On the Peak AoI of UAV-Assisted IoT Networks: A Stochastic Geometry Approach

Yujie Qin, Mustafa A. Kishk, Mohamed-Slim Alouini

TL;DR

This paper provides a rigorous mathematical framework using stochastic geometry to analyze the Peak Age of Information (PAoI) in UAV-assisted IoT networks. By comparing different resource allocation strategies, the authors demonstrate that time-splitting generally outperforms bandwidth-splitting for maintaining data freshness, especially in spatially clustered environments where drones can leverage high line-of-sight (LoS) probabilities.

What Problem Does This Paper Address?

In the Internet of Things (IoT), "freshness" is often more important than raw throughput. If a sensor is monitoring a forest fire or a chemical leak, a delayed packet is useless. This freshness is measured by the Age of Information (AoI). While many studies focus on average AoI, this paper targets Peak AoI (PAoI)—the maximum age reached just before a new update arrives—which represents the "worst-case scenario" for data freshness.

The challenge lies in the deployment: IoT devices are often clustered (e.g., in industrial zones), and using Unmanned Aerial Vehicles (UAVs) as mobile base stations is efficient but introduces complex interference and resource allocation problems. The authors set out to determine how to best split resources (time vs. bandwidth) among devices to keep the information as fresh as possible.

Methods

The authors employ a multi-disciplinary approach:

  • Stochastic Geometry: They model the spatial distribution of IoT devices using a Matérn Cluster Process (MCP), where devices are scattered around cluster centers, and UAVs hover directly above these centers.
  • Channel Modeling: The analysis incorporates a comprehensive channel model including Line-of-Sight (LoS) and Non-Line-of-Sight (NLoS) probabilities, Nakagami-m fading, and fractional path-loss inversion power control.
  • Queuing Theory: They use a non-preemptive queuing discipline to model how updates are generated and transmitted.
  • SINR Meta-Distribution: To handle the inherent randomness of device locations and interference, they use the SINR meta-distribution (approximated via a Beta distribution) to calculate the conditional success probability of transmissions.
  • Scenarios: They compare two load models (Bandwidth Splitting vs. Time Splitting) across two device types (Correlated—monitoring the same process; and Uncorrelated—monitoring different processes).

Datasets

The paper does not use external real-world datasets. Instead, it relies on Monte Carlo simulations to validate the derived analytical expressions. The simulation parameters (e.g., environment variables for urban vs. suburban areas, path-loss exponents) are sourced from established wireless communication literature (e.g., Al-Hourani's work on Low Altitude Platforms).

Key Insights and Ideas

  • Time-Splitting Wins: Counter-intuitively, even though time-splitting (where one device uses the full bandwidth for a short time) can lead to higher instantaneous interference, it results in a lower mean PAoI than bandwidth-splitting. This is because the shorter transmission duration per update is more beneficial for freshness than the interference reduction gained by splitting bandwidth.
  • The Power of Correlation: For "correlated" devices (multiple sensors watching the same thing), increasing the number of devices can actually improve PAoI at low data arrival rates. The UAV can effectively "pick" the freshest update from any of the devices in the cluster.
  • Environment Sensitivity: The system is highly sensitive to the "Highrise Urban" environment. In these scenarios, the LoS probability drops significantly, causing the PAoI to skyrocket because updates fail more frequently and require retransmissions.
  • UAV Advantage: The authors highlight that because UAVs hover close to clusters, the transmission distances are short and LoS links are common. This makes the "serving link" much stronger than the "interfering links" coming from other clusters, which is why the system can tolerate the higher interference of time-splitting.

Key Results

  • Analytical Accuracy: The proposed mathematical framework (using the Beta approximation for meta-distribution) matches simulation results with high precision.
  • Optimal Device Count: The study found that for correlated devices under certain arrival rates, there is an "optimal" number of devices (often around 2 per cluster) that minimizes PAoI.
  • Activity Probability: In Load Model 1 (Bandwidth Splitting), the activity probability of a device is limited by the time it takes to generate new data. In Load Model 2 (Time Splitting), the UAV is more efficient because it can use the "data generation time" of one device to serve another.

Limitations and Future Work

  • Static UAVs: The model assumes UAVs are stationary (hovering). Real-world applications might involve moving UAVs, which would require trajectory optimization to be integrated into the PAoI analysis.
  • Resource Constraints: The paper acknowledges that UAVs have finite battery life and processing power, but these constraints were not the primary focus of this specific mathematical model.
  • Mobility: The authors suggest that future research should explore "mobility-aware" IoT networks where both the devices and the collectors are in motion.
  • Limited Buffer: The study assumes a non-preemptive discipline; exploring different queuing disciplines (like Last-Come-First-Served) could yield different freshness results.

Summary updated

Abstract

In this article, we analyze the Peak Age of Information (PAoI) in unmanned aerial vehicle (UAV)-assisted Internet of Thing (IoT) networks, in which the locations of IoT devices are modeled by a () and UAVs are deployed at the cluster centers to collect the status updates from the devices. Specifically, we consider that IoT devices can either monitor the same physical process or different physical processes and UAVs split their resources, time or bandwidth, to serve the devices to avoid intercluster interference. Using tools from stochastic geometry, we are able to compute the mean activity probability of IoT devices and the conditional success probability of an individual device. We then use tools from queuing theory to compute the PAoI under two load models and two scenarios for devices, respectively. Our numerical results show interesting system insights. We first show that for a low-data arrival rate, increasing the number of correlated devices can improve the PAoI for both load models. Next, we show that even though the time-splitting technique causes higher interference, it has a limited impact on the mean PAoI, and the mean PAoI benefits more from the time-splitting technique. This is because of the nature of UAV communication, especially at places where devices (users) are spatially clustered: shorter transmission distances and better communication channels, comparing the links established by the cluster UAV and serving devices (users) to links established by interferers.
BibTeX
@article{qin2024peak,
  author = {Yujie Qin and Mustafa A. Kishk and Mohamed-Slim Alouini},
  title = {{On the Peak AoI of UAV-Assisted IoT Networks: A Stochastic Geometry Approach}},
  journal = {IEEE Internet of Things Journal},
  year = {2024},
  doi = {10.1109/jiot.2023.3320359},
}
Figure