UAV Communications for Sustainable Federated Learning

Quoc-Viet Pham, Ming Zeng, Rukhsana Ruby, Thien Huynh-The, Won-Joo Hwang

TL;DR

This paper proposes a novel approach to enable sustainable federated learning (FL) in wireless networks by using Unmanned Aerial Vehicles (UAVs) for wireless power transfer. The authors develop an efficient algorithm, UAV-SFL, that jointly optimizes UAV transmit power, transmission time, bandwidth allocation, and UAV placement to maximize power efficiency, significantly outperforming existing methods.

What Problem Does This Paper Address?

Federated learning (FL) allows for privacy-preserving machine learning by training models locally on user devices and only sharing model updates. However, a major bottleneck for FL in wireless networks is the limited battery life of mobile users, which can interrupt or prevent training. While wireless powered communication networks (WPCNs) offer a solution for charging users, traditional WPCNs face challenges like the "doubly near-far issue" and performance degradation over distance. Unmanned Aerial Vehicles (UAVs) have emerged as a promising solution due to their flexible deployment and line-of-sight connections. However, existing research hasn't fully explored the integration of UAVs for wireless power transfer to support sustainable FL networks, especially in scenarios with limited infrastructure. This paper aims to fill that gap by proposing a system that uses UAVs to power FL users and optimize resource allocation for energy efficiency.

Methods

The core of this paper is the development of the UAV for sustainable FL (UAV-SFL) algorithm. The problem of maximizing UAV transmit power efficiency, while jointly optimizing transmission time, bandwidth allocation, power control, and UAV placement, is formulated as a complex, non-convex optimization problem. To tackle this, the authors employ a Block Successive Upper-Bound Minimization (BSUM) method. This iterative approach divides the optimization variables into three blocks: (transmission time, CPU frequency), (UAV transmit power), and (user transmit power, bandwidth, UAV location). Each block is updated iteratively while keeping others fixed.

Key techniques include:

  • Decomposition and Successive Convex Approximation: Used to handle the non-convexity of the original problem.
  • Auxiliary Variables: Introduced to linearize certain constraints, particularly those involving the UAV's location and channel gains.
  • Bi-section Rule Analogy: Applied to update transmission time and UAV transmit power, ensuring they stay within feasible ranges.
  • Convex Feasibility Problem: The problem is reformulated into a series of convex feasibility problems that can be solved efficiently.
  • Lemma for Convexification: A crucial lemma is used to transform the non-convex data rate constraint into a convex form, enabling the use of convex optimization techniques.

Datasets

Not discussed in the paper. The paper relies on simulations with parameters representative of real-world scenarios.

Key Insights and Ideas

  • UAVs as Mobile Power Stations for FL: The paper demonstrates that UAVs can effectively act as mobile charging stations, powering the entire FL process for users with limited battery capacity.
  • Joint Optimization is Crucial: Simply optimizing one aspect (like UAV placement) isn't enough. The synergy achieved by jointly optimizing transmission time, bandwidth, power, and location is key to maximizing power efficiency.
  • Adaptive Resource Allocation: The UAV-SFL algorithm adaptively adjusts its transmit power based on the number of local iterations required by users, ensuring sufficient energy is provided for computation without excessive waste.
  • Sustainability through Efficiency: By maximizing the UAV's transmit power efficiency, the proposed method contributes to the overall sustainability of FL-based wireless networks, reducing energy consumption and potentially extending the operational lifespan of the system.

Key Results

The proposed UAV-SFL approach demonstrates significant improvements over benchmark schemes:

  • Reduced UAV Transmit Power: On average, UAV-SFL reduces UAV transmit power by 32.95% compared to a fixed UAV placement (FUP) scheme, 63.18% compared to a fixed CPU frequency (FF) scheme, and 78.81% compared to a fixed transmission time (FT) scheme.
  • Convergence: The algorithm converges to a stable solution within a reasonable number of iterations, as shown in Figure 1.
  • Impact of Local Iterations: Increasing the number of local iterations (Nk) leads to higher computation time but lower transmission time, and requires the UAV to increase its transmit power to meet energy demands.
  • Bandwidth Impact: The UAV's transmit power shows only a minor change with increasing system bandwidth, as higher bandwidths allow for more efficient data transmission, reducing the need for higher power.

Limitations and Future Work

  • Single Global Round: The current work focuses on a single global FL round. The authors suggest extending the framework to multiple global rounds as a promising future direction.
  • Other Multiple Access Techniques: While FDMA is used, the paper notes that other techniques like RSMA and NOMA could be explored.
  • Non-linear Energy Harvesting: The paper uses a linear energy harvesting model and suggests investigating a non-linear model as future work.
  • Other UAV Types and Scenarios: The authors mention exploring fixed-wing UAVs, UAV-assisted visible light communications, and multi-UAV WPFNs as potential extensions.

Summary updated

Abstract

Federated learning (FL), invented by Google in 2016, has become a hot research trend. However, enabling FL in wireless networks has to overcome the limited battery challenge of mobile users. In this regard, we propose to apply unmanned aerial vehicle (UAV)-empowered wireless power transfer to enable sustainable FL-based wireless networks. The objective is to maximize the UAV transmit power efficiency, via a joint optimization of transmission time and bandwidth allocation, power control, and the UAV placement. Directly solving the formulated problem is challenging, due to the coupling of variables. Hence, we leverage the decomposition technique and a successive convex approximation approach to develop an efficient algorithm, namely UAV for sustainable FL (UAV-SFL). Finally, simulations illustrate the potential of our proposed UAV-SFL approach in providing a sustainable solution for FL-based wireless networks, and in reducing the UAV transmit power by 32.95%, 63.18%, and 78.81% compared with the benchmarks.
BibTeX
@article{pham2021uav,
  author = {Quoc-Viet Pham and Ming Zeng and Rukhsana Ruby and Thien Huynh-The and Won-Joo Hwang},
  title = {{UAV Communications for Sustainable Federated Learning}},
  journal = {IEEE Trans. Vehicular Technology},
  year = {2021},
  doi = {10.1109/tvt.2021.3065084},
}
Figure