Search

Geo-decentralized federated learning (FL) can empower fully distributed model training for future large-scale 6G networks. Without the centralized parameter server, the peer-to-peer model synchronization in geo-decentralized FL would incur excessive communication overhead. Some existing studies optimized synchronization interval for communication efficiency, but may not be applicable to latency-constrained geo-decentralized FL. This paper first proposes the synchronization interval optimization for latency-constrained geo-decentralized FL. The problem is formulated to maximize the model training accuracy within a time window under communication/computation constraints. We mathematically derive the convergence bound by jointly considering data heterogeneity, network topology and communication/computation resources. By minimizing the convergence bound, we optimize the synchronization interval based on the approximated system consistency metric. Extensive experiments on MNIST, Fashion-MNIST and CIFAR10 datasets validate the superiority of the proposed approach by achieving up to 30% higher accuracy than the state-of-the-art benchmarks.

The development of offshore wind power is a crucial step in China’s energy transition and the achievement of carbon peaking and carbon neutrality goals. This paper first introduces a wireless communication technology for independent networking. Its applicability in the field of relay protection is analyzed, and a security encryption scheme for its use in relay protection services is designed. Considering the primary wiring form and protection function configuration of offshore wind power, a multi-terminal wireless differential protection method suitable for offshore wind power collection lines is proposed, achieving synchronization error less than 3° and typical delay of wireless self-organizing network communication less than 12ms. Utilizing multi-terminal data synchronization technology and differential protection algorithms, a scheme for fault isolation and recovery of non-faulty wind turbines is established, thereby achieving minimal range isolation of offshore wind power collection line faults and precise fault section location. The action time of multi-terminal differential protection is less than 40ms. Finally, experimental results confirm that this technology’s various metrics satisfy the demands of engineering applications and hold promotional value, offering a reference for the practical application of wireless communication technology in offshore wind power collection line protection services.

The radiometric calibration technique of spaceborne synthetic aperture radar (SAR) aims to establish a constant relationship between the SAR target pulse response energy and the actual radar cross-section (RCS) of the ground. It is an essential technology for achieving quantitative observations of the Earth using radar. In this article, the process of in-orbit radiometric calibration campaign for the SuperView Neo-2 SAR satellite is introduced, which is the distributed X-band SAR satellite in the new generation of Chinese SiWei commercial remote sensing satellite system. More specifically, the initial step involves deriving the total transfer function of the SAR system using the radar equation and establishing a stable calibration constant model for the SAR system. Further, by using the obtained calibration constant to invert the RCS value of the corner reflectors in the Otog Banner calibration field, while validating and analyzing the absolute radiometric accuracy. Considering that the elevation antenna pattern (EAP) predominantly affects the consistency of image radiometric correction, the method of extracting EAP and beam center bias angle from the rainforest is presented. Finally, the quality validation of impulse response function is performed over the California calibration field, which can effectively guarantee the quality of SuperView Neo-2 SAR satellite products.

To separate and unwrap the residual topographical and deformation components from the wrapped phases are the key issues for persistent scatterer interferometric synthetic aperture radar (PS-InSAR) technique. The current methods, such as grid-search method based on the assumption that the deformation in time-domain is linear, or the least squares ambiguity decorrelation adjustment method that needs priori information to construct the pseudo-observations, etc., are still difficult to tradeoff accuracy against efficiency. In this article, we propose a new method to separate and unwrap the differential phases for PS-InSAR by dividing them into residual topographical phase and differential deformation phases and performing the phase unwrapping separately. First, we calculate phase ambiguities caused by residual topographical error by zero-temporal baseline combination and single parameter grid-search method. The calculated residual topographical error is used to separate the differential phases dominated by deformations from the original differential phases of PS-InSAR. Then, we proposed time-domain differentiation to unwrap differential phases dominated by deformations without any assumption of deformation model in time-domain. The performance of the proposed method is tested by both simulated and real datasets. The experimental results show that, compared with classic PS-InSAR, the accuracy of residual topographical error estimation is comparable; however, there is an accuracy improvement of 34.4% for the deformation parameter estimation and 35.7% for nonlinear deformation time-series extraction, as well as an improvement of 20–40 times in computational efficiency.

Simultaneous localization and mapping (SLAM) is considered as a key technique in augmented reality (AR), robotics and unmanned driving. In the field of SLAM, solutions based on monocular sensors have gradually become important due to their ability to recognize more environmental information with simple structures and low costs. Feature-based ORB-SLAM is popular in many applications, but it has many limitations in complex indoor scenes. Firstly, camera pose estimation based on monocular images is greatly affected by the environment; secondly, monocular images lack scale information and cannot be used to obtain image depth information; thirdly, monocular based SLAM builds a fused map of feature points that lacks semantic information, which is incomprehensible for machine. To solve the aforementioned issues, this paper proposes an SDF-SLAM model based on deep learning, which can perform camera pose estimation in a wider indoor environment and can also perform depth estimation and semantic segmentation on monocular images to obtain an understandable three-dimensional map. SDF-SLAM is tested and verified using a CPU platform and two sets of indoor scenes. The results show that the average accuracy of the predicted point cloud coordinates reaches 90%, and the average accuracy of the semantic labels reaches 67%. Moreover, compared with the state-of-the-art SLAM frameworks, such as ORB-SLAM, LSD-SLAM, and CNN-SLAM, the absolute error of the camera trajectory on indoor data with more feature points is reduced from 0.436 m, 0.495 m, and 0.243 m to 0.037 m, respectively. On indoor data with fewer feature points, they decrease from 1.826 m, 1.206 m, and 0.264 m to 0.124 m, respectively.

The task of Point-of-Interest (POI) recommendation is to recommend the next interest locations for users. Gated Graph Neural Network (GGNN) has been proved to be effective on POI recommendation tasks. However, existing GGNN solutions rarely consider the spatiotemporal information between nodes in the sequence graph, which is essential for modeling user check-in behaviors in next POI recommendation. In this paper, we propose an attention-based spatiotemporal gated graph neural network model (ATST-GGNN) for next POI recommendation. Firstly, the user’s check-in sequence is represented as a graph structure. Secondly, we use spatiotemporal context information to dynamically update nodes in the sequence graph, and obtain the complex transfer relationships between the check-ins. Thirdly, each session is then represented as the composition of the long and short preference using an attention network. However, current short preference fails to model union-level sequential patterns, we improve the local embedding representation of graph nodes by window pooling method, as well as the global embedding representation of graph nodes by integrating it into attention mechanism. Finally, the objective function is constructed by cross entropy and the model parameters are learned. The experimental results show that the precision rate and mean reciprocal ranking of ATST-GGNN method are greatly improved compared with the state-of-art methods. It has good application prospect.

Traditional in-Silicon PN-junction-based on-chip electrostatic discharge (ESD) protection structures have inherent ESD-induced design overhead problems, including parasitic capacitance, leakage and Si area consumption. A potential solution to the ESD design overhead challenge is to use a non-conventional above-Si graphene-based nano-electrical mechanical system (gNEMS) transient switch structure to protect integrated circuits (ICs) against ESD failures. This paper reports investigation of materials and device structural impacts on gNEMS ESD protection structures. Transmission-line pulse (TLP) testing confirms that single-crystalline graphene gNEMS switch achieves a record high ESD current-handling capability of $\text{J}_{t2} {\mathbf {\sim }} 1.03\boldsymbol {\times } 10^{9}\text{A}$ /cm2, at least four times higher than $\text{J}_{t2} {\mathbf {\sim }} 0.24\boldsymbol {\times } 10^{9}\text{A}$ /cm2 for its poly-crystalline graphene counterpart. Transient 3D finite element method (FEM) simulation reveals that both device dimensions and shapes of the suspended graphene membranes can substantially affect gNEMS ESD discharging characteristics. The discovery offers guidelines for design optimization of gNEMS ESD switch structures.

This article reports design, fabrication and analysis of a novel under-transistor (under-FET) in-hole thermal sensor diode structure. Being able to accurately monitor self-heating of individual transistor in-operando, the under-FET temperature sensor enables smart full-chip run-time thermal management with spatial resolution down to single transistor level. The in-hole thermal sensors were fabricated in a CMOS process and validated in measurements. The new chip level thermal management technique was demonstrated using a prototype power amplifier (PA) IC designed in a foundry 40nm CMOS. It opens a door for self-learning based full-chip real-time intelligent thermal management for future ICs.