Link adaptation solutions based on reinforcement learning for 5G new radio

Abstract

In this work we propose two self-exploratory frameworks, based on reinforcement learning (RL) for link adaptation in fifth generation (5G) wireless communication systems. Firstly, a Q-learning solution for adaptive modulation and coding (AMC) is presented that allows the base station to learn the mapping between the modulation and coding scheme (MCS) and the channel quality indicator (CQI), in order to maximize the spectral efficiency of the system. Compared to classic AMC solutions, the proposed solution achieves superior performances in terms of spectral efficiency and block error rate (BLER). In the second part of this work, a broader problem is considered in the context of multiple-input multiple-output (MIMO) systems with spatial multiplexing. For this system, a solution based on Q-learning is presented for the joint selection of the MCS and the number of spatial transmission layers (spatial multiplexing factor), as well as the precoding scheme. In this case, the mapping is learned based on the information from the CQI and the rank indicator (RI). According to our simulation results, the proposed solution achieves a performance similar to that of the reference (genie-aided) solution, but with less signaling compared to the one specified in the 5G standard.