Scholarlink

<p>The IEEE 802.11bn (Wi-Fi 8) standard introduces Multi-AP Coordination (MAPC) to enable ultra-reliable low-latency communication (URLLC) in dense multi-AP wireless deployments. Existing centralized MAC-layer scheduling strategies impose significant control overhead, privacy risks, and scalability bottlenecks that preclude real-time coordination in high-density environments. This paper proposes and evaluates a decentralized Federated Reinforcement Learning (FRL) framework for MAC-layer scheduling across distributed IEEE 802.11bn access points (APs), designed to preserve local traffic privacy while enabling cooperative policy learning. Each AP independently trains a local tabular Q-learning agent using locally observed states (traffic load, queue length, interference estimate, and SNR), and periodically synchronizes Q-table parameters with a Federated Aggregation Server (FAS) via weighted averaging. The framework is evaluated using a custom Python-based discrete-event simulator implementing IEEE 802.11bn OFDMA resource allocation, Poisson traffic generation, random waypoint mobility, and inter-AP interference modeling. Experiments span 30 independent simulation runs across configurations of 4–12 APs and 5–20 STAs per AP. The proposed FRL scheduler achieves up to 31% lower average packet latency, 22% higher Jain’s Fairness Index, and 17% reduction in control signaling overhead compared to centralized RL and static OFDMA baselines under equivalent simulation conditions. Convergence is achieved within 500–600 training episodes, with only a 15% increase in convergence time as the AP count scales from 4 to 12. These results demonstrate the suitability of the FRL approach for scalable, privacy-aware MAC optimization in next-generation Wi-Fi networks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Federated reinforcement learning for distributed MAC optimization in IEEE 802.11bn networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月16日

查看全文
  • B. T. Vijay

摘要

The IEEE 802.11bn (Wi-Fi 8) standard introduces Multi-AP Coordination (MAPC) to enable ultra-reliable low-latency communication (URLLC) in dense multi-AP wireless deployments. Existing centralized MAC-layer scheduling strategies impose significant control overhead, privacy risks, and scalability bottlenecks that preclude real-time coordination in high-density environments. This paper proposes and evaluates a decentralized Federated Reinforcement Learning (FRL) framework for MAC-layer scheduling across distributed IEEE 802.11bn access points (APs), designed to preserve local traffic privacy while enabling cooperative policy learning. Each AP independently trains a local tabular Q-learning agent using locally observed states (traffic load, queue length, interference estimate, and SNR), and periodically synchronizes Q-table parameters with a Federated Aggregation Server (FAS) via weighted averaging. The framework is evaluated using a custom Python-based discrete-event simulator implementing IEEE 802.11bn OFDMA resource allocation, Poisson traffic generation, random waypoint mobility, and inter-AP interference modeling. Experiments span 30 independent simulation runs across configurations of 4–12 APs and 5–20 STAs per AP. The proposed FRL scheduler achieves up to 31% lower average packet latency, 22% higher Jain’s Fairness Index, and 17% reduction in control signaling overhead compared to centralized RL and static OFDMA baselines under equivalent simulation conditions. Convergence is achieved within 500–600 training episodes, with only a 15% increase in convergence time as the AP count scales from 4 to 12. These results demonstrate the suitability of the FRL approach for scalable, privacy-aware MAC optimization in next-generation Wi-Fi networks.