Privacy-Preserving Federated Learning for UAV-Enabled Networks: Learning-Based Joint Scheduling and Resource Management

Helin Yang, Jun Zhao, Zehui Xiong, Kwok-Yan Lam, Sumei Sun, Liang Xiao

Abstract

Unmanned aerial vehicles (UAVs) are capable of serving as flying base stations (BSs) for supporting data collection, artificial intelligence (AI) model training, and wireless communications. However, due to the privacy concerns of devices and limited computation or communication resource of UAVs, it is impractical to send raw data of devices to UAV servers for model training. Moreover, due to the dynamic channel condition and heterogeneous computing capacity of devices in UAV-enabled networks, the reliability and efficiency of data sharing require to be further improved. In this paper, we develop an asynchronous federated learning (AFL) framework for multi-UAV-enabled networks, which can provide asynchronous distributed computing by enabling model training locally without transmitting raw sensitive data to UAV servers. The device selection strategy is also introduced into the AFL framework to keep the low-quality devices from affecting the learning efficiency and accuracy. Moreover, we propose an asynchronous advantage actor-critic (A3C) based joint device selection, UAVs placement, and resource management algorithm to enhance the federated convergence speed and accuracy. Simulation results demonstrate that our proposed framework and algorithm achieve higher learning accuracy and faster federated execution time compared to other existing solutions.

Introduction

The deployment of AI at the network edge often requires training models on data generated by mobile devices such as smartphones and IoT sensors. However, two major hurdles impede this process: user privacy and resource constraints. Users are increasingly hesitant to share raw, sensitive data with central servers. Simultaneously, Unmanned Aerial Vehicles (UAVs), while versatile as "flying base stations," are limited by battery life and computing power.

Standard Federated Learning (FL) addresses privacy by keeping data on-device and sharing only model updates. However, traditional FL is typically synchronous, meaning the server must wait for all participating devices to complete local training before updating the global model. In dynamic UAV networks, "straggler" devices—those with poor connectivity or low processing power—become significant bottlenecks. This research proposes an asynchronous approach to ensure efficiency and reliability in these highly variable aerial environments.

System Architecture and Methodology

The proposed solution integrates two core components: an Asynchronous Federated Learning (AFL) framework and a Deep Reinforcement Learning (DRL) optimization algorithm.

Asynchronous Federated Learning (AFL)

Unlike synchronous FL, the AFL framework allows UAV servers to update the global model as soon as individual devices complete their local tasks. This eliminates the delay caused by the slowest devices, making the system more resilient to the intermittent connectivity typical of mobile networks.

Architecture of the proposed AFL framework in a multi-UAV network, illustrating the interaction between ground devices and aerial servers.

A3C-based Joint Optimization

To manage the network's complexity, the authors employ the Asynchronous Advantage Actor-Critic (A3C) algorithm. This DRL approach jointly optimizes three critical variables:

  1. Device Selection: Identifying and selecting high-quality devices with reliable data and sufficient hardware capabilities.
  2. UAV Placement: Dynamically adjusting the horizontal coordinates of UAVs to maximize signal strength for selected devices.
  3. Resource Management: Efficiently allocating subchannels and transmit power.

A3C is particularly effective here as it can handle continuous action spaces, such as the precise X and Y coordinates of UAVs, which traditional discrete methods like Q-learning struggle to optimize.

Experimental Evaluation

The framework was evaluated using a Convolutional Neural Network (CNN) trained on the MNIST dataset (60,000 training and 10,000 testing images). The simulation environment consisted of a 400m x 400m area with four UAVs and up to 300 ground devices, utilizing realistic air-to-ground path loss models (Line-of-Sight and Non-Line-of-Sight).

Key Insights and Results

  • Filtering for Quality: The A3C agent learns to exclude "low-quality" participants—devices with noisy data or poor local training—which prevents degradation of the global model and improves final accuracy.
  • Resilience through Asynchrony: Asynchronous updates prove essential for mobility. The system remains functional even when devices experience temporary connection drops, a common occurrence in UAV-enabled networks.
  • Performance Gains: The proposed A3C-AFL reached 90% accuracy in approximately 25 global rounds. In terms of scalability, the algorithm maintained high accuracy (~94%) as the device count increased to 300, whereas benchmarks without device selection saw performance stall.
  • Convergence: The A3C agent demonstrated steady convergence after approximately 500 episodes, validating the DRL approach for complex aerial resource management.

Conclusion and Future Work

This research demonstrates that combining asynchronous federated learning with DRL-based joint optimization significantly enhances the efficiency and privacy of UAV-enabled networks. While the results using MNIST are promising, future work should explore more complex, non-IID datasets. Additionally, extending the UAV placement to 3D trajectories (optimizing altitude) and incorporating explicit energy-efficiency objectives would further enhance the practical utility of this framework for long-term deployments.

Summary updated

Abstract

Unmanned aerial vehicles (UAVs) are capable of serving as flying base stations (BSs) for supporting data collection, artificial intelligence (AI) model training, and wireless communications. However, due to the privacy concerns of devices and limited computation or communication resource of UAVs, it is impractical to send raw data of devices to UAV servers for model training. Moreover, due to the dynamic channel condition and heterogeneous computing capacity of devices in UAV-enabled networks, the reliability and efficiency of data sharing require to be further improved. In this paper, we develop an asynchronous federated learning (AFL) framework for multi-UAV-enabled networks, which can provide asynchronous distributed computing by enabling model training locally without transmitting raw sensitive data to UAV servers. The device selection strategy is also introduced into the AFL framework to keep the low-quality devices from affecting the learning efficiency and accuracy. Moreover, we propose an asynchronous advantage actor-critic (A3C) based joint device selection, UAVs placement, and resource management algorithm to enhance the federated convergence speed and accuracy. Simulation results demonstrate that our proposed framework and algorithm achieve higher learning accuracy and faster federated execution time compared to other existing solutions.
BibTeX
@article{yang2021privacy,
  author = {Helin Yang and Jun Zhao and Zehui Xiong and Kwok-Yan Lam and Sumei Sun and Liang Xiao},
  title = {{Privacy-Preserving Federated Learning for UAV-Enabled Networks: Learning-Based Joint Scheduling and Resource Management}},
  journal = {IEEE Journal on Selected Areas in Communications},
  year = {2021},
  url = {https://openreview.net/forum?id=dYduCOK2mq},
}
Figure