As part of the COREnext project, researchers demonstrated an AI-enabled physicallayer security solution based on radio frequency (RF) fingerprinting – a technique that enables the authentication of wireless devices by exploiting subtle, hardware-based signal characteristics. The demonstration showcased how machine learning can identify, authenticate, and protect against impersonation attacks in communication network systems.
Enhancing Physical Layer Security with RF Fingerprinting
Every radio transmitter, even when manufactured to the same specifications, exhibits unique hardware imperfections – such as variations in amplifiers, oscillators, and other circuits. These imperfections leave a distinct “fingerprint” on the transmitted signal.
The RF fingerprinting method leverages these differences to create a unique identifier for each device, making it possible to authenticate transmitters not only by their software credentials or classical cryptography- based methods, but by their physical-layer signal signatures too.
In this demonstration, 5G New Radio (NR) signals at 2.5 GHz are used as authentication signal. By using techniques such as operating the radio amplifiers at high gain, the researchers amplified the nonlinearities in the signal — the very characteristics that make each device’s transmission unique.
Device Authentication through Machine Learning
The setup included three software-defined radio devices from the same manufacturer, two of which were identical models — a challenging scenario for classification. A convolutional neural network (CNN) was trained to analyse short (1.4 ms) signal transmissions and classify each device based on its RF fingerprint.
Despite the physical similarity between devices, the trained model achieved near 100% accuracy in distinguishing between all three. Visualisations of the model feature space showed well-separated clusters for each device, confirming that the neural network successfully captured subtle signal-level distinctions invisible to traditional security mechanisms.
Testing the System’s Robustness – Impersonation Attack Scenario
To assess the system’s resilience, the researchers simulated an impersonation attack where a malicious actor attempts to mimic a legitimate device signal using identical transmission settings, with a device of the same model as the target. This test represented an open-set scenario, meaning the attacker’s device had not been part of the model’s training data.
When the impersonating device transmitted its signal, the machine learning model successfully detected it as an unknown or anomalous device, labelling it as unauthorised.
Towards AI-Driven Physical Layer Security
This demonstration highlights the potential of combining AI techniques with physical-layer security to protect wireless systems from spoofing and impersonation attacks. In addition to traditional cryptographic methods, RF fingerprinting provides a hardware-rooted form of authentication that is extremely difficult to forge or tamper with.
As the COREnext project continues to advance secure communication technologies, AI-enabled RF fingerprinting stands out as a powerful approach to ensure device authenticity, resilience against attacks, and trustworthy operation in next-generation networks.