With the development of deep learning (DL), several fields have ushered in leapfrog development, such as image classification and natural language processing. Combing the powerful feature extraction tool, DL-based Automatic Modulation Classification (AMC) emerges. However, most DL-based AMC methods require massive labeled samples for training, which is difficult for non-cooperative scenarios. In this paper, we propose a novel self-supervised AMC method called GAF-MAE. Gramian Angular Field (GAF) is applied for domain transformation from time series to images first, then a Masked Autoencoder (MAE) is built for self-supervised reconstruction tasks through unlabeled samples. After self-supervised pretrain, a small number of labeled samples are used for downstream classification finetuning. We conduct a comprehensive evaluation of a public dataset RadioML.2016.10a. The simulation results show that GAF-MAE can achieve a relatively high average accuracy of 54.85% even if the label proportion is 5% and 61.38% when the label proportion reaches 100%, which outperforms other well-known models.

Automatic modulation classification (AMC) is the key technique in both military and civilian wireless communication. However, the performance is unsatisfactory, even several deep learning-based methods are involved. Targeting its low accuracy at low SNR, high computational cost and label overdependence, we propose a novel AMC framework, where the autoencoder (AE) serves as the backbone and Convolution-AE and LSTM-AE are combined in a parallel way as temporal and spatial feature extractors. The comparisons with serval algorithms on the radioML2016.10a show that our proposed network can achieve higher classification accuracy at low SNR with a low cost. In addition, it suits the semi-supervised scenario since the dependence on labels is loosen.

The advancement of deep learning (DL) techniques has led to significant progress in Automatic Modulation Classification (AMC). However, most existing DL-based AMC methods require massive training samples, which are difficult to obtain in non-cooperative scenarios. The identification of modulation types under small sample conditions has become an increasingly urgent problem. In this paper, we present a novel few-shot AMC model named the Spatial Temporal Transductive Modulation Classifier (STTMC), which comprises two modules: a feature extraction module and a graph network module. The former is responsible for extracting diverse features through a spatiotemporal parallel network, while the latter facilitates transductive decision-making through a graph network that uses a closed-form solution. Notably, STTMC classifies a group of test signals simultaneously to increase stability of few-shot model with an episode training strategy. Experimental results on the RadioML.2018.01A and RadioML.2016.10A datasets demonstrate that the proposed method perform well in 3way-Kshot, 5way-Kshot and 10way-Kshot configurations. In particular, STTMC outperforms other existing AMC methods by a large margin.

Specific emitter identification (SEI) aims to distinguish different emitter individuals based on the subtle differences in the signals. The technology can be widely applied to various wireless communication systems. Existing individual identification methods often ignore the radio frequency (RF) fingerprint information carried by the waveforms and thus can be interfered by unreliable RF features. To alleviate these issues, we devise the deep bidirectional long short-term memory (DBi-LSTM) and the one-dimensional residual convolution network with dilated convolution and squeeze-and-excitation block (Conv-OrdsNet). We exploit the combination of DBi-LSTM and Conv-OrdsNet (CoBONet) to extract temporal structure features directly from baseband in-phase and quadrature (I/Q) samples. The proposed network is able to capture the fine-grained details of signals and combine different information extracted by two networks. Moreover, we propose a data augmentation method to solve the interference of unreliable features by randomly changing the values of noise, power, frequency offset (FO), and phase offset (PO). It is worth noting that our method only needs a short slice of a steady-state signal without complex preprocessing, which reduces the cost of acquisition and calculation. Extensive experiments show that our method can effectively extract reliable RF fingerprinting features from I/Q samples and the classification results are better than most existing methods.

In modulation classification domain, handcrafted feature based method can fit well from a few labeled samples, while deep learning based method require a large amount of samples to achieve the superior classification performance. In order to improve the modulation classification accuracy under the constraint of limited labeled samples, this paper proposes a few-shot modulation classification method based on feature dimension reduction and pseudo-label training (FDRPLT), which combines handcrafted feature based method with deep learning based method. First, an optimal low-dimensional feature subset is created by the combination of the handcrafted features and autoencoder-extracted features post-processed by a feature selection algorithm. Then, a fully connected network (FCN), trained on a small number of labeled signals, is designed for the automated annotating, where unlabeled samples can be annotated and used for the later convolution neural network (CNN) training. The simulation results show that the classification accuracy of eight kinds of modulation types can reach to 98.3% when the SNR is 20dB.

Wideband signal detection and recognition (WSDR) is considered an effective technical means for monitoring and analyzing spectra. The mainstream technical route involves constructing time–frequency representations for wideband sampled signals and then achieving signal detection and recognition through deep learning-based object detection models. However, existing methods exhibit insufficient attention on the prior information contained in the time–frequency domain and the structural features of signals, leaving ample room for further exploration and optimization. In this paper, we propose a novel model called TFDP-SANet for the WSDR task, which is based on time–frequency distribution priors and structure-aware adaptivity. Initially, considering the horizontal directionality and banded structure characteristics of the signal in the time–frequency representation, we introduce both the Strip Pooling Module (SPM) and Coordinate Attention (CA) mechanism during the feature extraction and fusion stages. These components enable the model to aggregate long-distance dependencies along horizontal and vertical directions, mitigate noise interference outside local windows, and enhance focus on the spatial distributions and shape characteristics of signals. Furthermore, we adopt an adaptive elliptical Gaussian encoding strategy to generate heatmaps, which enhances the adaptability of the effective guidance region for center-point localization to the target shape. During inference, we design a Time–Frequency Clustering Optimizer (TFCO) that leverages prior information to adjust the class of predicted bounding boxes, further improving accuracy. We conduct a series of ablation experiments and comparative experiments on the WidebandSig53 (WBSig53) dataset, and the results demonstrate that our proposed method outperforms existing approaches on most metrics.