Unsupervised Drones Essaim Characterization Using RF Signals Analysis and Machine Learning Methods is an open-access work by Nerya Ashush, Shlomo Greenberg, Erez Manor, and Yehuda Ben-Shimol.

Over the past decade, autonomous unmanned aerial vehicles (UAVs) have garnered significant attention from academic and industrial sectors. Drones offer numerous advantages, including civil and military applications, aerial photography and videography, mapping and surveying, agriculture, and disaster management. However, recent advancements in UAV technology have also led to its malicious use, such as breaching secure areas like airports and facilitating terrorist activities. The potential for autonomous weapon systems to deploy drone swarms for complex military operations further raises concerns.

Using large numbers of drones simultaneously enhances mission reliability through redundancy, survivability, scalability, and improved performance in complex environments. This research proposes a novel approach for characterizing and detecting drone swarms by analyzing RF (radio frequency) signals combined with various machine learning techniques. Unlike existing methods focusing on single-drone detection using supervised learning, this study introduces an unsupervised approach to drone swarm characterization.

The proposed method leverages diverse RF signatures from drone transmitters. The authors apply various frequency transforms, including continuous, discrete, wavelet scattering transforms, to extract RF features from radio frequency fingerprints. These features are then used as inputs for unsupervised classifiers. To manage the dimensionality of the input data, the authors employ dimensionality reduction techniques such as Principal Component Analysis (PCA), Independent Component Analysis (ICA), Uniform Manifold Approximation and Projection (UMAP), and t-Distributed Stochastic Neighbor Embedding (t-SNE).

Their clustering approach utilizes common unsupervised methods, including K-means, Mean Shift, and X-means algorithms. Its effectiveness has been evaluated using both custom-built and standard drone swarm datasets. Results indicate a classification accuracy of approximately 95% even under additive Gaussian white noise conditions with varying signal-to-noise ratios (SNR).

Date de publication– February 2023

Unsupervised Drones Swarm Characterization Using RF Signals Analysis and Machine Learning Methods contains the following major sections:

  • Introduction
  • Contexte et travaux connexes
  • Proposed Approach
  • Résultats expérimentaux
  • Summary and Conclusions

Crédit photo : Adobe Stock par lililia (Generated with AI)