Federated Learning in Small-Cell Networks: Stochastic Geometry-Based Analysis on the Required Base Station Density

Area of Science:

• Wireless Communications
• Machine Learning
• Network Analysis

Background:

• Federated learning (FL) is gaining traction for secure, distributed machine learning.
• Wireless communication is crucial for FL in edge networks, yet its impact is understudied.
• Existing research lacks comprehensive analysis of wireless network effects on FL.

Purpose of the Study:

• To investigate the impact of geographic node deployment on federated learning (FL) in small-cell networks.
• To analyze the effects of wireless network characteristics on FL model aggregation.
• To provide insights for designing efficient wireless FL systems.

Main Methods:

• Stochastic geometry-based analysis to model small-cell networks with homogeneous Poisson point process (PPP) node distributions.
• Derivation of closed-form expressions for coverage probability with approximations.
• Analysis of model aggregation performance based on node densities.

Main Results:

• Geographic node deployment significantly affects FL model aggregation in small-cell networks.
• Closed-form expressions for coverage probability were derived, offering tractable approximations.
• Minimum base station (BS) density requirements for target model aggregation rates were identified.

Conclusions:

• The study offers crucial insights into FL behavior within small-cell wireless networks.
• Findings can guide the design of networks to enhance wireless FL performance.
• Understanding node deployment is key to optimizing FL in edge computing environments.