Federated Deep Reinforcement Learning-Based Intelligent Channel Access in Dense Wi-Fi Deployments

TL;DR

The paper tackles high collision and latency in dense Wi-Fi deployments by coupling Federated Learning with Deep Deterministic Policy Gradient to adapt the contention window. It introduces a pruning-based STA selection and an exponential aggregation rule to improve FL efficiency and model robustness, validated in ns-3/ns3-ai simulations. Results show substantial MAC delay reductions (up to ~25% in static and up to ~46% in dynamic settings) and meaningful throughput gains compared to baseline methods, including standard DRL and RTS/CTS. The approach advances practical FL-based optimization for IEEE 802.11 MAC in heterogeneous, dynamic dense networks, with implications for fairness and scalability in real deployments.

Abstract

The IEEE 802.11 MAC layer utilizes the Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) mechanism for channel contention, but dense Wi-Fi deployments often cause high collision rates. To address this, this paper proposes an intelligent channel contention access mechanism that combines Federated Learning (FL) and Deep Deterministic Policy Gradient (DDPG) algorithms. We introduce a training pruning strategy and a weight aggregation algorithm to enhance model efficiency and reduce MAC delay. Using the NS3-AI framework, simulations show our method reduces average MAC delay by 25.24\% in static scenarios and outperforms A-FRL and DRL by 25.72\% and 45.9\% in dynamic environments, respectively.

Figures (5)

• Figure 1: Static scenario delay performance vs. Training eposides.
• Figure 2: Delay comparison at different traffic arrival rates in static scenario.
• Figure 3: Dynamic scenario delay performance vs. Training eposides.
• Figure 4: Delay comparison at different traffic arrival rates in dynamic scenario.
• Figure 5: Throughput comparison at different traffic arrival rates in dynamic scenario.