Deep Reinforcement Learning–Enhanced Non-Orthogonal Multiple Access Framework for UAV-Assisted Terahertz 6G Communication Networks: A Large-Scale Monte Carlo Simulation Study

Abstract

This paper presents a Deep Reinforcement Learning (DRL)-enhanced Non-Orthogonal Multiple Access (NOMA) framework for UAV-assisted Terahertz (THz) 6G communication networks. The proposed framework introduces three key novelties: (i) a K-means clustering algorithm for 300-node UAV swarm coordination that reduces inter-cluster interference through intelligent cluster-head selection; (ii) an adaptive NOMA user-pairing strategy based on real-time channel-quality estimation that dynamically assigns weak and strong users to optimal power-allocation levels; and (iii) a Q-learning-based distributed power-control mechanism that continuously optimizes UAV transmission power in response to time-varying THz channel conditions and network load. A large-scale Monte Carlo simulation (M = 10 runs, 200 rounds, 3000 × 3000 m² deployment area) validates the proposed scheme against traditional NOMA and Orthogonal Multiple Access (OMA) benchmarks. Results demonstrate that the proposed DRL-NOMA achieves an average system throughput of 185.4 Mbps — a 29.5 % improvement over traditional NOMA and a 124.2 % gain over OMA — while maintaining a Packet Delivery Ratio (PDR) of 0.8812 and a Jain’s Fairness Index of 0.9321. Energy-efficiency and outage-probability analyses further confirm the superiority and practical deployability of the proposed scheme for next-generation heterogeneous 6G networks.