Delay-aware dynamic access control for mMTC in wireless networks using deep reinforcement learning

Abstract

The success of the applications based on the Internet of Things (IoT) relies heavily on the ability to process large amounts of data with different Quality-of-Service (QoS) requirements. Access control remains an important issue in scenarios where massive Machine-Type Communications (mMTC) prevail, and as a consequence, several mechanisms such as Access Class Barring (ACB) have been designed aiming at reducing congestion. Although this mechanism can effectively increase the total number of User Equipments (UEs) that can access the system, it can also harm the access delay, limiting its usability in some scenarios. In this work, we propose a delay-aware double deep reinforcement learning mechanism that can dynamically adapt two parameters of the system in order to enhance the probability of successful access using ACB, while at the same time reducing the expected delay by modifying the Random Access Opportunity (RAO) periodicity. Results show that our system can accept a simultaneously massive number of machine-type and human-type UEs while at the same time reducing the mean delay when compared to previously known solutions. This mechanism can work adequately under varying load conditions and can be trained with real data traces, which facilitates its implementation in real scenarios.