Utilizing Reinforcement Learning and Dynamic Modulation in 5G New Radio to Improve MIMO QoS on a Physical Downlink Shared Channel

This research work presents a reinforcement learning (RL)-based algorithm for dynamically modulating schemes in 5G New Radio (NR) downlink transmissions over a Multiple Input Multiple Output (MIMO) channel. By emulating a real scenario of an 8×8 MIMO Physical Downlink Shared Channel (PDSCH) under Additive White Gaussian Noise (AWGN) and 3GPP TDL-C fading, we apply a Deep Q-Network (DQN) agent for adaptively learning to select types of modulation such as QPSK, 16QAM, 64QAM, 256QAM, and 1024QAM from monitored Signal-to-Noise Ratio (SNR) and Channel State Information (CSI). Simulation results show that the RL-based adaptive modulation scheme improves Bit Error Rate (BER) and relative throughput in various Signal Noise Ratio (SNR) conditions, showing better Quality of Service (QoS) than fixed modulation schemes. Compared to general sampling, in preliminary results, the RL model shows a 34% improvement in BER using DQN and about 29% improvement using PPO. For relative throughput, the DQN agent improved by about 6.5%.