Reinforcement Learning-based Multi-channel Random Access for Massive Machine-Type Communication in 5G Networks

This paper takes into account the development trend of random access methods in wireless communication networks, traffic features and random access methods for massive Machine-Type Communications (mMTC). Existing multi-channel random access methods based on artificial intelligence mechanisms use neural networks due to a large number of system states. Many mMTC applications require a simple implementation of the random access method and low power consumption of mMTC devices. The paper proposes a simple multi-channel random access method based on reinforcement learning, called RL-MCSA. The states of the system (radio channel) are the total number of collisions during access. Only the previous state of the radio channel is required to make a decision. The reward for the decision is the total number of successful accesses. The decision is made in order to maximize the total number of successful packet transmissions in frequency-time channels (resource blocks) in the uplink direction of the 5G access network. Simulation-based experiment results show that the proposed method can improve the rate of successful access in each simulation cycle and the rate of utilization of the radio channel by 1.5-2 times more than existing, well-known algorithms.