Cognitive radio systems have a recognized need for automatic modulation recognition (AMR). Prior studies on recognizing modulation classes have faced challenges in efficiently extracting reliable and robust features from complex-valued signals. To conquer the issue, we propose a dual-channel transformer (DualFormer) to integrate temporal dependence characteristics from in-phase and quadrature (IQ) channels. Specifically, the complex-valued signal is decomposed into two orthogonal channels, and each channel is divided into fixed-length patches for acting as tokens. Then, parameter-shared transformers embed each patch into a fixed-dimension feature. Lastly, a tailored projection header fuses patch features and outputs modulation-scheme probability. The proposed DualFormer with two independent I/Q channels as inputs can theoretically achieve lower generalization error boundaries than a single real channel and a complex-valued channel. Experiments on a public dataset RadioML 2018.10A reveals that the proposed method can achieve 85.5% with signal-to-noise ratio (SNR) at 8 dB.

Specific emitter identification (SEI) is a novel remote sensing method capable of detecting the external features of intercepted signals to recognize emitter identities. In order to accommodate a wider range of application scenarios, a novel SEI framework based on self-supervised learning (SSL) is proposed, which incorporates close-set classification, open-set recognition, and clustering. The whole procedure can be divided into two main steps, i.e., the pretext task and the downstream task. The pretext task involves the development of a novel phase space reconstruction-based FoldingNet (PSR-FoldingNet) that can extract latent fingerprint features from the reconstructed phase space. The downstream task involves the designation of several predictors containing multi-layer perception (MLP), OpenMax, and K-means. In this study, radar signals collected in the laboratory are employed as the objects of analysis. Based on the experiment results, it is evident that the proposed architecture can efficiently extract fingerprint features without requiring the ground truth in advance. Additionally, fingerprint features outperform other state-of-the-art (SOTA) methods in the SEI task.

The advancements in Automatic Modulation Classification (AMC) have propelled the development of signal sensing and identification technologies in non-cooperative communication scenarios but also enable eavesdroppers to effectively intercept user signals in wireless communication environments. To protect user privacy in communication links, we have optimized the adversarial example generation model and introduced a novel framework for generating adversarial perturbations for transmitted signals. This framework implements dual-domain constraints in both the time and frequency domains, ensuring that the adversarial perturbation cannot be filtered out. Comparative experiments confirm the superiority of the proposed method and the concealment of the adversarial examples it generates.

Automatic modulation classification (AMC) is crucial for ensuring secure and efficient operation of the Internet of Things (IoT). In this study, we focus on addressing the challenge of cross-domain modulation classification, which arises due to differences in data distribution between the source and target domains. Current research focuses on leveraging multimodal information to improve performance, but struggles with mapping diverse domains into a common feature space for effective classification. Additionally, most existing domain-adversarial approaches measure the distribution differences between the source and target domains using batch sample average differences, without considering the labels of the samples. To tackle these issues, we propose a progressive multisource partial domain adaptation modulation classification (PMSPDMC) method. PMSPDMC utilizes multiple sources comprising three sequence modalities and an auxiliary modality. It consists of two main components: 1) feature space alignment and 2) decision space alignment. In feature space alignment, we design a hybrid metric function to extract distinctive features that maintain tight intraclass distances and separable interclass distances, achieving fine-grained alignment between the source and target domains in the feature space. To achieve decision space alignment, a weighted fusion strategy and prediction consistency regularization are applied to the final classification prediction. This design of distribution difference measurement criteria effectively addresses the practical situation where the target domain has fewer modulation types than the source domain. The simulation results demonstrate that the proposed PMSPDMC exhibits performance gains in the cross-domain modulation classification tasks.

Interference cancellation (IC) is crucial for ensuring the continuous operability of wireless communication systems based on the Internet of Things (IoT). This study focuses on blind signal separation (BSS)-based IC for co-channel multiuser systems. Considering humans’ selective auditory attention abilities, we propose a novel top-down auto-focusing target signal extraction (TSE) method, which has been developed according to our recently established data-driven BSS scheme. In the signal separation stage, we input clues regarding the target signal into the separation system to guide it toward the target signal; as a result the final output is only the desired signal. This approach mitigates the global permutation ambiguity and eliminates the need for prior knowledge or estimation of signal numbers in existing BSS schemes. This study focuses on the clue encoder and clue fusion layer, which can be integrated into existing data-driven single-channel BSS schemes. In the clue encoding process, we use multiple modal clues as inputs to leverage their complementary advantages. Additionally, we apply a multimodal clue feature space alignment method to reduce the impact of feature space distribution differences on the interactions of multimodal clue information. In the clue fusion process, we propose an attention-based clue fusion scheme to provide more informative target signal clues for at each time frame. Finally, during the training process, we implement a multitask learning approach that considers the losses in different modal clues, thereby enabling the separation system to function properly even when some clues are unavailable. Numerical results confirm the effectiveness of the proposed scheme.

Automatic Modulation Classification (AMC) plays a vital role in enhancing spectrum utilization efficiency in Cognitive Internet of Things (CIoT) scenarios. However, it introduces security vulnerabilities by enabling eavesdroppers to accurately identify modulation schemes of intercepted signals through deep neural network (DNN)-based classifiers. Considering that DNN performance is highly susceptible to adversarial examples, this work aims to design adversarial perturbations for signal transmitters to effectively prevent eavesdroppers from intercepting and identifying target user signals. Unlike image classification tasks, existing methods for generating adversarial perturbations in wireless communications face three critical limitations: instability in perturbation power control, vulnerability to signal filtering, and high perceptibility of perturbations. To address these challenges, we reformulate the adversarial example generation problem with specific target perturbation power constraints. By jointly leveraging feature vectors from the final DNN layer and frequency-domain characteristics of transmitted signals, we propose a Cross-Domain Power Constrained Adversarial Attack (CPCAA) method that further optimizes perturbation generation. Experimental results demonstrate that the proposed method generates highly imperceptible adversarial examples with specified energy constraints while achieving significant attack performance, even when eavesdroppers deploy various types of filters.

In non-cooperative environments, deep learning-based automatic modulation recognition techniques often struggle with the situations with insufficient or even no training data accessible. In this letter, we investigate this problem in the amplitude-phase-modulation recognition task and introduce a knowledge-informed waveform description for zero-shot recognition generalization. Specifically, drawing inspiration from constellation association knowledge, we define a constellation-based semantic attribute set to describe waveform structures and employ graph formulation to model attributes’ symmetric dependency for improving representations. Subsequently, we align the waveform and semantic spaces by associating waveform and attribute compositional representations, facilitating the transfer of knowledge from the seen to unseen domain. Our scheme can reason the labels of unseen waveform types with the guidance of the attribute description outputting, beyond merely distinguishing test instances as unseen. Experiments validate the efficacy of the proposed method across few-shot and zero-shot recognition tasks.

Cognitive radio systems have a recognized need for Zero-Shot Automatic Modulation Recognition (ZSAMR). Previous research aimed at identifying unobserved modulation types struggles to incorporate text-format expert insights into zero-shot learning approaches for unseen modulation schemes. To overcome these limitations, we propose a novel ZSAMR framework utilizing a vision-language model (ZSAMR-VLM). The ZSAMR-VLM leverages the pre-trained vision-language model to describe multiple views of signals with preset prompts. Modulation scheme prediction is achieved by matching multi-view signal characteristics with manually defined semantic prototypes. By facilitating linguistic interactions, the ZSAMR-VLM allows signal processing experts to embed domain-specific knowledge directly into the modulation recognition process, effectively bridging the gap between modulation recognition and natural language. Our simulation experiments demonstrate that the ZSAMR-VLM can effectively implement zero-shot recognition without relying on instances from seen classes. Furthermore, comparative experiments suggest that incorporating multiple views of communication signals significantly enhances the performance of the ZSAMR-VLM. The source code is available at https://github.com/zhaoyurui/ZSAMR-VLM.

Automatic Modulation Classification (AMC) aims to identify the modulation type of received signals, playing a crucial role in ensuring secure dynamic spectrum access in the cognitive radio-enabled Internet of Things (CR-IoT). Most existing algorithms assume that the receiver is affected by only one desired transmitter, solely focusing on identifying the modulation type of that specific transmitter. However, owing to the broadcast nature of wireless communication, the received signal is often a time-frequency overlapped co-channel signal, requiring identification of the modulation type of all transmitted signals. Notably, current single-user modulation classification (SUMC) methods cannot implement co-channel multi-user modulation classification (MUMC). To address this problem, this paper proposes the first end-to-end MUMC scheme to support blind signal separation (BSS). This scheme consists of two stages: signal recovery and modulation type identification. First, we propose a data-driven sparse component analysis-based BSS model to recover target signals from the received signals. Subsequently, the modulation types of the recovered signals are classified using an SUMC model. Additionally, a number counter is developed to cope with changes in the number of target signals. The proposed scheme pioneers the development of a BSS model based on physical characteristics of the communication signal. Numerical results verify that it outperforms existing competitive MUMC methods, particularly for high-order modulation types.

Spectrum sensing (SS) is essential in the cognitive radio-enabled Internet of Things (CR-IoT) to enable spectrum resource allocation. In this study, we propose a novel semantic segmentation-based deep SS (SSDSS) scheme for co-channel signals. Through the extraction of the interleaved signal-wise differential feature (SWDF) sequence, interference among the target signals is efficiently alleviated. Our approach formulates the SS task as a sequence semantic segmentation problem using the interleaved SWDF sequence, which encodes symbol information from latent signals. By constructing three semantic segmentation networks, we successfully extract the SWDF sequence for each latent signal by classifying sequence elements based on their semantics. This aspect enables us to detect the activity time of different target signals within the mixed signal. Notably, our method achieves co-channel signals detection with multiple targets, all accomplished within a single step and a single network. Numerical results confirm strong detection performance against co-channel signals with varying power ratios. Additionally, the SWDF sequence demonstrates robustness against element loss and noise. Furthermore, the offline-trained SSDSS performs well in diverse open-set tests. Moreover, this research highlights the superiority of self-attention mechanisms and recurrent neural networks (NNs) over convolutional NNs for this sequence modeling task.

Although Deep learning (DL) provides state-of-art results for most spectrum sensing tasks, it is vulnerable to adversarial examples. Based on this phenomenon, we consider a non-cooperative communication scenario where an intruder tries to recognize the modulation type of the intercepted signal. Specifically, this paper aims to minimize the intruder’s accuracy while guaranteeing that the intended receiver can still recover the underlying message with the highest reliability. This process is implemented by adding adversarial perturbations to the channel input symbols at the encoder. In image classification, the perturbation is limited to be imperceptible to a human observer by minimizing the ℓp norm, while in this work, we enriched the connotation of adversarial examples, and first proposed that the imperceptibility of adversarial examples in the field of wireless signals is the imperceptibility of filters. Based on this perspective, we optimized the model of adversarial examples and constrained the adversarial perturbation to a narrow frequency band so that filters cannot filter it out. We also define a new set of metrics to describe the imperceptibility of the wireless signal adversarial example. The simulation results demonstrate the viability of our approach in securing wireless communication against state-of-the-art DL-based intruders while minimizing communication performance reduction.

With the popularization of the Internet of Things (IoT), its security has become increasingly prominent. Radio-frequency fingerprinting (RFF) is a promising approach to identify a specific emitter by extracting the intrinsic physical layer characteristics from transmitted signals, adopted as a lightweight noncryptographic access authentication technique. The realization of RFF relies on unintentional modulation of the pulse (UMOP). In this article, we discover the locality and inhomogeneity of RFF, that is, UMOP is concentrating in the radio-frequency fingerprint distribution subregion (RFDR), instead of evenly distributed over the entire feature. This property is demonstrated in the cepstral domain based on the transmitter distortion models. This is the first time that the inhomogeneity of UMOP and the cepstral domain analysis are employed in RFF. First, to automatically search the RFDRs, a method based on the dispersion characteristic index (DCI) and emitter-specific information index (ESI) is proposed, where DCI provides constraints on the scattering characteristics of RFF feature clusters and ESI evaluates the local differences of RFF features. Second, combining the RFDR selection method with the cepstrum analysis, a new feature CepH is obtained. Systematic experiments are conducted on the simulated data sets containing multiple modulation patterns and received data sets from multiple sources, which show that the RFDR is not affected by the modulation pattern and noise, and has high practicability. It also demonstrates that the proposed algorithm outperforms the state-of-art algorithms, especially in the case of a low signal-to-noise ratio (SNR). It can achieve a recognition accuracy of more than 90% under 0 dB.

With the advance in wireless network technique, its security becomes of paramount importance. Radio Frequency Fingerprint (RFF) is the underlying characteristic of hardware chains in transmitters, which can be used as a unique ID for specific emitter identification (SEI) and a non-cryptographic access authentication technology in physical layer to enhance wireless network security. To date, few studies have extracted the inevitable non-linearity in the transmitter as RFF features. Hence, this letter provides a novel nonlinear dynamics approach based on Multi-dimension Approximate Entropy (MApEn) for SEI. Specifically, this method utilizes the steady-state portion of the preamble structure of Standard IEEE802.11b/g to obtain the nonlinear properties of wireless network cards. The experimental results demonstrate that the proposed identification algorithm outperforms the existing steady-state methods in terms of the identification accuracy.

Despreading signal at receiver side requires prior knowledge of the spreading sequences in direct‐sequence code division multiple access (DS‐CDMA) system. However, the knowledge of spreading sequences is always unknown to the receiver in non‐cooperative communication. Therefore the receiver has to estimate the spreading sequences in a blind manner. This study presents a novel algorithm to estimate the spreading sequences for asynchronous long‐code (LC) DS‐CDMA signals. By investigating the projection characteristics of the signal subspace, the authors treat the spreading sequence estimation problem as a discrete optimisation search problem that can be solved efficiently by modern optimisation search methods. Moreover, they design a successive subspace deflation scheme to alleviate interference from other users. Numerical experiments demonstrate that the proposed algorithm provides good estimation of the spreading sequences for LC‐DS‐CDMA signals. Compared with the existing algorithms, the proposed algorithm exhibits much better performance in large system load scenario with low signal‐to‐noise ratio.

The beneficial characteristics of signal cyclostationarity have been widely exploited in varied domains. However, one common problem is that algorithms adopting cyclostationarity would suffer from severe performance degradation when there is an error or mismatch in the cycle frequency (CF) involved. Hence, the authors attempt to propose a single channel correction method for CF error (CFE). First, a new cyclostationary spectral feature function is introduced which reaches its maxima at CFs; second, the steepest descent algorithm is deduced for the correction of CFE, the convergence property of which is also discussed. The main contributions of the novel scheme are that first, the estimation accuracy is remarkably improved compared with most existing algorithms adopting cyclic feature functions. Second, the steepest descent scheme holds an advantage in computational complexity compared with previous works based on linear search. Moreover, the proposed algorithm is robust against the pulse shape variation and multipath fading. Simulation works are carried out, which demonstrate the performance of the algorithm under varied environment conditions, and results fit well with theoretical analysis. Comparisons with existing methods are also presented, together with examples to illustrate how the correction scheme can be applied in real systems.

A novel specific emitter identification method based on transient communication signal's time–frequency–energy distribution obtained by Hilbert–Huang transform (HHT) is proposed. The transient starting point is detected using the phase‐based method and the transient endpoint is detected using a self‐adaptive threshold based on the HHT‐based energy trajectory. Thirteen features that represent both overall and subtle transient characteristics are proposed to form a radio frequency (RF) fingerprint. The principal component analysis method is used to reduce the dimension of the feature vector and a support vector machine is used for classification. A signal acquisition system is designed to capture the signals from eight mobile phones to test the performance of the proposed method. Experimental results demonstrate that the method is effective and the proposed RF fingerprint can represent more subtle characteristics than the RF fingerprints based on instantaneous amplitude, phase, frequency and energy envelope. This method can be equally applicable for any wireless emitter to enhance the security of the wireless networks.