Caching Placement and Resource Allocation for AR Application in UAV NOMA Networks

The cache-enabling unmanned aerial vehicle (UAV) cellular networks with massive access capability supported by non-orthogonal multiple access (NOMA) are investigated in this paper. The delivery of multi-media contents for the mixed augmented reality (AR) and normal multi-media application is assisted by multiple mobile UAV base stations, which cache popular contents for wireless backhaul link traffic offloading. To cope with the dynamic content requests and mobility of users in practical scenarios, the dynamic optimization problem for user association, caching placement of UAVs, real-time deployment of UAVs, and power allocation of NOMA is modeled as a stackelberg game to minimize the long-term content delivery delay. Specifically, the game is decomposed into a leader level problem and a number of follower level problems. A correction mechanism is added in deep reinforcement learning (DRL) to optimize the user association in leader level. A meta actor network is proposed in DRL to jointly optimize the UAVs caching placement, real-time UAVs deployment and power allocation of NOMA in follower level. Then, a dynamic caching placement and resource allocation algorithm based on multi-agent meta deep reinforcement learning is proposed to minimize the long-term content delivery delay. Finally, we demonstrate that the considerable gains are achieved by the proposed algorithm.