Multi-Agent Reinforcement Learning for a Multichannel Uplink Random Access: Congestion Game Perspective

Abstract

In this paper, we propose a time-slotted multichannel uplink random access (RA) game model where players do not cooperate. We first analyze its sum throughput from the congestion game (CG) perspective and obtain the pure strategy Nash equilibria (PNEs) that fully utilize each slot. Then, we propose an Upper Confidence Bound (UCB)-based multi-agent reinforcement learning (MARL) algorithm to realize the PNEs, where UCB is one of the multi-armed bandit algorithms that work by assigning a confidence level for each action. Finally, via simulation, we show that our proposed algorithm can obtain near-optimal average sum throughput in the long run.