) for end users. However, due to the inherent time-varying nature of wireless environments, ensuring stable and controllable QoE remains a significant challenge in 5G networks. This dissertation addresses this issue by proposing a set of strategies driven by Artificial Intelligent (AI) to improve QoE performance within 5G and B5G ecosystems. These approaches are developed and validated using empirical data collected from real-world driving scenarios over a commercial 5G in Taiwan. The first proposed strategy is an innovative QoE prediction framework based on logistic regression, which detects QoE degradation in video applications by monitoring wireless signal quality. A corresponding QoE sustainability architecture is also introduced, integrating the predictor with advanced 5G network components such as the Network Data Analytics Function (NWDAF). This architecture enables real-time cross-layer data collection and facilitates dynamic network resource adjustments in response to application demands—thereby supporting QoE-sensitive services, fostering business innovation, and improving energy efficiency. The second strategy focuses on enhancing connection reliability during handovers, which is critical for supporting 5G Ultra-Reliable Low-Latency Communications (URLLC) applications. While packet duplication is effective for improving reliability during unstabled communication conditions, it incurs considerable bandwidth and energy overhead. To address this, a reliability-predicting AI model is developed to assess the reliability degradation probability of source links and target links during handover, enabling reliability-aware packet duplication decisions. This predictor is further integrated with the proposed handover architecture—an AI-driven conditional handover (CHO) method designed for 5G Multi-Radio Dual Connectivity (MR-DC) networks. Although reliability of connection is a network-level quality of service (QoS) metric, our approach also associates the reliability degradation threshold with the application-level latency, which means the method can also assist to keep the user experienced latency. Overall, the proposed AI-driven approaches shift the focus from traditional network-centric performance metrics toward user-centric QoE optimization, laying the foundation for more intelligent mobile service delivery and efficient network resource management.
顯示於類別:	[資訊工程研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	13	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....