Robust Channel Assignment for Hybrid NOMA Systems with Condition Number Constrainted DRL

The Hybrid Non-Orthogonal Multiple Access (NOMA) is an alternative solution for future multiple access techniques, and the performance of hybrid NOMA systems relies on the quality of channel assignment. Conventional optimization approaches rely on the perfect Channel State Information (CSI), which hinders the deployment of the Hybrid systems. Deep Reinforcement Learning (DRL) approaches are robust to uncertain environments, and have been applied to deal with the dynamic channel assignment in hybrid NOMA systems. In this paper, a novel DRL approach based on condition number constraint is proposed to further enhance the robustness of the model. The simulation results show that the proposed approach achieves higher average spectral efficiency under imperfect CSI, compared to unconstrained DRL approaches and conventional approaches. This is useful for critical infrastructure systems such as base stations that require a high degree of robustness.