Deep Reinforcement Learning Mechanism for Dynamic Access Control in Wireless Networks Handling mMTC

[EN] One important issue that needs to be addressed in order to provide effective massive deployments of IoT devices is access control. In 5G cellular networks, the Access Class Barring (ACB) method aims at increasing the total successful access probability by delaying randomly access requests. This mechanism can be controlled through the barring rate, which can be easily adapted in networks where Human-to-Human (H2H) communications are prevalent. However, in scenarios with massive deployments such as those found in IoT applications, it is not evident how this parameter should be set, and how it should adapt to dynamic traffic conditions. We propose a double deep reinforcement learning mechanism to adapt the barring rate of ACB under dynamic conditions. The algorithm is trained with simultaneous H2H and Machine-to-Machine (M2M) traffic, but we perform a separate performance evaluation for each type of traffic. The results show that our proposed mechanism is able to reach a successful access rate of 100 % for both H2H and M2M UEs and reduce the mean number of preamble transmissions while slightly affecting the mean access delay, even for scenarios with very high load. Also, its performance remains stable under the variation of different parameters. (C) 2019 Elsevier B.V. All rights reserved.The research of D. Pacheco-Paramo was supported by Universidad Sergio Arboleda, P.t. Tecnologias para la inclusion social y la competitividad economica. 0.E.6. The research of L Tello-Oquendo was conducted under project CONV.2018-ING010. Universidad Nacional de Chimborazo. The research of V. Pla and J. Martinez-Bauset was supported by Grant PGC2018-094151-B-I00 (MCIU/AEI/FEDER,UE).Pacheco-Paramo, DF.; Tello-Oquendo, L.; Pla, V.; Martínez Bauset, J. (2019). Deep Reinforcement Learning Mechanism for Dynamic Access Control in Wireless Networks Handling mMTC. Ad Hoc Networks. 94:1-14. https://doi.org/10.1016/j.adhoc.2019.101939S1149