A Multi-Agent Deep Q-Learning Framework for Uplink Congestion Control in Communication Networks

Abstract

Congestion control is crucial for reliable and efficient communication in networks. Current systems mainly focus on managing download traffic. However, upload communication is becoming increasingly important with the rise of 5G/6G networks, IoT devices, cloud services, and real-time apps. Upload channels have limited resources, lack predictability, and are shared by multiple users, making congestion management challenging. This paper proposes an AI and machine learning-based framework to address uplink congestion. It uses reinforcement learning to dynamically adjust transmission rates and incorporates fairness mechanisms for multi-user scenarios. The results demonstrate that this approach improves throughput, reduces delays, and enhances fairness compared to traditional methods, suggesting that AI-driven solutions hold promise for future networks.