< Back to previous page

Project

Drone detection, classification and neutralization through artificial intelligence based RF sensing and protocol aware jamming

Recent technological advancements and the growing popularity of drones have brought about many beneficial applications across various industries. However, their widespread availability has resulted in a rise in illegal activities. Implementing effective Counter-UAS (C-UAS) is essential to protect critical infrastructure, enforce regulations, and prevent contraband deliveries. By promptly identifying unauthorized drones, authorities can take the necessary actions to ensure responsible use of drones and mitigate potential liabilities and risks associated with their misuse. Radio Frequency (RF)-based drone detection is of significant importance in countering the illegal use of drones due to its effectiveness and versatility. RF detection systems monitor the RF signals transmitted by a drone and its Ground Control Station (GCS) in order to detect and classify them. However, conventional classification algorithms face significant challenges in accurately classifying drone RF protocols due to the prevalence of widely used RF communication systems, such as Wi-Fi and Bluetooth, operating in the same frequency bands (2.4 GHz and 5 GHz ISM bands) and utilizing similar modulation techniques like spread spectrum modulation. This study aims to address these challenges by utilizing Artificial Intelligence (AI)-based algorithms to automatically detect and classify drone RF protocols. We achieve our study objective with the following five key contributions: (1) First, a novel and detailed RF database is created using commercial drone RF and Wi-Fi communication signals. This database facilitates the development of a Deep Learning (DL)-based detection and classification framework, addressing the absence of an open-source drone RF database for training such frameworks. (2) Second, this thesis proposes RF signal detection, feature extraction, and classification simultaneously using the You Only Look Once (YOLO) framework. We develop a YOLO framework to detect and classify multiple signals simultaneously, even at low SNR conditions. (3) Third, following signal detection and feature extraction, novelty detection is performed to determine whether the detected signal is known to the classifier or novel. (4) Fourth, this thesis proposes a Residual network-based model to classify the drone RF protocols. (5) Finally, we develop a spectrum prediction framework to predict the future time and frequency sequences for smart RF jamming.

Date:9 Aug 2019 →  6 Dec 2023
Keywords:Spectrum monitoring, Protocol aware jamming, Deep learning
Disciplines:Microwave and millimetre wave technology, Signal processing
Project type:PhD project