Radio Frequency Fingerprint Based Wireless Transmitter Identification Against Malicious Attacker: An Adversarial Learning Approach
Hao Han, Li Cui, Wen Li, Luying Huang, Yuan Cai, Jihao Cai, Yuli Zhang
Abstract
This paper investigates the problem of wireless transmitter identification in the presence of malicious attacker based on radio frequency (RF) fingerprint. A robust wireless transmitter identification scheme is proposed using Generative Adversarial Networks (GAN). First, to build unique RF fingerprint of different wireless transmitter, in-phase (I) and quadrature (Q) imbalance model is introduced. Then, a classifier which consists of multiple discriminators is proposed to detect attacker and classify trusted transmitters based on RF fingerprint. Every discriminator is trained in adversarial learning framework and serves as one binary classifier to verify RF fingerprint. Wasserstein GAN with gradient penalty (WGAN-GP) is modified to train the discriminator. Finally, a simulation dataset is generated to validate the proposed approach. Experimental results reveal that the approach achieves 99.98% accuracy for detecting attacker and an average accuracy of 97.36% for classifying trusted transmitters.