Machine Learning Based Spectrum Fingerprinting of Drones for Defensive Cyber Operations

Abstract

The rapid development of unmanned aerial vehicle (UAV) or drone capabilities in the past decade has significantly expanded the commercial, military, and consumer applications for these innovative airborne devices. Characterized by fixed wings or multiple rotors, drones are valued for their long-range flight, their lightweight de-sign, and their imaging and sensory capabilities. Traditionally operated by radio controllers on dedicated channels, modern drones are evolving towards autonomous, machine- controlled swarms of tactical UAVs that are capable of fulfilling an array of complex purposes. However, due to the size and relatively limited battery power of these devices, the computing capabilities and onboard software embedded in these functional tools remain extremely limited. As a growing number of malicious actors seek to disrupt, hijack, and misdirect drone flight paths, the challenge of securing drones is an important academic problem. From drone hijacking to denial of service (DoS) to signal interference, the common techniques for affecting drone flight reliabil-ity are simple, high-powered, and widely available to the general public. The current study analyzes the relationship between drone risk management capabilities and the opportunities afforded by trained machine learning models. This study demonstrates the viability of algorithmic in-flight data monitoring and security threat detection for future onboard applications by applying a Python-based semi-supervised training set to several machine learning solutions. Further extension of these findings to swarm-based, multi-drone fingerprinting and flight monitoring demonstrates the potential for networked threat identification and security management. Ultimately, these findings propose a novel model that incorporates both onboard and offline machine learning capabilities into a shield-based software solution that can detect and respond to flight anomalies and changing threat patterns of malicious actors.