This paper provides a comprehensive study on the learning models' power violation, sum-rate performance while taking into consideration power constraint, and computational efficiency in terms of training and execution times over a dynamic wireless channel. We propose a reward shaping method and modify learning models with the output scaling strategy to enforce them to fully respect the power constraints while optimizing the sum-rate performance. The proposed approach reaches close-to-optimal accuracy, i.e., up to 99.15%, while satisfying the predefined power constraint of the base station. Moreover, learning models are shown to be more computationally efficient compared to the traditional algorithm. However, solving the power allocation problem within the Orthogonal Frequency Division Multiplexing (OFDM) symbol duration of $16.7\mu \mathrm{s}$ is a remaining challenge.

The polarized massive multiple-input multiple-output (MIMO) technique has been regarded as a promising solution to millimeter wave (mmWave) communication systems, because it experiences more degrees-of-freedom than the scalar configuration, and it represents a significant opportunity for secure communication. To deliver smart service to terminals, it is essential to provide base stations (BS) with the capability of terminal’s direction-of-arrival (DOA) awareness. In this paper, a compressive sampling (CS) framework is proposed for two-dimensional (2D) DOA and polarization estimation in mmWave polarized massive MIMO systems. The proposed approach first reduces the data volume via a reduced-dimension matrix. Then it computes the signal subspace via the eigendecomposition of the compressed array measurement. Thereafter, the rotational invariance characteristic is utilized to form a normalized polarization steering vector. Finally, 2D-DOA and polarization are estimated by incorporating the Poynting vector and the least squares (LS) techniques. The proposed architecture is computationally much more economical than existing algorithms. Besides, it allows a mmWave BS to provide comparable estimation performance with arbitrary sensor geometry, which is more flexible than most of the existing architectures. Furthermore, it is robust to the sensor position error. Numerical simulations verify the advantages of the proposed framework.

As one of the important enabling techniques for 6G, wireless caching network (WCN) attracts significant attentions. In this paper, we jointly apply unmanned aerial vehicle (UAV), millimeter wave (mmWave) multiple input multiple output (MIMO) and non-orthogonal multiple access (NOMA) in WCN. Our aim is to minimize the user delay, which is decomposed into three sub-problems, i.e., UAV deployment, hybrid beamforming and power allocation. Firstly, to improve the user rate, we apply K-means to reduce the distance between UAVs and users and propose a user pairing method to maintain the channel gain gap in each pair. Then, for increasing UAV hit probability, particle swarm optimization (PSO) and zero forcing are used for analog beamforming and digital beamforming, respectively. Finally, to further improve the user data rate, the genetic algorithm (GA) is applied to calculate the optimal NOMA power allocation factors. Simulation results confirm that the proposed schemes can achieve lower user delay compared with baseline schemes.

The Angle-of-Arrival (AoA)-based approach is an appealing solution for unmanned aerial vehicle (UAV) positioning, and has received significant interest recently. In this article, we propose a novel framework for UAV three-dimensional (3-D) positioning, the core of which is to measure the two-dimensional (2-D) Angle-of-Departure (2D-AoD) and 2D-AoA via a bistatic multiple-input multiple-output (MIMO) radar. Unlike the existing positioning architectures, the MIMO radar is equipped with polarized array antennas. An estimator based on the parallel factor (PARAFAC) decomposition is developed. It first obtains the direction matrices via performing the PARAFAC decomposition of the array data. Thereafter, the rotational invariance characteristic is utilized to form a normalized polarization response vector, from which the 2D-AoD, 2D-AoA, and polarization status of the UAVs are achieved via incorporating the vector cross-product method and the least squares (LSs) technique. Finally, the 3-D positions of the UAVs are easily calculated via the location relationship between the 2D-AoD, 2D-AoA, and the coordinates of transmitting/receiving (Tx/Rx) array. The proposed framework is computationally friendly, and is capable of positioning anonymous UAV. Moreover, it is insensitive to the geometry of the Tx/Rx array, indicating that the proposed framework supports configurable Tx/Rx antennas. Simulation results are provided to verify our theoretical advantages.

Due to the computing capability and memory limitations, it is difficult to apply the traditional deep learning (DL) models to the edge devices (EDs) for realizing lightweight automatic modulation classification (AMC). Recently, many works attempt to use different ways to realize lightweight AMC methods for EDs. However, the lightweight seems to be a contradiction with the classification performance in these lightweight networks. In this article, we propose an efficient lightweight decentralized-learning-based AMC (DecentAMC) method using spatiotemporal hybrid deep neural network based on multichannels and multifunction blocks (MCMBNN). Specifically, the lightweight network is designed from the perspectives of comprehensive consideration of lightweight and classification performance, which is composed of three parts to extract different features for realizing high classification performance and they are phase estimator and transformer (PET) block, spatial feature extraction block and temporal feature extraction & Softmax block. In addition, we use a multichannel input to extract complementary features of different channels for a better classification performance. The proposed DecentAMC method is an efficient training method, which is achieved by the cooperation in which multiple EDs update and upload the model weight to a central device (CD) for model aggregation to avoid the data privacy disclosure and reduce the computing power and storage pressure of CD. Experimental results show that the proposed MCMBNN can obtain an improved classification accuracy while reducing model complexity with the contributions of three blocks. Moreover, the proposed DecentAMC method can be deployed on EDs efficiently. Thus, the method has the advantages of avoiding data leakage on EDs and relieving the computing pressure of CD with relatively lower communication overhead. The simulation code and datasets are shared on GitHub.

With the explosive growth of advanced wireless technologies and computing device platforms, mobile sensing has gained huge attention. Indoor localization is actually considered as one of most valuable techniques in the field of contactless sensing. In this paper, we propose a novel graph convolutional network (GCN) empowered indoor localization method, which aggregates channel state information (CSI) features extracted from multiple multiple-input multiple-output (MIMO) links. CSI features from multiple antennas are basically converted into graph nodes in order to adopt GCN classification model. At the same time, graph attention mechanism is introduced to study and transfer spatial and frequency of CSI features. Eventually, output of graph is mapped with multiple measurement points through prediction network to provide final estimate position. 5GHz commercial Wi-Fi equipment is respectively utilized for data collection and experimental evaluation in two representative indoor scenarios. Experimental result shows that the proposed method has better performance in robust localization compared to other state-of-the-art deep learning methods.

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 a potential physical layer authentication technology, which is one of the most critical complements of upper layer authentication. Radio frequency fingerprint (RFF)-based SEI is to distinguish one emitter from each other by immutable RF characteristics from electronic components. Due to the powerful ability of deep learning (DL) to extract hidden features and perform classification, it can extract highly separative features from massive signal samples, thus enabling SEI. Considering the condition of limited training samples, we propose a novel few-shot SEI (FS-SEI) method based on hybrid data augmentation and deep metric learning (HDA-DML) which gets rid of the dependence on auxiliary datasets. Specifically, HDA consisting rotation and CutMix is designed to increase data diversity, and DML is used to extract high discriminative semantic features. The proposed HDA-DML-based FS-SEI method is evaluated on an open source large-scale real-world automatic-dependent surveillance-broadcast (ADS-B) dataset and a real-world WiFi dataset. The simulation results of two datasets show that the proposed method achieves better identification performance and higher feature discriminability than five latest FS-SEI methods.

With the ubiquitous deployment and applications of Internet of Things (IoT), security issues pose a critical challenge to IoT devices. External attackers often utilize vulnerable IoT devices to invade the target’s internal network and then further cause a security threat to the whole network. To prevent such attacks, it is necessary to develop a security mechanism to control the access of suspicious IoT devices and manage the internal devices. In recent years, deep learning (DL) algorithm has been widely used in the field of edge device identification (EDI), and has made great achievements. However, these previous methods are essentially centralized learning-based EDI (CentEDI) that trains all data together, which can not guarantee data security and not conducive to deployment on edge devices. To address this problem, we introduce a federated learning-based EDI (FedeEDI) method via network traffic to automatically identify edge devices connected to the whole network. Experimental results show that the training efficiency of our proposed FedeEDI method is much higher than that of the CentEDI method, although its classification accuracy is slightly reduced. In contrast to the CentEDI method, the proposed FedeEDI method has two main advantages: faster training speed and safer training process.

Rogue emitter detection (RED) is a crucial technique to maintain secure internet of things applications. Existing deep learning-based RED methods have been proposed under friendly environments. However, these methods perform unstably under low signal-to-noise ratio (SNR) scenarios. To address this problem, we propose a robust RED method, which is a hybrid network of denoising autoencoder and deep metric learning (DML). Specifically, denoising autoencoder is adopted to mitigate noise interference and then improve its robustness under low SNR while DML plays an important role to improve the feature discrimination. Several typical experiments are conducted to evaluate the proposed RED method on an automatic dependent surveillance-Broadcast dataset and an IEEE 802.11 dataset and also to compare it with existing RED methods. Simulation results show that the proposed method achieves better RED performance and higher noise robustness with more discriminative semantic vectors than existing methods.

Specific emitter identification (SEI) plays an increasingly crucial and potential role in both military and civilian scenarios. It refers to a process to discriminate individual emitters from each other by analyzing extracted characteristics from given radio signals. Deep learning (DL) and deep neural networks (DNNs) can learn the hidden features of data and build the classifier automatically for decision making, which have been widely used in the SEI research. Considering the insufficiently labeled training samples and large-unlabeled training samples, the semi-supervised learning-based SEI (SS-SEI) methods have been proposed. However, there are few SS-SEI methods focusing on extracting the discriminative and generalized semantic features of radio signals. In this article, we propose an SS-SEI method using metric-adversarial training (MAT). Specifically, pseudo labels are innovatively introduced into metric learning to enable semi-supervised metric learning (SSML), and an objective function alternatively regularized by SSML and virtual adversarial training (VAT) is 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) data set and Wi-Fi data set and is compared with the state-of-the-art methods. The simulation results show that the proposed method achieves better identification performance than existing state-of-the-art methods. Specifically, when the ratio of the number of labeled training samples to the number of all training samples is 10%, the identification accuracy is 84.80% under the ADS-B data set and 80.70% under the Wi-Fi data set. Our code can be downloaded from https://github.com/lovelymimola/MAT-based-SS-SEI.

Drone-aided ubiquitous applications play important roles in our daily lives. Accurate recognition of drones is required in aviation management due to their potential risks and disasters. Radiofrequency (RF) fingerprinting-based recognition technology based on deep learning (DL) is considered an effective approach to extracting hidden abstract features from the RF data of drones. Existing deep learning-based methods are either high computational burdens or have low accuracy. In this paper, we propose a deep complex-valued convolutional neural network (DC-CNN) method based on RF fingerprinting for recognizing different drones. Compared with existing recognition methods, the DC-CNN method has a high recognition accuracy, fast running time, and small network complexity. Nine algorithm models and two datasets are used to represent the superior performance of our system. Experimental results show that our proposed DC-CNN can achieve recognition accuracies of 99.5% and 74.1%, respectively, on four and eight classes of RF drone datasets.

As a typical Internet of Things application, network traffic prediction (NTP) plays a decisive role in congestion control, resource allocation, and anomaly detection. The trend of network traffic is different at different scales, so multiscale is an important characteristic of network traffic. In addition, the network traffic is nonlinear on each scale and dependent between scales. The existing NTP methods cannot comprehensively consider these characteristics, which limits their performance. In view of the characteristics of network traffic, such as multiscale, nonlinearity, and scale dependence, this article proposes a new multiscale NTP method based on a deep echo-state network (ESN). First, a multiscale parallel layered structure based on deep ESN is designed to fully consider the influence of each scale on the prediction result and then reduce the prediction error. Second, a feature extraction algorithm is proposed to improve the nonlinear approximation ability by extracting more abundant dynamic features with multiple reservoirs. Third, an NTP model based on scale dependence is proposed to reduce the influence from partial scale missing and then improve the prediction accuracy. Finally, simulation results demonstrate that compared with the state-of-the-art NTP methods, the proposed method significantly improves the prediction performance of network traffic with a slight increase in running time.