DDPG-driven RIS-assisted Data Acquisition with a Solar-Powered Multi-Rotor UAV  

Abstract: The unmanned aerial vehicle (UAV) serves as a mobile data collector aiding the internet of things (IoT) in data acquisition. However, in urban environments, the communication link between UAV and IoT devices faces obstacles due to buildings, resulting in seriously compromised data freshness. To tackle this issue, the reconfigurable intelligent surface (RIS), comprising numerous low-cost reflective elements, emerges as a promising and economical solution. Additionally, the limited battery capacity of the UAV significantly impacts network performance. We propose a RIS-assisted solar-powered drone optimization (RSDO) framework, encompassing two primary stages: RIS phase shift optimization and decision-making on drone resource allocation. This allocation includes transmission power, flight speed, and solar energy harvesting mechanisms. Traditional methods are challenging to use for complex and dynamic environments. To overcome this challenge, we employ deep deterministic policy gradient (DDPG) techniques to optimize decision-making, considering constraints such as limited battery capacity and task execution time, aimed at minimizing total task execution time. Simulation results underscore the superiority of our proposed scheme over baseline methods.

多無人機通訊網路之效率與安全：基於輕量區塊鏈的聯邦學習 

摘要 —本研究開發一種專門為多無人機(Unmanned Aerial Vehicle, UAV)設計的創新聯邦學習(Federated Learning)架構，利用區塊鏈(Blockchain)技術實現輕量級、安全和高效的數據處理。在這個架構中，無人機直接參與聯邦學習，無需中央服務器，而是透過一系列基地台執行區塊鏈節點來處理梯度訊息和維護系統。我們透過將梯度聚合從中央服務器轉移到區塊鏈節點來消除單點故障，以提高系統的穩定性和可靠性。此外，利用私有星際文件系統（Inter Planetary File System, IPFS）降低了存儲和通訊成本。並採用拜占庭容錯（Byzantine Fault Tolerance, BFT）共識機制以提高系統效率和安全性。實驗證實，即使在惡意區塊鏈節點占多數的情況下，我們提出的架構亦能有效防止模型中毒攻擊，確保學習效能。最後，引入基於Multi-Krum算法的節點聲譽(Reputation)選擇機制，以提升節點提交高質量梯度的機率，從而增強整體系統效能。