Device-to-Device (D2D) communication is confronted with critical security threats arising from its inherently short link distances. To address this, the Active Reconfigurable Intelligent Surface (ARIS) is employed to actively amplify reflected signals so as to reinforce the legitimate link, while simultaneously leveraging the near-field spherical wavefront characteristic to suppress eavesdropping directions. In this paper, we investigate an ARIS-assisted D2D secure transmission system operating in near-field channels and formulate a security energy efficiency maximization problem. Given the high coupling among variables in the objective function and constraints, an Alternating Optimization (AO) based algorithm is proposed to decouple the original problem into several subproblems. To tackle the non-convexity, we successively adopt Successive Convex Approximation (SCA) and Semidefinite Relaxation (SDR) to iteratively solve each subproblem. Simulation results verify that the proposed scheme significantly enhances the system security energy efficiency, thereby corroborating the synergistic benefits of active amplification and coherent phase control of ARIS in near-field D2D secure communications.
To address the issues of limited network coverage and weak computing capabilities in maritime environments, a task offloading scheme in the space-air-ground-sea integrated network scenario was introduced. Considering factors such as the offloading success rate of computational tasks, energy consumption constraints, environmental differences between near-shore and offshore scenarios, and the dynamic nature of the integrated space-air-ground-sea environment, a task offloading framework suitable for space-air-ground-sea integrated networks was constructed. A multi-agent collaborative task offloading scheme based on deep reinforcement learning was proposed. Experimental results demonstrate that, compared to offloading schemes based on the MADQN algorithm, the DDPG algorithm, and random policy, the proposed scheme improves the offloading success rate by 3.09%, 18.42%, and 66.42%, respectively, reduces delay by 19.07%, 21.53%, and 65.02%, respectively, and lowers energy consumption by 10.59%, 8.20%, and 8.75%, respectively.
Aiming at the problems of data security and low management efficiency in the operation of plant-station side intelligent power distribution and utilization terminals, a blockchain-based management system is proposed and designed in this paper. The system applies blockchain technology to construct an Alliance Chain Network (ACN), where the control units of plant-station side intelligent power distribution and utilization terminals are taken as the network nodes. Public key data verification and data submission for intelligent terminals are realized through a Trust Authority (TA). Besides, Smart Contracts (SCs) are adopted to automatically complete the public key registration, update and revocation of terminals, and a random number proof consensus mechanism is designed by combining with trusted chips. Experimental results demonstrate that, in comparison with the traditional management systems, the proposed system reduces the communication latency by 30%, makes the data tampering probability close to 0, and shortens the equipment fault response time by 40%. It effectively improves the information sharing efficiency among control unit nodes and guarantees the security of information transmission in the power distribution and utilization system.
Accurate prediction of electric vehicle charging demand is crucial for alleviating pressure on the power grid. To address the issues of slow convergence and vulnerability to unreliable nodes in existing prediction methods when processing non-independently and identically distributed (non-IID) data, this paper proposes a Double-Layer Credible Federated Learning (DCFL) strategy for energy demand prediction. The proposed method employs heuristic association rules to automatically mine charging information without requiring additional data collection. It selects benign local updates based on weight loss variation to accelerate convergence and resist interference from unreliable nodes. Furthermore, a multi-channel attention mechanism is introduced to design the loss function and the weighted aggregation scheme of federated learning. Experimental results demonstrate that, compared with traditional federated learning methods, the proposed method reduces training time by 71% and increases convergence speed by 17.6%, showing significant advantages in both prediction accuracy and convergence efficiency.
Sixth generation mobile communications (6G) is driving the evolution of wireless networks from traditional information transmission toward integrated systems that combine environmental perception with spatial intelligence. As the central medium through which 6G networks interact with the physical environment, electromagnetic space and its precise cognition directly determines the synergistic efficiency of perception, communication, and decision making; thus, it serves as a critical prerequisite for realizing on demand services in 6G networks. However, the inherent complexity of electromagnetic space makes it challenging to achieve high precision, real-time, and predictable digital representations, thereby constraining the development of intelligent digital twin technologies for 6G electromagnetic environments. Traditional modeling approaches struggle to simultaneously balance accuracy, real-time performance, and generalizability; moreover, existing research often focuses on only one of the three fundamental elements-radiation sources (“Source”), propagation environments (“Environment”), and electromagnetic fields (“Field”)-without systematically characterizing their coupled relationships. To address these issues, an integrated “Source-Environment-Field” cognitive model for intelligent electromagnetic space twins was proposed. The model achieves a fine-grained representation of the “Source” through techniques such as parameter estimation and behavior recognition, completes the physical reconstruction of the “Environment” through 3D geometric reconstruction and electromagnetic propagation-effect modeling, and performs digital-twin modeling of the “Field” using multiscale strategies and physics-data fusion methods. Furthermore, from an integrated cognitive perspective, this paper summarizes the roles of existing techniques in three collaborative links: existing techniques such as differentiable ray tracing, neural radiance fields, and generative radio map (RM) can mainly support the “Source-Environment-to-Field” forward prediction; source localization and parameter inversion methods can support the “Field-Environment-to-Source” inverse solving; and environmental structure and electromagnetic material inversion methods can support the “Field-Source-to-Environment” reverse inference. Subsequently, the paper explores the potential applications of intelligent electromagnetic space twins across various typical scenarios, including 6G networks, integrated space-air-ground communication systems, emergency rescue operations, and national defense security. Finally, the paper summarizes the major challenges currently facing this field of research and outlines future directions for the development of unified electromagnetic space modeling theories, physics-constrained da-ta-driven methodologies, and collaborative evolutionary approaches involving large-scale models-all geared toward the 6G era.
To address the problems of high small-object proportion, severe dense occlusion, shallow-detail degradation during downsampling, and limited onboard edge resources in aerial images captured by unmanned aerial vehicles (UAVs), a lightweight small-object detection method for UAV edge devices was proposed. YOLO11s was adopted as the baseline. The original P3/P4/P5 detection structure was reconstructed into a P2/P3/P4 structure, so that high-resolution shallow features could directly participate in tiny-object prediction. Meanwhile, the P5 branch with limited benefit for UAV small-object scenes was removed, and the high-level semantic modeling module was moved to the P4 layer to reduce redundant parameter overhead. A selective guidance block, a P2-guided cross-layer re-injection mechanism, and an adaptive fusion module were further designed to enhance shallow details and alleviate multi-source feature fusion conflicts. Experimental results on the VisDrone validation set showed that mAP (mean average precision)@0.5 and mAP@0.5:0.95 were improved by 4.40 and 3.22 percentage points, respectively, compared with the YOLO11s baseline, while the number of parameters was reduced from 9.46M to 3.82M. When deployed with TensorRT using 32-bit floating-point precision (FP32) on a Jetson Orin NX 16 GB platform, an inference speed of 52.63 frames per second (FPS) was achieved. The results indicate that the proposed method improves detection accuracy while maintaining a compact model size and real-time edge inference capability.
In reconfigurable intelligent surface (RIS)-assisted cell-free (CF) massive multiple-input multiple-output (m-MIMO) systems, the central processing unit (CPU) is required to perform large-scale computations for uplink signal detection. Concurrently, frequent and iterative RIS phase shift design and optimization significantly increase the computational burden on the CPU. This issue becomes particularly pronounced in scenarios involving RIS equipped with hundreds or thousands of reflecting elements, potentially leading to severe hardware overload. In light of this, for RIS-CF-m-MIMO systems featuring a substantial number of RIS reflection units, with the aim of reducing the CPU load and simultaneously enhancing the detection effectiveness of the uplink, this paper incorporates a data processing center to undertake the regulatory tasks of the RIS. On this foundation, a detection approach for the uplink of RIS-CF-m-MIMO based on the group-convolutional-neural-network (G-CNN) is put forward. Moreover, the closed-form expressions of the theoretical bit error rate (BER) and the achievable sum-spectral efficiency (ASSE) that can be achieved are deduced. The simulation results demonstrate the effectiveness and validity of the proposed method. In the uplink RIS-CF-m-MIMO with 4 transmit antennas, 32 receive antennas, 4-quadrature amplitude modulation signaling, and a 0.05 channel estimation error, where the CPU has no prior knowledge of the cascaded channel state information, and at an average received signal-to-noise ratio of 14 dB: The G-CNN detection method proposed in this paper can achieve its theoretical values in both bit error rate (BER) and average symbol square error (ASSE) performance, specifically, compared with the conventional zero-forcing (ZF) detection method and the standard CNN detection method, the BER of the proposed G-CNN method is reduced by approximately 5 orders of magnitude, and in terms of ASSE, it exhibits an improvement of about 5.5-fold relative to the ZF detection method and a 64.2% enhancement compared to the CNN detection method.
To address issues in class-incremental object detection models such as noise interference during detection, instability of old class boundaries, and imbalance between learning old and new classes, this paper proposes improvements to the CL-DETR model and introduces a class-incremental object detection model CC-DETR based on curriculum learning distillation. First, a noise-resistant collaborative purification module is constructed to filter label noise and suppress background noise interference while performing pseudo-label purification and foreground-guided distillation. Secondly, a boundary distillation method based on knowledge consolidation is introduced, applying KL distillation only to old class channels to stabilize the decision boundaries of old classes. Finally, we designed a curriculum-style learning distillation strategy, enabling the model to quickly learn new classes in the early stages of training, and dynamically adjust the distillation weights as training progresses to stabilize old classes while learning new ones. Experiments show that under the 70+10 and 40+40 splits of the COCO dataset, detection accuracy reached 40.9% and 39.1%, respectively, improving over CL-DETR by 1.9% and 2.4%, and outperforming current mainstream class-incremental object detection algorithms. Meanwhile, with the same inference structure, FPS increased by 9.7% and training time decreased by 25.4%, making it better suited for practical class-incremental object detection scenarios.
To solve the problems of poor flexibility of ground-based edge servers and insufficient computing power of servers in high-density traffic corridors to provide reliable offloading services, a UAV-assisted edge computing system for the Internet of vehicles (IoV) was proposed, which uses multiple UAVs as mobile base stations to assist ground-based edge servers in providing offloading services for vehicles. The system models the joint optimization problem of computational offloading and resource allocation as a mixed-integer nonlinear programming problem to optimize the offloading decision of vehicles and the resource allocation of edge devices to minimize the total system cost. To solve the problem, an algorithm incorporating a genetic algorithm and a deep deterministic policy gradient was proposed, which has a better optimization finding capability in the telematics scenario by introducing a genetic algorithm to pre-train the learning rate of the Actor module and the Critic module. Simulation results show that CORA_GADDPG reduces the system cost by 18.2%, 18.8%, 30.7%, and 78.3% compared to four algorithms for computing offloading and resource allocation problems, respectively, when offloading delay and energy consumption are considered. Therefore, CORA_GADDPG has a superior reduction effect on system cost.
To enable high-quality wireless communications in low-altitude economy scenarios, this paper investigates a rotatable antenna (RA)-assisted unmanned aerial vehicle (UAV) communication system. By leveraging the high directional gain of directional antennas and the spatial degrees of freedom provided by three-dimensional rotation, the UAV can efficiently receive uplink signals from multiple ground users. First, we formulate an optimization problem to maximize the minimum spectral efficiency among all ground users. Then, we investigate two RA rotation modes, namely the dynamic and static rotation modes, which aim to enhance beam-steering capability and reduce hardware implementation complexity, respectively. To tackle the resulting non-convex problem, we propose an efficient iterative algorithm based on successive convex approximation, where the RA rotation angles, time allocation, and UAV trajectory are jointly optimized. In addition, a closed-form solution for the optimal RA rotation angles is derived for the dynamic rotation mode. Finally, simulation results verify that the proposed RA-assisted UAV communication scheme achieves significantly better performance than conventional movable-antenna, isotropic-antenna, and fixed-antenna schemes.
Under the background of global population growth and environmental change, to improve the efficiency and accuracy of crop identification, this paper systematically reviews the characteristics and application scenarios of multi-source remote sensing satellite data, including optical, radar, and hyperspectral data, and discusses preprocessing methods such as radiometric correction, geometric processing, and resolution normalization. It focuses on spatiotemporal fusion, hierarchical fusion, and anti-interference fusion strategies, as well as the application of traditional machine learning methods (e.g., support vector machines and random forests) and deep learning models (e.g., convolutional neural networks, recurrent neural networks, and Transformers) in crop recognition. The results show that the combination of multi-source data fusion and machine learning can significantly improve identification accuracy. Spatiotemporal fusion enhances the ability to reconstruct crop growth curves, decision-level fusion achieves classification accuracy above 92%, and deep learning models exceed 93% accuracy in complex scenarios. This paper points out that data quality, model generalization, and computational resources remain major challenges, and future research should focus on novel fusion algorithms, deep integration technologies, and model optimization to promote the digitalization and intelligent development of precision agriculture.
To solve the problems existing in conventional passive Internet of Things energy harvesting schemes, including poor adaptability of single-band RF energy harvesting, low efficiency of multi-source integration, inaccurate impedance matching, and unstable output under variable environments, a dual-channel RF–solar hybrid energy harvesting circuit with Maximum Power Point Tracking (MPPT) was proposed in this paper. An improved voltage multiplier with single-ended grounded cascaded output was designed to realize efficient DC stacking of RF voltages. An open-circuit voltage of nearly 14.8 V and a maximum output power of 1.363 mW were achieved at 0 dBm input power. Compared with traditional structures, the open-circuit voltage was increased nearly fourfold, and the output power was improved by approximately 7.5%. A dual-input energy management module based on BQ25504 was adopted, in which independent MPPT control was implemented for RF and solar inputs with diode reverse-current protection. Coordinated operation and stable energy output are realized in the whole system.
Integrated sensing and communication (ISAC) achieves promising applications in target localization and environment sensing by sharing spectrum, hardware and software resources. Cell-free massive multiple-input multiple-output (CF-mMIMO) can suppress inter-cell interference and improve spectral efficiency. The CF-mMIMO ISAC system is expected to support both high-rate communication and high-precision sensing. Most existing studies model sensing targets as a single reflection point ones in CF-mMIMO ISAC systems. Massive distributed access point (AP) deployed in such systems shorten the distance between AP and targets, resulting in limited sensing information and failure to characterize target spatial distribution. This paper proposes a multiple reflection points target model to enrich sensing information and improve sum-rate, and focuses on precoding design.. For maximizing the system sum rate under sensing performance constraints, a two-stage framework is proposed: transform the non-convex problem via weighted minimum mean square error, derive a closed-form precoding solution using semi-definite relaxation, and solve optimal precoding via alternating optimization. Simulation results demonstrate that under sensing constraints, more reflection points boost the system sum-rate by around 30%. Moreover, the multiple reflection points model achieves greater communication gains with more users or AP, which aligns with future communication scenarios trends of massive users and ultra-dense CF-mMIMO, further underscoring its research significance.
Millimeter-wave Synthetic Aperture Radar (SAR) capable of three-dimensional (3D) imaging, achieved through two-dimensional scanning, has become a significant research direction in radar signal processing. However, 3D reconstruction performance is often limited by the transmitted signal bandwidth and aperture length, resulting in a range resolution significantly lower than the azimuth resolution. In multi-target scenarios, this disparity leads to severe image degradation from sidelobe interference. To address these limitations, this paper proposes a super-resolution 3D imaging method that combines a joint sparse constraint with the CLEAN algorithm. Within this framework, the 3D joint sparse constraint suppresses sidelobe energy and enhances reconstruction resolution, while the CLEAN process effectively separates signals from distinct scatterers using data-domain projection cancellation and amplitude correction. Validation with measured data shows that the proposed method substantially improves 3D image reconstruction, mitigates multi-target sidelobe interference, and enhances overall image quality. This work provides an effective solution for applying millimeter-wave 3D imaging in complex scenarios and shows considerable promise for practical engineering applications
In open vocabulary object detection, the traditional region proposal paradigm performs independent feature cropping and extraction on candidate regions, ignoring potential dependencies between regions, which leads to the loss of contextual information in feature representation. To address this issue, this paper proposes a Global-Local Context Synergy Network (GLCS-Net). First, a Global Context Aggregation (GCA) module is constructed, which aggregates global semantics by introducing a self-attention mechanism to reconstruct semantic dependencies between candidate regions. Second, a Local Spatial Integration (LSI) module is designed, which integrates spatial cues around the target by dynamic neighborhood sampling to complete the missing local geometric details of the object. Finally, to achieve the synergy between global and local contexts, a Context Synergy Purification (CSOP) mechanism is further proposed. This mechanism fuses the enhanced features from GCA and LSI, and eliminates noise by recalibrating the confidence of pseudo-labels, thereby constructing reliable supervision signals to guide model training. Experimental results show that on the COCO dataset, the core metric of novel class detection accuracy reaches 40.5%, which is 1.8% higher than the existing state-of-the-art methods. On the more challenging fine-grained LVIS dataset, the AP of rare, common, frequent categories and overall AP reach 25.4%, 32.9%, 37.7% and 33.5% respectively, which are 0.8%, 0.4%, 2.1% and 1.1% higher than the previous state-of-the-art methods. The results demonstrate that GLCS-Net effectively alleviates the information bottleneck caused by independent cropping of candidate regions, and provides a general feature enhancement paradigm for region-proposal-based detection frameworks.
To address the challenges of strong node heterogeneity, frequent dynamic topology changes, and difficult cache coordination in Space-Air-Ground Integrated Vehicular Network (SAGVN), this paper constructs a directed heterogeneous graph model for SAGVNs, designs a distributed caching decision framework that integrates Graph Attention Network (GAT) and Multi-Agent Deep Reinforcement Learning (MADRL), and proposes a caching strategy optimization method based on GAT and MADRL. Experimental results show that, compared to the Random strategy, GA strategy, MADQN strategy, and MADDPG strategy, the proposed strategy reduces the average task delay by 51.61%, 38.12%, 26.83%, and 22.12%, lowers the total system energy consumption by 71.68%, 67.53%, 43.23%, and 24.21%, and improves the cache hit rate by 261.19%, 49.11%, 21.43%, and 6.21%, respectively.
Object recognition in complex scenes is an important task in the Internet of Things (IoT) based edge perception systems. Infrared polarization images show superior foreground–background contrast in security monitoring and camouflage detection. However, due to the limited hardware resources of imaging systems, it is difficult for long-wave infrared division-of-focal-plane (DoFP) polarization imaging systems to achieve high-quality images in practical applications, which adversely affects edge perception performance. To this end, a lightweight joint denoising and demosaicking network for infrared polarization images is developed via spatial-frequency collaborative learning. The proposed method employs a three-stage learning network to enhance the quality of polarization image reconstruction while maintaining low computational complexity. The dual-domain interactive denoising block leverages the statistical properties of the frequency domain to suppress noise while preserving polarization features. After coarse demosaicking, a lightweight fine reconstruction module is adopted to generate the final results. Depthwise grouped convolutions are used to reduce model parameter count and improve reconstruction quality. Extensive experiments are performed on the IR-DoT and IR-DoFP datasets. The results show that the proposed method achieves superior performance compared with other leading approaches while maintaining low computational overhead, making it a low-power, high-quality data preprocessing approach for IoT-based edge devices.
The short message communication service (SMCS) is an integrated communication and navigation (ICAN) advantage service of China's BeiDou Navigation Satellite System (BDS). It not only enables ten million times/h high-capacity SMCS, but also features ten meter level high-precision positioning capability. The next-generation BDS will build a global short message communication service (GSMCS) system combining high and low orbit satellites, which service performance will upgraded from regional high capacity to global high capacity, to better support various Satellite Internet of Things (SIoT) applications. Under the existing frequency resource constraints of SMCS, to address the mutual interference problem of user inbound signals caused by satellite overlapping coverage and frequency sharing, while maintaining the ranging accuracy not less than 8ns, a multi narrowband subcarrier spectrum reuse method is proposed, with two schemes: equal division into 4 subcarriers and equal division into 8 subcarriers. Comparative analysis shows that the low orbit inbound performance using multi-narrowband subcarrier spectrum is significantly better than directly using the existing high orbit inbound spectrum scheme. Under the equal division into 8 subcarriers scheme, the inbound information rate range supported by low orbit SMCS covers the existing inbound information rate categories, which can better accommodate existing users of SMCS, while achieving a maximum increase of 4 times in inbound capacity and a maximum increase of 46.3% in spectrum efficiency.
Based on the 3GPP specifications and ITU-R standardized channel models, in direct-to-satellite smartphone scenarios, the satellite channel is dominated by the energy of the line-of-sight (LoS) path. Consequently, its delay spread is significantly smaller than that in terrestrial non-line-of-sight (NLoS) environments, exhibiting strong sparsity in the discrete delay domain. Due to the lack of structural constraints, the conventional least mean squares algorithm is highly sensitive to noise fluctuations on zero taps in the steady state, resulting in slow convergence and high steady-state error. Therefore, building upon the zero-attracting LMS (ZA-LMS) and reweighted zero-attracting LMS (RZA-LMS), this paper proposes an improved sparse least mean squares algorithm with a rational-function penalty, termed rational least mean squares (Rational-LMS), and derives its stochastic gradient descent-based weight update equation. By exploiting the high-sensitivity gradient of the rational-function penalty near the origin and its rapid attenuation in the large-coefficient region, the proposed algorithm effectively suppresses noise components while ensuring stable convergence of the equalizer coefficients, thereby yielding a regularization effect that is mathematically closer to the ideal 𝓁0-norm constraint. Furthermore, an adaptive parameter update mechanism is introduced to dynamically adjust the sparsity penalty factor according to the instantaneous error, thereby achieving faster convergence and lower steady-state mean squared error (MSE). Theoretical analysis and simulation results demonstrate that, in sparse channel environments, the proposed Rational-LMS outperforms ZA-LMS and RZA-LMS in terms of both convergence rate and steady-state MSE performance.
Wireless radio frequency backscatter is characterized by shared spectrum, low power consumption, and low cost, and it has important application value in the terminal devices and repeater markets of communication and sensing. Mobile communication technology has always been a hotspot of development. The transition from 5G to 6G is not only about speed but also about evolving from the Internet of things to the Internet of intelligent things. Both mobile communication technology and Internet of things technology face the proldem of spectrum scarcity, there is also demand for low-power energy-saving design, creating intersections with wireless radio frequency backscatter communication. Firstly, by exploring research hotspots such as mobile communication and Internet of things technologies, the wireless radio frequency backscatter communication was introduced in this study. Secondly, through reviewing seminal papers on the environmental backscatter communication, recent research, and fundamental principles, a clear developmental trajectory was mapped out and the current state was categorized across different dimensions. Then, base on this developmental trajectory and classification, existing issues and partial solutions were identified. Finally, combining the analysis of the full text, the next phase of development trends were predicted in this study.
