Fine-Grained Radio Frequency Fingerprint Recognition Network Based on Attention Mechanism

Area of Science:

• Cybersecurity
• Machine Learning
• Wireless Communications

Background:

• The Internet of Things (IoT) connects millions of devices, increasing vulnerability to cyber threats.
• Traditional encryption is unsuitable for IoT due to complexity and high communication overhead.
• Radio Frequency (RF) fingerprinting offers a promising alternative, leveraging unique transmitter hardware defects.

Purpose of the Study:

• To address challenges in RF fingerprint recognition, specifically low inter-class and high intra-class variations.
• To propose a novel fine-grained RF fingerprint recognition network (FGRFNet) inspired by computer vision techniques.
• To enhance the security and identification accuracy of diverse IoT devices.

Main Methods:

• Developed a fine-grained RF fingerprint recognition network (FGRFNet).
• Incorporated a top-down feature pathway hierarchy for pyramidal feature generation.
• Utilized attention modules for discriminative region localization and a fusion module for adaptive multi-scale feature integration.

Main Results:

• Achieved 89.8% recognition accuracy on 100 ADS-B devices.
• Demonstrated 99.5% recognition accuracy on 54 Zigbee devices.
• Obtained 83.0% recognition accuracy on 25 LoRa devices.

Conclusions:

• FGRFNet effectively overcomes limitations of traditional RF fingerprinting methods.
• The proposed network shows high recognition accuracy across different IoT device types.
• This approach offers a robust solution for securing the Internet of Things.