Deep reinforcement learning-based access class barring for mMTC access control in cellular networks

Internet-of-things (IoT) is generating great income to countries’ economy by interconnecting billions of Users Equipment(UEs). Massive Machine Type Communication (mMTC), a use case of IoT, enables this huge number of UEs interconnections using cellular networks (LTE-A and 5G NR). However, when sporadic large number of M2M UEs try access the eNB or gNB, which primarily supports H2H communication, there will be random access channel congestion. Access Class Barring (ACB) is an access control scheme to reduce the congestion. But, the static ACB scheme does not give the best access success probability performance in such dynamic environment. As a result, we proposed a mechanism which dynamically adjusts the barring factor parameter of the scheme based on the arrival traffic using deep reinforcement learning to improve its access success performance. The expermental results show that the mechanism achieves maximum access success probability (0.9998), meeting the theoretical optimum ACB performance (Ps≥0.95). In addition, for number of M2M UEs NM≤30,000, the mechanism achieves similar access success (99.98%) as a RL-based ACB scheme, but with lower access delay and energy consumption; and under high traffic load (NM>30,000), the mechanism achieves better access success than the RL-based ACB scheme.