The use of supervised deep neural network (DNN) for automatic modulation classification, offering an end-to-end diagram, has gained significant attention in military and civilian field, e.g., spectrum monitoring, specific emitter identification and cognitive radio. However, this approach suffers from issues such as generalization error and spurious correlations. In an effort to capture and extract more abstract and useful concepts that can enhance performance on downstream tasks, one of the promising representation learning methods, known as “contrastive learning”, has achieved notable success in computer vision and natural language processing. This approach maximizes the similarities between different views of the same data example in the latent space to learn useful features. In this paper, we propose a contrastive-based objective for improving the transferability performance on the lower signal-to-noise ratios (SNR) dataset. Compared to the previous denoised-based methods used for classifying noisy signal data, we eliminate the constraints of pairwise input. This means that our model can leverage arbitrary combination of noisy and clean signal examples within same category. Additionally, the introduction of the noise level estimation enhances the robustness to the uncertain noise conditions. Simulation results on both the synthetic radar signal dataset and public communication signal dataset demonstrate that the proposed method exhibits minimal generalization error and showcases promising performance on signal data with different noise type.

Unmanned aerial vehicles (UAVs), also known as drones, have already been widely used in wireless networks. UAV-assisted integrated communication and sensing (ISAC) networks are feasible solutions to many challenging scenarios in which ground users (GUs) are inaccessible by terrestrial networks. However, the uncertainty of GU's locations undermines the performance of UAV-assisted networks, especially for UAV trajectory designs. To tackle this issue, we formulate this optimal UAV trajectory design problem to a catenary shape determination problem, which transforms the objective of maximizing the overall performance to that of minimizing the potential of the catenary. In the proposed scheme, an arbitrary partial distribution of GU's locations is represented by a matter with areal mass density in an artificial potential field (APF). To obtain an optimal solution, we derive a second-order mechanical equation representing the shape of this catenary, by analyzing its static equilibrium state when achieving minimal potential. Different from conventional path discretization methods, the obtained trajectory solution in this paper is a continuous-form second-order equation with remarkable path compression. The numerical results show that, when compared to conventional UAV trajectory optimization methods, the proposed approach can achieve an optimal solution of UAV trajectory with low computational complexity. It further demonstrates that the proposed approach can flexibly and continuously adjust the UAV trajectory under the scenario of uncertain GU's locations.

The radar signal intrapulse clustering (RSIPC) can help achieve unsupervised radar emitter identification, which is of great significance in the field of electronic warfare. In order to address the poor performance of traditional clustering methods in handling RSIPC tasks, we propose a contrastive learning-based RSIPC method called CLIPC. Since the single-domain information of radar signal intrapulses may result in the loss of important features, we integrate the multidomain information of radar signal intrapulses to obtain information fusion samples. By training a contrastive learning network on these information fusion samples, the network can extract deep features of radar signal intrapulses. Subsequently, we realize RSIPC using these deep features. To enhance the adaptability of the used contrastive learning network, we optimize the data augmentation methods in the network through experimental analysis. Additionally, we optimize the dimension of the deep features extracted by the network to reduce information redundancy and improve the efficiency of features clustering. Experimental results demonstrate that our improvements in contrastive learning network lead to better clustering performance and efficiency. We also investigate the clustering performance and reliability of the CLIPC under different signal-to-noise ratios (SNRs) through experiments. When the SNR is 0 dB, the proposed method has a clustering accuracy that is 0.1 higher than the contrastive learning method based on single-domain information, 0.3 higher than the traditional clustering method, and 0.2 higher than the clustering method based on autoencoder.