Reinforcement Learning-Based ACB in LTE-A Networks for Handling Massive M2M and H2H Communications

Using cellular networks for providing machine-to- machine (M2M) connectivity offers numerous advantages regarding coverage, deployment costs, security and management, among others. Nevertheless, having a large number of M2M devices activated simultaneously is difficult to tackle at the evolved Node B, and it causes complications in the connection establishment. The random access channel (RACH) in LTE-A is adequate for handling human-to-human (H2H) communications. However, for the efficient provision of simultaneous H2H/M2M communications, it is necessary to optimize the available access control mechanisms so that network overload is avoided and a better QoS can be offered. Access Class Barring (ACB) has shown to be effective in reducing the number of simultaneous users contending for access. However, it is still not clear how to dynamically adapt its parameters, especially in highly changing scenarios with bursty traffic as it can occur when M2M communications are involved. We propose a dynamic algorithm based on reinforcement learning to adapt the barring rate parameter of ACB. This algorithm can adapt it to different traffic conditions, reducing congestion and hence the number of collisions in the RACH. The results show that our proposed mechanism increases the access success probability for all the users while barely impacting H2H users and other key performance indicators.