Exploring Detection of Unmanned Aerial Systems on 5G Networks Via Machine Learning
Abstract
The development and implementation of 5G network technologies, including beamforming andsidelinking, will significantly improve the capabilities of adversaries performing drone operations beyond line of sight. This research explores methods for addressing the challenges presented by 5G for drone detection, such as when data is encrypted. The approach generates datasets for drone network traffic by capturing packets between a ground control station and a simulated drone. Then the individual communication flows are separated, and statistical fingerprints are constructed from the extracted temporal features: mean, median, and standard deviation of inter-arrival times, and packet direction ratio. These fingerprints are used to train and test a random forest classifier, which distinguishes drone traffic flows simulated over WiFi or Ethernet from normal 5G traffic flows with 99% accuracy and an F1 score greater than 98% in less than one tenth of a second. The classifier also detects drone traffic from data sent across a different transmission system than it was trained on with an F1 score greater than 97%. While due to tool limitations the drone data was not tested over 5G, detection aspects between drone data and other normal 5G data such as the data directional rate show promise regardless of transmission method. The proposed method's high performance and exclusive use of temporal features make it a promising direction to explorefor 5G drone detection.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2023
- Accession Number
- AD1213284
Entities
People
- Alexander D. Gore
Organizations
- Naval Postgraduate School