Fairness-Based Multi-AP Coordination Using Federated Learning in Wi-Fi 7.

Area of Science:

• Computer Science
• Wireless Communication
• Machine Learning

Background:

• Federated learning (FL) faces challenges in short-range wireless environments due to limited devices per access point (AP).
• Decentralized FL struggles with data imbalance in non-independent and identically distributed (non-IID) settings.
• IEEE 802.11be standards offer multi-AP coordination potential for enhanced FL.

Purpose of the Study:

• To propose a federated learning method addressing device limitations and data distribution issues in multi-AP environments.
• To ensure uniform global model performance irrespective of data heterogeneity.
• To enhance fairness and communication efficiency in decentralized FL.

Main Methods:

• Leveraging multi-AP coordination characteristics of IEEE 802.11be for decentralized FL.
• Implementing primary-AP selection based on device count for initial communication efficiency.
• Utilizing training time and energy consumption for subsequent primary-AP selection to ensure fairness.

Main Results:

• Achieved up to 97.6% prediction accuracy on MNIST and FMNIST datasets.
• Demonstrated fair global model updates in non-IID multi-AP environments.
• Successfully addressed data transmission imbalance and ensured fairness among multi-APs.

Conclusions:

• The proposed method effectively enables decentralized federated learning in resource-constrained wireless networks.
• Fairness and communication efficiency are maintained through adaptive primary-AP selection.
• High prediction accuracy is attainable even with non-IID data across multiple access points.