In recent years, neural network algorithms have demonstrated tremendous potential for modulation classification. Deep learning methods typically take raw signals or convert signals into time–frequency images as inputs to convolutional neural networks (CNNs) or recurrent neural networks (RNNs). However, with the advancement of graph neural networks (GNNs), a new approach has been introduced involving transforming time series data into graph structures. In this study, we propose a CNN-transformer graph neural network (CTGNet) for modulation classification, to uncover complex representations in signal data. First, we apply sliding window processing to the original signals, obtaining signal subsequences and reorganizing them into a signal subsequence matrix. Subsequently, we employ CTGNet, which adaptively maps the preprocessed signal matrices into graph structures, and utilize a graph neural network based on GraphSAGE and DMoNPool for classification. Extensive experiments demonstrated that our method outperformed advanced deep learning techniques, achieving the highest recognition accuracy. This underscores CTGNet’s significant advantage in capturing key features in signal data and providing an effective solution for modulation classification tasks.

Radio-frequency fingerprinting identification (RFFI), based on hardware inherent imperfections, is an effective technique for physical authentication in the Internet of Things (IoT). However, due to the nonstationary and open factors in realistic scenes, problems, such as inaccurate fingerprinting extractions and unreliable transmitter identifications, often occur. Specifically, in actual scenarios, there are challenges, such as varying channel status, variable transmission content, insufficient data set acquisition, and the unknown number of rogue devices. In this article, a robust and protocol-agnostic RFFI system is proposed for open-set recognition, which is achieved by a slicing-enhanced preprocessing with noise augmentation and a well-designed bone network with a modified open layer. Moreover, in order to verify the effectiveness and transferability of the designed RFFI system in applications of IoT, comparative experiments are conducted using a real-world data set of automatic-dependent surveillance-broadcast (ADS-B) signals and the open-source FIT/CorteXlab data set. The performance of the designed extractor and classifier, under varying channel and openness conditions, verifies the channel feasibility and openness generalization of our designed system. The results of the ablation experiment indicate that the designed RFFI system can achieve an accuracy of 96.28% in the classification of 50 registered IoT transmitters and 87.52% in open-set recognition of mixed 50 registered and 10 unregistered IoT transmitters.

Multi-signal detection is of great significance in civil and military fields, such as cognitive radio (CR), spectrum monitoring, and signal reconnaissance, which refers to jointly detecting the presence of multiple signals in the observed frequency band, as well as estimating their carrier frequencies and bandwidths. In this work, a deep learning-based framework named SigdetNet is proposed, which takes the power spectrum as the network’s input to localize the spectral locations of the signals. In the proposed framework, Welch’s periodogram is applied to reduce the variance in the power spectral density (PSD), followed by logarithmic transformation for signal enhancement. In particular, an encoder-decoder network with the embedding pyramid pooling module is constructed, aiming to extract multi-scale features relevant to signal detection. The influence of the frequency resolution, network architecture, and loss function on the detection performance is investigated. Extensive simulations are carried out to demonstrate that the proposed multi-signal detection method can achieve better performance than the other benchmark schemes.

This paper proposes a new digital signal recovery (DSR) technique for next-generation power efficient low Earth orbit (LEO) satellite-to-ground communication systems, which feature additive white Gaussian noise (AWGN) channel and significant power variation. This technique utilizes the prior knowledge (i.e., nonlinearities of radio frequency power amplifiers (RF-PAs)) of space-borne transmitters to improve the quality of the signal received at ground stations by modeling and mitigating the imperfection using deep neural networks (DNNs). Benefiting from its robustness against noise and power variation, the proposed DNN based DSR technique (DNN-DSR), can correct high signal distortions caused by the nonlinearities and hence allows RF-PAs to work close to their saturation region, leading to a high power efficiency of the LEO satellites. This work has been validated by both simulations and experiments, in comparison with the power back-off technique as well as memory polynomial based DSR solutions. Experimental results show that the DNN-DSR technique can increase the drain efficiency of the space-borne RF-PA from 32.6% to 45% while maintaining the same error vector magnitude as the power back-off technique. It has also been demonstrated that the proposed DNN-DSR technique can handle a signal power variation of 12 dB, which is challenging for conventional solutions.

The parameter estimation and sorting of frequency-hopping (FH) signals are of great importance in both civilian and military applications. In this article, the estimation of hopping time, hopping period, hopping frequency, instantaneous bandwidth, and 2-D direction of arrival (2-D-DOA) of multiple FH signals in the uniform circular array (UCA) is studied. First, a novel lightweight model based on You Only Look Once (YOLO), named FH-GYOLO, is proposed to detect FH signals and estimate their time–frequency (TF) parameters. The proposed FH-GYOLO takes advantage of the ghost module, which can significantly reduce the computational cost and, simultaneously, achieve an efficient detection with high performance. Then, the 2-D-DOA estimation method of FH signals in UCA is proposed with the spatial TF distribution (STFD). In the proposed method, the steering vector of UCA and the STFD matrix are transformed into beamspace by beamspace transformation, and the closed-form solution is derived for the estimation of autopaired elevation and azimuth angles without the computationally expensive spatial–spectral search. Furthermore, in the proposed method, an iterative compensation process is developed, which can greatly improve the accuracy of DOA estimation without significantly increasing the computational complexity. Extensive simulation results and analysis demonstrated that the proposed algorithms exhibit better or similar performance than the other benchmark methods, and most importantly, the computational cost is substantially reduced.

Radio frequency fingerprints are commonly exploited as a unique signature in the physical layer for distinguishing transmitters in transmitter identification systems (TISs). In response to the growing demand for TIS in open dynamic scenarios, this article proposes the incremental open-set recognition (IOSR) framework to address IOSR tasks, which involve changes in transmitter categories, component replacements, and open-set recognition (OSR). To overcome the limitations of traditional methods, the proposed framework focuses on enhancing the security, adaptability, reliability, and efficiency of TIS. Specifically, a well-designed data representation and a lightweight extractor based on supervised contrastive learning are introduced to improve interclass discriminative ability and intraclass compactness for feature extraction. The incorporation of MobileNetV3 reduces the training parameters of the extractor while improving computational efficiency. Moreover, an adaptive evolved block is designed to mitigate catastrophic forgetting in incremental learning, preserving historical knowledge and enhancing system scalability and adaptability. Additionally, an enhanced open-set recognizer is proposed to establish a suitable open-set decision boundary through output calibration. The performance evaluation of the framework on the WiFi data set showcases its superiority over existing approaches in the closed-set recognition, achieving an accuracy of over 99.6%. It also performs well in incremental tasks, with an accuracy exceeding 98.9%. In the OSR, the framework achieves an accuracy improvement of approximately 8%. Moreover, it demonstrates superior accuracy in the IOSR task, outperforming other algorithms by more than 5.8%. Furthermore, ablation experiments provide further evidence of the effectiveness of the proposed framework, while a complexity comparison demonstrates its ability to balance computational load and accuracy.