A Reinforcement-Learning-Based Access Scheme for Low-Latency and Correlated-Traffic MTC Networks

This paper presents an access scheme for machine-type communication (MTC) networks where the base station (BS) is equipped with a massive antenna array and devices have correlated traffic and delay constraints. We formulate an optimization problem to allocate resources and calculate the access probabilities to maximize the throughput with delay constraints. Since the traffic model parameters and event locations are not available to the BS and the throughput with delay constraints is hard to be derived, we propose a reinforcement-learning-based algorithm to solve the problem. Our simulation reveals that our proposed algorithm is superior to a random scheduling baseline both in terms of throughput and delay. More importantly, our proposed algorithm achieves comparable throughput and lower average delay compared to the algorithm that has full information of traffic model parameters and event locations but optimizes throughput without delay constraints.