Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improving Energy Efficiency in Federated Learning Through the Optimization of Communication Resources Scheduling of Wireless IoT Networks

Renan R. de Oliveira, Kleber V. Cardoso, Antonio Oliveira-Jr

TL;DR

This work addresses energy efficiency and reliability in Federated Learning over wireless IoT networks by introducing FL-E2WS, a MILP-based framework for joint device selection and uplink resource scheduling. By selecting data-rich devices and optimally allocating bandwidth and power, FL-E2WS minimizes energy consumption while preserving global model accuracy, outperforming channel-ignorant FL and comparable fixed-resource baselines. The approach demonstrates substantial energy savings (up to 70.12%) and accuracy improvements (up to 10.21%) in simulated non-IID FL scenarios, and highlights improved resource utilization and transmission success. Its practical impact lies in enabling scalable, energy-aware FL on bandwidth-constrained edge networks, with open-source reproducibility and directions for future mobility-aware, heterogeneous FL deployments.

Abstract

Federated Learning (FL) allows devices to train a global machine learning model without sharing data. In the context of wireless networks, the inherently unreliable nature of the transmission channel introduces delays and errors that compromise the regularity of updating the global model. Furthermore, limited resources and energy consumption of devices are factors that affect FL performance. Therefore, this work proposes a new FL algorithm called FL-E2WS that considers both the requirements of federated training and a wireless network within the scope of the Internet of Things. To reduce the energy cost of devices, FL-E2WS schedules communication resources to allocate the ideal bandwidth and power for the transmission of models under certain device selection and uplink resource block allocation, meeting delay requirements, power consumption, and packet error rate. The simulation results demonstrate that FL-E2WS reduces energy consumption by up to 70.12% and enhances the accuracy of the global model by up to 10.21% compared to the FL algorithms that lacks transmission channel knowledge. Additionally, when compared to FL versions that scale communication resources, FL-E2WS achieves up to a 38.61% reduction in energy consumption and improves the accuracy of the global model by up to 1.61%.

Improving Energy Efficiency in Federated Learning Through the Optimization of Communication Resources Scheduling of Wireless IoT Networks

TL;DR

This work addresses energy efficiency and reliability in Federated Learning over wireless IoT networks by introducing FL-E2WS, a MILP-based framework for joint device selection and uplink resource scheduling. By selecting data-rich devices and optimally allocating bandwidth and power, FL-E2WS minimizes energy consumption while preserving global model accuracy, outperforming channel-ignorant FL and comparable fixed-resource baselines. The approach demonstrates substantial energy savings (up to 70.12%) and accuracy improvements (up to 10.21%) in simulated non-IID FL scenarios, and highlights improved resource utilization and transmission success. Its practical impact lies in enabling scalable, energy-aware FL on bandwidth-constrained edge networks, with open-source reproducibility and directions for future mobility-aware, heterogeneous FL deployments.

Abstract

Federated Learning (FL) allows devices to train a global machine learning model without sharing data. In the context of wireless networks, the inherently unreliable nature of the transmission channel introduces delays and errors that compromise the regularity of updating the global model. Furthermore, limited resources and energy consumption of devices are factors that affect FL performance. Therefore, this work proposes a new FL algorithm called FL-E2WS that considers both the requirements of federated training and a wireless network within the scope of the Internet of Things. To reduce the energy cost of devices, FL-E2WS schedules communication resources to allocate the ideal bandwidth and power for the transmission of models under certain device selection and uplink resource block allocation, meeting delay requirements, power consumption, and packet error rate. The simulation results demonstrate that FL-E2WS reduces energy consumption by up to 70.12% and enhances the accuracy of the global model by up to 10.21% compared to the FL algorithms that lacks transmission channel knowledge. Additionally, when compared to FL versions that scale communication resources, FL-E2WS achieves up to a 38.61% reduction in energy consumption and improves the accuracy of the global model by up to 1.61%.
Paper Structure (14 sections, 26 equations, 9 figures, 3 tables, 2 algorithms)

This paper contains 14 sections, 26 equations, 9 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Mathematical Modeling of the System
  4. Network Model
  5. ML Model
  6. Communication Model
  7. Energy Consumption Model
  8. Algorithm Formulation
  9. Device Selection
  10. Communication Resource Scheduling
  11. FL-E2WS Algorithm
  12. Performance Evaluation
  13. Simulation Results
  14. Conclusion and Future Work

Figures (9)

  • Figure 1: Distance from devices to the BS in meters. (a) Absolute Frequency. (b) NIID R-MNIST dataset. (c) NIID R-FMNIST dataset. In (b) and (c), each device is represented by a circle, with the circle's size being proportional to the amount of data.
  • Figure 2: Accuracy of $w_i$ using local data from other devices. (a) NIID R-MNIST dataset. (b) NIID R-FMNIST dataset.
  • Figure 3: Evolution of accuracy and $f(w_{global})$ of FL algorithms with NIID R-MNIST and NIID R-FMNIST datasets. In (a) and (b), it is the FL algorithms that do not know the transmission channel. FedAvg and POC are unable to successfully transmit a significant number of local models. In (c) and (d), it is the FL algorithms that incorporate knowledge of the transmission channel enabling them to successfully transmit a significantly greater number of local models. FedAvg-w$_{Opt}$ and POC-w$_{Opt}$ incorporate knowledge of the transmission channel using the FL-E2WS algorithm operating with fixed bandwidth and transmission power. FL-E2WS prioritizes minimizing the energy consumption of devices by optimizing the allocation of uplink RBs, bandwidth and transmission power.
  • Figure 4: Sum of the RBs occupation times in each communication round. FL-E2WS enhances communication resource utilization by minimizing the models transmission time of local models to the BS.
  • Figure 5: Bandwidth and power allocation. FL-E2WS dynamically allocates communication resources based on the demand of the devices selected for each round in order to maximize energy efficiency.
  • ...and 4 more figures