Robust Design for IRS-Assisted MISO-NOMA Systems: A DRL-Based Approach

In this letter, we propose a robust design for an intelligent reflecting surface (IRS)-aided multiple-input single-output (MISO) non-orthogonal multiple access (NOMA) system. In particular, the ergodic sum-rate maximization problem is formulated by taking into account the channel uncertainties of both direct links and the reflected links through IRS elements. The unbounded channel uncertainties with imperfect channel estimation are mathematically modelled based on the statistical channel state information (CSI) error model. However, the formulated ergodic sum-rate maximization problem with the outage-constraints is not jointly convex in terms of the beamforming vectors and the phase shifts of IRS elements, and hence it cannot be solved with the conventional optimization algorithms. To address the non-convexity issues and develop a joint design, the challenging robust design is reformulated as a reinforcement learning (RL) environment. Two deep RL agents are developed to jointly optimize the beamforming vectors and phase shifts of the IRS elements with the channel uncertainties and quality of service constraints. Simulation results are provided to validate the performance of the proposed agents for both fixed and dynamic channels.