This paper proposes a novel deep learning-based radio frequency fingerprint (RFF) identification method for internet of things (IoT) terminal authentications. Differential constellation trace figure (DCTF), a two-dimensional (2D) representation of differential relationship of signal time series, is utilized to extract RFF features without requiring any synchronization. A convolutional neural network (CNN) is then designed to identify different devices using DCTF features. Compared to the existing CNN-based RFF identification methods, the proposed DCTF-CNN possesses the merits of high identification accuracy, zero prior information and low complexity. Experimental results have demonstrated that the proposed DCTF-CNN can achieve an identification accuracy as high as 99.1% and 93.8% under SNR levels of 30 dB and 15 dB, respectively, when classifying 54 target ZigBee devices, which significantly outperforms the existing RFF identification methods.

Physical layer security key generation exploits unpredictable features from wireless channel to achieve high security, which requires high reciprocity in order to set up symmetric keys between two users. This paper investigates enhancing the channel reciprocity using a loop-back scheme with multiple frequency bands in time-division duplex (TDD) communication systems, in order to mitigate the effect of hardware fingerprint interference and synchronization offset. The scheme is evaluated to be robust to passive eavesdropping and active Man-in-the-Middle attack through both theoretical analyses and practical measurements. A secret key generation protocol is subsequently designed. The performance of the proposed secret key generation method is then evaluated through both numerical simulation and hardware experiments. Results demonstrate that the proposed scheme can effectively mitigate non-reciprocity and outperforms the classical TDD scheme in both key disagreement rate and key generation rate.

Recently, semantic communication (SC) has been regarded as one of the most potential paradigms of 6G. Current SC frameworks require the physical layer channel state information (CSI) in order to handle the severe signal distortion induced by channel fading. Since practical CSI cannot be obtained accurately and the overhead of channel estimation cannot be neglected, we therefore propose a generative adversarial network (GAN) based SC framework (Ti-GSC) that doesn’t require CSI. In Ti-GSC, there are two main modules, i.e., an autoencoder-based encoder-decoder module (AEDM) and a GAN-based non-CSI signal distortion suppression (SDS) module (GSDSM), where SDS only relies on learning the syntactic distribution and the semantics of the transmitted data, so no prior information such as CSI is needed by GSDSM. In order to measure signal distortion, a novel loss function is proposed where two terms, i.e., a syntactic distortion loss term and a semantic distortion loss term, are newly added, and a differentiable semantic measurement method is designed based on the intermediate layers of the AEDM decoder. To achieve better training results of Ti-GSC, two training schemes, i.e., the joint optimization based training (JOT) and the alternating optimization based training (AOT) are designed for the proposed Ti-GSC. Experimental results show that JOT is more efficient for Ti-GSC, and Ti-GSC outperforms conventional communication frameworks in terms of bilingual evaluation understudy (BLEU) score in both Rician and Rayleigh fading channels. Moreover, without CSI, the BLEU score achieved by Ti-GSC is about 40% and 62% higher than that achieved by existing SC frameworks in Rician and Rayleigh fading, respectively. Besides, each term of the presented loss function has a great impact on the BLEU performance of Ti-GSC, where in Rician fading syntactic learning has the greatest impact, and in Rayleigh fading, the adversarial learning becomes important.

The radio frequency fingerprint (RFF)-based device identification is a promising physical layer authentication technique. However, the wireless channel significantly affects the RFF features of the wideband wireless devices. In this paper, we extensively investigate the impact of channel variation on RFF identification using 20 MHz IEEE 802.11 signal. A time domain least mean square (LMS) equalization-based feature extraction method has been proposed. This method progressively restores the transmitted signal and preserves more details of RFF features than the classical frequency domain equalization (FDE) method. Moreover, a hybrid identifier is proposed to take advantage of both LMS-based and FDE-based methods. With the equalized samples, a four-layer convolutional neural network is designed for device identification. An experimental system has been set up to capture the waveform of 68 802.11 devices at different positions. The experimental results show that the LMS-based method outperforms others when the acquisition positions of the training dataset are the same as those of the testing dataset. On the other hand, the FDE-based method is shown to be more effective when the acquisition positions of the training dataset do not fully include those of the testing dataset. Moreover, the hybrid identifier achieves an improvement of 2% for overall identification accuracy.

Radio frequency fingerprint identification (RFFI) is a promising authentication mechanism for physical layer security. In this paper, we thoroughly validate the feasibility of using RFFI for cellular long-term evolution (LTE) devices. Firstly, we conduct simulations to examine the subtle impacts of hardware impairments on LTE signals. The simulated results reveal that I/Q imbalance and power amplifier non-linearity introduce significant distortions within in-band spectrum, forming unique hardware fingerprints. We then leverage the strong channel correlation between adjacent subcarriers and separate the channel-robust radio frequency fingerprints (RFF) from uplink demodulation reference signal (DMRS) in Msg3. Subsequently, we construct a hybrid feature matrix to serve as input for a shallow long short-term memory (LSTM) network. Due to the more effective channel mitigation strategy, our method outperforms three benchmarks in terms of classification accuracy under cross-scenario testing. Additionally, we explore the impacts of bandwidth configuration on RFFI, and experimental findings demonstrate that LTE terminals will exhibit more distinct RFF when occupying a larger number of physical resource blocks (RB) during transmission. We also investigate the stability of RFF towards frequency band variations. The results suggest that there will be a significant accuracy loss under training with one band but testing with another, indicating the importance of frequency band-independent feature extraction in practical environments. Lastly, we expose four key implications to pave the way for exploring corresponding solutions. To the best of our knowledge, it is the first performance evaluation of the RFFI system on different frequency bands and with multiple bandwidth configurations.

Radio frequency fingerprint identification (RFFI) is a promising technique that exploits the transmitter-specific characteristics of the RF chain for identification. Disregarding its massive deployment, long-term evolution (LTE) systems have not fully benefited from RFFI. In this paper, an RFFI technique is designed to authenticate LTE devices. Three segments of the LTE physical layer random access channel (PRACH) preambles are captured, namely the transient-on, transient-off, and modulation parts. The segments are first converted into differential constellation trace figures (DCTFs), and then a specific type of neural network called multi-channel convolutional neural network (MCCNN) is used for identification. Additionally, the protocol is able to be applied for open-set identification, i.e., unknown device detection. Experiments are conducted with ten LTE mobile phones. The results show that the proposed RFFI scheme is robust against location changes. In the known device classification problem, the classification accuracy can reach 98.70% in the line-of-sight (LOS) scenario and 89.40% in the non-line-of-sight (NLOS) scenario. In the open-set unknown device detection problem, the identification equal error rate (EER) and area under the curve (AUC) reach 0.0545 and 0.9817, respectively, among six known devices and four unknown devices.