Deep Reinforcement Learning-Driven Dynamic Spectrum Access in Dense Wi-Fi Environments

Abstract

The rapid growth of wireless devices and bandwidth-intensive applications in urban environments has exacerbated spectrum congestion in Wi-Fi networks, resulting in performance degradation in terms of latency, throughput, and fairness. Traditional mechanisms such as CSMA/CA and static channel allocation often fail to adapt effectively in high-density scenarios. To address this challenge, we propose a novel framework that leverages Deep Reinforcement Learning (DRL)—specifically Deep Q-Networks (DQN) and Proximal Policy Optimization (PPO)—to enable intelligent, dynamic spectrum access in IEEE 802.11ax/be (Wi-Fi 6/7) environments. Using the NS-3 simulator integrated with PyTorch-based agents, we model dense deployments with multiple access points and varied traffic patterns, including VoIP, video, FTP, and AR/VR flows. Our DRL agents are trained to select frequency channels in real-time based on environmental observations, such as interference levels, traffic load, and delay, to maximize overall network quality of service. Comparative evaluations against legacy MAC schemes demonstrate that PPO improves average throughput by up to 38%, reduces end-to-end latency by 48%, and enhances fairness, while maintaining minimal switching overhead. Furthermore, we discuss the potential for prototyping the system using OpenWRT-enabled access points, demonstrating the framework’s feasibility for real-world deployments in smart cities and enterprise networks.