Federated Learning Over LoRa Networks: Simulator Design and Performance Evaluation

TL;DR

This paper presents a detailed simulator for federated learning over LoRa networks, analyzing how transmission parameters and interference affect FL performance and convergence.

Contribution

It introduces a comprehensive Python-based simulator integrating FL and LoRaSim frameworks, modeling LoRa-specific channel effects and evaluating FEC's role in FL.

Findings

FEC significantly improves FL convergence over LoRa.

Transmission parameters like spreading factor impact FL performance.

Interference levels critically affect FL update success.

Abstract

Federated learning (FL) over long-range (LoRa) low-power wide area networks faces unique challenges due to limited bandwidth, interference, and strict duty-cycle constraints. We develop a Python-based simulator that integrates and extends the Flower and LoRaSim frameworks to evaluate centralized FL over LoRa networks. The simulator employs a detailed link-level model for FL update transfer over LoRa channels, capturing LoRa's receiver sensitivity, interference characteristics, block-fading effects, and constraints on the maximum transmission unit. It supports update sparsification, quantization, compression, forward frame-erasure correction (FEC), and duty cycling. Numerical results illustrate the impact of transmission parameters (spreading factor, FEC rate) and interference on FL performance. Demonstrating the critical role of FEC in enabling FL over LoRa networks, we perform an in-depth evaluation of the impact of FEC on FL convergence and device airtime, providing insights for communication protocol design for FL over LoRa networks.