最新刊期
- 摘要：As the core resource for wireless communications and sensing, the electromagnetic spectrum is undergoing a paradigm shift from passive management to active cognition under the impetus of intelligent technologies. Electromagnetic-spectrum embodied intelligence deeply couples the agent with the electromagnetic environment to achieve autonomous sensing, cognition, and control of spectrum resources. This paper provides a systematic review of the theoretical framework, enabling technologies, and application prospects of electromagnetic-spectrum embodied intelligence. Starting from embodied-cognition theory, it argues for the legitimacy of “electromagnetic organs” as components of the body of an agent, reconceptualizes the electromagnetic spectrum as a key environmental dimension on par with three-dimensional physical space, constructs a mathematical model of the electromagnetic embodied closed loop, and reveals the essence that intelligence emerges from the dynamic coupling between body and environment. Building on a three-layer “sensing-cognition-action” architecture, we analyze core enabling technologies: at the electromagnetic sensing layer, across the signal, spatial, and multi-domain-fusion dimensions, we explain how software-defined radio, cognitive radar, cooperative spectrum sensing, and multimodal fusion enable a leap from physical signals to semantic understanding; at the cognition and decision-making layer, for both platform-constrained and multi-platform collaborative scenarios, we examine how lightweight reinforcement learning, online meta-learning, multi-agent collaborative decision-making, and game-theoretic optimization support autonomous decision-making and collective coordination in dynamic, adversarial environments; at the action-and-feedback layer, from electromagnetic-environment interaction to embodied-platform reconfiguration, we discuss how dynamic spectrum access, jamming and counter-jamming, waveform reconfiguration, and hardware reconfigurable technology translate cognitive policies into electromagnetic behaviors and close the loop. We summarize application practices and advantages in representative scenarios such as 6G mobile communications, intelligent electromagnetic system of LEO mega-constellations spectrum governance for smart cities, and electromagnetic operations on intelligent battlefields. In view of current challenges—including real-time constraints, sample efficiency, safety and robustness, and interpretability—we outline future trends involving standardization, ecosystem development, cross-disciplinary theoretical integration, and large-scale engineering deployment. The study offers a systematic theoretical reference and a technical roadmap for intelligent spectrum management, contributing to the modernization of spectrum-governance systems and supporting the construction of national information infrastructure.关键词：electromagnetic spectrum;embodied artificial intelligence;electromagnetic organ;reinforcement learning;multi-agent systems10|0|0更新时间：2026-06-16
- 摘要：Compared to traditional hard disk drives, flash memory-based solid-state drives (SSDs) have many attractive characteristics, such as high performance, low power consumption, small size, and shock resistance, and have been widely adopted in various storage systems including embedded devices, personal computers, data center servers and supercomputers. SSDs implement a software layer called the flash translation layer (FTL), which abstracts the underlying flash memory characteristics and provides management functions, enabling SSDs to act as traditional block interface devices. Data placement technology refers to the strategies or mechanisms for determining the storage locations of different types of data within the SSD based on varying storage requirements. It is crucial for optimizing SSD performance, lifespan, and quality of service. Traditional SSDs typically implement hot-cold data separation algorithms within the FTL, identifying data types based on information such as start address, length, and time of write requests to determine storage locations. However, limited available information within SSDs leads to low data classification accuracy. In recent years, various novel SSDs have emerged, such as multi-streamed SSDs, open-channel SSDs, zoned namespace SSDs, and flexible data placement SSDs. They adopt a hardware-software co-design approach, broadening the data semantic path between the host and the SSD, and providing new design ideas for SSD data placement technology. This article reviews the evolution of data placement technology in these novel SSDs, provides a detailed introduction to the research status of each technique, and focuses on analyzing their design principles, implementation methods, advantages and disadvantages, as well as deployment status. Additionally, we discuss the challenges and future research directions of data placement technology.关键词：flash memory;solid-state drives;flash translation layer;data placement;novel solid-state drives;hardware-software co-design5|0|0更新时间：2026-06-16
- 摘要：Airborne radio frequency (RF) systems are rapidly evolving toward intelligent skins, multifunctional integration, and intelligence, with exponential growth in system data volume, computational demands, and product iteration speed. Conventional FPGA+DSP custom computing architectures based on SRIO interconnection suffer from strong coupling between computing resources and tasks, inefficient aggregation of heterogeneous computing power, and limited system scalability, making them inadequate to support the requirements of next-generation airborne RF processing. To overcome these bottlenecks, this paper draws on the computational architecture concepts of mainstream intelligent computing centers and introduces them into the airborne RF computing domain, proposing a data processing unit (DPU)-based integrated RF processing system. This system employs a high-precision clock synchronization mechanism combining IEEE 1588v2 and synchronous ethernet (SyncE), addressing the high-precision spatiotemporal alignment of massive multi-source RF data in elastic distributed architectures. Based on remote direct memory access (RDMA), domain-specific architecture direct (DSA Direct) data passthrough is implemented, supporting end-to-end zero-copy transmission from data acquisition and processing to storage, significantly reducing system transmission latency. Through a heterogeneous computing power interconnection mechanism integrating PCIe and RDMA, aggregation and flexible configuration of heterogeneous computing power under a unified memory view are achieved. Leveraging nanosecond-level synchronous access, collaborative heterogeneous computing, and distributed storage capabilities, a data-centric integrated RF processing architecture is constructed. The architecture consists of RF access nodes and a heterogeneous computing resource pool interconnected through high-speed Ethernet switches, integrating intelligent, parallel, general-purpose, and reconfigurable heterogeneous computing nodes together with NVMe-oF distributed storage nodes into a unified computing power network, thereby supporting task-oriented elastic resource composition and dynamic deployment. Using a domestic DPU FPGA prototype, an experimental environment for the integrated RF processing system was established, and typical data flows including multi-channel RF data access, cross-node heterogeneous computing collaboration, distributed storage access, as well as RF data transmission, storage, and computation were validated. Test results indicate that compared to conventional architectures, the system reduces single-copy latency in data transmission to approximately 1/360 of the original; the peak sequential read/write bandwidth of NVMe over fabrics (NVMe-oF) storage nodes with three disks exceeds 5 700 MB/s, representing a 6 to 8 fold performance improvement over conventional methods; and the clock synchronization accuracy based on IEEE 1588v2+SyncE is approximately 8.33 ns. Furthermore, by constructing typical radar and communication signal processing chains, the system demonstrated an access latency as low as 24.92 ms for 128 MB radar raw data in the radar scenario, while stably supporting the real-time access of 38 parallel baseband data streams in the communication scenario, thereby validating its real-time service capability under high-throughput raw data injection and multi-level heterogeneous computing. Experiments demonstrate that this architecture provides a feasible technical pathway for next-generation airborne integrated RF processing systems.关键词：DPU;RF integration;computing power network;signal processing;edge computing5|0|0更新时间：2026-06-16
- 摘要：With characteristics of instantaneity, openness, and widespread dissemination, social networks facilitate information exchange but have simultaneously become a hotbed for the proliferation of rumors. Integrating heterogeneous information forms such as text and images, multimodal rumors exhibit greater deceptiveness and provocativeness, thereby significantly complicating network governance. Consequently, achieving efficient and accurate multimodal rumor detection has emerged as a critical mission for maintaining a clean cyberspace and ensuring public safety. In recent years, the research community has proposed diverse detection methodologies tailored to the data characteristics of multimodal rumors, leading to an increasingly diversified technical landscape. To systematically review current research, this paper constructs a multimodal rumor detection taxonomy from the perspective of information composition and sources, encompassing four dimensions: content, external knowledge, social context, and external environment. Unlike existing reviews with a singular emphasis on the content dimension, this paper, through its four-dimensional analytical framework, reveals a profound paradigm shift in rumor detection from localized static content verification to an integrated approach that incorporates dynamic social evolution and external environmental cross-validation. Within this framework, it analyzes the core mechanisms, evolutionary trajectories, and representative models of four methodological categories: visual-textual interaction, external knowledge enhancement, social context information, and external environment perception, summarizing their technical advantages and limitations. Additionally, this paper summarizes commonly used datasets and evaluation metrics, and comparatively analyzes the performance disparities of various models from the three dimensions of longitudinal technical evolution trajectories, cross-category statistical trends, and model performance under specific detection scenarios. Finally, this paper discusses five critical challenges: bottlenecks in modality heterogeneity and feature alignment, risks of external knowledge dependence and large language model (LLM) hallucinations, complexities of dynamic social and environmental modeling, cross-domain biases, and the authentication of highly realistic artificial intelligence generated content (AIGC) rumors. Furthermore, it outlines future research directions.关键词：social networks;multi-modal rumor detection;multi-modal fusion;knowledge enhancement;social context;external environment perception4|0|0更新时间：2026-06-16
- 摘要：In the distributed estimation framework based on wireless sensor networks, the conventional periodic time-triggered mechanism suffers from inherent drawbacks such as high communication frequency and massive redundant data exchange, which easily leads to excessive consumption of network communication resources and further degrades the performance of distributed estimation. To address this issue, this paper proposes a stochastic event-triggered communication mechanism and investigates the event-triggered distributed nonlinear estimation method for wireless sensor networks. This paper focuses on the event-triggered distributed nonlinear filtering over wireless sensor networks by introducing a stochastic event-triggered communication scheme. Given the pronounced nonlinear characteristics of the system’s observation model, traditional linear estimation methods fail to match the state evolution rules of nonlinear systems and cannot guarantee satisfactory estimation performance. Thus, within the framework of unscented kalman filtering (UKF), this paper adopts the unscented transformation to accurately approximate the nonlinear state propagation process, and develops a UKF-based local estimator tailored for nonlinear observation models to effectively improve local estimation performance. To reduce excessive estimation interactions caused by continuous and frequent information interaction among neighboring nodes, a local estimation-based stochastic event-triggered communication strategy is constructed according to the characteristics of local state estimation errors. Meanwhile, a recursive event-triggered local estimator is designed based on Bayesian estimation theory. By further employing the covariance intersection fusion technique, event-triggered local estimates are fused in real time to derive an event-triggered distributed estimator, which guarantees the consistency of distributed estimation under incomplete information exchange among neighboring nodes. Furthermore, the convergence of the proposed event-triggered distributed estimator is analyzed under the global observability of the network, and a sufficient condition is established to ensure the boundedness of the distributed covariance. Finally, simulation experiments are conducted using a typical maneuvering target tracking scenario to verify that the proposed algorithm can significantly reduce the network communication load while ensuring estimation accuracy.关键词：stochastic event-triggered communication;unscented Kalman filtering (UKF);covariance intersection fusion;nonlinear estimation;wireless sensor networks7|0|0更新时间：2026-06-15
- 摘要：Advanced Persistent Threats (APTs) have emerged as one of the most severe challenges to modern cybersecurity defense systems due to their extreme stealthiness, prolonged duration, and multi-stage nature. Although provenance-based analysis of host logs provides structured support for threat detection by correlating isolated system events into granular behavioral auditing paths, existing research still faces a core bottleneck: in complex system environments, attackers often disguise malicious activities as benign behavior through low-frequency operations, rendering traditional detection schemes based on signatures or static rules highly susceptible to failure when encountering zero-day attacks. To address these challenges, this paper proposes a Hierarchical-Aware Graph Masked Autoencoder framework for APT detection. The primary innovation of this framework lies in the introduction of hierarchical topological knowledge to guide the masking process, fundamentally overcoming the limitations of blind random masking. Specifically, the model integrates three targeted strategies: Global-Aware Masking (GAM), Local-Aware Masking (LAM), and Element-Aware Masking (EAM). GAM aims to preserve the macro-structural stability of the provenance graph; LAM focuses on characterizing neighborhood interaction logic between entities; and EAM addresses fine-grained entity attributes. This hierarchical design effectively filters out non-structural system noise during the pre-training phase while maximizing the retention of critical causal logic chains. Notably, the node-level consistency constraint models at an atomic scale, effectively circumventing the risk of signal dilution caused by global aggregation in traditional graph representation learning. This ensures that even under extremely imbalanced sample distributions, faint attack signals can still obtain sufficient gradient responses through the loss function, thereby mathematically guaranteeing the logical alignment between training objectives and point-wise anomaly detection tasks. During the detection phase, the framework employs an unsupervised anomaly detection algorithm to quantify node anomaly scores based on the embedding distributions of entity types, enabling the precise identification of malicious behaviors that disrupt local causal chains. Comprehensive evaluations were conducted on multiple authoritative public datasets, including StreamSpot, Unicorn Wget, and DARPA E3. Experimental results demonstrate that the proposed framework achieves an average precision of 98.49% and an F1-score of 98.97%. Compared to state-of-the-art baselines, our method exhibits superior robustness and recall in scenarios with extremely low attack base rates, effectively identifying subtle anomalous signals throughout the entire APT lifecycle.关键词：advanced persistent threats;anomaly detection;graph masked autoencoder;graph neural network;self-supervised learning;System Provenance Graph7|0|0更新时间：2026-06-15
- 摘要：In recent years, the growing demand for global sustainable energy development has accelerated the deployment of photovoltaic (PV) systems across residential, commercial, and industrial sectors. To ensure the safe operation of power grids and improve energy dispatch efficiency, accurately forecasting user-side power load variations has been increasingly critical. However, existing methods often overlook that the data collected from user-side smart meters typically contain both PV generation (source) and power consumption (load) components, which are not effectively separated, leading to biased prediction results. To address this issue, this paper proposes a multi-agent reinforcement learning-based source-load decoupling method, namely SoLED, that mitigates prediction bias caused by the unseparated mixed data of load and PV generation, as well as insufficient modeling of complex spatiotemporal dependencies among users. Specifically, we first design a feature extraction module to capture both short-term and long-term variations in user load, thereby enhancing the representation of spatiotemporal characteristics within the power grid. Then, based on the physical topology of the grid, we construct a simulation environment that models voltage responses under different load and PV power conditions. Finally, leveraging the feedback generated from this simulation, we train a decoupling model within a multi-agent reinforcement learning framework to achieve accurate source-load separation. Experimental results demonstrate that the proposed method improves prediction accuracy by 5.7%~24.6% and 6.6%~24.2% under different weather conditions on two real-world urban power grid datasets. These results confirm the effectiveness of the proposed approach in accurately decoupling user-side mixed power data and enhancing the modeling capability of complex spatiotemporal dependencies, thereby reducing the prediction bias in user-side load forecasting.关键词：photovoltaic generation system;spatiotemporal feature modeling;prototype learning;reinforcement learning;multi-agent reinforcement learning;source-load decoupling7|0|0更新时间：2026-06-15
- 摘要：Tensor-based radiance field methods established a mapping between inputs (3D spatial positions) and outputs (volume density and appearance features) via tensor regression. These methods relied on compact scene representations and significantly improved the efficiency of novel view synthesis while maintaining high-quality rendering results. However, existing approaches, whether based on conventional tensor decomposition or tensor train (TT) decomposition, are unable to fully exploit structural information in 3D scene space, thereby limiting the representation of deep-level features. To address this issue, we introduced tensor ring (TR) decomposition into the vector-matrix (VM) decomposition framework and proposed a vector-matrix tensor ring radiance fields (VMTR-RF) model for novel view synthesis. Unlike existing tensor radiance field methods, VMTR decomposition adopted a hierarchical modeling strategy: VM decomposition was first used to represent the scene as a combination of outer products of multiple vector and matrix factors, enabling an initial compact representation of the 3D scene. The vector-matrix factors were then reorganized into high-order tensors and further decomposed using TR decomposition, resulting in a tensor ring network composed of multiple core tensors, thereby enabling more effective capture of deep-level features in 3D scenes. Benefiting from the VMTR decomposition, VMTR-RF exhibited stronger modeling capability in volume density estimation and appearance feature learning. Finally, novel view synthesis was performed using volume rendering by combining the learned volume density and appearance features. Experimental results demonstrated that VMTR-RF outperformed existing state-of-the-art methods, particularly in detail preservation, enabling better reconstruction of sharp edges, complex structures, and natural textures, while achieving higher-quality novel view synthesis with a compact scene representation.关键词：neural radiance fields;tensor network;novel view synthesis;compact representation;VMTR decomposition5|0|0更新时间：2026-06-15
-
Radio Map Construction and Non-cooperative Radiation Source Localization Based on Agents Interaction
摘要：Radio map construction and non-cooperative emitter localization are pivotal tasks for accurate perception and dynamic control of the electromagnetic environment, characterized by high coupling and mutual constraints. Drawing on cognitive learning and embodied intelligence theories, this paper proposes a unified dual-task framework for radio map construction and emitter localization featuring multi-agent collaborative interaction and endogenous intelligence, thereby establishing a "perception-cognition-action" feedback loop. First, leveraging the concept of crowdsourced probabilistic consensus, we integrate Gaussian Process Regression (GPR) with 2D-positional encoding to transform discrete sparse sampling data from multiple agent into spatial attention maps. These maps encompass continuous signal field predictive distributions, geometric topologies, and probability estimates of emitter spatial presence, establishing a consensus perceptual prior regarding the target emitters. Building upon this, a consistent joint representation for emitters is constructed within a unified multi-agent semantic space. A multi-head self-attention mechanism is then utilized to design task-driven and target-driven agent interaction strategies, enabling the simultaneous execution of both tasks. Furthermore, a dual-task information interaction feedback strategy based on Proximal Policy Optimization (PPO) is designed to achieve self-organization and self-learning, forming an embodied behavioral loop of perception, cognition, and interactive decision-making. Simulation results validate the feasibility and superiority of the proposed scheme. Under a 5% sparse sampling condition, the synergistic optimization of both tasks achieves accuracy improvements of 18.2% and 43.5% for radio map construction and emitter localization, respectively, compared to independent task execution. Specifically, the RMSE of radio map construction improves by 60.95% and 32.55% over Kriging and Full Convolutional Auto-Encoder, respectively. Meanwhile, the localization error is reduced by 65.7% compared to the RSS-based baseline, maintaining a stable relative accuracy within 1.2% even when extended to a 10 km×10 km large-scale scenario.关键词：agent interaction;embodied intelligence;consensus perception;radio map;non-cooperative radiation source localization5|0|0更新时间：2026-06-15 - 摘要：The missile-borne terminal guidance radar employs monopulse angle measurement to achieve precise angle tracking of a ship target during the tracking phase. Centroid corner reflector jamming exploits the inherent limitation of monopulse angle measurement — that it points to the energy centroid of the echo rather than the geometric center of the target — by deploying corner reflectors with high radar cross section (RCS) in the same range resolution cell as the ship, thereby decoying the tracking point and causing the missile to miss. In the range-Doppler image, the ship and the centroid corner reflector occupy the same range-Doppler cell, making identification and suppression of the corner reflector extremely difficult. The fundamental reason is that the two are closely spaced and aliased in the range-azimuth parameter domain, while polarization, spatial, and other multi-dimensional radar features have long been processed in isolation, leading to insufficient joint identification capability. Forward-looking imaging can integrate range-azimuth-polarization information and offers the potential to identify and localize the target and the jammer in the image domain. However, under forward-looking conditions, the platform motion direction is nearly aligned with the beam pointing direction, so the target Doppler frequency variation is minimal, rendering high-resolution methods that rely on the Doppler effect, such as synthetic aperture and Doppler beam sharpening, ineffective. Methods such as monopulse imaging, scanned beam deconvolution imaging, and array spectral estimation imaging face performance bottlenecks in close-range terminal centroid jamming scenarios, including pointing errors, scan dependence, and low signal-to-clutter ratio, respectively. Polarization, as a vector property of electromagnetic waves, contains rich information about the target’s geometric structure and scattering mechanisms. The polarization scattering characteristics of a ship and a corner reflector are inherently different. However, classical polarization processing methods such as polarization matching and polarization filtering are designed based on the enhancement or suppression of echo energy and fail to fully exploit the separability differences of signal components between targets in the polarization domain. This paper, grounded in the polarization modulation super-resolution theory, derives a polarization array echo model, establishes an adaptive modulation criterion, and proposes a parallel iterative optimization method based on polarization state screening. On this basis, by exploiting the differences in polarization scattering characteristics between the ship and the corner reflector, a corner reflector suppression method based on polarization principal components is proposed, thus forming a forward-looking imaging method against centroid corner reflector jamming. Simulation results show that at a signal-to-clutter ratio of 10 dB, for a combined target consisting of a trihedral and a dihedral corner reflector separated by 0.2 times the beamwidth, the proposed method achieves an angle measurement accuracy of 0.05 times the beamwidth. Under conditions of an 8 km missile-to-target range and a 20 dB signal-to-clutter ratio, the proposed forward-looking imaging method improves the ship-to-corner-reflector signal-to-interference ratio by 23.3 dB and 11.4 dB compared with the single-polarization method and the polarization-spatial joint method, respectively, before ship maneuvering, and by 14 dB and 2.4 dB, respectively, after ship maneuvering.关键词：polarization modulation;forward-looking;super-resolution;centroid-based jamming;corner reflectors suppression4|0|0更新时间：2026-06-15
- 摘要：Alzheimer’s Disease (AD) is a latent and irreversible progressive neurodegenerative disorder. Early precise identification is crucial for delaying disease progression and supporting clinical intervention. Structural Magnetic Resonance Imaging (sMRI) can reveal anatomical abnormalities such as brain atrophy and gray matter degeneration, while Functional Magnetic Resonance Imaging (fMRI) captures functional activities and dynamic interactions between brain regions. Both provide complementary information for AD diagnosis from structural and functional perspectives. Multimodal joint diagnosis for sMRI and fMRI faces three key challenges. First, 3D sMRI and 4D fMRI differ significantly in spatial resolution, temporal dimension, and signal distribution, making it difficult to construct a unified end-to-end cross-modal modeling framework. Second, although Vision Transformer-based sMRI feature extraction offers global modeling capabilities, the standard multi-head attention suffers from redundant heads, insufficient inter-head collaboration, and limited representation of local structural details, reducing sensitivity to AD-relevant regions such as the hippocampus and olfactory cortex. Third, most multimodal fusion methods rely on feature concatenation, unidirectional attention, or dense interaction strategies, which are insufficient to screen key regions and establish fine-grained bidirectional associations between structural and functional features in high-dimensional heterogeneous image data. To address these issues, this paper proposes BiSparseFusion, a cross-modal bidirectional sparse interaction fusion model. The sMRI branch employs a 3D Vision Transformer enhanced with a dynamic composable multi-head attention mechanism (DCMHA) and a multi-level feature fusion module (MFFM). DCMHA reduces redundant attention outputs by dynamically combining attention heads, and MFFM aggregates multi-level features to enhance local lesion details and global semantic representation. The fMRI branch uses SwiFT to directly model spatio-temporal dependencies of the original 4D fMRI, avoiding information loss caused by conventional region of interest- or connectivity-based methods. During cross-modal fusion, a bidirectional sparse cross-attention fusion module (BSCAF) suppresses redundant features within modalities and enables deep complementary interaction between sMRI structural and fMRI functional representations. The proposed method is validated on the ADNI dataset across three classification tasks: AD/NC, MCI/NC, and AD/MCI. BiSparseFusion achieves classification accuracies of 97.67%, 93.27%, and 96.72%, respectively, surpassing various single- and multi-modal comparison models. Visualization results indicate that the model effectively focuses on brain regions associated with AD pathology, including the hippocampus, olfactory cortex, and amygdala, forming a more discriminative fusion feature space. These results demonstrate the effectiveness of BiSparseFusion in multi-modal neuroimaging feature modeling, cross-modal fine-grained fusion, and AD auxiliary diagnosis.关键词：Alzheimer’s disease;multimodal fusion;Structural magnetic resonance imaging;Functional magnetic resonance imaging;Dynamic Composable Multi-Head Attention;Bidirectional Sparse Cross-Attention Fusion3|0|0更新时间：2026-06-15
- 摘要：Low-altitude unmanned aerial vehicle (UAV) swarms in urban airspace operate over rapidly time-varying multipath channels, where severe Doppler shifts and doubly selective fading can degrade data links, sensing reliability, and closed-loop cooperative control. To address these issues, this paper proposes an integrated sensing-communication-control framework based on MIMO-OTFS-non-orthogonal multiple access. Specifically, orthogonal time frequency space (OTFS) modulation is utilized to provide a quasi-static representation in the delay-Doppler (DD) domain, enhancing robust transmission and parameter estimation (e.g., range and velocity) under doubly selective channels. Simultaneously, multiple-input multiple-output (MIMO) and power-domain non-orthogonal multiple access (NOMA) are combined to achieve massive access and spectral efficiency. Furthermore, we construct a DD-domain micro-Doppler structured-mapping mechanism, where rotor micro-Doppler signatures serve as physical-layer fingerprints for identity association, providing a reliable index for robust tracking and cooperative control under dense or crossing trajectory conditions. For the uplink backhaul, a superposition transmission architecture is designed with full-frequency multiplexing for the master UAV and frequency-division access for slave UAVs. By decomposing complex cluster interference into parallel two-user NOMA models and applying a joint detection scheme based on the message passing algorithm and successive interference cancellation, the proposed method effectively mitigates inter-user interference, ensuring air-ground link reliability. Finally, a sensing-aided cooperative trajectory control loop is established using the sensed range/velocity and micro-Doppler feature identity association results to realize the integrated sensing-communication-control process. Simulation results demonstrate that the proposed scheme avoids the bit-error-rate floor observed in baselines based on orthogonal frequency division multiplexing and non-orthogonal multiple access, and achieves a signal-to-noise ratio gain of approximately 7 dB. Moreover, the sensing module realizes stable estimation of range and velocity, with accuracy trends consistent with the Cramér-Rao bound analysis. In the closed-loop cooperative validation, the sensing-feedback-based UAV swarm achieves three-dimensional formation trajectory tracking. During continuous maneuvering cruise, the cooperative position root mean square errors of the four slave UAVs are controlled at 0.13, 0.15, 0.19, and 0.23 m, respectively, and the relative errors do not diverge. This preliminarily verifies the closed-loop stability of the proposed scheme in the integrated sensing-communication-control process at the simulation level.关键词：integrated sensing-communication-control system;orthogonal time frequency space;micro-Doppler;non-orthogonal multiple access;control compensation mechanism16|4|0更新时间：2026-06-12
- 摘要：With the development of 5G wireless communication, the Internet of Things, high-speed interconnection, and artificial intelligence technologies, radio-frequency electronic devices have imposed higher requirements on the high-frequency transmission capability, flexible integration characteristics, and process compatibility of interconnect materials. Conventional metallic conductors suffer from problems such as high elastic modulus, relatively large mass, and additional losses induced by surface roughness in miniaturized, high-frequency, and flexible electronic applications, making it difficult for them to meet the development requirements of next-generation integrated systems. Two-dimensional layered titanium carbide (Ti₃C₂Tₓ, MXene) exhibits promising application potential in radio-frequency transmission lines and high-frequency interconnect devices owing to its relatively high electrical conductivity, lightweight nature, flexibility, and solution processability. However, studies on the effects of the film-forming process, surface morphology, and roughness characteristics of MXene films on radio-frequency transmission performance remain relatively limited. In this study, MXene thin-film coplanar waveguide radio-frequency transmission lines were investigated. The conductor surface roughness was regulated by controlling the preparation process, and its influence on radio-frequency transmission performance was studied. MXene films with thicknesses of 1.4 ± 0.2 μm,2.6 ± 0.3 μm, and3.6 ± 0.2 μm were prepared using doctor-blading and drop-casting methods. Atomic force microscopy, scanning electron microscopy, vector network analyzer measurements, and electromagnetic simulation were combined to analyze surface roughness, laser-cut edge morphology, insertion loss, and attenuation rate per unit length. The results show that different film-forming processes significantly affect the stacking state of MXene nanosheets, film uniformity, and surface roughness. The drop-casting process mainly relies on capillary-force-driven random assembly of nanosheets, which tends to form a rough surface with large height fluctuations. In contrast, the doctor-blading process induces the in-plane alignment of nanosheets through shear action, facilitating the formation of a flatter and denser film structure. For all three thicknesses, the surface roughness of the doctor-bladed films was only approximately 30%~40% of that of the drop-cast films. Radio-frequency test results show that lower surface roughness helps reduce the insertion loss and attenuation rate of MXene transmission lines, and this advantage becomes more pronounced at smaller film thicknesses and higher operating frequencies. At a thickness of 1.4 μm and a frequency of 18 GHz, the doctor-bladed sample exhibited an insertion loss approximately 2.7 dB lower than that of the drop-cast sample, with transmission performance improved by approximately 36.5% and attenuation rate reduced by approximately 1.7 dB/cm. Further results indicate that the roughness of the laser-cut edge also affects transmission performance, and a laser power of 25W with a cutting speed of 150 mm/s produced a more regular edge morphology and better transmission performance. Finally, by combining the roughness-gradient conductivity model, the influence of surface roughness on the transmission coefficient was revealed. This study provides experimental and theoretical bases for the process optimization and device design of radio-frequency transmission lines in electronic devices based on non-metallic, novel layered conductive materials.关键词：MXene;Radio-Frequency transmission;surface roughness;thin-film fabrication;Transmission coefficient17|1|0更新时间：2026-06-11
- 摘要：With the deep integration of 6G communication, sensing, and computing, as well as the continuous penetration and deployment of large models toward the edge, emerging application scenarios such as crowd intelligence sensing have witnessed an explosive growth in demand for edge computing power. Against this backdrop, the imbalance between computing power supply and demand has become increasingly prominent, and computing power pooling and virtualization technologies have emerged as core enabling technologies to address this challenge. Targeting complex scenarios characterized by dynamic access, high mobility, and strong access uncertainty of computing power devices, this paper proposes a computing resource discovery and pooling mechanism based on the collaboration of near-field sensing and semantic-driven strategies. In the phase of computing power device access, the spherical wave characteristics of near-field propagation are exploited to endow the receiving array with high-precision spatial resolution capability. The device detection task in the scenario of random access of multiple devices is modeled as an energy-aware task with distance and orientation awareness, thereby achieving highly reliable spatial sensing of unknown mobile computing nodes. In the phase of computing power pooling management, a unified semantic model of computing resources is constructed to eliminate discrepancies in resource description among heterogeneous devices, and a fast resource indexing architecture incorporating spatial information is established. Moreover, a two-level pooling organization scheme consisting of spatial micro-pools and semantic capability hierarchical pools is adopted to realize low-latency matching and flexible orchestration of computing resources. Experimental results demonstrate that the proposed near-field sensing method significantly outperforms conventional approaches in detection performance under multi-user random access and complex propagation environments. Meanwhile, the semantic-driven resource management mechanism exhibits superior performance in terms of resource matching accuracy, recall rate, and response latency. The proposed scheme effectively adapts to the dynamic edge computing environment, providing technical support for the large-scale development of edge deployment of large models and dynamic edge computing power pooling.关键词：near-field sensing;computing resource discovery;random access detection;semantic modeling;dynamic computing power pooling;edge computing power network9|2|0更新时间：2026-06-11
- 摘要：To address the collaborative requirements of high-security transmission and high quality of service (QoS) in dispersed computing network under harsh wireless communication environments, this paper proposes a secure communication and resource optimization scheme for intelligent reconfigurable surface (IRS)-assisted dispersed computing network. Firstly, due to the energy limitations of unmanned aerial vehicle (UAV) nodes, this paper studies a novel energy harvesting (EH) scheme. By dividing the IRS passive reflection array in geometric space, some reflection elements are used for information reflection, and some elements are used for EH, so as to realize the cooperation of information transmission and EH. Secondly, the secure sum rate maximization optimization model in IRS-assisted dispersed computing network is formulated. The model jointly optimizes multiple coupling variables such as user transmission power, IRS reflection element phase shift, EH constraint and communication QoS, while improve the overall system security performance and resource utilization efficiency. Since the formulated optimization problem is highly non-convex and the variables are strongly coupled, traditional optimization methods are difficult to directly obtain the global optimal solution. Furthermore, considering the characteristics of dispersed computing network, such as high user mobility, rapidly varying wireless channels, and uncertain environmental states, a robust deep reinforcement learning (DRL)-based dynamic resource optimization algorithm is designed to guarantee QoS in dynamic dispersed computing environments. Simulation results show that the performance of the IRS-assisted dispersed computing network scheme based on robust DRL proposed in this paper not only outperforms existing learning-based solutions but also achieves performance close to that of the exhaustive search method, verifying the effectiveness and superiority of the proposed scheme.关键词：dispersed computing network;intelligent reflecting surface (IRS);energy harvesting (EH);secure sum rate;robust deep reinforcement learning (DRL)40|4|0更新时间：2026-06-09
- 摘要：Physical reservoir computing (PRC), characterized by its low training cost and high parallel efficiency, has emerged as a promising paradigm for temporal signal processing and complex dynamical system modeling. By leveraging the intrinsic nonlinearity and short-term memory dynamics of physical devices, PRC maps temporal inputs into a high-dimensional state space, generating rich reservoir states that enable accurate solving and prediction of complex state equations. With advantages such as readout-layer-only training, low inference power consumption, and few-shot data driven, PRC offers a novel pathway for real-time solving of chaotic differential and partial differential equations on edge hardware. However, effectively co-designing high-order temporal processing algorithms with the intrinsic modulation mechanisms of physical devices remains a key bottleneck hindering practical deployment. Here, we report a highly efficient PRC system based on indium gallium zinc oxide (IGZO) electrolyte transistors, achieving the solving and prediction of time-dependent chaotic differential equations and partial differential equations. Leveraging the low-power characteristics and continuous nonlinear mapping capability of IGZO devices, we designed a lightweight algorithm co-optimizing reservoir training, inference, and equation solving. The electrical response and ion migration dynamics of the IGZO transistors were accurately modeled and characterized, followed by system-level validation on the field programmable gate array (FPGA). Resource evaluation demonstrates that, compared to a conventional fourth-order Runge-Kutta (RK4) implementation, our approach reduces digital signal processor (DSP) and flip-flop utilization by approximately 64% and 50%, respectively, while maintaining equivalent accuracy. In performance benchmarks using the Mackey-Glass equation, Lorenz chaotic sequence, and one-dimensional heat diffusion equation, our system achieves an inference latency of 11 μs and reduces power consumption by 15%, with a normalized root mean square error (NRMSE≈0.04) comparable to the Runge-Kutta method. For the partial differential equation (PDE) task, computation time is reduced to 1/400 of that required by the second-order Runge-Kutta (RK2) method, while maintaining a comparable NRMSE≈0.008. This work not only validates the efficiency and accuracy of IGZO-based PRC in solving general differential equations, but also provides a scalable architectural foundation and practical engineering reference for next-generation lightweight, low-power edge scientific computing and neuromorphic hardware platforms.关键词：indium gallium zinc oxide (IGZO) electrolyte transistor;artificial synaptic device;reservoir computing;chaotic differential equation solving;partial differential equation solving13|1|0更新时间：2026-06-09
- 摘要：With the rapid development of deep learning technology, convolutional neural networks (CNNs) have demonstrated exceptional performance in image recognition and processing tasks. However, as the network depth increases, the massive transmission of intermediate data imposes tremendous pressure on the on-chip memory and memory access bandwidth of hardware accelerators. The increasingly prominent “memory wall” problem has severely constrained the overall throughput and energy efficiency of the system.To address this issue, existing inter-layer feature data compression methods are mainly divided into two categories. The first category focuses on lightweight hardware implementation: despite low area overhead, their compression ratio is limited by algorithm complexity, making it difficult to effectively alleviate the off-chip bandwidth pressure in high-throughput scenarios. The second category pursues superior compression performance, but incurs excessive hardware area overhead, which is hard to deploy on resource-constrained edge devices.Aiming at the above challenges, this paper proposes a statistic-aware hybrid compression method for CNN inter-layer feature maps, with the core design goal of achieving high compression ratio and low hardware overhead to resolve the difficulty in balancing compression performance and resource consumption. By deeply exploiting the sparsity and distribution characteristics of the data, this method realizes hardware-friendly data coding combined with a hardware-software co-design mechanism of “offline analysis-online compression”. In the offline analysis stage, statistical analysis is performed on the CNN inter-layer feature data to generate the required coding tables and baseline values. In the online compression stage, the feature data are classified into zero-value data and non-zero-value data. For zero-value data, an enhanced zero run-length encoding combined with entropy coding is adopted; for non-zero data, dynamic baseline-delta encoding is applied. This differentiated coding mechanism reduces the hardware area overhead by 58.7% to 72.9% while maintaining a high compression ratio, which solves the problem of high hardware complexity in traditional compression algorithms.We conduct a systematic evaluation of the compression performance of the proposed method under different network structures and data formats, based on compression experiments on inter-layer feature maps of four representative CNNs: AlexNet, VGG16, ResNet34, and MobileNetV2. Experimental results show that, compared with similar studies, the proposed data compression method achieves a maximum improvement of 58.5% in compression ratio under the INT8 quantization format, and a maximum improvement of 36.7% under FP32/FP16 formats. When deploying the VGG16 model on the ALINX AXU5EV target platform, the accelerator based on the proposed data compression method reaches an inference throughput of 242.8 GOPS. Compared with the compression-free baseline architecture, the computing performance and energy efficiency are improved by 41.4% and 27.8%, respectively.The experimental results demonstrate that the proposed method balances the compression ratio and hardware overhead for CNN inter-layer feature map compression, and provides a new solution for the design of CNN accelerators in resource-constrained edge scenarios.关键词：field programmable gate array;convolutional neural network accelerator;hardware accelerator;interlayer feature map;data compression18|5|0更新时间：2026-06-09
- 摘要：Passive tracking of uncooperative maneuvering radiation sources by a single unmanned aerial vehicle (UAV) using only received signal strength indicator (RSSI) measurements is of considerable strategic and engineering value for missions such as electronic reconnaissance, spectrum monitoring, and emergency search and rescue. However, existing RSSI localization methods rely heavily on environmental propagation priors and treat parameter estimation, state filtering, and path planning as isolated serial processes, leaving them unable to handle dynamic scenarios in which core propagation parameters such as transmit power and path loss exponent are unknown and the target keeps maneuvering. The inherent low spatial resolution of scalar RSSI from a single channel further intensifies the structural conflict between exploration and exploitation. To address these challenges, this paper proposes an embodied intelligence-based closed-loop tracking method for a single UAV, which adopts structured uncertainty metrics as a unified carrier of cross-layer information flow and deeply couples perception, cognition, and decision-making through bidirectional propagation of uncertainty, forming a cooperative tracking architecture driven by closed-loop information flow. At the perception layer, a parameter prior inference network is built to extract the empirical distribution of propagation parameters from historical spatio-temporal observation sequences, providing data-driven initialization for the subsequent Bayesian inference. At the cognition layer, a dual-chain decoupled inference mechanism based on Rao-Blackwellized conditional decomposition is designed for the joint estimation of unknown propagation parameters and maneuvering states, in which the outer particle filter estimates large-scale fading parameters online while the inner adaptive interacting multiple model filter accurately infers maneuvering states of the target. A particle revival mechanism guided by effective sample size monitoring is further introduced together with a belief inheritance mechanism to prevent parameter resampling from disrupting the continuity of state tracking. At the decision layer, to mitigate the exploration-exploitation conflict noted above, a hierarchical planning strategy fusing information gain with model predictive control is constructed, and a learnable ranking gating network is introduced to perform real-time utility evaluation over candidate strategies including tracking, probing, and conservative options, allowing the UAV to adaptively transition from broad-area exploration to tight escort flying according to belief uncertainty. Simulation results show that the proposed method outperforms the compared baseline approaches across all core metrics under cold-start conditions with completely unknown propagation parameters, and the tracking performance under the mode without manual calibration priors approaches and in certain metrics surpasses the ideal performance of several baselines operating with known parameters, which verifies the adaptability and robustness of the proposed architecture under degraded sensing conditions.关键词：embodied intelligence;target navigation;RSSI;active path planning;information-driven decision18|1|0更新时间：2026-06-09
- 摘要：In dense urban environments, blockage by high-rise buildings, concurrent radiation from multiple wireless systems, and the superposition of multi-source interference make the electromagnetic environment highly time-varying and spatially heterogeneous. Traditional radio maps, which are constructed offline and updated at a low frequency, cannot reflect the real channel state in time. As a result, they are unable to meet the requirements of dynamic spectrum management, wireless resource scheduling, and low-altitude communication support in terms of both accuracy and timeliness. This challenge is more pronounced in complex mission scenarios, where online sensing and rapid decision-making are especially important. To address this issue, this paper proposes an uncertainty-driven active sensing method for UAV-based spectrum mapping. First, the online radio map updating and path planning processes are jointly modeled as a Markov decision process. The state explicitly includes statistical measures of map uncertainty, UAV position, flight time, target distance, and mission progress, so as to improve the policy's ability to represent environmental changes and mission stages. Second, a pre-trained U-Net is employed to reconstruct the SINR radio map online and to output pixel-level uncertainty estimates. The reduction in uncertainty is used to characterize information gain, which guides the UAV to sense high-value unknown regions with priority. Furthermore, a FiLM-D3QN decision network based on Feature-wise Linear Modulation is designed. It uses uncertainty and mission-progress conditions to dynamically modulate intermediate features and value estimation, thereby achieving an adaptive balance between information acquisition and trajectory efficiency. Simulation results show that, in urban scenarios with dense interference, the proposed method effectively reduces map reconstruction error while maintaining an arrival rate above 90%. The RMSE is reduced by about 7.1% compared with the classical IPP method and by about 9.8% compared with the original D3QN. The proposed method also achieves a shorter average episode length. These results verify its effectiveness for online radio map updating and active sensing in complex urban electromagnetic environments.关键词：urban electromagnetic environment;unmanned aerial vehicle(UAV);radio map;active sensing;uncertainty estimation;deep reinforcement learning22|4|0更新时间：2026-06-08
- 摘要：Air-sea cross-medium communication is a crucial technology for enabling high-speed air-sea cross-domain interconnection. Owing to its excellent oceanic penetration characteristics, the blue-green light band (450 nm~550 nm) has emerged as an ideal transmission window for underwater wireless optical communication links, providing vital support for high-speed cross-medium communication. However, subjected to the nonlinear coupling effects of the atmosphere, random sea surface waves, and seawater scattering and absorption, the cross-medium channel exhibits significant non-stationarity and strong time-varying characteristics. In such rapidly time-varying channel environments, traditional fixed coding and modulation strategies, which rely on quasi-static channel assumptions, suffer from degraded system error performance due to mismatches with the actual channel conditions. To address this issue, this paper proposes a transmission scheme based on dynamic channel feature learning for air-sea cross-medium communication. By constructing a closed-loop feedback mechanism of "periodic probing—dynamic learning—reconstruction," the coding structure can be dynamically optimized in response to varying sea states. Specifically, the transmitter periodically transmits pre-designed pilot sequences. The receiver utilizes these pilot signals for channel estimation, compiles real-time statistics on the instantaneous characteristics of the current channel, and subsequently generates a polar code construction scheme that optimally matches the prevailing sea conditions. This construction information is then fed back to the transmitter via a feedback link. Based on this information, the transmitter executes code reconstruction and applies adaptive polar coding to the information sequence to be transmitted. This closed-loop feedback process operates periodically, ensuring that the code reconstruction closely tracks the non-stationary variations of the cross-medium time-varying channel, thereby overcoming the mismatch between fixed coding strategies and the actual physical channel. To verify the performance of the proposed system, a cross-medium communication channel model encompassing the atmosphere, the air-sea interface, and the seawater medium was established. Three typical water types—Jerlov IB, Jerlov II, and Jerlov Ⅲ—were selected to simulate oceanic environments with weak to strong absorption and scattering effects, respectively. Simulation results demonstrate that the proposed dynamic channel feature learning strategy can effectively adapt to variations in the seawater medium. Compared to uncoded systems, applying polar coding under the same Jerlov water type yields a significant coding gain, with the bit error rate (BER) markedly reduced. Furthermore, compared to the fixed polar code construction strategy adopted in the 3GPP-5G standard, the proposed dynamic channel feature learning strategy achieves substantial performance improvements across all three Jerlov water conditions. Notably, the performance gain increases as water turbidity rises and channel conditions deteriorate. These results validate that adaptive coded transmission based on dynamic channel feature learning can perceive and adapt to the dynamic changes of the seawater medium, providing a solid theoretical foundation and technical support for the development of highly reliable air-sea cross-medium adaptive communication systems.关键词：air-sea cross-medium communication;polar codes;channel characteristic dynamic learning;channel estimation;closed-loop feedback;adaptive coding11|0|0更新时间：2026-06-08
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0