Differential Complex-Valued Convolutional Neural Network-Based Individual Recognition of Communication Radiation Sources

Abstract

Data assaults from unauthorized access to the Internet of Things will induce severe intrusion and hazard to the whole network. Employing only traditional application layer password authentication approaches cannot guarantee the security of the communication system. Therefore, it is critical to develop a capable and efficient radio frequency fingerprints based physical layer authentication system. To incorporate the domain knowledge in more capable feature extracting and reduce information loss caused by converting RF baseband I/Q signals, we propose a novel differential complex-valued convolutional neural network based individual recognition approach of communication radiation sources in the paper. The proposed method can fully capture the nonlinearity of the RF baseband I/Q signals while decreasing the unfavorable impact of phase rotation induced by carrier frequency offset, which also significantly reduces the required data length of the collected steady-state data transmission section. The recognition performance evaluation on 20 wireless network card devices with duplicate batch, type, and manufacturer shows that the proposed approach has the best recognition performance compared with two conventional approaches whose recognition accuracies are lower than 95%, achieving the total recognition accuracy of 99.7%. Moreover, compared with constellation based approaches, which require at least 5,000 to 10,000 data points as input parameters, the proposed method can reduce the required data length of the collected steady-state data transmission section effectively, which is easier to implement in practical applications. INDEX TERMSRadio frequency fingerprint; constellation figure; physical layer authentication; differential process; complex-valued convolutional neural network This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.

Specific Emitter Identification of Frequency Hopping Signals Based on Feature Extraction and Deep Residual Network

“…When extracting steady-state features, sample fragments need to have a long duration. The transmitter needs to extract a large amount of RF information, which can be extracted in time domain, frequency domain, and transform domain [8], such as frequency features obtained through Empirical Mode Decomposition (EMD) [9], fractal features, high-order cumulants of signals nonlinear characteristics of power amplifier, I/Q offset characteristics [10], etc.…”

Unsupervised specific emitter identification of bispectrum features based on contrastive learning

“…Xie Cunxiang et al [5] established a model integrating Hilbert-Huang transform and adversarial training in the research process, which can achieve good recognition results even when the training samples are small. Qu Lingzhi et al [6] proposed a communication radiation source identification method by embedding a double-layer attention mechanism in the residual network, which improved the recognition accuracy and stability. In literature [7] , the network is first trained on labeled samples, and then semi-supervised learning is used to detect unlabeled samples and automatically label new samples, thus realizing dynamic identification of individual unknown radiation sources.…”

Identification of Shortwave Radio Communication Behavior Based on Autocorrelation Spectrogram Features

“…In recent years, machine learning (ML) has been applied in the field of signal processing, including signal modulation pattern recognition [13][14][15][16], CRS individual recognition [17][18][19], communication specific signal type recognition [20,21], and other fields, and a series of achievements have been achieved. However, research on CRS behavior recognition is still in the beginning stage.…”