Reinforcement Learning Aided Link Adaptation for Downlink NOMA Systems With Channel Imperfections

Non-orthogonal multiple access (NOMA) is a promising candidate radio access technology for future wireless communication systems, which can achieve improved connectivity and spectral efficiency. Without sacrificing error rate performance, link adaptation combining with adaptive modulation and coding (AMC) and hybrid automatic repeat request (HARQ) can provide better spectral efficiency and reliable data transmission by allowing both power and rate to adapt to channel fading and enabling re-transmissions. However, current AMC or HARQ schemes may not be preferable for NOMA systems due to the imperfect channel estimation and error propagation during successive interference cancellation (SIC). To address this problem, a reinforcement learning based link adaptation scheme for downlink NOMA systems is introduced in this paper. Specifically, we first analyze the throughput and spectrum efficiency of NOMA system with AMC combined with HARQ. Then, taking into account the imperfections of channel estimation and error propagation in SIC, we propose SINR and SNR based corrections to correct the modulation and coding scheme selection. Finally, reinforcement learning (RL) is developed to optimize the SNR and SINR correction process. Comparing with a conventional fixed look-up table based scheme, the proposed solutions achieve superior performance in terms of spectral efficiency and packet error performance.