WiFi Network Congestion Control Using Deep Reinforcement Learning

The increasing heterogeneity of networks across applications, protocols, and devices highlights the need for dynamic control, aligned with the rise of machine learning for intelligent adaptability. In home WiFi networks, shared bandwidth heavily impacts user experience, especially with bandwidth-intensive applications like torrents that aggressively create connections, causing overload and congestion. Hence, this paper investigates congestion control in home WiFi networks using the TD3 deep reinforcement learning algorithm, addressing a gap in intelligent network management. The system transforms a traditional router into an intelligent device capable of dynamically regulating torrent download bandwidth based on available resources and other applications’ needs. This solution acts as a preventive measure, curbing congestion and preventing full bandwidth saturation. Integrating the trained model, WHCC, into the network results in a $43 \%$ reduction in queued packets, a $17 \%$ decrease in packet loss, a 3\% latency improvement, and a 7\% enhancement in users’ Quality of Experience (QoE).