Deep reinforcement learning for AI-driven successive ınterference cancellation in hybrid NOMA-UAV networks

Abstract In this paper, we propose a deep reinforcement learning (DRL)-based approach to maximize the efficacy of successive interference cancellation (SIC) in hybrid non-orthogonal multiple access (NOMA)-UAV networks. In fact, The proposed framework provides an integrated DRL-based optimization approach to achieve simultaneously high spectral efficiency, low latency, low power consumption, and low financial cost. The DRL model was trained for 250,000 episodes in real-time, interacting with the environment to learn the interference maps, motion of the UAVs, and load variations in the network. It is shown through simulations that DRL-optimized SIC significantly outperforms classical heuristic methods, with higher SIC successful rates, lower bit error rates, and enhanced network throughput. Results show that convergence occurred shortly after 150,000 episodes — a sufficient number of episodes to balance the complexity of the model with optimal performance. Moreover, the optimization of UAV mobility allows for coverage of mobile users, as well as minimizing Doppler impacts. This study approves the deployment of AI-driven SIC in practical scenarios, showing that AI technologies have the potential to transform the manner in which wireless communication takes place in NOMA-UAV networks and deliver performance while maximizing energy efficiency as well as cost-effectiveness.