Multi-Agent Quantum Reinforcement Learning for Adaptive Transmission in NOMA-Based Irregular Repetition Slotted ALOHA

This paper proposes a Multi-Agent Quantum Deep Reinforcement Learning (MA-QDRL) framework to optimize uplink access in Irregular Repetition Slotted ALOHA with Non-Orthogonal Multiple Access (IRSA-NOMA) systems under practical constraints such as finite frame lengths and fading channels. IRSA enhances reliability by allowing devices to transmit multiple packet replicas across random time slots, while NOMA increases spectral efficiency through power-domain multiplexing with successive interference cancellation (SIC). As a benchmark, Contention Resolution Diversity Slotted ALOHA (CRDSA) improves traditional ALOHA through packet repetition and interference cancellation, maintaining solid performance under moderate network loads; however, its efficiency gradually declines under heavier traffic due to increased collisions and limited adaptability. Conventional multi-agent deep reinforcement learning (MA-CDRL) approaches have been explored to address coordination challenges in such environments. Nevertheless, these models often experience scalability limitations and unstable convergence as the number of agents increases. To overcome these challenges, MA-QDRL integrates variational quantum circuits into agents’ policy networks to improve learning efficiency and convergence. Simulation results demonstrate that MA-QDRL reduces packet loss rate by 63.2% compared to classical DRL and by 39.2% compared to CRDSA-NOMA under high network load conditions $(G = 1.0)$ , confirming its effectiveness in highly congested IoT environments. In addition, it reduces overall computational cost compared to MA-CDRL. These results highlight the potential of MA-QDRL as a scalable and efficient solution for dynamic multi-agent wireless access in next-generation IoT networks.