Missing fact prediction in temporal multimodal knowledge graphs (TMKGs) remains challenging because temporal dynamics are often relation-specific, and multimodal evidence may contain cross-modal noise. We propose MDRTCF, a modality-decoupled framework with lightweight relation-aware temporal context fusion for interpolation-oriented TMKG reasoning. MDRTCF combines two complementary designs: modality-decoupled time-aware modeling, which separately encodes structural, textual, and visual modalities through temporal modulation, temporal translation, and global temporal bias to suppress cross-modal interference and generate robust time-aware embeddings; and relation-aware temporal context fusion, which adaptively aggregates past, present, and future temporal signals to refine relation-specific temporal dependency modeling and improve the interpretability. We construct five TMKG benchmark datasets, including TM_ICEWS14, TM_ICEWS05–15, and three TM_Wikipedia variants, comprising 775,238 temporal facts, 64,155 images, and 64,731 textual descriptions in total. Extensive experiments show that MDRTCF achieves state-of-the-art performance across all five datasets, with an average relative MRR improvement of 5.39% over FTPComplEx. On TM_ICEWS05–15, MDRTCF improves MRR by 6.25%. In addition, MDRTCF can be plugged into five tensor factorization-based TKGE backbones, yielding an average relative MRR gain of 3.65% and demonstrating its effectiveness, scalability, and generality for TMKG reasoning.

Spatiotemporal knowledge graph reasoning (STKGR) is challenging due to the complex temporal and spatial dependencies in real-world data. Existing approaches are often limited by the narrow coverage of symbolic rule-based methods and the lack of interpretability in embedding-based models. Although large language models (LLMs) have demonstrated strong capabilities in knowledge understanding and temporal reasoning, they are generally opaque and difficult to interpret. To address these limitations, we propose STEL-LLM, a Spatio-Temporal Evolving Logic framework augmented with LLM-based reasoning. STEL-LLM introduces an STLogic module that integrates bidirectional temporal rule sampling and entity activity-range learning to extract interpretable spatiotemporal logical rules, and is capable of automatically generating spatiotemporal predicates to describe the spatiotemporal relations embedded in the rules. These rules are iteratively refined through a three-stage Chain-of-Thought (CoT) prompting pipeline that generates, validates, and enhances rule quality. The refined rules are subsequently combined with graph embeddings, forming a closed-loop evolutionary system that unifies symbolic, embedding-based, and LLM-driven textual reasoning. Experiments on four spatiotemporal datasets derived from ICEWS and GDELT demonstrate that STEL-LLM achieves an average improvement of 4%–5% in MRR and 5%–7% in Hits@1 over state-of-the-art baselines, with a notable 9% Hits@1 gain on the GDELT dataset. These results demonstrate that STEL-LLM consistently outperforms state-of-the-art methods, offering significant gains in accuracy, interpretability, and adaptability.

Weak magnetic video recording with warm atomic ensembles constitutes a non-cryogenic and non-contact methodology for the magnetic source identification and failure reproduction. However, the spatial resolution, imaging speed, and shooting mode from traditional optical-pumping systems have constrained the real recording for ever-changing magnetic phenomena. This work reports a 684-pixel Bell-Bloom atomic magnetic-videorecorder with the global shutter and two-dimensional differential readout, for the real recording of changing gradient fields, which implements the free Larmor precession of Cs atoms to infer local magnetic information, employs a high-speed dual-quadrant Complementary Metal Oxide Semiconductor (CMOS) sensor with the global shutter and the extra microlens focusing to simultaneously detect differential optical rotations on all pixels. Also, a digital micro-mirror device (DMD) is employed to weigh the pixel crosstalk and the spatial resolution, and to facilitate the one-to-one pairing of the profiles for each differential probe beam pair projected onto the two CMOS quadrants. Furthermore, the average sensitivity is demonstrated to be 194 pT/ Hz @0.5-178 Hz, with a high spatial resolution of 137 μm2 and a frame rate of 205 fps in a field of view up to 5 × 2.6 mm2. Finally, the magnetic distributions from a moving source have been experimentally measured and found to be in good agreement with the simulation results.

Fiber-coupled ensembles of nanodiamonds containing nitrogen-vacancy (NV) centers are widely used for microscopic magnetic field measurements. However, the random distribution of NV crystal axes in nanodiamond ensembles causes resonance frequencies from differently oriented NV centers to overlap in the ODMR spectrum, distorting the spectral line shape and breaking the linear mapping between frequency shift and magnetic field. As a result, conventional magnetic field measurements based on Lorentzian fitting suffer from significant systematic errors. Here, we establish a fiber-nanodiamond ensemble model to analyze the impact of random NV orientations on the ODMR linear response and define an effective ensemble-averaged linear response coefficient K. Fisher information analysis identifies an information-optimal intrinsic magnetic field of approximately 640 µT, at which the ensemble response can be reliably approximated as linear in its vicinity. Building on this condition, we propose a model-assisted regression scheme, in which the physical fiber-nanodiamond ensemble model defines the optimal operating condition and the associated linear frequency-field scaling, while intrinsic-field-guided Gaussian regression (IFGGR) extracts magnetic-field-induced frequency shifts from distorted ensemble ODMR spectra, enabling accurate magnetic field estimation via the response coefficient K. Under identical acquisition conditions, the proposed method reduces the RMSE from 37.21 µT to 8.99 µT and significantly suppresses systematic bias, while achieving an absolute sensitivity approaching the shot-noise limit at optimal averaging time. The relative slope error reduction (≈15.6-fold) further confirms improved local linearity near the optimal operating point.

Nitrogen-vacancy-center (NVC) vector magnetometers have emerged as one of the most promising quantum magnetic-sensing technologies owing to their high sensitivity and wide dynamic range. However, most reported integrated or portable NVC magnetometers are scalar systems, whereas vector implementations often remain optical-table-based and suffer from resonance-frequency drift on open-loop operation. This article presents a portable closed-loop NVC vector magnetometer using four-channel proportional–integral (PI) frequency tracking. The system consists of a miniaturized sensing probe (163 × 55 × 62 mm³) and a control-electronics unit (440 × 274 × 177 mm³). Four independent frequency-locking loops track the optically detected magnetic resonance (ODMR) frequencies of four NVC axes, allowing real-time three-dimensional magnetic-field vector reconstruction. Experimental results show that the closed-loop linear dynamic range reaches approximately ±200 μT for all three Cartesian axes. The measured magnetic-field sensitivities, evaluated at 10 Hz, are 0.897, 0.783, and 0.809 nT/Hz1/2 for the X, Y, and Z axes, respectively. Furthermore, stable three-dimensional vector reconstruction is achieved through four-channel closed-loop frequency tracking, and the closed-loop architecture enables robust resonance-frequency tracking over a wide range of magnetic-field conditions. These results confirm that the proposed portable closed-loop NVC vector magnetometer can achieve stable vector-field reconstruction and has potential for applications in magnetic-field monitoring and geomagnetic-anomaly detection.

Reliable underwater tactile sensing technology can promote the effective exploitation of marine resources, which is of significant importance for sustainable human development. Such technology typically requires a combination of waterproofing, intelligence, and high efficiency. Herein, we leverage the in-sensor computing (ISC) neuromorphic device architecture as an innovative platform to construct a diamond nitrogen-vacancy (NV) center-based wireless magnetic tactile sensor (ISC-NVTS). This sensor achieves highly linear force-to-magnetic signal conversion through an array of NdFeB magnetized flexible films and accomplishes ultra-fast magnetic signal detection using diamond NV center magnetic sensing unit arrays operating in a fixed-frequency mode. This wireless connection approach effectively solves the waterproofing issue for tactile electronic devices. Furthermore, we enable adjustable pressure responsivity of the sensing unit array through a microwave multi-parameter NV center electron spin resonance control method, endowing it with the advanced intelligence to execute ISC architecture-based artificial neural network algorithms. Finally, the ISC-NVTS achieved a recognition rate of 95.1% (Random noise 0.1, offline simulation) in a classification task involving five types of marine debris and organisms, with a recognition time of only 6.45 ms. We anticipate that this work will advance the further development of tactile sensors and provide support for the sustainable exploitation and utilization of marine resources.

Fault prediction offers an efficient means to minimize losses resulting from equipment failure. Among existing prediction models, deep learning-based methods lack semantic information when analyzing time-series data of equipment condition, while traditional knowledge-based models have significant prediction errors and overlook the application of operational data. To address these issues, this paper proposes a reasoning framework (named TA-ComplEx) based on knowledge graphs (KGs) for machinery fault prediction. Particularly, different from existing KG-based methods for dealing with equipment fault, the paper explicitly incorporates temporal dynamics of equipment condition in the reasoning framework. Specifically, the paper first constructs static and temporal knowledge graphs to encode semantic relationships and temporal dynamics from equipment time-series data. Then, attribute embedding (via LSTM (Long Short-Term Memory)-based character encoding) and time-aware embeddings (via ComplEx) are integrated to learn the vector representations of the constructed knowledge graphs. Finally, the paper uses attention-based LSTM with Extra-Tree classifiers for failure mode prediction. Experiments on the XJTU-SY bearing dataset report 100 % F1-scores in fault classification/prediction, and knowledge reasoning infers root causes/maintenance measures.

Two-dimensional imaging of microwave electric field gradients was performed using a MEMS cesium vapor cell based on Rydberg atoms. The variations in probe laser fluorescence intensity, continuously monitored via high-sensitivity CCD imaging at 24.038 GHz, enabled detailed analysis of electromagnetically induced transparency (EIT) and Autler-Townes (AT) splitting effects. By establishing the quantitative relation between the splitting interval (Δf) and the electric field intensity, spatial electric field gradient distributions were accurately measured. The experimental results demonstrated a minimum detectable electric field strength of 1.22 V/m and a spatial resolution of 151.2 µm, with the measurement error in electric field strength maintained below 2%.

Accurate phase measurement of microwave signals constitutes a critical prerequisite for enhancing communication network capacity and improving radar target recognition capabilities. Here, a microwave phase measurement method based on the nitrogen-vacancy (NV) centers in diamond was proposed. Based on the principle of microwave self-coherent reference, a solution model for microwave phase measurement was established. By interrogating the fluorescence intensity of spin resonance peaks under varying phase conditions, we identified a high-sensitivity phase measurement regime. Combined with modulation and demodulation technology, the phase measurement sensitivity was enhanced from 0.0217 mV/degrees (without modulation) to 46.4163 mV/degrees (after applying lock-in amplification technology), representing a 2139-fold improvement. Correspondingly, the shot-noise-limited sensitivity improved from 2.7844 degrees/Hz1/2 (without modulation) to 0.0065 degrees/Hz1/2 (after applying lock-in amplification technology)-a 428-fold enhancement. These results demonstrated that based on the NV center ensembles self-coherent microwave phase measurement enables high precision and high sensitivity microwave phase detection. This methodology offers an important reference scheme for defense-critical and communication applications requiring high fidelity microwave signal phase characterization.

Wearable technology is rapidly moving towards intelligent, flexible, and multifunctional integration. Its core challenge lies in achieving a balance between high sensitivity, autonomous power supply, data processing, and comfortable wear under conditions of limited volume, complex deformation, and dynamic interaction. Two-dimensional (2D) materials, with their atomic-level thickness, exceptional flexibility, rich surface chemistry, and tunable electrical and optical properties, provide a revolutionary material platform for building next-generation intelligent wearable systems. This paper first describes the characteristics of 2D materials (such as graphene, transition metal dichalcogenides (TMDs), MXenes, and black phosphorus), then introduces the research progress in wearable intelligent systems, including design strategies and technical paths for 2D materials in intelligent sensing interfaces, drive structures and human-computer interaction, neuromorphic computing and flexible circuits, energy and thermal management systems, and system integration. Next, it analyzes the challenges that 2D materials will face in the future, and finally provides a systematic summary of the current application of 2D materials in wearable technology, and looks forward to future development directions.

Conventional radar antennas are constrained by phase noise and physical aperture size, which limits further improvement in velocity detection resolution. To address these challenges, this study demonstrates a high-precision velocity detection method based on Rydberg atoms. By constructing a quantum precision measurement system utilizing Rydberg atoms as microwave mixers, high-accuracy frequency measurement of automotive radar-band microwave signals is achieved. Through optimization of signal acquisition time, the frequency resolution is enhanced to the millihertz (mHz) level. Subsequently, based on a frequency-velocity conversion model, precise detection within a velocity range from static conditions to 120 cm/s is realized, achieving a resolution of 62.4 mu m/s. Furthermore, by measuring the electric field sensitivity of the system (516 nV & sdot;cm-1 & sdot;Hz-1/2), the maximum detectable distance is theoretically calculated to be approximately 2000 m, which far exceeds the detection range requirements for long-range radar in the automotive radar industry. Finally, leveraging the atomic system's bandwidth, the upper limit of measurable velocity can be extended to 310 m/s. This research establishes a critical technical foundation for high-precision velocity detection based on atomic quantum systems, with broad application prospects and significant socio-economic value in fields such as inertial navigation, precision manufacturing, and aerospace.

The characterization of microwave field holds indispensable engineering value for miniaturized integrated circuit design and defect detection. This study presents high-fidelity microwave field imaging through concentration compensation of nitrogen-vacancy (NV-) centers in diamond. By establishing an NV-fluorescence efficiency model and developing scanning technology utilizing complementary metal-oxide-semiconductor (CMOS) camera, we achieved quantitative concentration gradient imaging and reconstructed microwave fields over interdigital electrode surfaces within a 0.5 x 0.5 mm2 area. Experimental results confirm that we achieve 98.9 % effective concentration homogeneity and 92.76 % microwave field imaging fidelity. This work provides an essential reference for chip-scale precision characterization of microwave fields.

With the continuous progress of artificial intelligence, the application of large language models (LLMs) has provided completely new possibilities for knowledge graph reasoning and semantic understanding. The proposed method focuses on solving the multi-hop link prediction problem in spatiotemporal knowledge graphs through reasoning. For this purpose, an innovative framework Spatiotemporal Graph Multi-hop Reasoning Based on Large Language Model (STMH-LLM) is proposed. This method transforms the structured knowledge graph data into natural language descriptions and utilizes these prompts to fine-tuning LLMs, thereby enhancing the performance of multi-step relational inference. The proposed framework aims to capture the latent representations of entities and their relationship networks through natural language prompts. To verify the effectiveness of STMH-LLM, the commonly used open-source large language models Qwen2.5 and GLM4 were selected for fine-tuning tests. In addition, this framework also demonstrates the potential to provide zero-shot reasoning capabilities for LLMs, enabling them to handle previously unseen prompts. Experiments were conducted on three datasets, and the results show that, compared with traditional models, STMH-LLM significantly improves the generalization ability of the model and achieves more accurate predictions in unfamiliar scenarios.

Rapid non-invasive 3D tracking and velocimetry of picotesla-level droplet flows in optically inaccessible environments remains a major challenge for high-throughput biosensing, drug delivery and industrial process monitoring. This work proposes a volumetric quantum tracking architecture based on a warm Rb-Cs hybrid ensemble with π/2 radio-frequency (RF) pulse modulation. It achieves uniform 3D magnetic-voxel sampling through Rb-to-Cs indirect pumping and tomographic scanning. Robust inverse localization and velocimetry are further realized by an improved heuristic algorithm, while neither acoustical vibrations nor extra electrical signals are introduced to the measured droplets. Firstly, a 795-nm linear-shaped pump laser tomographically and indirectly polarizes Cs atomic layers, avoiding polarization non-uniformity and fictitious magnetic fields. Then, after a π/2 RF pulse, the resulting free-induction-decay Faraday-rotation signals from a probe beam array are detected via 2D differential measurement. The beam array transmits through sequential Cs atomic layers, enabling the reconstruction of common-mode-noise-free 3D magnetic field distribution outside the flow channel. Finally, based on the dynamic 3D field data, a teaching–learning-based optimization algorithm with adaptive parameters and elitist preservation is adopted to estimate the optimal droplet positions and velocities on all frames. This system achieves a 24-fps 3D-tracking frame rate, a sensitivity of 18.3pT/Hz@1–200 Hz, a spatial resolution of 500 × 857 × 1000 µm3, and average velocity errors of 5.54% and 8.84% at 4.2 mm/s and 8.4 mm/s, respectively. This work establishes a high-bandwidth, electroacoustic-safe sensing strategy for optically-blind magnetic sources and provides a scalable pathway toward volumetric tracking in microfluidics, biomedical sensing, and concealed industrial environments.

The characterization of microwave fields holds significant engineering importance in high-frequency chip design and performance optimization. In this study, we propose a noninvasive wide-field imaging method based on the spectral hole-burning effect in optically detected magnetic resonance (ODMR). This wide-field imaging method enables imaging of microwave fields at different frequencies under near-field radiator operating conditions. Its feasibility is confirmed by characterizing near-field radiator-generated microwave fields across multiple frequency and power settings. Furthermore, the effectiveness of the method is validated through a comparative analysis of simulation results and experimental measurements. Under the current experimental conditions, the results show that the approach achieves a frequency resolution of 100 kHz over the range of the resonance region, a microwave detection sensitivity of 0.78 nT/Hz (1/2 ), a spatial imaging resolution of 1.55 mu m, and the minimum detectable power is 1 mW. The above methods indicate great potential for application in the fields of integrated circuit fault diagnosis, electromagnetic compatibility testing, and radio equipment parameter calibration.

Flexible strain sensors have attracted considerable attention in gait recognition owing to their ability to adhere directly to the skin near joints and transduce local deformation. In existing work, however, sensor placement and orientation are largely determined by anatomical experience, while multi-channel classification still relies on back-end digital processors, whose power consumption and latency constrain system practicality in wearable scenarios. This paper presents an integrated design path that proceeds from skin-mechanics theory through sensor-layout optimization to analog-domain front-end inference. On the layout side, the lines-of-non-extension (LoNE) theory is employed to convert the selection of sensor attachment angles from empirical judgment into a calculable mechanics problem; guided by the spatial course of LoNE in the ankle and knee regions, the positions and angles of the nine sensors are determined individually—channels perpendicular to the LoNE capture maximum strain, channels offset by 45 degrees supplement non-sagittal-plane information, and a channel aligned along the LoNE provides a near-zero-strain reference. On the circuit side, the mathematical equivalence between the weighted summation of a linear classifier and Kirchhoff’s current law (KCL) nodal current superposition is exploited to map the classification operation onto current aggregation in an analog circuit, yielding an in-sensor computing (ISC) front end in which the nine-channel weighted summation is completed in a single analog step. The sensors are fabricated by screen-printing a liquid-metal–polymer composite conductive ink onto a TPU film substrate, with a gauge factor RSD of 6.8% and a tensile linearity R2>0.99. Using walking, running, and stair descent as verification targets, the analog classifier reaches 99% accuracy at the circuit-level functional-verification stage. On real multi-subject data, it achieves 87.0%±8.4% accuracy under intra-subject cross-session validation, with an analog-domain inference response faster than 100μs. This design path is not bound to a specific joint or sensor material; when the layout methodology is extended to additional joint regions and the circuit architecture incorporates multiple outputs to cover more classification categories, the same workflow remains applicable, offering a promising low-power, lightweight technical solution for wearable motion monitoring.

Multi-parameter quantum sensing based on Nitrogen-Vacancy (NV) centers in diamond is often limited by microwave cross-interference and intermodulation distortion in multi-channel excitation schemes, which elevate noise levels and hinder accurate magneto-thermal decoupling. In this work, we propose a low-interference dual-port microwave synthesizer integrated into a compact NV hybrid quantum sensor with fiber-coupled optical excitation and efficient fluorescence collection. Two independently generated sinusoidally modulated microwave signals address distinct NV spin transitions, while continuous-wave optically detected magnetic resonance (CW-ODMR) combined with dual lock-in demodulation enables real-time separation of magnetic field and temperature responses. Compared with conventional power-combiner-based dual-frequency delivery, the proposed architecture suppresses third-order intermodulation distortion from 37 dBc to 56 dBc and improves the signal-tonoise ratio from 47.51 dB to 55.47 dB. Owing to independent microwave excitation and a highly uniform near-field distribution in the optical interrogation region, a magneto-thermal isolation of 42.63 dB is achieved with negligible crosstalk. Temperature-induced magnetic drift is effectively suppressed during long-term operation, with the temperature-drift coefficient reduced from 188.2 ppm/K to 3.3 ppm/K. The sensor achieves sensitivities of 866.48 pT/Hz1/2 for magnetic field sensing and 1.44 mK/Hz1/2 for temperature sensing. These results demonstrate that a low-interference dual-port microwave synthesizer combined with CW-ODMR readout provides an effective route toward compact and stable NV quantum hybrid sensing platforms.

ABSTRACT We propose a method for simultaneously measuring photon shot noise (PSN), spin projection noise (SPN), and technical noise in an atomic magnetometer using a spatial light modulator. A digital micromirror device (DMD) is placed behind the cell to modulate the area of the probe beam, thereby enabling the detected optical power and the number of polarized atoms to be varied while keeping the light intensity inside the cell constant—thus avoiding additional perturbations. Based on this experimental approach, we adapt the conventional models and fit the total equivalent magnetic‐field noise power spectral density as a function of the modulation area to obtain the key model parameters. Using the effective transverse relaxation time T 2 extracted from the free‐induction‐decay (FID) signal, the three noise components are then separated. The method is validated on a Bell‐Bloom magnetometer operated with a bias magnetic field of 2–3 µT. The results show that our model can simultaneously measure PSN, SPN, and technical noise with high accuracy, and outperforms conventional methods in terms of fitting robustness and goodness of fit. This work provides a practical approach for quantum‐noise diagnostics in operating atomic magnetometers without requiring ideal conditions such as weak perturbations or low technical noise.