Wireless Devices Identification with Light-Weight Convolutional Neural Network Operating on Quadrant IQ Transition Image
Hiro Tamura, Kiyoshi Yanagisawa, Atsushi Shirane, Kenichi Okada · 2020
This paper presents a wireless device identification method that uses a convolutional neural network (CNN) operating on a quadrant IQ transition image. The proposed method can identify IoT wireless devices by exploiting their RF fingerprints, which is a technology to identify wireless devices using variation in analog signals. We proposed a quadrant IQ image technique to reduce the size of CNN while maintaining the accuracy. The CNN utilizes the IQ transition image, which is cut out into four-part. The over-the-air measurement was performed with six Zigbee wireless devices to confirm the validity of the proposed identification method. The measurement results demonstrate that the proposed method can achieve 99% accuracy with the light-weight CNN model with 36,500 trainable parameters. Furthermore, the proposed threshold algorithm can realize the detection of unknown devices that are not trained with 80% accuracy for further secured wireless communication.