Deep Reinforcement Learning based Dynamic Resource Allocation Method for NOMA in AeroMACS

To overcome the constraints posed by the scarcity of spectrum resources in the dedicated frequency band and the challenge of fulfilling real-time requirements across various services in civil airport surface operations, we propose a dynamic resource allocation method for airport communication system. This innovative approach is based on the non-orthogonal multiple access (NOMA) architecture. To account for variations in service priority among different entities on the surface, we design a multi-objective utility function that considers both transmission rate and service priority. We establish a joint optimization problem model for sub-channel allocation and power control in the scenario of airport uplink communication. Since the problem model exhibits non-convexity and highly coupled parameters, the multi-agent proximal policy optimization based on multi-discrete (MD-MAPPO) algorithm is introduced. Simulation results demonstrate that the NOMA architecture significantly improves the spectral efficiency of the airport communication system. Furthermore, our proposed algorithm effectively meets the requirements of multiple services by achieving dynamic and efficient wireless resource allocation, surpassing traditional reinforcement learning algorithms in terms of cumulative reward, convergence, and learning efficiency.