Koopman-based Prediction of Connectivity for Flying Ad Hoc Networks

TL;DR

This work addresses predicting connectivity in highly dynamic FANETs using data-driven Koopman approaches to model UAV trajectories and forecast SINR-based outages. It develops two complementary frameworks: a centralized Graph-KeKoopman Autoencoder (GKAE) and a distributed per-UAV Koopman Autoencoder (KAE), leveraging Koopman operator theory to linearize nonlinear UAV dynamics in a latent space. The methods enable long-horizon SINR predictions and isolation-event detection, with simulations showing that the centralized approach substantially improves predictive accuracy (about 75–80% on average) at the expense of higher computational load, while the distributed approach offers scalable, privacy-preserving predictions with lower energy costs. The results demonstrate the potential of Koopman-based connectivity prediction to inform routing and transmission scheduling in FANETs, enhancing reliability and efficiency in dynamic aerial networks.

Abstract

The application of machine learning (ML) to communication systems is expected to play a pivotal role in future artificial intelligence (AI)-based next-generation wireless networks. While most existing works focus on ML techniques for static wireless environments, they often face limitations when applied to highly dynamic environments, such as flying ad hoc networks (FANETs). This paper explores the use of data-driven Koopman approaches to address these challenges. Specifically, we investigate how these approaches can model UAV trajectory dynamics within FANETs, enabling more accurate predictions and improved network performance. By leveraging Koopman operator theory, we propose two possible approaches -- centralized and distributed -- to efficiently address the challenges posed by the constantly changing topology of FANETs. To demonstrate this, we consider a FANET performing surveillance with UAVs following pre-determined trajectories and predict signal-to-interference-plus-noise ratios (SINRs) to ensure reliable communication between UAVs. Our results show that these approaches can accurately predict connectivity and isolation events that lead to modelled communication outages. This capability could help UAVs schedule their transmissions based on these predictions.

Figures (8)

• Figure 1: A diagram illustrating the deployment of FANETs for surveillance operations, where each UAV is assigned a specific area (denoted by colored circles) to monitor and follows predetermined flight paths. Intra-UAV links (dotted lines) ensure seamless communication and coordination among UAVs,
• Figure 2: Two approaches are proposed for proactively detecting isolation events: 1) In the distributed approach, each UAV utilizes its local observations to train a KAE, enabling it to predict connectivity with other UAVs. 2) In the centralized approach, a central unit gathers information from all UAVs within the operational area to collectively assess the connectivity of each UAV.
• Figure 3: (Left) The dynamics of $L = 4$ UAVs with varying cycle durations; (right) Quasi-periodic SINR for UAV $1$ over $2000$ time steps.
• Figure 4: (Left) Prediction error over varying UAVs when $P = 50$ time steps, aggregated over 1900 trajectory predictions, each starting from a unique initial point. (Right) Computational complexity comparison between training the GKAE and the KAE approaches, where the per UAV computational cost is highlighted using a black line.
• Figure 5: (Top) Comparison of 20-step predictions up to 2000 time steps for UAV 1 using two different approaches. (Bottom) Aggregated prediction error comparison for the two approaches over 1000 20-step predictions.