Reinforcement Learning-Based QoS Optimization in SDN-Driven Wireless Mesh Networks

The research develops a method to optimize Quality of Service (QoS) in Software-Defined Networking (SDN)-driven Wireless Mesh Networks (WMNs) by using Deep Reinforcement Learning techniques. The research integrates SDN central control features with Deep Q-Network (DQN) learning to create an adaptive routing framework that handles simultaneous optimization of multiple QoS metrics. The new system implements a reward function composed of latency, throughput, bandwidth utilization and packet delivery ratio while maintaining equilibrium among them. The experimental findings show that the DQN-based system delivers superior results than traditional routing protocols such as OSPF and shortest path and round-robin and random selection when deployed across different traffic patterns especially during conditions involving hotspot traffic and bursty traffic behavior. The results of hyperparameter tuning demonstrated that the proper balance between exploitation and exploration together with future reward weighting leads to the best convergence. This study fills essential knowledge gaps through an optimization strategy which applies SDN's network-wide monitoring capabilities to improve time-sensitive quality of service in mesh wireless networks.