The rapid increase in wireless devices has raised significant security and privacy concerns, positioning Specific Emitter Identification (SEI) as a crucial physical-layer security technology. Although Deep Learning (DL) methods have been widely applied to SEI due to their powerful end-to-end nonlinear mapping capabilities, they generally require large amounts of high-quality signal examples, which are difficult to obtain in adversarial environments. Moreover, wireless channel perturbations induce a distribution shift between the training and testing signal examples from the same emitter. This shift prevents DL-enabled SEI models from learning emitter-specific, channel-agnostic features, leading to a severe degradation in identification performance. In this work, we propose a cross-channel SEI method based on Meta-Feature Augmentation-Enhanced Few-Shot Learning (MFA-FSL) to efficiently address the aforementioned challenges. To overcome the data scarcity, we use signal examples from base emitters with physical-layer characteristics similar to target emitters for pre-training. To overcome the channel perturbations, we employ meta-learning as the pre-training technique to learn a channel-agnostic feature embedding function. Considering that the function does not perform well in scenarios where signal examples of target emitters are extremely scarce, we approximate the target emitter’s feature distribution and sample augmented features from it. These augmented features, together with the raw features extracted from a few signal examples of target emitters, provide sufficient supervision to train a simple classifier. Experiments conducted on a Wi-Fi dataset comprising 16 emitter categories—10 as base emitters and 6 as target emitters—demonstrate that our proposed method achieves more than 85.63% identification accuracy with only 5 signal examples per target emitter, maintaining more than 83.35% accuracy even under varying wireless channel conditions. The code can be downloaded from https://github.com/lovelymimola/MFA-FSLIoTJ-Version.

With the increasing demand for secure communication in 5G and beyond, authentication of wireless devices has become a crucial task for communication security. Radio frequency fingerprint identification (RFFI) leverages the hardware-specific features in radio frequency (RF) signals, known as radio frequency fingerprints (RFF), to achieve highprecision device identification. However, the dependence of RFFI on the physical characteristics of devices makes it vulnerable to physical spoofing attacks. This paper proposes an innovative physical spoofing attack framework that combines spoofed transmitter and legitimate transmitter models. It performs RFF modeling, RFF concealment (RFFC), and RFF spoofing (RFFS) sequentially to achieve precise spoofing of the original baseband signal. We validate the effectiveness of the proposed physical spoofing mechanism through simulations of seven types of transmitters using MATLAB Simulink. The performance is further evaluated on an RFFI model based on complexvalued convolutional neural networks (CVCNN). Experimental results show that neural networks (NN) significantly outperform the generalized memory polynomial (GMP) model in nonlinear data fitting and temporal relationship modeling. Consequently, NN-based physical spoofing methods exhibit superior attack effectiveness. Specifically, under the signal-to-noise ratio (SNR) of 15 dB, the NN-based physical spoofing method achieves a target attack success rate (TSR) as high as 98%, which is superior to adversarial attack methods. NN-based methods also enhanced performance in terms of stealthiness metrics.

Specific emitter identification (SEI) is a crucial technique for recognizing individual emitters based on the unique characteristics of their radio frequency signals. Deep learning (DL)-based SEI has become the dominant method for identifying and authenticating wireless devices. However, in real-world applications, electromagnetic signals are subject to dynamic channel conditions and variations across different receivers, leading to significant performance degradation when models trained on specific datasets are applied to new, unseen environments. This variability challenges traditional DL methods, making domain generalization (DG) an essential approach to tackle this issue. In this paper, we propose a robust SEI method, latent distribution mixing (LDM), to enhance model generalization in receiver-agnostic scenarios. Our approach first applies the Fourier transform to convert time-domain signals into frequency-domain representations. Then, it mixes latent feature distributions across domains in the feature space to improve robustness, enabling the model to adapt to domain shifts effectively. We evaluated our method on a cross-receiver dataset, achieving a peak performance of 88.52%, surpassing other domain generalization methods. The experimental results demonstrate that the proposed LDM method offers a promising solution for SEI tasks in cross-receiver scenarios. Our code can be downloaded from https://github.com/frownean/LDM.

Radio frequency fingerprinting (RFF) presents a promising solution for advancing specific emitter identification (SEI) methods, which are crucial for securing the Internet of Things (IoT). While deep learning (DL)-based SEI approaches have demonstrated strong potential, they heavily depend on large, labeled datasets, which are often difficult to obtain in real-world scenarios. This reliance limits the robustness of existing SEI methods. To overcome this challenge, we propose a robust few-shot SEI (FS-SEI) method leveraging multi-view feature fusion with attention (MFFA). By integrating interpretable signal processing (SP) features with DL features and incorporating an attention mechanism for adaptive multi-view fusion, the proposed approach enhances both identification accuracy and robustness in few-shot scenarios. Experimental results validate the effectiveness of the method, showing consistent robustness under noisy conditions and significant gains in identification accuracy. These findings highlight its strong potential for practical applications in dynamic and challenging environments.

Radio Frequency Fingerprinting (RFF) is a critical technology for enhancing physical-layer security by leveraging the unique RF characteristics of hardware, enabling authentication and anti-counterfeiting for wireless communication devices. In recent years, Deep Learning (DL) has been extensively applied in$R$FF, significantly improving identification accuracy and efficiency. However, DL- based RFF methods still encounter challenges regarding robustness, particularly in cross-channel scenarios. To address these challenges, we propose a channel-robust RFF method based on a Multi-Scale Convolutional Attention Network (MSCAN) with Random Overlay Augmentation (ROA). Specifically, MSCAN extracts and fuses features at different scales, allowing it to capture more comprehensive signal characteristics. Additionally, ROA is a combinatorial data augmentation (DA) strategy designed to simulate diverse characteristics of wireless propagation environments, thereby enhancing the adaptability and robustness of RFF in complex channel conditions. Experiments conducted on the ORACLE dataset demonstrate that our proposed method achieves over 92 % accuracy in cross-channel scenarios, outperforming the previously proposed DA strategy. The codes will be published in GitHub11https://github.com/BeechburgPieStar/SDG-for-Robust-SEI

The rapid increase in wireless devices has raised significant security and privacy concerns, positioning Specific Emitter Identification (SEI) as a crucial physical-layer security technology. While Deep Learning (DL) has been widely applied to SEI, it often requires large amounts of high-quality signal examples, which are laborious and expensive to obtain. Moreover, the DL-enabled SEI models have difficulties in extracting features from the signal examples in the testing process that are consistent with those from the signal examples in the training phase due to the wireless channel variations, further resulting in a significant reduction in identification performance. To address these challenges, we propose a channel-robust Few-Shot SEI (FS-SEI) method based on Meta-Feature Augmentation (MFA). Our approach utilizes datasets from base emitters to construct a meta-feature embedding function that can extract generalizable features from a few signal examples of target emitters. We then calculate and calibrate the statistics of these extracted features to describe the feature distribution of target emitters. A Multi-Layer Perceptron (MLP) is subsequently trained on both original and augmented features derived from this distribution, achieving a robust FS-SEI model. Experiments conducted on a Wi-Fi dataset comprising 16 emitter categories - 10 as base emitters and 6 as target emitters - demonstrate that our method achieves 93.75% identification accuracy with only 5 examples per target emitter, maintaining 92.56% accuracy even under varying wireless channel conditions. Code is available at https://github.com/lovelymimola/MFA-based-FS-SEI.

Specific Emitter Identification (SEI) is pivotal for ensuring the security of the Internet of Things (IoT). Traditional deep learning-based SEI techniques often falter in real-world applications, particularly when distinguishing between legitimate and rogue devices amid noisy conditions and low Signal-to-Noise Ratios (SNR). To surmount these challenges, we propose a novel open-set SEI (OS-SEI) strategy that utilizes a Metric-enhanced Denoising Auto-encoder (MeDAE) architecture. This advanced framework incorporates a deep residual shrinkage network, significantly augmenting the denoising autoencoder’s capability, thereby bolstering its resilience against noisy environments. Further, the integration of discriminative metrics, such as center loss, markedly enhances feature discrimination, resulting in heightened accuracy of device identification. Our comprehensive experimental assessments, conducted on an Automatic Dependent Surveillance-Broadcast (ADS-B) dataset, underscore the superiority of our proposed OS-SEI method over existing models. The findings confirm our approach’s enhanced robustness to noise and its superior accuracy in device identification within open-set scenarios.

Unmanned Aerial Vehicle (UAV) recognition using Deep Learning (DL) is critical for ensuring the safety of low-altitude airspace. However, the limited availability of labeled UAV signal data poses significant challenges to achieving high recognition accuracy and robustness. To address this, we propose a novel method, Self-Supervised learning with Self-Adaptive Pseudo-Labeling (SS-SAPL), designed to enhance UAV recognition performance. The method operates in two stages: a self-supervised pre-training stage and a semi-supervised fine-tuning stage. In the pre-training stage, contrastive learning with weak and strong data augmentations is employed to extract generic feature representations from all UAV signal samples. In the fine-tuning stage, Pseudo-Labeling (PL) is combined with a Self-Adaptive Threshold (SAT) and Self-Adaptive Fairness (SAF) mechanism to improve the accuracy of PSeudo-Labels (PSLs) and leverage both labeled and unlabeled data for refining feature representations. Simulation results demonstrate the effectiveness of our method. For UAV signals at 2.4 GHz with only 30 labeled samples, our approach achieves a recognition accuracy of 82.38%, outperforming state-of-the-art methods by at least 6.63%. In mixed-frequency scenarios (2.4 GHz and 5.8 GHz) with only 10 labeled samples, our method exceeds 92.13% accuracy, surpassing competitors by at least 4.63%. These results highlight the robustness and practical value of the proposed method in challenging environments.

Automatic modulation classification (AMC) is a fundamental technology for identifying modulation types in non-cooperative communication systems. It plays a crucial role in various applications, including spectrum monitoring, cognitive radio, and signal intelligence. Recently, deep learning (DL) based AMC methods have achieved remarkable classification accuracy. However, their practical deployment in resource-constrained edge devices remains challenging due to their high computational complexity and excessive model size. To address this limitation, we propose an ultra-lightweight AMC method based on filter pruning via geometric median (FPGM). The key idea is to leverage the geometric median as a robustness-driven filter selection criterion, effectively eliminating redundant convolutional kernels while preserving essential model representations. Specifically, we first determine the geometric median of the filters in each layer, which effectively represents the distribution of filters within that layer. Then, filters near the geometric median are identified and filtered out through the characteristics of the geometric median. Finally, the performance degradation of the model caused by the removal of filters can be restored through fine-tuning. Experimental results demonstrate that the proposed AMC method achieves a 99% reduction in model size while limiting the classification accuracy drop to merely 1.61%, significantly outperforming other lightweight AMC techniques. These results highlight the feasibility of deploying the proposed AMC model on edge Internet of Things (IoT) devices, enabling efficient real-time modulation classification with minimal computational overhead.

Accurate streamflow (SF) forecasting is crucial for effective water-resource management amid global climate change. Traditional ensemble SF-forecasting methods, relying on historical data and watershed characteristics, often produce uncertainties in their input, structure, and parameters, reducing their forecasting accuracy. This study introduces a variance-combined bias-correction (VCB) method, integrated with Internet of Things (IoT) technology to improve ensemble SF forecasts’ accuracy and responsiveness. The VCB method significantly improves the SF-forecasting performance by incorporating variance information from ensemble forecasts along with the ensemble mean. We apply the method to the Shiquan Reservoir in China’s Han River basin, and the results show that the VCB method outperforms the Bayesian joint probability (BJP) method, achieving an increases of 8.8% in the Nash-Sutcliffe efficiency (NSE), 0.7% in the Pearson correlation coefficient (PCC), 2.1% in the qualified rate (QR), and a 7.2% reduction in the mean absolute percentage error (MAPE). Furthermore, IoT technology integration improves method inputs’ accuracy and timeliness, showing the strongest performance during extreme weather events. Thus, by improving uncertainty management and forecasting accuracy, the IoT-integrated VCB method provides more effective support for water-resource management. Future research should apply this approach to diverse hydrological contexts and explore deeper integration with machine-learning techniques.

Specific emitter identification (SEI) is a technique that identifies individual emitters based on the inherent characteristics reflected in the radio frequency signals due to the individual differences of the emitters. Deep learning (DL) has become the primary research method for identifying and authenticating wireless devices in SEI. However, in the real world, electromagnetic signals continuously change with the channel environment and time, causing models trained on datasets collected from known specific domains to exhibit significant performance degradation when applied to unknown channel environments. This limitation makes general DL methods unsuitable, and domain generalization (DG) becomes a key method to address this issue. To overcome the limitations of SEI identification performance across different scenarios, we propose a robust SEI method by mixing augmentation, named SigMix. Specifically, we innovatively introduce the Mixup method into the SEI task, mixing data from different source domains and then performing pairwise linear interpolation before using it for training the neural network. The SigMix method helps the model learn more comprehensive features by generating new samples in the training data, thereby improving the model’s generalization ability. To validate the effectiveness of the SigMix method, while also considering the impact of different receivers on identification performance, we evaluate a dataset spanning both time and receivers. The experimental results indicate that the average identification accuracy of the proposed SigMix method in unknown domains reaches 84.40%, significantly outperforming existing DG methods, demonstrating the robustness and generalization of our proposed SigMix method in SEI tasks. Our code is available for download at://github.com/frownean/SigMix.

Network intrusion detection (NID) is a critical area of research in network security. While deep learning based NID methods have recently achieved advanced detection performance, they often struggle with limited labeled traffic and the resource constraints of edge Internet of Things (IoT) devices. To address these challenges, we propose an advanced few-shot NID method using lightweight transfer learning (LTL), termed NID-LTL. Our approach begins by pre-training a detection model on the large-scale auxiliary dataset to learn universal representations of network traffic characteristics. Then, an automatic pruning strategy is crafted to prune the pre-trained model, which uses a kernel based nonlinear traffic feature selection algorithm to filter out the key information most relevant to the original traffic. Finally, the layer-wise knowledge distillation method is combined to transfer the useful knowledge learned by the pre-trained model to a lightweight student model. This method can not only quickly adapt to novel few-shot NID tasks, but also further compress the model size, reduce computational and storage overhead. Experimental results demonstrate that the proposed NID-LTL method has excellent classification performance with small model sizes, low parameter counts, and low floating point operations (FLOPs). Especially, in the 1-shot scenario, the NID-LTL method achieves 89.38% classification accuracy with only 1.41% of the parameters in the original model.

In millimeter-wave and terahertz band communication systems, precise beam prediction is crucial for optimizing network performance and enhancing signal transmission efficiency. Traditional beam prediction methods have primarily relied on single-modal data, which often fails to capture the comprehensive environmental information necessary for optimal accuracy. In contrast, multi-modal data-based approaches offer a more promising solution by leveraging the strengths of diverse data sources. However, many existing fusion methods are static, inadequately accounting for variations in information content across different modalities, which can hinder the full utilization of each modality’s advantages. To address these limitations, this paper proposes an advanced multi-modal beam prediction method that integrates multipath-like data augmentation (MLDA), cross-modal feature enhancement (CMFE), and an uncertainty-aware dynamic fusion mechanism. Our approach combines image and radar data to predict beam indices, dynamically adjusting the weights of different modalities to accommodate varying information densities. The proposed method employs ResNet34 for feature extraction from the multi-modal data, followed by a cross-modal feature enhancement module that aggregates complementary information from the image and radar data. Finally, the dynamic fusion mechanism integrates the predictions from the single-modal data. Experimental results demonstrate that our method significantly improves the accuracy and robustness of beam prediction, achieving an overall accuracy of 89.72%. The performance of the proposed method is further validated through comparisons with various existing methods and comprehensive ablation studies, highlighting its superiority in multi-modal assisted beam prediction scenarios.

Accurate road network extraction from remote sensing images (RSIs) is essential for applications such as urban planning, map updates, and autonomous navigation. However, challenges such as complex backgrounds, varying spatial resolutions, and occlusions hinder traditional single-modality approaches, which often fail to capture comprehensive contextual information. To address these limitations, we propose DEFNet, a novel Dual-Layer Evidential Fusion Network for robust multimodal road extraction. DEFNet features two key modules: Cross-Attention Feature Interaction (CAFI) and Dual-Layer Evidential Fusion (DEF). The CAFI module facilitates adaptive multimodal interaction at both pixel and superpixel levels, enhancing feature fusion while mitigating noise. The DEF module, leveraging the Dirichlet framework and Dempster-Shafer Theory, performs uncertainty-aware fusion, improving prediction reliability and robustness. Extensive experiments on multiple benchmark datasets demonstrate that DEFNet consistently outperforms state-of-the-art methods in both accuracy and robustness, making it highly effective for multimodal road extraction in remote sensing applications. The codes can be downloaded from GitHub1.

Radio frequency fingerprint identification (RFFI) is regarded as one of the most promising techniques for managing and regulating Internet of Things (IoT) devices. This technology analyzes the unique electromagnetic signals emitted by wireless devices to enable precise identification and authentication. Most existing RFFI methods focus on RF signals collected in specific scenarios. However, in real-world applications, signals are often collected at different times or from varying deployment locations, leading to differences between the training and testing distributions. The study of RFFI methods under these conditions remains underexplored. To address this gap, this paper introduces a cross-domain RFFI framework centered on adaptive semantic augmentation (ASA). The framework integrates a computationally efficient multi-resolution spectrogram decomposition strategy with a feature-sensitive multi-scale network. The ASA method enhances RFFI accuracy in cross-domain settings by linearly interpolating between two distinct semantic features to create new semantics for further identification. The proposed approach leverages two-dimensional discrete wavelet transform (2D-DWT) to decompose the raw spectrogram into four sub-bands, followed by a multi-scale network to extract critical semantic features for the ASA method. Simulation results show that the proposed ASA method significantly improves Unmanned Aerial Vehicle (UAV) identification performance, achieving accuracies of 93.05% and 98.90% on two different cross-domain datasets, respectively, outperforming existing data augmentation (DA) methods. Furthermore, generalizability validation demonstrates that the proposed method performs outstandingly across other Internet of Things (IoT) applications.

Wireless Technology Recognition (WTR) distinguishes different wireless technologies by analyzing characteristic features extracted from radio signals. While deep learning (DL)-based methods are extensively used in WTR due to their ability to extract hidden data features and make accurate classification decisions, their application is often limited by excessive power consumption. In this paper, we propose a novel WTR method that addresses this challenge using a time-frequency feature fusion spiking neural networks (TFSNN) framework. Our approach combines information from both the time and frequency domains to enhance feature extraction. Experimental results demonstrate that our model performs exceptionally well at high signal-to-noise ratios on open-source datasets. Specifically, at a sampling rate of 15 Msps, our method achieves a recognition accuracy of 99.85%. Even when the sampling rate is reduced to 10 Msps, the average accuracy remains 1.61% higher than the best existing method. Additionally, our method reduces energy consumption by about half compared to most current methods. These results emphasize the effectiveness and necessity of time-frequency domain feature fusion (TFSF) in WTR.

Radio Frequency Fingerprinting (RFF) has emerged as a vital technique for enhancing Physical Layer Authentication (PLA) in New Radio (NR) networks. Unlike cryptographic methods, RFF leverages device-specific signal impairments to uniquely identify transmitters. Deep Learning (DL) advances have improved PLA, though challenges persist due to communication channel dynamics and device state changes. In this study, we propose a novel framework that integrates 5G NR protocol-specific structures and channel knowledge via SRS-based CSI to generate relative RFF features. Through a tailored frame design and carefully engineered processing pipeline, we achieve cross-domain stability and improved robustness against time-varying conditions. By applying regularization techniques (e.g., mixup) during training, our method further mitigates model overfitting and domain bias. Simulation and real-world SDR experiments, using data from 9 ADALM-PLUTO devices, validate the approach’s effectiveness. The proposed system attains recognition accuracies of 99.878%, 93.376%, 86.325%, and 66.558% in intra-domain, cross-channel, cross-time, and cross-scenario tests, respectively, highlighting its potential to substantially enhance physical layer security in NR-based networks.

Malware traffic classification (MTC) is a crucial step in network intrusion detection, which is significant for network security and management. With the continuous evolution of malware traffic, traditional MTC methods are difficult to adapt efficiently to new traffic categories, and manually designed neural network structures suffer from performance bottlenecks and low design efficiency. Hence, we propose an enhanced MTC method based on expandable class incremental learning (CIL) with architecture search. The architecture search can automatically design the optimal neural network structure tailored to different network traffic characteristics, avoiding the limitations of manually designing network structures and improving classification performance. Meanwhile, expandable CIL allows the MTC model to gradually learn new traffic categories without forgetting previous knowledge, avoiding the computational overhead and efficiency loss caused by frequent retraining of the model. The experimental results demonstrate that the proposed CIL-MTC approach surpasses advanced incremental learning methods on both the Edge-IIoTset and ISCX VPN-nonVPN datasets, achieving superior classification performance while maintaining lower average trainable parameters and training costs. Especially, it achieves an average incremental accuracy of 98.55% and 99.09% on the Edge-IIoTset dataset with incremental tasks of 5 and 2, respectively.

By extracting radio frequency (RF) fingerprints from received signals, specific emitter identification (SEI) becomes a promising technique for physical layer identification of wireless devices. Recently, channel-robust SEI has attracted increasing attention due to the weak robustness exhibited by deep learning (DL)-based SEI methods in cross-channel conditions. To address these limitations, we propose a novel channel-robust SEI framework based on single-source domain generalization (SDG). Initially, we analyze the weak robustness of existing SEI methods from the perspective of the “shortcut learning” phenomenon in DL. Shortcut learning may lead traditional SEI methods to prioritize easily-mined, yet transient, channel characteristics in signal samples, rather than focusing on the more stable RF fingerprints derived from hardware differences. Next, from the perspective of SDG, we outline the optimization goal to rectify the shortcut learning in SEI. Inspired by this optimization goal, we then propose a channel-robust SEI method. This method consists of feature embedding through a multi-scale convolutional attention network (MSCAN), domain expansion using random overlay augmentation (ROA) to generate multiple virtual domains, and dual alignment strategy based on contrastive learning. Specifically, supervised contrastive learning is implemented for category-wise alignment, while supervised contrastive adversarial learning is utilized for domain-wise alignment. This dual alignment strategy can optimize the MSCAN to learn discriminative and domain-invariant feature representations, thereby enhancing the robustness of SEI. Simulation experiments on the ORACLE dataset and the WiSig dataset have demonstrated the superiority of our method compared to state-of-the-art techniques. The codes can be downloaded from GitHub (https://github.com/BeechburgPieStar/SDG-for-Channel-Robust-SEI).

In this paper, we investigate the effects of severe channel fading in the scenario of distributed deep learning-based joint source-channel coding (DJSCC) for image transmission without perfect channel state information (CSI). To tackle the challenges posed by imperfect CSI, we propose a robust DJSCC (RDJSCC) scheme that operates at three levels: modulation, encoding, and decoding, respectively. Firstly, at the modulation level, we adopt orthogonal frequency division multiplexing (OFDM) modulation for exploring the tradeoff between reconstruction performance and peak-to-average power ratio (PAPR). Secondly, at the encoding level, two parameter-efficient operators are introduced to combat channel fading with low encoding complexity. Finally, at the decoding level, we divide the decoding process into two stages, i.e., denoising and recovery, aiming to maximize the correlation between the encoded representations. Theoretic analysis and simulation results show that our proposed RDJSCC can effectively alleviate the effects of severe fading with imperfect CSI, leading to an improved reconstruction performance while maintaining low PAPR and encoding complexity.

As unmanned aerial vehicles (UAVs) become increasingly prevalent in various environments, their detection is essential for ensuring safety and effective management. For non-standard transmitter waveforms, identifying a specific model poses challenges, and most researchers focus on developing deeper and broader architectures to improve performance at the expense of computational burden and speed, which is not suitable for the deployment of UAV radio frequency (RF) fingerprinting identification. In this paper, we propose a novel multi-resolution analysis-based method for UAV identification. We develop a lightweight, multi-scale convolutional network that utilizes various receptive fields to extract unique hardware intrinsic features. We employ the two-dimensional Discrete Wavelet Transform (2D-DWT) to innovatively decompose the low-frequency and the high-frequency components of the RF signal spectrogram into four distinct sub-bands. Simulation results indicate that our proposed methods outperforms other state-of-the-art UAV identification methods in terms of both performance and computational complexity.

Specific emitter identification(SEI) plays an integral role in network security. In recent years, deep neural networks (DNNs) have demonstrated significant success in various application scenarios. The robust feature extraction capabilities of DNNs have led to advancements in SEI. However, it has been shown that DNNs are susceptible to adversarial attacks. The proposal of well-performing adversarial attacks is conducive to improving the security of SEI with DNN-based models. This paper introduces an universal black-box adversarial attack algorithm, named UBBA, for SEI with DNN-based models. The experimental findings indicate that this universal black-box adversarial attack algorithm substantially reduces the identification accuracy of SEI models. Given a sufficient number of queries, the proposed algorithm achieves an attack effect similar to that of the universal adversarial perturbations (UAP), a universal white-box attack algorithm. Additionally, the results demonstrate that when the perturbation signal is not synchronized with the signal under attack, the proposed algorithm outperforms the fast gradient sign method (FGSM).

In this paper, we introduce an innovative Radio Frequency Fingerprinting (RFF)-based device authentication scheme for the Internet of Things (IoT), a network marked by extensive interconnections and interactions among various entities. Our approach, designed for an open and dynamic communication environment, not only identifies devices encountered during training but also effectively rejects those not previously seen. The scheme employs a hypersphere projection for feature embedding, strategically avoiding the need to optimize intra-device variations in the radial direction. It uses a K-Means-based binary classifier for initial device assessment based on cosine similarity scores, followed by a SoftMax classifier for precise identification of known devices. Our extensive numerical analysis confirms that this method delivers superior performance, setting a new benchmark in RFF authentication for IoT security.

In the context of emerging 6G technology challenges, this paper introduces the LSMFF-AMC approach, leveraging multimodal feature fusion (MFF) with Long-Short range attention (LSRA) to enhance automatic modulation classification(AMC). The method significantly boosts classification accuracy by employing convolutional neural networks (CNN) for diverse modal feature extraction and integrating LSRA for comprehensive feature combination. Our experiments demonstrate an increase in accuracy from 88% to nearly 97%, outperforming traditional single-modal approaches. Additionally, a convergence analysis of the training loss function reveals LSMFF-AMC's superior and faster convergence compared to standard AMC methods.

Deep learning-based recognition of Unmanned Aerial Vehicles (UAVs) has become a critical tool for enhancing UAV control through improved accuracy and efficiency. However, the practical deployment of these systems is often hampered by the costly acquisition and scarcity of annotated data, which challenges the generalizability of the models. To address this bottleneck, our study employs semi-supervised (SS) learning strategies to exploit the untapped potential of unlabeled data effectively. We introduce a novel semi-supervised approach for UAV recognition that utilizes a self-adaptive threshold mechanism. This technique features Self-adaptive Threshold (SAT) and Self-adaptive Fairness (SAF) mechanisms, designed to dynamically optimize threshold values and guarantee a balanced distribution of labels among various classes. Our method is rigorously evaluated against a comprehensive, open-source UAV dataset. The findings indicate that our semi-supervised model significantly outperforms existing supervised learning models, static threshold SS approaches, and generative models, especially in scenarios with a limited amount of labeled data. These results underscore the effectiveness of our approach in enhancing the practicality and applicability of UAV recognition systems.

Specific emitter identification (SEI) has emerged as a notable device authentication technology, distinguishing various emitters through the unique radio frequency fingerprint (RFF) inherent in wireless devices. Traditional SEI methods, often hindered by time-consuming manual feature extraction, struggle with complex encrypted signals. The advent of deep learning, with its robust feature extraction capabilities, has significantly advanced SEI, yet it typically demands extensive radio frequency signal samples and falters with limited (i.e., few-shot) samples. Our proposed few-shot SEI (FS-SEI) approach, integrating neural architecture search (NAS) and deep transfer learning (DTL), adeptly identifies few-shot long range (LoRa) devices. This method begins with NAS to autonomously tailor optimal network architectures for SEI tasks, followed by pre-training on extensive auxiliary datasets to extract general RFF features of LoRa devices. Transfer learning then fine-tunes these features for distinctiveness with compact intra-class distances. By only utilizing few-shot LoRa data for final parameter adjustments, the classifier rapidly assimilates new categories. Simulations confirm our FS-SEI method's superior accuracy over classical approaches, with visualized feature analysis underscoring its distinguishing and generalizing prowess.

Radar emitter identification (REI) is a crucial function of electronic radar warfare support systems. The challenge emphasizes identifying and locating unique transmitters, avoiding potential threats, and preparing countermeasures. Due to the remarkable effectiveness of deep learning (DL) in uncovering latent features within data and performing classifications, deep neural networks (DNNs) have seen widespread application in REI. In many real-world scenarios, obtaining a large number of annotated radar transmitter samples for training identification models is essential yet challenging. Given the issues of insufficient labeled datasets and abundant unlabeled training datasets, we propose a novel REI method based on a semi-supervised learning (SSL) framework with virtual adversarial training (VAT). Specifically, two objective functions are designed to extract the semantic features of radar signals: computing cross-entropy loss for labeled samples and virtual adversarial training loss for all samples. Additionally, a pseudo-labeling approach is employed for unlabeled samples. The proposed VAT-based SS-REI (SS-VAT) method is evaluated on a radar dataset. Simulation results indicate that the proposed SS-VAT method outperforms the latest SS-REI method in recognition performance.

In the field of Industrial Internet of Things (IIoT) security, Specific Emitter Identification (SEI) plays a crucial role. Recent advancements have seen a rise in the adoption of machine learning (ML) and deep learning (DL) techniques in SEI methodologies, recognized for their impressive effectiveness. However, DL-based SEI methods often incur significant computational costs, making them less suitable for IIoT environments. Similarly, conventional ML-driven SEI approaches depend heavily on feature extraction and employ complex, often redundant classifiers. These methods typically lack in optimizing feature integration and computational efficiency. To overcome these limitations, we introduce an advanced DL-based SEI methodology that focuses on harnessing signal features more effectively. Our method centers around an Adaptive Feature Combination (AFC) strategy, enhanced by an attention mechanism, to develop a more efficient SEI classifier. The essence of our approach is the strategic exploration of adaptive feature combinations, aiming to fine-tune the SEI classifier for peak performance. Simulation results demonstrate that our AFC algorithm outperforms existing SEI methods in both identification accuracy and computational efficiency. This breakthrough offers a viable and promising solution for implementing SEI in IIoT scenarios, achieving heightened effectiveness without sacrificing computational resources.

Automatic modulation classification (AMC) plays an important role in the field of physical layer security, providing a new way to enhance the security of data transmission and anti-interference ability. Recently, deep learning (DL) has been widely applied in radar and communication signal classification, which requires sufficient labeled training samples to achieve high classification accuracy. However, in non-cooperative situations, it is difficult to obtain a large number of labeled signal samples. Therefore, we propose a novel few-shot AMC method using architecture search and knowledge transfer. This method first utilizes the state-of-the-art neural architecture search algorithm, A-DARTS, to automatically search for the optimal network structure (i.e., Auto-MCNet) based on the auxiliary sample set. Then, the Auto-MCNet model is pre-trained on the auxiliary dataset to explore prior knowledge about signal classification. Finally, we transfer this knowledge to a few-shot training dataset and fine-tune the Auto-MCNet model to enhance its generalization ability. The simulation results show that compared to advanced competitors, Auto-MCNet achieves higher classification accuracy with lower model complexity.

The rapid changes in high-mobility vehicle environments make it challenging for base stations (BS) to obtain comprehensive channel state information. Furthermore, road and traffic safety require communication with low latency and high reliability, posing significant challenges to spectrum resource allocation in vehicular networks. To address these challenges, this paper proposes a method combining dueling double deep-Q network (D3QN) reinforcement learning (RL) with long short term memory (LSTM) network. By using a Manhattan Grid Layout City Model as the foundational environment, a multi-agent model is constructed, with each vehicle-to-vehicle (V2V) link acting as an individual agent. These agents collaborate and interact with the environment, receiving feedback, and then determining the optimal resource allocation to ensure both superior mobile service and a safe driving environment. The experimental results indicate that our proposed method outperforms the conventional D3QN network in both the vehicle-to-infrastructure (V2I) links and the V2V links.

Malware traffic classification (MTC) plays an important role in cyber security and network resource management for the secure Internet of Things (IoT). Many deep learning (DL)-based MTC methods have been proposed due to their robustness and effectiveness with self-designed model architecture. However, to completely adjust complex parameters in the DL model, the architecture design of the DL model requires substantial professional knowledge and effort from human experts. To solve these problems, we propose an automatic and efficient MTC method using neural architecture search via proximal iterations (NASP), which can automatically and efficiently search the optimal model architecture according to the network traffic in the realistic environment. Specifically, we first describe NAS as a constrained optimization problem by keeping the search space differentiable and forcing the architecture to be discrete in the search process. Second, a suitable regularizer is introduced to balance the complexity and performance of the model architecture. Finally, the simulation results show that the proposed NASP-aided MTC method not only can efficiently and accurately search the optimal classification model architecture on the USTC-TFC2016 data set and the Egde-IIoTset data set but also compared with the typical MTC methods it can achieve the optimal classification performance with the fewer parameters as well as the floating-point operations (FLOPs).

The inherent flexibility of small unmanned aerial vehicles (UAVs) enables their deployment across various emerging markets. Unauthenticated UAVs pose a significant threat if they intrude into aviation-sensitive areas. To address this issue, deep learning (DL)-based radio frequency fingerprint identification (RFFI) has been developed as a promising approach for identifying illegal UAVs. However, these commonly used DL-based methods demand high computation and storage requirements, which are not suitable for the deployment of RFFI. In this paper, we propose an efficient and low-complexity RFFI method for UAV identification. Specifically, we design a lightweight backbone network consisting of lightweight multi-scale convolution (LMSC) blocks that can significantly reduce the model size and enhance the feature extraction ability. The simulation results indicate that our proposed UAV RFFI method outperforms other state-of-the-art and popular DL-based RFFI methods in terms of both identification performance and complexity. The identification accuracy surpasses that of all other methods at low signal-to-noise ratios (SNRs) and achieves nearly 100% accuracy at high SNRs. To further enhance model efficiency, we employ data truncation in our experimental simulations, demonstrating that a sample length of 2000 is sufficient to retain high identification performance. Additionally, we incorporate the Mixup regularization strategy, which improves accuracy without increasing the complexity, especially as sample length decreases.

Radiofrequency signal identification (RSI) provides a critical security solution for device authentication in the Internet of Things (IoT), characterized by extensive interconnections and interactions among numerous entities. By analyzing received radiofrequency signals, device-specific features are extracted at the receiver and used for identification. In a dynamic and ever-changing communication environment, where some devices not visible during the training process may appear during testing, a robust RSI method must not only identify devices encountered during training but also reject those that were not. In this article, we propose an open-set RSI method based on hypersphere manifold embedding. This approach leverages hypersphere projection for radiofrequency signal feature extraction on a hypersphere manifold, thereby avoiding the need to optimize intradevice variation in the radial direction. Additionally, we introduce an open-set identification approach based on generalized Pareto distribution, which does not rely on any radiofrequency signals from unknown devices. Extensive experimental results demonstrate that the proposed method achieves state-of-the-art identification performance.

Deep learning (DL)-based automatic modulation classification (AMC) has made breakthroughs and is generally used for signal detection and recognition in wireless communication systems, unmanned aircraft vehicle (UAV) systems, and other fields. However, high storage and computational demands limit its use in resource-constrained UAV systems. This paper presents an AMC method featuring a streamlined design with lower computational needs, using the ultra-lite convolutional neural network (ULCNN). This innovative model combines data augmentation, complex-valued convolution, separable convolution, channel attention, and shuffling techniques for enhanced performance. The proposed ULCNN model balances efficiency and accuracy, with simulations showing it achieves 62.47% accuracy on the RML2016.10a dataset using only 9,751 parameters. Furthermore, we evaluated the actual speed of ULCNN on a Raspberry Pi, an edge platform with roughly equivalent computing power to a conventional UAV, achieving an inference speed of only 0.775 ms per sample. This high performance, coupled with a significantly smaller model size, underscores the potential of ULCNN for integration into resource-constrained UAV systems, thereby enabling rapid and efficient data processing.

Specific Emitter Identification (SEI) is a critical technology for physical layer authentication in wireless communications and the Internet of Things. Leveraging the inherent and hard-to-forge characteristics of Radio Frequency Fingerprinting (RFF), SEI has gained significant attention. Recent advancements in deep learning have propelled SEI methods to new heights of identification performance. However, these methods are often constrained by their reliance on large datasets, posing challenges in real-world scenarios with limited samples. Addressing this issue, this paper proposes an enhanced SEI approach tailored for limited sample environments, employing Double Implicit Regularization (DIR). Our proposed method, DIR-MRAN, utilizes a Multi-Scale Residual Attention Network (MRAN) to extract features effectively from limited samples. The DIR strategy enhances model generalizability by incorporating Sample-wise Implicit Regularization (SIR) and Label-wise Implicit Regularization (LIR), which respectively facilitate sample expansion and label smoothing. We evaluated DIR-MRAN on two real-world datasets, achieving an impressive 95.34% accuracy on the PA dataset and outperforming comparative methods by 26.4% on the ADS-B dataset.

With the development of wireless technology and the Internet of Things (IoT), managing limited spectrum resources has become crucial. As the IoT landscape grows, more effective wireless technique classification (WTC) is imperative. Traditional deep learning (DL) methods for WTC, while robust, suffer from high computational complexity, making them less practical for real-time applications. Addressing this, our article introduces a novel solution, the multifeature fusion broad learning network (MFBLN) for WTC, which employs broad learning (BL). Here, several features of the wireless technique are inputted into a multibranch module to obtain classification information from different perspectives. Then, those features are integrated, which performs better than the typical BL structure. Our simulation results show that our proposed MFBLN method performs well on the classic WTC data sets in the intelligent transportation system (ITS) band. The performance of MFBLN at a 25 Msps sampling rate shows an improvement of approximately 0.67%, coupled with a significant reduction in floating-point operations by 81.93%, and 72.53% decrease in training time. Additionally, the ablation studies further affirm the necessity of each module within the MFBLN framework, underscoring their collective contribution to its enhanced efficiency and effectiveness.

Deep learning (DL), renowned for its superior feature extraction capabilities, has remarkably succeeded in specific emitter identification (SEI), especially when supported by high-quality labeled data. However, obtaining accurate signal labels in complex electromagnetic environments is challenging, and manual labeling is prone to errors, underscoring the need for robust DL-based SEI methods that can handle label noise. These methods prevent neural networks from overfitting noisy labels, thereby boosting identification performance. Yet, research in this area is still limited. Our study introduces a robust label-noise SEI approach and the sample selection and regularization (SSR) method. This involves a two-stage adaptive sample selection (ASS) driven by confidence learning. The first stage entails coarse-grained separation of true and false labels through direct deep neural network (DNN) training. In the second stage, semi-supervised learning (SSL) utilizes a regularization-inspired loss, incorporating label smoothing regularization (LSR) and entropy minimization (EM), for fine-grained sample selection. The DNN is ultimately trained on precisely selected true-labeled samples. Comparative experiments on the automatic dependent surveillance-broadcast (ADS-B) and Wi-Fi data sets demonstrate that our SSR method outperforms the existing methods in identification accuracy, particularly at a 20% label-noise ratio, achieving 86.00% accuracy with the ADS-B data set, and 99.38% with the Wi-Fi data set. The code is available at:https://github.com/sleepeach/SSR-SEI.

Malware traffic classification (MTC) is one of the important techniques to ensure the security of cyberspace, which aims to detect anomalies and classify different types of network traffic. Recently, MTC methods based on deep learning (DL) have shown their excellent performance. However, these DL-based methods rely on datasets with manually labeled samples for training, which are costly and hard to obtain. To address this problem, this paper proposes a novel self-supervised MTC method based on the framework of masked auto-encoder (MAE). Specifically, MAE first constructs a reasonable unsupervised pretext task with a random masking strategy, which reduces the redundant information in samples and speeds up the pre-training process. The transformer-based backbone network then efficiently extracts features from the non-redundant traffic data efficiently. The proposed MTC-MAE method employs self-supervised learning on a large-scale unlabeled dataset to acquire unbiased features, and fine-tunes on specific datasets to adapt to diverse traffic classification scenarios. Simulation experiments show that our proposed MTC-MAE method is able to learn universal features with high quality and has excellent classification performance on various downstream datasets. The datasets we used, code implementation, and pre-trained models are available on GitHub.

Ensuring network security, effective malware detection is of paramount importance. Traditional methods often struggle to accurately learn and process the characteristics of network traffic data, and must balance rapid processing with retaining memory for previously encountered malware categories as new ones emerge. To tackle these challenges, we propose a cutting-edge approach using self-paced class incremental learning (SPCIL). This method harnesses network traffic data for enhanced class incremental learning (CIL). A pivotal technique in deep learning, CIL facilitates the integration of new malware classes while preserving recognition of prior categories. The unique loss function in our SPCIL-driven malware detection combines sparse pairwise loss with sparse loss, striking an optimal balance between model simplicity and accuracy. Experimental results reveal that SPCIL proficiently identifies both existing and emerging malware classes, adeptly addressing catastrophic forgetting. In comparison to other incremental learning approaches, SPCIL stands out in performance and efficiency. It operates with a minimal model parameter count (8.35 million) and in increments of 2, 4, and 5, achieves impressive accuracy rates of 89.61%, 94.74%, and 97.21% respectively, underscoring its effectiveness and operational efficiency.

Deep learning (DL)-based automatic modulation classification (AMC) is increasingly utilized in wireless applications, particularly within the Internet of Things (IoT) ecosystem. However, the open data collection for these systems can lead to vulnerabilities, as the data sets are susceptible to malicious manipulations, potentially resulting in backdoor attacks. In this article, we propose a novel wavelet domain frequency steganography (WDFS) backdoor attack method to demonstrate this security flaw, designed explicitly for misleading AMC. This method employs discrete wavelet transform and singular value decomposition to segment signals into distinct wavelet domain frequency components. We embed the backdoor trigger directly into these components, ensuring it is sample-specific and undetectable. Extensive testing shows that our WDFS method outperforms existing methods in terms of attack efficiency and stealth and successfully evades several advanced backdoor defense mechanisms, demonstrating its robustness. These findings highlight the urgent need for enhanced security measures in AMC systems within the artificial intelligence domain.

Specific emitter identification (SEI) has emerged as a notable device authentication technology, distinguishing various emitters through the unique radio frequency fingerprint (RFF) inherent in wireless devices. Traditional SEI methods, often hindered by time-consuming manual feature extraction, struggle with complex encrypted signals. The advent of deep learning, with its robust feature extraction capabilities, has significantly advanced SEI, yet it typically demands extensive radio frequency signal samples and falters with limited (i.e., few-shot) samples. Our proposed few-shot SEI (FS-SEI) approach, integrating neural architecture search (NAS) and advanced deep transfer learning (DTL), adeptly identifies few-shot long-range (LoRa) devices. This method begins with NAS to autonomously tailor optimal network architectures for SEI tasks, followed by pre-training on extensive auxiliary datasets to extract general RFF features of LoRa devices. Transfer learning then fine-tunes these features for distinctiveness with compact intra-class distances. By only utilizing few-shot LoRa data for final parameter adjustments, the classifier rapidly assimilates new categories. Simulations confirm our FS-SEI method’s superior accuracy over classical approaches, with visualized feature analysis underscoring its distinguishing and generalizing prowess.

Automatic modulation classification (AMC) holds a significant position in physical-layer security, offering an innovative method to enhance the security of data transmission and anti-interference ability. Recently, deep learning (DL) has seen extensive application in radar and communication signal classification, which requires sufficient labeled training data to ensure great classification performance. However, obtaining a significant amount of labeled samples is extremely challenging in complex and ever-changing electromagnetic environments. Therefore, we propose a novel few-shot AMC method using architecture search and knowledge transfer. This method first utilizes an advanced neural architecture search algorithm,$\Lambda $-DARTS, to automatically search for the optimal network structure (i.e., Auto-MCNet) based on the auxiliary sample set. Then, the Auto-MCNet model is pretrained on the auxiliary data set to explore prior knowledge about signal classification. Finally, we transfer the knowledge to a few-shot training data set and fine-tune the Auto-MCNet model to enhance its generalization ability. The simulation results indicate that when the signal-to-noise ratio (SNR) is greater than 0 dB and the shot of each class is 3 and 10, the average accuracy of the proposed Auto-MCNet is higher than 81% and 90%, respectively. Moreover, compared to advanced competitors, Auto-MCNet achieves higher classification performance with lower model complexity.

Specific emitter identification (SEI) based on radio frequency fingerprinting (RFF) is a physical layer authentication method in the field of wireless network security. RFFs are unique features embedded in the electromagnetic waves, which come from the hard imperfections in the wireless devices. Deep learning has been applied to many SEI tasks due to its powerful feature extraction capabilities. However, the success of most methods hinges on massive and labeled samples, and few methods focus on a realistic scenario, where few samples are available and labeled. In this paper, to overcome data limitations, we propose a few-shot SEI (FS-SEI) method based on self-supervised learning and adversarial augmentation (SA2SEI). Specifically, to overcome the limitation of label dependence for auxiliary dataset, a novelty adversarial augmentation (Adv-Aug)-powered self-supervised learning is designed to pre-train a RFF extractor using unlabeled auxiliary dataset. Subsequently, to overcome the limitation of sample dependence, knowledge transfer is introduced to fine-tune the extractor and a classifier with target dataset including few samples (5-30 samples per emitter in this paper) and corresponding labels. In addition, auxiliary dataset and target dataset are come from different emitters. An open-source large-scale real-world automatic-dependent surveillance-broadcast (ADS-B) dataset and a Wi-Fi dataset are used to evaluate the proposed SA2SEI method. The simulation results show that the proposed method can extract more discriminative RFF features and obtain higher identification performance in the FS-SEI. Specifically, when there are only 5 samples per Wi-Fi device, it can achieve$83.40\%$identification accuracy, in which$38.63\%$identification accuracy improvement comes from the Adv-Aug of pre-training process. The codes are available athttps://github.com/LIUC-000/SA2SEI.

The evolution of wireless communication technologies has brought significant conveniences but also raised security concerns. Radio frequency fingerprint (RFF) is a potential feature, which can uniquely identify a specific emitter. The integration of Deep Learning (DL) has further enhanced the reliability of RFF identification. However, DL methods often struggle in dynamic communication environments. In this paper, we propose a dynamic adaptive RFF identification method leveraging Cognitive Representation Learning (CRL). Our proposed method is capable of recognizing and storing cognitive knowledge from historical environments. Furthermore, it dynamically adapts to current situations through its cognitive module, offering enhanced adaptability in dynamic environments. Specifically, we analyze the causes of RFF and define the RFF identification problems at first. Secondly, our cognitive module evaluates current data by examining both data distribution and feature distribution distances. Concurrently, our representation learning strategy enhances feature reuse and focuses on feature space. Finally, we implement an unsupervised ensemble module, combining unsupervised clustering with model ensemble techniques to boost performance. Simulation results validate our method’s robust generalization in dynamic settings, with an improvement of 7.66% in controlled environments and 5.98% in more challenging scenarios on PA dataset. Furthermore, the high identification ratio and ablation study results underscore the efficacy and necessity of each module in our approach.

Radio frequency fingerprinting (RFF) serves as a distinctive hardware trait in transmitters, forming the cornerstone of transmitter identification. While recent advancements led to significant improvements in identification accuracy, these developments also inadvertently simplify the process for adversaries to detect our transmitters. This vulnerability is particularly concerning in secure communications, as the exposure of device information could potentially result in the compromise of communication content, posing significant security threats. To counteract such risks and safeguard transmitters against unauthorized identification, this paper proposes a novel RFF concealment (RFFC) method based on IQ imbalance compensation and digital pre-distortion (DPD) techniques. This method not only effectively conceals the RFF, preventing malicious detection of the transmitter, but also enhances the system’s linearization performance. The effectiveness of the proposed RFFC framework is validated through MATLAB Simulink and a software and hardware test platform. Experimental results show that using the blind generalized linear structure-based IQ imbalance and deep neural network (DNN)-based PA nonlinearity joint concealment method performs best, reducing transmitter identification accuracy to only 17% under various signal-to-noise ratio conditions. Additionally, this method performs the best in system linearization performance.

Malware traffic classification (MTC) is one of the important research topics in the field of cyber security. Existing MTC methods based on deep learning have been developed based on the assumption of enough high-quality samples and powerful computing resources. However, both are hard to obtain in real applications especially in availability of IoT. In this paper, we propose a few-shot MTC (FS-MTC) method combining knowledge transfer and neural architecture search (i.e. NAS-based FS-MTC) with limited training samples as well as acceptable computational resources, in order to mitigate the identified challenges. Specifically, our proposed method first converts the raw network traffic into traffic images through data pre-processing to serve as input data for the neural network. Second, we use neural architecture search to adaptively search for the effective feature extraction model on the source domain (including Edge-IIoTset, Bot-IoT, and benign USTC-TFC2016). Third, the searched model is pre-trained on source task to achieve the generic feature representation of malware traffic. Finally, we only use few-shot malware traffic samples to fine-tune the pre-trained model to quickly adapt to new types of MTC tasks in realistic network environments. The experimental results show that the proposed NAS-based FS-MTC method has great scalability and classification performance in different FS-MTC tasks, including 5-wayK-shot USTC-TFC2016 dataset and 10-wayK-shot CIC-IoT dataset. Compared with state-of-the-art methods in the field of malware classification, the proposed NAS-based FS-MTC has higher classification accuracy. Especially in the 1-shot case of the USTC-TFC2016 dataset, its average accuracy is as high as 86.91%.

Specific emitter identification (SEI) uses advanced techniques to identify radio equipment by analyzing unique characteristics in radio frequency signals. Recently, deep learning (DL) has been considered a promising tools for designing various intelligent SEI methods. This is primarily due to its ability to fully exploit hidden data features and make autonomous classification decisions, leading to effective performance. The existing DL-SEI methods are based on the availability of extensive labeled datasets, however, collecting and annotating such data is challenging and time-consuming in real-world scenarios. Furthermore, these datasets often contain both device-specific and irrelevant features, which limits the adaptability of models to fixed channels. To overcome these challenges, we propose a robust variable-channel SEI (VC-SEI) method. This method uses semantic consistency-powered semi-supervised domain adaptation (SSDA). We introduce domain adversarial training to ensure global semantic consistency (GSC), allowing the extraction of channel-irrelevant features. Additionally, we design two loss functions to maintain local semantic consistency (LSC) for extracting category-relevant features. This approach enables effective domain adaptation. Our SSDA-based VC-SEI method has been rigorously evaluated using the ORACLE RF fingerprinting datasets from 16 USRP X310 radios. When only 1% of training samples in the target domain are labeled, our method achieves 84.20% identification accuracy in the target domain and 92.00% identification accuracy in the source domain. These results surpass those of current state-of-the-art methods. Simulation results confirm the robust identification performance of our proposed VC-SEI method in both source and target domains across all scenarios. Our code can be downloaded fromhttps://github.com/frownean/VC-SEI-based-SSDA.

Contactless Wi-Fi sensing, using channel state information (CSI) fingerprints, plays a pivotal role in communication, smart healthcare, and industrial automation. Deep learning has revolutionized the efficiency of non-contact sensing technology. Owing to its robust feature extraction capabilities and the interconnectedness of diverse sensing tasks, methods that address multiple tasks at once, like joint activity recognition and indoor localization (JARIL), have gained prominence. The primary goal of JARIL is to improve performance while reducing computational demands. Nevertheless, there remains substantial potential for enhancing its effectiveness through additional refinement and optimization measures. To address this, we introduce a regularized multi-label learning (RML) framework specifically designed for JARIL. This framework combines a parameter-efficient backbone network based on multi-scale separable convolution with residual connections, and a regularization training strategy. The latter strategy boosts performance by linearly combining two distinct CSI samples with their labels, creating new training instances in the training process. Simulation results show that the proposed method boasts a recognition accuracy of 91.73% and a localization precision of 99.64%. This marks an improvement of 4.32% and 3.60% respectively, in comparison to the prior ResNet1D+-based JARIL method. The codes can be downloaded fromhttps://github.com/BeechburgPieStar/JARIL.

Radio frequency fingerprint (RFF) has been widely used in wireless transceivers as an additional physical security layer. Most of the existing RFF extraction methods rely on a large number of labeled signal samples for model training. However, in real communication environments, it is usually necessary to process timely received signal samples, which are limited in quantity and are difficult to obtain labels, the performance of most RFF methods is generally poor. To effectively extract features from the limited and unlabeled signal samples, we propose an efficient RFF extraction method using an asymmetric masked auto-encoder (AMAE). Specifically, we design an asymmetric extractor-decoder, where the extractor is used to learn the latent representation of the masked signals and the decoder as light as a convolution layer reconstructs the unmasked signal from the latent representation. Using commercial off-the-shelf LoRa datasets and WiFi datasets, we show that the proposed AMAE-based RFF extraction method achieves the best performance compared with four advanced unsupervised methods whether in the case of large data size or small data size, or under line of sight (LOS) and non line of sight (NLOS) channel scenarios. The codes of this paper can be downloaded from Github: https://github.com/YZS666/AnEfficient-RFF-Extraction-Method.

Specific emitter identification (SEI) refers to a process to determine the category of emitters by extracting, analyzing and matching the characteristics of received emitter signals. With the increasingly complex environment, traditional SEI methods, such as parameter matching, become difficult to meet the needs of robust and effective signal identification. Deep learning (DL) possesses powerful feature extraction ability and has been widely used in SEI. The superior performance of DL-based SEI methods also brings problems of redundant model parameters and high feature dimensionality, which further causes slow convergence rate, high storage requirements, and ever-increasing computational complexity. In this paper, we propose an SEI method based on end-to-end sparse feature selection (SFS) to make model pay more attention to features with good identification performance. Specifically, we add sparse parameters to features and design loss function composed of cross-entropy loss and sparse regularization. Several experiments are conducted on ADS-B, WiFi and LoRa datasets. From the simulation results, our proposed SFS-SEI method improves feature sparsity, speeds up loss convergence, reduces model parameters on the premise of ensuring accuracy. Code is available at: https://github.com/sleepeach/SFS-SEI.

Specific emitter identification (SEI) extracts the fingerprint characteristics of emitters according to the subtle differences of transmitted signals, to distinguish different emitter individuals and prevent unauthorized network access. Deep learning (DL) based SEI methods have been proposed to achieve a good identification performance in recent years. However, these methods highly rely on expert experience to design network structures. These hand-designed fixed network structures lack flexibility, which often leads to insufficient model generalization. Neural architecture search (NAS) can be seen as a subdomain of automatic machine learning (AutoML), which can automatically adjust network structure and parameters according to a specific task. In this paper, we propose a neural architecture search-based SEI method, which can achieve an efficient search of the architecture with the use of a gradient descent algorithm. Experimental results show that the proposed NASEI method both improves the accuracy and reduces the parameter quantity when compared with state-of-the-art methods. Code available at https://github.com/huangyuxuan11/NASEI.git.

Deep learning (DL)-based specific emitter identification (SEI) is a potential physical layer authentication technique for Industrial Internet-of-Things (IIoT) Security, which detects the individual emitter according to its unique signal features resulting from transmitter hardware impairments. The success of DL-based SEI often depends on sufficient training samples and the integrity of samples’ labels. The extensive deployment of wireless devices generates a huge amount of signals, but signals labeling is quite difficult and expensive with the high demand for expertise. In this article, we present an SEI method based on dual consistency regularization (DCR), which enables feature extraction and identification using a few labeled samples and a large number of unlabeled samples. With the help of pseudo labeling, we leverage consistency between the predicted class distribution of weakly augmented unlabeled training samples and that of strongly augmented training unlabeled samples, and consistency between semantic feature distribution of labeled samples and that of pseudo-labeled samples, which takes the unlabeled samples into account to model parameter tuning for a more accurate emitter identification. Extensive numerical results demonstrate that compared with well-known semi-supervised learning-based SEI methods, our method obtains 99.77% identification accuracy on a WiFi data set and 90.10% identification accuracy on an automatic dependent surveillance-broadcast (ADS-B) data set when only 10% of training samples are labeled, and improves the identification accuracy on the WiFi data set and the ADS-B data set by more than 19.07% and 5.30%, respectively. Our codes are available athttps://github.com/lovelymimola/DCR-Based-SemiSEI.

Wi-Fi-based passive sensing is considered as one of the promising sensing techniques in advanced wireless communication systems due to its wide applications and low deployment cost. However, existing methods are faced with the challenges of low sensing accuracy, high computational complexity, and weak model robustness. To solve these problems, we first propose a robust channel state information (CSI)-based Wi-Fi passive sensing method using attention mechanism deep learning (DL). The proposed method is called as convolutional neural network (CNN)-ABLSTM, a combination of CNNs and attention-based bi-directional long short-term memory (LSTM). Specifically, CSI-based Wi-Fi passive sensing is devised to achieve the high precision of human activity recognition (HAR) due to the fine-grained characteristics of CSI. Second, CNN is adopted to solve the problems of computational redundancy and high algorithm complexity which are often occurred by machine learning (ML) algorithms. Third, we introduce an attention mechanism to deal with the weak robustness of CNN models. Finally, simulation results are provided to confirm the proposed method in three aspects, high recognition performance, computational complexity, and robustness. Compared with CNN, LSTM, and other networks, the proposed CNN-ABLSTM method improves the recognition accuracy by up to 4%, and significantly reduces the calculation rate. Moreover, it still retains 97% accuracy under the different scenes, reflecting a certain robustness.

The use of deep learning (DL) based hyperspectral image (HSI) classification has been made remarkable progress in recent years. However, obtaining sufficient labeled samples for training DL models remains a challenge. Transfer learning is effective in addressing the problem of HSI classification with limited labeled samples. However, cross-domain HSI classification using transfer learning remain difficult, as differences in ground object categories between two datasets make it challenging to transfer and learn accurate. To address this issue, we propose a simple yet effective method for HSI classification using Model-Agnostic Meta-Learning (MAML) and Regularized Fine-tuning (MRFSL). Our method uses optimized 3-Dimension Convolutional Neural Networks (3D-CNNs) model, aided by MAML and cutout data augmentation to enable cross-domain transfer learning and carry out the HSI classification with limited target samples. Experiments conducted on three HSI datasets demonstrate that the MRFSL method achieves excellent results compared to existing methods. Specifically, the overall accuracy of our proposed MRFSL method reached 91.81%, 71.04%, and 88.35%, when only five labeled samples for each category were randomly extracted from the Salinas, Indian Pines, and University of Pavia datasets, respectively.

Radio frequency-based device-free passive perception (RF-DFPP) is considered as one of the most promising techniques for ubiquitous smart applications in the WiFi field due to its extremely low deployment cost. Existing RF-DFPP methods typically employ received signal strength indicator (RSSI), ignoring the potential benefits of fine-grained sensing accuracy of channel state information (CSI). In addition, the robustness of such sensing methods is not good at present. To solve the problem, in this paper, we propose a robust CSI-based RF-DFPP method using a combination network of convolutional neural networks (CNN) and attention-based bi-directional long short term memory (LSTM). The combined network can extract the signal features of the collected CSI through CNN, and then realize RF-DFPP recognition through the training of LSTM and attention layers. Simulation results show that the proposed method significantly improves the recognition accuracy compared with the existing methods. Moreover, it performs robustly even if the model training is done under the different datasets.

Radio frequency fingerprint (RFF) identification is an emerging physical layer security technique, which provokes many promising applications in the internet of things (IoT). However, traditional machine learning-based RFF identification methods rely on complex manual feature extraction, while it is difficult for methods based on deep learning to deal with RFF identification under different channel environments. To solve these problems, we propose three different transfer learning-based RFF identification methods based on ConvMixer network, which is a mixture of different convolutional layers, using pre-trained model in the previous channel environment to assist in training under the new channel environment. Experimental results show that, compared with the previous retraining method, our proposed method reduces the number of training parameters and improves the identification performance at low SNR. Moreover, the proposed method can still have a certain performance guarantee with less training data.

Specific emitter identification (SEI) is developed as a potential technology against attackers in cognitive radio networks and authenticate devices in Internet of Things (IoT). It refers to a process to discriminate individual emitters from each other by analyzing extracted characteristics from given radio signals. Due to the strong capability of deep learning (DL) in extracting the hidden features of data and making classification decision, deep neural networks (DNNs) have been widely used in the SEI. Considering the insufficiently labeled training dataset and large unlabeled training dataset, we propose a novel SEI method using semi-supervised (SS) learning framework, i.e., metric-adversarial training (MAT). Specifically, two object functions (i.e., cross-entropy (CE) loss combined with deep metric learning (DML) and CE loss combined with virtual adversarial training (VAT)) and an alternating optimization way are designed to extract discriminative and generalized semantic features of radio signals. The proposed MAT-based SS-SEI method is evaluated on an open source large-scale real-world automatic-dependent surveillance-broadcast (ADS-B) dataset. The simulation results show that the proposed method achieves a better identification performance than four latest SS-SEI methods.

Blind signal recognition (BSR) is a significant research topic in the field of intelligent signal processing. However, existing BSR of space-time block codes (STBC) mainly depends on conventional algorithms, which require priori information and can only identify a relatively limited amount of STBC. Although deep learning (DL) has been widely used in signal recognition, so far there are few studies on BSR of STBC in multiple-input multiple-output (MIMO) systems using DL. In this paper, a blind recognition approach for STBC based on multichannel convolutional neural network (MCNN) is proposed. By leveraging the structure of multiple input channel, the in-phase and quadrature (IQ) channel information of STBC signals can be comprehensively extracted. Simulation results demonstrate that the proposed algorithm extends the recognizable STBC codes to 6, and can also improve the recognition accuracy in comparison to traditional convolutional neural network (CNN). The model proposed in this paper has been validated with two datasets and experimentally proved to be well generalized.

Malware traffic classification (MTC) is a very important component of cyber security, and a number of the MTC techniques are based on deep learning (DL) with a strong capability of feature mining and classification. However, these DL-based MTC methods are heavily dependent on a large amount of network traffic samples. In the few-shot scenarios, these methods usually overfit and have poor classification performance. Considering that the update cycle of malware is faster and faster, and there are more and more types of malware, collecting enough training samples for all malware is very challenging, if not impossible. In this paper, a novel few-shot MTC(FS-MTC) method is proposed based on convolutional neural network (CNN) and model-agnostic meta-learning (MAML) algorithm. Specifically, the CNN is trained on samples from normal softwares by MAML rather than the conventional optimization methods, then the CNN is finetuned by a few samples from malware for MTC. Simulation results show that our proposed MAML-based FS-MTC can outperform the traditional MTC methods. The performance of our proposed method can reach up to 95.69%.

To address the problem of spectrum resources and transmitting power for vehicular networks, this paper proposes a resource allocation (RA) method based on dueling double deep-Q network (D3QN) reinforcement learning (RL). Due to the high mobility of the vehicle, the channel changes rapidly which makes it difficult to accurately collect high-accuracy channel state information at the base station and to perform centralized management. In response of this difficulty, we construct a multi-intelligence model, using Manhattan Grid Layout City Model as the basis of environment and with each vehicle-to-vehicle (V2V) link as an intelligence. They work together to interact with the environment, receive appropriate observations, get rewards, and finally learn to improve the allocation of power and spectrum to enable users to achieve a better entertainment experience and a safer driving environment. Experimental results demonstrate that with proper training mechanism and reward function construction, cooperation among multiple intelligence can be performed in a distributed manner, with improvements in both the capacity of total vehicle-to-infrastructure links and the effective payload delivery success rate of the V2V links compared to common Q-network.

Specific emitter identification (SEI) is a promising physical layer authentication technique based on unintentionally hardware impairments of transmitters. These impairments are independent of the data’s content, so they are difficult to forge and analyze. Recently, most deep learning (DL) based SEI methods have been proposed, and have shown their great performance. However, these methods are big data-driven which means they have poor performance with limited training samples, and the vulnerability of neural networks to adversarial attacks is also a problem worth considering. In this paper, we propose an innovative few-shot SEI method based on class-reconstruction classification network and adversarial training (CRCN-AT) without the support of auxiliary dataset. Simulation results show that the proposed method achieves better identification performance and robustness in few-shot scenarios compared to traditional methods. The Pytorch code is released at https://github.comLIUC-000/CRCN-AT.

With the increasing popularity of Internet of things (IoT), the emergence of many IoT devices has led to security vulnerabilities. The classification of wireless signals is very important for secure communications. Most of existing signal classification tasks only focus on single signal classification task, while ignoring the relationship between radio frequency fingerprinting identification (RFFI) and automatic modulation classification (AMC). To solve the multi-task classification problem, this paper designs a multi-task learning convolutional neural networks (MTL-CNN). Real-radio datasets are generated by Signal Hound VSG60A and collected by Signal Hound BB60C to solve the lack of RFF samples with numerous modulation types. Experimental results confirm that the MTL-CNN method can work well by using the generated dataset. The MTL network designed in this paper improves the accuracy of RFFI by 1xs% relative to the single-task learning (STL) network. The keras code is released at https://github.comLiuK1288/1hw-000.

Human activity recognition (HAR) is one of the most promising technologies in the smart home, especially radio frequency (RF-based) method, which has the advantages of low cost, few privacy concerns and wide coverage. In recent years, deep learning (DL) has been introduced into HAR and these DL-based HAR methods usually have outstanding performance. However, as the recognition scenarios and target change, the model performance drops sharply. To solve this problem, we propose a generalized method for cross-person activity recognition (CPAR), which is called snapshot ensemble learning based an attention with bidirectional long short-term memory (SE-ABLSTM). Specifically, by defining the cosine annealing learning rate, the models with diversity are saved and integrated in the same training process. In addition, we provide a dataset for CPAR and simulation results show that our method improves generalization performance by 5% compared to the original method. The source code and dataset for all the experiments can be available at https://github.com/NJUPT-Sivan/Cross-person-HAR.

Specific emitter identification (SEI) extracts the fingerprint characteristics of emitters according to the subtle differences of transmitted signals, to distinguish different emitter individuals and prevent unauthorized network access. Deep learning (DL) based SEI methods have been proposed to achieve a good identification performance in recent years. However, the existing methods need a massive specific emitter dataset to alleviate model overfitting during the training stage. In this paper, we propose data augmentation (DA) aided few-shot learning method and validate the proposed method using automatic dependent surveillance-broadcast (ADS-B) signals. Specifically, according to the characteristics of ADS-B signals, four DA methods, i.e., flip, rotation, shift, and noise are studied for the proposed method. Experimental results are provided to show that the proposed method improves the recognition accuracy and the model robustness.

Specific emitter identification (SEI) is a highly potential technology for physical-layer authentication that is one of the most critical supplements for the upper-layer authentication. SEI is based on radio frequency (RF) features from circuit difference, rather than cryptography. These features are inherent characteristics of hardware circuits, which are difficult to counterfeit. Recently, various deep learning (DL)-based conventional SEI methods have been proposed, and achieved advanced performances. However, these methods are proposed for close-set scenarios with massive RF signal samples for training, and they generally have poor performance under the condition of limited training samples. Thus, we focus on few-shot SEI (FS-SEI) for aircraft identification via automatic dependent surveillance-broadcast (ADS-B) signals, and a novel FS-SEI method is proposed, based on deep metric ensemble learning (DMEL). Specifically, the proposed method consists of feature embedding and classification. The former is based on metric learning with a complex-valued convolutional neural network (CVCNN) for extracting discriminative features with compact intracategory distance and separable intercategory distance, while the latter is realized by an ensemble classifier. Simulation results show that if the number of samples per category is more than 5, the average accuracy of our proposed method is higher than 98%. Moreover, feature visualization demonstrates the advantages of our proposed method in both discriminability and generalization. The code and the dataset can be downloaded fromhttps://github.com/BeechburgPieStar/FS-SEI.

Malware traffic classification (MTC) is a key technology for anomaly and intrusion detection in secure Industrial Internet of Things (IIoT). Traditional MTC methods based on port, payload, and statistic depend on the manual-designed features, which have low accuracy. Recently, deep-learning methods have attracted a significant attention due to their high accuracy in terms of classification. However, in practical application scenarios, deep-learning methods require a large amount of labeled samples for training, while the available labeled samples for training are very rare. Furthermore, the preparation of a large amount of labeled samples requires a lot of labor costs. To solve these problems, this article proposes three methods based on semisupervised learning (SSL), transfer learning (TL), and domain adaptive (DA), respectively. Our proposed methods use a large amount of unlabeled data collected in the Internet traffic, which can greatly improve the classification accuracy with few labeled samples. Then, we use the DA method to solve the mismatch problem between the source domain and the target domain in the TL process. The proposed method is not only applicable to the shallow network but also to the deep neural network structure, and can achieve better classification results. Experimental results show that our proposed methods can satisfy the requirement of MTC in the case of few labeled samples in IIoT. The source code for all the experiments is available at GitHub.The code of this article can be downloaded from GitHub link:https://github.com/yzjh/Keras-MTC-DA-Ladder.

Malware traffic classification (MTC) is a key technology for solving anomaly detection and intrusion detection problems. And hence it plays an important role in the field of network security. Traditional MTC methods based on port, payload and statistic depend on the manual-designed features, which have low accuracy. Recently, deep learning methods have attracted significant attention due to their high accuracy in terms of classification. However, in practical application scenarios, deep learning methods require a large amount of labeled samples for training, while the available labeled samples for training are very rare. Furthermore, the preparation of a large amount of labeled samples requires a lot of labor costs. To solve these problems, this paper proposes two methods based on semi-supervised learning (SSL) and transfer learning (TL), respectively. Our proposed methods use a large amount of unlabeled data collected in the Internet traffic, which can greatly improve the accuracy classification with few labeled samples. Through experiments, we obtained the best method to improve the accuracy of few labeled samples in different situations. Experiment results show that our proposed methods can satisfy the requirement of MTC in the case of few labeled samples.

Convolutional neural network (CNN) is one of the most important tools to accomplish high-spatial-resolution remote sensing (HSRRS) image classification tasks with their unique feature extraction and feature expression capabilities. However, the CNN-based classification method is very limited due to the acquisition of HSRRS images is difficult and the sample size is limited. In addition, the extraction of features by a single model is very limited, which limits the further improvement of classification performance. To solve the above problems, we propose ResNet50-InceptionV3 based on deep transfer learning and multi-feature fusion (TLMFFRI) model to apply for high-spatial-resolution remote sensing image classification. First, both ResNet50 and InceptionV3 are trained on the ImageNet dataset. Then, transfer the trained convolutional layers weights to the TLMFFRI model to fuse the features and realize the HSRRS image classification. Finally, we evaluate the method on the HSRRS dataset. Compared with ResNet50 based on transfer learning (TL-ResNet50) and InceptionV3 based on transfer learning (TL-InceptionV3), the proposed method achieved better classification performance.

Deep learning-based radar signal recognition is considered one of the important technologies in the field of electronic countermeasure (ECM). However, existing deep learning-based methods require much time to design a specific neural network by experts for recognizing radar signals. It is difficult to employ these methods in real application scenarios. To solve this problem, we proposed an effective radar signal recognition method using neural architecture search (NAS) to automatically design convolutional neural networks (CNN). Experiments are given to validate the proposed method via comparing with both machine learning and deep learning-based methods. Experimental results show that the proposed method can achieve the optimal accuracy with low parameters and floating-point operations.

Scenario identification (SCI) is one of key techniques for intelligent vehicular communications (IVC) to maintain an effective and reliable operating state. Based on the deep learning (DL), it is a hotspot to identify scenarios of wireless communication using the characteristic quantity inherent in wireless channels. This paper proposes a decentralized learning-based SCI (DecentSCI) for IVC, relying on the algorithm of lightweight and model aggregation. By improving training efficiency and meanwhile reducing model complexity, the proposed method achieves low computing and communication, which is applicable for vehicular devices. Simulation results show that the training efficiency is upgraded by 97.15% and the model complexity is decreased by 90.25% at the cost of slight performance loss, i.e., 0.15%.

This paper proposes a decentralized automatic modulation classification (DecentAMC) method using light network and model aggregation. Specifically, the lightweight network is designed by separable convolution neural network (S-CNN), in which the separable convolution layer is utilized to replace the standard convolution layer and most of the fully connected layers are cut off, the model aggregation is realized by a central device (CD) for edge device (ED) model weights aggregation and multiple EDs for ED model training. Simulation results show that the model complexity of S-CNN is decreased by about 94% while the average CCP is degraded by less than 1% when compared with CNN and that the proposed AMC method improves the training efficiency when compared with the centralized AMC (CentAMC) using S-CNN.

Specific emitter identification (SEI) is a promising technology to discriminate the individual emitter and enhance the security of various wireless communication systems. SEI is generally based on radio frequency fingerprinting (RFF) originated from the imperfection of emitter's hardware, which is difficult to forge. SEI is generally modeled as a classification task and deep learning (DL), which exhibits powerful classification capability, has been introduced into SEI for better identification performance. In the recent years, a novel DL model, named as complex-valued neural network (CVNN), has been applied into SEI methods for directly processing complex baseband signal and improving identification performance, but it also brings high model complexity and large model size, which is not conducive to the deployment of SEI, especially in Internet-of-things (IoT) scenarios. Thus, we propose an efficient SEI method based on CVNN and network compression, and the former is for performance improvement, while the latter is to reduce model complexity and size with ensuring satisfactory identification performance. Simulation results demonstrated that our proposed CVNN-based SEI method is superior to the existing DL-based methods in both identification performance and convergence speed, and the identification accuracy of CVNN can reach up to nearly 100% at high signal-to-noise ratios (SNRs). In addition, SlimCVNN just has 10% ~ 30% model sizes of the basic CVNN, and its computing complexity has different degrees of decline at different SNRs; there is almost no performance gap between SlimCVNN and CVNN. These results demonstrated the feasibility and potential of CVNN and model compression.

Automatic modulation classification (AMC) is a critical algorithm for the identification of modulation types so as to enable more accurate demodulation in the non-cooperative scenarios. Deep learning (DL)-based AMC is believed as one of the most promising methods with great classification accuracy. However, the conventional CNN-based methods are lack of generality capabilities under time-varying signal-to-noise ratio (SNR) conditions, because these methods are merely trained on specific datasets and can only work under the corresponding condition. In this paper, a novel multi-task learning (MTL)-based generalized AMC method is proposed, and a more realistic scenario is considered, including white non-Gaussian noise and synchronization error. Its generalization capability stems from knowledge-sharing-based MTL in varying noise scenarios. In detail, multiple CNN models with the same structure are trained for multiple SNR conditions, but they share their knowledge (e.g. model weight) with each other. Thus, MTL can extract the general features from datasets in different noise scenarios. Simulation results show that our proposed architecture can achieve higher robustness and generalization than the conventional ones.

Signa1 modulation recognition (SMR) is an essential and challenging topic in orthogonal frequency-division multiplexing (OFDM) systems, and also it is the fundamental technique for signal detection and recovery. However, traditional feature extraction based SMR methods cannot effectively acquire the characteristics of the OFDM signals. Hence, the modulated OFD-M signal cannot be reliably identified. In this paper, we propose a deep learning (DL) based SMR method for recognizing OFDM signals, which is combined with a convolutional neural network (CNN) trained on in-phase and quadrature (IQ) samples. In the network model, the batch normalization (BN) layer and dropout layer are used to speed up model training and prevent overfitting, respectively. Three convolution layers with different convolution kernels perform well than traditional feature extraction methods in obtaining intrinsic properties of OFDM signals. The same number of multiple modulated signals are mixed and sent to the trained model for identification. Experiments are conducted to show that the method we proposed performs better than the traditional methods, mainly reflected in a higher probability of correct classification (PCC) and better consistency.

In this paper, we propose a convolutional neural network (CNN) aided automatic modulation recognition (AMR) method for a multiple antenna system. We also present two specific combination strategies, such as the relative majority voting method and arithmetic mean method to improve the classification performance in comparison with the state of the art. Our results are given to verify that the proposed method dominant exploits features and classify the modulation types with higher accuracy in comparison with the AMR employing high order cumulants (HOC) and artificial neural networks (ANN).