Analysis of Data Harvesting by Unmanned Aerial Vehicles

Chang-Sik Choi, François Baccelli, Gustavo de Veciana

Introduction

As the Internet of Things (IoT) expands into remote and infrastructure-poor environments—such as rural agricultural zones, environmental monitoring sites, and tactical battlefields—traditional connectivity solutions often fail. Fixed infrastructure like cellular towers is prohibitively expensive for low-density areas, and terrestrial "data mules" are limited by existing road networks.

Unmanned Aerial Vehicles (UAVs) present a compelling alternative, acting as mobile hubs that can traverse any terrain. However, their mobility introduces significant network modeling challenges. This research establishes a rigorous mathematical framework to optimize data "harvesting" by a fleet of UAVs, focusing on managing interference and maximizing throughput.

Methodology: Modeling the Mobile Network

The study utilizes Stochastic Geometry to model the random spatial distribution of network components. IoT devices are represented as a stationary Poisson Point Process (PPP), while UAVs are modeled as a "shifted comb process"—a line of points moving at a constant velocity $v$ and fixed altitude $h$.

The Activation Window

A central innovation in this model is the "activation window" ($w \times l$). To manage interference, a UAV only communicates with an IoT device when that device falls within a specific rectangular footprint directly beneath it. This prevents the network from being overwhelmed by simultaneous transmissions from every sensor in range.

The authors apply Palm calculus to analyze the network from the perspective of a "typical" UAV, deriving the Laplace transform of the aggregate interference (shot-noise). Furthermore, the Mass Transport Principle is used to link the instantaneous data rate experienced by a UAV to the total data successfully uploaded by an individual ground sensor during a UAV's passage.

Key Insights and Optimization

The analysis reveals several critical properties of these dynamic networks:

  1. Time-Invariant Interference: Despite the constant motion of the UAV fleet, the statistical distribution of interference remains constant over time, simplifying the task for network designers.
  2. The "Goldilocks" Window: There is an optimal size for the activation window ($w$). If the window is too large, interference from multiple active devices degrades the signal. If it is too small, the UAV spends too much time over areas with no active sensors, reducing overall efficiency.
  3. Intermittent Universal Coverage: Unlike fixed base stations, a mobile fleet ensures that every ground device eventually enters an activation window, providing universal coverage over time.

Key Results

The analytical models developed in this paper show exceptional alignment with Monte Carlo simulations across various fading conditions.

Comparison of analytical and simulation results for the typical data rate against the SIR threshold. The curves demonstrate the existence of an optimal activation window width (w) for a given UAV inter-distance (μ).

The researchers found that for typical device densities (10–30 devices/km²), the optimal activation window size is approximately 20% to 40% of the distance between UAVs. The study also confirms that these networks are primarily interference-limited; adding more sensors leads to diminishing returns unless the activation windows are narrowed to compensate.

Limitations and Future Directions

While providing a robust foundation, the current model assumes fixed, straight-line trajectories and constant speeds. It does not yet account for adaptive path planning, where UAVs might linger over high-density clusters, or complex urban channel models involving Line-of-Sight (LoS) transitions. Future research could explore dynamic altitude adjustments and more sophisticated Medium Access Control (MAC) protocols to further enhance harvesting efficiency in real-world deployments.

Summary updated

Abstract

This paper explores an emerging wireless architecture based on unmanned aerial vehicles (UAVs), i.e., drones. We consider a network where UAVs at fixed altitude harvest data from Internet-of-Things (IoT) devices on the ground. Each UAV serves IoT devices within its coverage area. In such a system, the UAVs' motion activates IoT uplink transmissions and so the motion triggers the interference field and determines the network performance. To analyze the performance, we propose a stochastic geometry model. The coverage area of each UAV, referred to as the activation window, is modeled for simplicity as a rectangle where at most one IoT device is scheduled to transmit at any given time. In this setting, we analyze the signal-to-interference and data rate from two typical perspectives, namely from a typical UAV's and from a typical IoT device's points of view.
BibTeX
@article{choi2019analysis,
  author = {Chang-Sik Choi and François Baccelli and Gustavo de Veciana},
  title = {{Analysis of Data Harvesting by Unmanned Aerial Vehicles}},
  journal = {IEEE Transactions on Wireless Communications},
  year = {2019},
  doi = {10.1109/twc.2019.2958807},
}
Figure