Deep Q Network-Enpowered Resource Allocation Optimisation in a Low-Altitude Wireless Caching Network

With the advantages of high flexibility and low cost, unmanned aerial vehicles (UAV)-assisted edge caching has emerged as a promising technology for future low-altitude wireless networks. This paper proposes a low-altitude single-cell base station wireless network, where multiple UAVs serve as edge caching nodes to offer caching services to different ground Internet of Things (IoT) users. Specifically, each UAV follows a Zipf-based content popularity model, and the requested content can be delivered from a UAV edge caching node to an IoT user through an air-to-ground (A2G) uplink, which includes both Line-of-Sight (LoS) and Non-Line-of-Sight (NLoS) components. In order to minimise the content delivery delay, while improving the caching hit ratio, we propose a deep reinforcement learning (DRL)-based joint optimisation algorithm, which jointly optimises the trajectories of each UAV, the transmission power, and the communication resource. Simulation results demonstrate that our solutions achieve 95% lower delay compared to traditional FDMA schemes while improving the cache hit rates by 60% at each UAV edge node.