Automatic modulation recognition (AMR) is a widely used technique in various communication systems. In this work, we propose a complex-valued transformer (CV-TRN) network for AMR. Considering the in-phase (I) and quadrature (Q) components of the signal are two consistent data with only a phase difference of $\pi $ /2, they can teach the network independently which in disguise augment the training data, but the I/Q components are collectively needed to measure similarity in the multihead self-attention (MHSA). We input the I/Q data individually into the network with shared parameters, and they are transmitted independently in the network except in the MHSA, where a complex-valued MHSA (CMHSA) is proposed to let the information from I/Q components integrate. Moreover, CV-TRN adopts the relative position embedding, with a mathematical analysis of its advantages for AMR. A data augmentation method of random phase offset is introduced to further improve the robustness. Experimental results on RML2016.10a and RML2018.01a data sets demonstrate that the proposed CV-TRN outperforms state-of-the-art AMR methods and is parameter efficient.

This letter addresses the issue of underfitting or failure of deep learning models caused by insufficient training samples. Unlike previous supervised methods, a new few-shot learning semi-supervised automatic modulation recognition method based on multimodal information and domain adversarial network is proposed herein. The fusion input of multimodal information realizes the joint utilization of modulated signal modal features in the time and frequency domains. Domain adversarial training mines the potential knowledge information of a large number of unlabeled target domain data and introduces the convolutional block attention module (CBAM) to enhance the ability of the network to represent the key features of data. Numerical results validate the high-average classification accuracy of the proposed scheme compared to that of state-of-the-art schemes using fewer samples, particularly in high-order modulation classification. Alternative network structures are compared to confirm the applicability of multimodal information, domain adversarial training, and CBAM.

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

Automatic modulation classification (AMC) based on deep learning (DL) is gaining increasing attention in dynamic spectrum access for 5G/6G wireless communications. However, inconsistent feature parameters between the training (source) and testing (target) data lead to performance degradation or even failure of existing DL-based AMC. The primary reason for this is the difficulty in obtaining sufficient labeled training data in the target domain. Therefore, we propose a novel cross-domain AMC algorithm based on multimodal information and transfer learning, utilizing abundant unlabeled target domain data. We achieve complementary gains by fusing multimodal information such as amplitude, phase, and spectrum, which are used to train a network. Additionally, we apply domain adversarial neural network technology from transfer learning to learn from a large number of unlabeled data samples in the target domain to address the issue of decreased accuracy in cross-domain AMC caused by differences in sampling rate, signal-to-noise ratio, and channel variations. Furthermore, we introduce class weight weighting and entropy weighting to solve the partial domain adaptation problem, considering that the target domain has fewer modulation signal classes than the source domain. Experimental results on two designed modulation datasets demonstrate improved performance gains, thus validating the effectiveness of the proposed method.

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

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