Automatic modulation classification (AMC) is gaining greater significance in both military and civilian contexts. However, the diversity and dynamics of actual wireless communication environments can cause shifts in signal data distribution, posing risks of encountering unfamiliar modulations and negatively impacting recognition performance. To tackle these challenges, we propose the Open set domain adaptation for AMC (OSDA-AMC) algorithm. The approach utilizes a partial iterative separation technique, consisting of a pre-processing unit, a deep feature extractor, and meticulously crafted unknown, known, and domain classifiers. OSDA-AMC innovatively introduces an unknown class to the source classifier, facilitating the differentiation between known and unknown class features. Through multiple binary classifiers, the algorithm estimates the similarity between target data and each source class, differentiating unknown and known class data. Iteratively, it separates unknown signals from the target domain and labels them as unknown class. Simultaneously, the known class domain adaptation unit utilizes classifiers and domain discriminator for adversarial domain adaptation within the known class, ensuring similar feature distribution across the two domains. The proposed OSDA-AMC method can enhance the adaptability of AMC for recognizing unknown signals in dynamic channels in real-world environments. The experimental results demonstrate that the algorithm performs better in dynamic communication environments. By utilizing unknown class sample information effectively, we improve the accuracy of recognition and overall robustness.

Deep neural networks are used as effective methods for the Low Probability of Intercept (LPI) radar waveform recognition. However, existing models’ performance degrades seriously at low Signal-to-Noise Ratios (SNRs) because the effective features extracted by the networks are insufficient under noise jamming. In this paper, we propose a multi-resolution deep feature fusion method for LPI radar waveform recognition. First, we apply the enhanced Fourier-based Synchrosqueezing Transform (FSST), which shows good performance at low SNRs, to convert radar signals into time-frequency images. Then, we construct a multi-resolution deep convolutional network to extract more deep features from each resolution channel. Next, we explore an interactive feature fusion strategy for deep feature fusion. By some down-sampling or up-sampling blocks, different resolution features are fused to generate new features. Finally, we apply a fusion algorithm to the fully connected layer to achieve classification fusion for better performance. Simulation experiments on twelve kinds of LPI radar waveforms show that the overall recognition accuracy of our method can reach 95.2% at the SNR of −8 dB. It is proved that our approach does indeed improve the recognition accuracy effectively at low SNRs.

With the rapid development of wireless communication, automatic modulation recognition (AMR) plays a key role in spectrum management of cognitive radio (CR). However, the dynamic attributes of real-world communication environments, characterized by variations in channels, noise, and other factors, present formidable challenges to AMR systems based on deep learning (DL) technologies. Conventional DL-based AMR approaches, which presuppose data independence and identical distribution (i.i.d.), typically falter in adapting to these perturbations, thereby impeding their efficacy. To rectify this predicament, In this paper, a novel semi-supervised domain-adaptive automatic modulation recognition (SSDA-AMR) method is proposed. The proposed framework seamlessly combines labeled source domain data, sparsely labeled target domain data, and employs semi-supervised domain-adaptive techniques to harmonize features across domains. Data preprocessing encompasses the transformation of in-phase/quadrature (I/Q) signals into enhanced gray-scale contour stellar images (GCSI). By optimizing through the application of adversarial domain-adaptive loss and constraint functions, effective adaptation both inter-domain and intra-domain is achieved. Comprehensive experimentation, conducted on public datasets and custom dataset, conclusively affirms the remarkable generalization capabilities of the SSDA-AMR algorithm for disparate data distributions across various channels.

The YOLOv3 algorithm is directly used to detect aerial targets with complex backgrounds, which leads to low detection accuracy and high missed detection rate. This paper proposes an improved YOLOv3 algorithm to detect aerial targets. Because the aerial target is small and its pixel is low, and its features are not obvious, this paper first uses the K-means clustering algorithm to analyze the dataset, selecting the appropriate number and size of the anchor boxes, and then improving the network detection scale to establish the feature fusion target detection layer. In the dataset of PETS200S-Tracking and DOTA, the improved YOLOv3 algorithm and the YOLOv3 algorithm are compared. The results show that the improved network can effectively detect the aerial target, and its precision and recall rate are significantly improved.

As known to us all, the Cramér–Rao bounds (CRBs) is a fundamental tool of performance assessment which expresses the lowest possible mean‐squared error among all unbiased methods. Although existing results display the closed‐form expression of CRBs for co‐prime array and give the conditions under which the CRBs exist, they are not applicable when the authors consider two or more kinds of parameters estimation in multi‐dimensional sparse sensing, especially parameters estimation in the spatio‐temporal coarray domain. The authors first derive the CRBs for joint angle‐frequency estimation based on spatio‐temporal co‐prime sampling and present the closed form of the joint CRBs. Then, the sufficient and necessary conditions under which the CRBs exist are explored. Finally, they discuss and analyse the asymptotic behaviour of the joint CRBs as a function of the number of snapshots and signal‐to‐noise ratio, respectively. The correctness of their results is demonstrated by simulation experiments.