Dynamic allocation of RACH slots for minimizing collisions in NB-IoT networks based on reinforcement learning algorithms

The subject of the research is access control to random access channels (RACH) in narrowband Internet of Things (NB-IoT) networks experiencing congestion at high device densities. The object of the study is the procedure for dynamic allocation of RACH slots using machine learning methods. The main focus is on the use of reinforcement learning (RL) algorithms, in particular Q-learning and Deep Q-Network (DQN) methods. The authors examine in detail the problem of channel congestion and associated collisions, which lead to data transmission delays and increased device power consumption. The insufficient efficiency of traditional static slot management methods is analyzed and the need to introduce a dynamic approach that can adapt to changing network conditions is substantiated. The study used machine learning methods including Q-learning and DQN, as well as simulation in the NS-3 environment with the integration of an RL agent for dynamic redistribution of RACH slots. The scientific novelty of the research lies in the development and integration of a specialized RL agent that allows adaptive allocation of RACH slots based on the current state of the NB-IoT network. The main conclusions of the study are the confirmed high efficiency of the proposed dynamic approach, which reduced the number of collisions by 74%, increased the number of successful connections by 16% and improved the energy efficiency of devices by 15% compared to traditional static methods. The prospects for further research include scaling the proposed approach to networks with a large number of devices, studying multi-agent RL approaches, experimental testing and integration of the developed method with real NB-IoT networks, as well as developing hybrid control models that combine reinforcement learning algorithms with other machine learning methods.