Reinforcement Learning Based Congestion Control Technique for Wireless Mesh Networks

Due to the Internet’s rapid expansion, substantial user population, and growing reliance on broadband networks, congestion is one of the most significant network challenges. Congestion occurs when the number of packets on link exceeds the network’s capacity. Wireless Mesh Networks (WMNs) belong to self-configuring, multi-hop, and self-healing network types and are simple to use. The present research provides a novel reinforcement-based TCP-Int (Intelligent TCP) method aimed at improving WMN performance, in which the size of the congestion window is maintained to preserve WMN performance while maintaining throughput. The peculiarity of this work is that it integrates the recently published Open AI gym library, ns3-gym, with a conventional network simulator, ns3. This connection enables the Reinforcement Learning (RL) agent to interact with and learn from the environment provided by ns3. In this work, TCP variations such as TCP New Reno, TCP Yeah, TCP Vegas, TCP Hybla, and TCP High-speed are implemented in the ns3 simulator. These algorithms generate the data (throughput), an important performance indicator of the algorithm. The proposed TCP-Int shows an improvement in the throughput of the WMN of 81 %. This result shows that by regulating the CWND size using the TCP-Int, the wireless mesh network’s performance can be improved significantly. © 2025 Elsevier B.V., All rights reserved.