Benchmarking Multi-agent Reinforcement Learning-based Access Control using Real-world IoT Traffic

Abstract

In recent years, the proliferation of Internet of Things (IoT) devices and applications has given rise to massive Machine-Type Communication (mMTC), a critical domain within nextgeneration wireless communication. mMTC is characterized by a massive number of low-power devices generating data sporadically, posing a formidable challenge in efficiently allocating the limited physical wireless channel resources among them. Conventional centralized allocation methods prove infeasible in mMTC scenarios due to the overwhelming volume of control messages relative to the actual data transmissions  required by mmtc devices. To address this challenge, there has been a notable upsurge in research endeavors dedicated to enhancing fully distributed channel access protocols for mMTC, leveraging the power of cooperative multi-agent reinforcement learning. However, an existing research gap becomes evident when we observe that prior studies predominantly tested their algorithms under conditions of either saturated traffic or simulated traffic models. This limited scope fails to accurately capture the dynamics of real-world mMTC networks. In this study, we bridge this gap by utilizing real-world IoT traffic data from a network operator in Vietnam.