无线VR场景下基于U-RIS的资源分配算法研究

his article investigates resource management in a mobile edge computing (MEC) network for wireless virtual reality (VR) applications, which is faced with challenges such as limited network resources and unstable transmission. A distributed resource allocation algorithm based on deep reinforcement learning is proposed to optimize the average latency, bandwidth, and utilization efficiency of computational resources for multiple VR gamers. MEC servers deployed in the VR gaming scenario render high-resolution and high-dynamic immersive VR game worlds based on real-time virtual coordinates of the VR users, and transmit rendered data via terahertz (THz) frequencies. In practical scenarios, the downlink transmission channel between the base station and the VR user may be affected by human body blockage due to the user\'s movements and head rotation, resulting in a decrease in transmission rate. To address this problem, the use of unmanned aerial vehicle mounted with reconfigurable intelligent surfaces (U-RIS) technology is introduced, which combines the flexibility of unmanned aerial vehicle (UAV) and the channel reconstruction capability of reconfigurable intelligent surfaces (RIS) to establish an auxiliary transmission link between the base station and the user, thereby improving the stability of the downlink channel. However, the real-time optimization effect of U-RIS on the user\'s channel is also affected by factors such as user movements and head rotation, as well as the directional gain of the THz receive antenna. Furthermore, considering the randomness of user movements and the sensitivity of VR applications to latency, optimizing the flight trajectory of U-RIS in three-dimensional space in the absence of sufficient user location information is a challenging task. To address this issue, a multi-level reinforcement learning model based on fast proximal policy optimization (PPO) and deep dueling Q-network (DDQN) is proposed in this article. Firstly, to solve the problem of U-RIS flight trajectory optimization without real-time user location information, the first level of the proposed model establishes a connection between the user\'s historical movement information and the real-time flight vector of U-RIS using fast PPO. Secondly, the next level of the model comprehensively analyzes the impact of U-RIS trajectory and the current remaining resources of the base station on user latency using DDQN, and provides a bandwidth and computational resource allocation decision for the user. Simulation results demonstrate that the proposed algorithm significantly improves the utilization of bandwidth and computational resources in the network, while reducing user latency.