Mobility‐Aware IRS‐Assisted Hybrid NOMA With Deep Reinforcement Learning for THz Enabled 6G Networks

ABSTRACT
Conventional orthogonal multiple access (OMA) and non‐orthogonal multiple access (NOMA) face serious limitations in ultra‐dense terahertz (THz) enabled 6G networks: OMA suffers from low spectral efficiency, while NOMA incurs high SIC complexity under user mobility. To address these challenges, this paper proposes a mobility‐aware intelligent reflecting surface‐assisted hybrid NOMA (M‐IRS‐HNOMA) framework for THz downlink transmission in 6G networks. The proposed design integrates three components: (i) IRS‐enhanced THz propagation to mitigate severe path loss and extend coverage, (ii) a mobility and quality‐of‐service (QoS) aware hybrid clustering strategy that applies NOMA within clusters and OMA across clusters to balance spectral efficiency and SIC complexity, and (iii) a deep reinforcement learning (DRL) model that jointly optimizes power allocation, IRS phase shifts and hybrid access parameters under time‐varying channels. A unified optimization problem is formulated to maximize the weighted sum‐rate subject to power, QoS, and IRS hardware constraints and is decomposed into tractable power and phase‐shift subproblems embedded within the DRL control loop for real‐time adaptation. Simulation results demonstrate that, compared with conventional HNOMA, IRS‐OMA, cooperative NOMA, and static IRS‐NOMA baselines, the proposed M‐IRS‐HNOMA achieves significantly higher spectral efficiency, improved energy efficiency and reduced end‐to‐end latency in THz mobility scenarios, confirming its potential as a practical solution for ultra‐dense 6G communications.