Enhancing GF-NOMA Spectral Efficiency Under Imperfections Using Deep Reinforcement Learning

In this letter, we present a deep reinforcement learning (DRL) based multi-carrier grant-free (GF) non-orthogonal multiple access (NOMA) scheme for Internet of Things networks to solve the power and sub-carrier allocation problem. Compared to existing work in this area, the proposed scheme is more practical, takes into account the imperfections in successive interference cancellation (SIC), and allows for unrestricted user sub-carrier selection. In the proposed DRL framework, each GF user acts as an agent and tries to find the optimal resources selection policy. To search for optimal policies, a good trade-off between exploration and exploitation is achieved. A 60% exploration and 40% exploitation provides better rewards. Numerical results show the significance of imperfection in the SIC on spectral efficiency. As compared to the benchmark schemes, the proposed scheme increases the user fairness up to 62.1% and outperform the single-carrier GF-NOMA in terms of spectral efficiency.