Electronics doi: 10.3390/electronics15215170

Authors: Huimin Wang Xiaoyan Zhou Jun Hu Xuefeng Kong Jun Pan

DC–DC converters are widely used in data centers, aerospace systems, electric vehicles, and renewable energy systems, where their reliability directly affects overall system performance. Among their critical components, capacitors are particularly susceptible to degradation and failure, making accurate capacitor health assessment and remaining useful life (RUL) prediction essential for converter-level reliability. However, capacitor degradation exhibits nonlinear and condition-dependent behavior under coupled electrical and environmental stresses, posing significant challenges to accurate RUL prediction. This paper proposes an online capacitor RUL prediction method for DC–DC converters based on an adaptive physics-informed neural network (PINN). First, capacitor equivalent series resistance (ESR) is estimated online from the converter output voltage ripple and used as a capacitor degradation indicator. An adaptive PINN with learnable physical degradation parameters is then developed to address the limited adaptability of conventional models to nonlinear and varying degradation dynamics. Finally, a step-stress temperature–humidity accelerated degradation experiment was conducted on capacitors embedded in an industrial integrated product, and the extracted ESR trajectories were used for model training and RUL prediction. Experimental results showed that the proposed method substantially improved the accuracy of capacitor degradation-trajectory and RUL predictions. Across the three target-domain training-data cutoffs (800, 1000, and 1400h), the proposed PINN reduced the mean percentage error of degradation trajectory prediction by 44.9% compared with LSTM and by 38.7% compared with the particle filter (PF). For RUL prediction with 800h of target-domain training data, the proposed method achieved an RMSE of 99.8 h, compared with 116.5h for LSTM and 149.1h for PF, corresponding to reductions of 14.3% and 33.1%, respectively. These results demonstrate the effectiveness of the proposed method for online capacitor health prognostics and RUL prediction in DC–DC converters.

Electronics doi: 10.3390/electronics15215169

Authors: Itır Köymen

Memristors are commonly employed as emerging devices for in-memory computing. As the variety of memristive devices and materials grows, investigation of transport mechanisms and resulting memristive dynamics becomes more important. This work investigates electrical characterization and resistive switching of a memristive device. The device is observed to exhibit a reproducible piecewise power-law response with varying exponents across different regions of the switching cycle. Higher power-law exponents exhibit characteristics consistent with space-charge-limited conduction, whereas exponents closer to unity indicate weaker nonlinear behavior. The power-law parameters are quantitatively extracted from the measured I–V characteristics and then the derived expressions utilized to emulate memristive dynamics with a sub-threshold circuit, employing the inherent nonlinear current–voltage dynamics. The simple circuit models experimental behavior through the understanding of the memristive conduction mechanism. The characterization, modeling, analysis and emulation of memristive devices presented in this work can shed light on potential applications of memristive devices in analog CMOS circuits.

Electronics doi: 10.3390/electronics15215168

Authors: Jiakai Geng Chenyang Zhang Xufeng Zhou Linjing Li Daniel Zeng

Fine-grained Multimodal Named-Entity Recognition and Grounding (FMNERG) aims to extract entity–type–object triples from image–text pairs. However, existing methods still struggle with two major challenges: distinguishing semantically similar labels in an expanded fine-grained label space and achieving robust entity–object alignment under complex image–text relations in social media. To address these issues, we propose a Multi-level Knowledge Integration Framework (MKIF) for FMNERG, which leverages multi-level knowledge to jointly improve fine-grained entity-type disambiguation and robust entity grounding in a unified manner. Specifically, we employ multimodal large language models (MLLMs) to obtain global-level knowledge, enriching the semantics of raw multimodal inputs and serving as a shared semantic anchor across subtasks. To improve fine-grained entity-type discrimination, we further derive conceptual and discriminative descriptions for each type as label-level knowledge and utilize both global- and label-level knowledge as supervision via contrastive learning. In addition, we discover potential visual concepts in the dataset as object-level knowledge and design a dynamic gating mechanism to incorporate the global- and object-level knowledge into object embeddings for robust entity–object alignment. Comprehensive experiments demonstrate that MKIF outperforms state-of-the-art methods. Furthermore, detailed analyses validate the effectiveness of each core component.

Electronics doi: 10.3390/electronics15215165

Authors: You-Shyang Chen Mike Yau-Jung Lee Ying-Hsun Hung Ching-Hsue Cheng Cai-Yun Qiu

The Internet of Things (IoT) is widely used in modern business, industry, and daily life, but its rapid advancement also brings security challenges. Processing large amounts of IoT security discussion data in a short period is difficult, especially for users who lack relevant awareness or skills. To address this issue, this study proposes an advanced MPNet-CTM-VR method using an experimental dataset collected from IoT-related literature in the Web of Science (WoS) database between 2018 and 2023. The method extracts key information for subsequent researchers to further analyze IoT and IoT security issues. The proposed model has four main components with several stages and procedures. (1) In the research topic, a masked and permuted net (MPNet) model was employed for text mining during the feature extraction stage, addressing the challenges posed by word order and polysemy. (2) For the main topic model (TM), Latent Dirichlet Allocation (LDA) was used for comparison, followed by the selection of a correlated topic model (CTM) as the core tool for text feature extraction. Afterward, we used a consistency score value (Cv) to determine the optimal number of topics for the best TM. Five well-known machine learning models, adaptive boosting (AdaBoost), convolutional neural network (CNN), decision tree (DT), random forest (RF), and extreme gradient boosting (XGBoost), were subsequently used for further classification and evaluation. (3) For the experimental validation stage, LDA and CTM processed by the bidirectional encoder representations from transformers (BERT) were compared with LDA and CTM processed by MPNet, as proposed in this study. Experimental results show that, when XGBoost is used as the classifier, the CTM processed by MPNet achieves the best evaluation results in terms of accuracy, precision, F1 score, and AUC, with 0.9508, 0.9364, 0.9361, and 0.9816, respectively. (4) Finally, VOSviewer and word clouds were used to visualize the extracted results, highlighting trends in IoT security and helping interested groups better understand the importance and challenges associated with the IoT security issue. The research findings help identify IoT and IoT security technologies, enhance knowledge of information security protection, and provide guidance for interested groups. In conclusion, this study presents a best practice with effective empirical results to benefit this related IoT and IoT security issue.

Electronics doi: 10.3390/electronics15215167

Authors: Ying Li Ruishan Deng Qing Liu Fengping Wang Qiong Yang

The multiobjective flexible job shop scheduling problem (MOFJSP) seeks to jointly minimize makespan and maximum machine workload under coupled opera-tion-sequencing and machine-assignment decisions. Although the nondominated sorting genetic algorithm II (NSGA-II) is widely used for the MOFJSP, its performance can be constrained by imbalanced machine workloads, low-quality initial populations, and limited adaptation to evolving population states. This study develops a Hy-brid-Initialized Load-Aware Adaptive NSGA-II (HLA-NSGA-II) that integrates load-aware generation of machine-assignment vectors, dynamic bottleneck-aware hy-brid initialization, and multi-operator neighborhood search with closed-loop adaptive operator selection. These mechanisms respectively incorporate current workloads into machine assignment, prioritize jobs according to the evolving partial schedule, and update neighborhood-operator selection probabilities using relative hypervolume im-provement, objective-space diversity, and search stagnation. Experiments on the Ka-cem and Brandimarte benchmark sets show that the best solutions obtained by HLA-NSGA-II attain the known makespan lower bounds on all four Kacem instances. On the ten Brandimarte instances, HLA-NSGA-II achieves the lowest mean makespan on nine and the lowest or tied-lowest mean maximum machine workload on seven. Ablation studies further support the effectiveness of the proposed mechanisms. Over-all, HLA-NSGA-II achieved competitive performance on the tested instances, with workload outcomes varying across instances and higher computational cost than HSGA. This reflects the trade-off between solution quality and computational cost.

Electronics doi: 10.3390/electronics15215166

Authors: Kamel Saleh

This article presents a novel fault-tolerant control (FTC) scheme for a seven-phase PMSM drive to enhance its performance when two phases experience an open-circuit failure at the same time. For example, phases “a” and “b” are considered first-neighbor phases, “a” and “c” are second-neighbor phases, and “a” and “d” are third-neighbor phases. This approach makes the drive system more reliable and helps to reduce torque and speed fluctuations. To maintain the Magnetomotive Force (MMF) and improve overall performance, the FTC method focuses on controlling the current’s magnitude and angle after the faults. The process starts by calculating the angles and magnitudes after the fault occurs in each scenario. Based on this new current distribution, a specific multidimensional SVPWM technique is designed for each fault case. When simulating the fault in phases “a” and “b”, the torque ripple decreased from 0.4 N·m to 0.1 N·m, and the speed ripple dropped from 10 rpm to 4 rpm after applying the new FTC. For the fault in phases “a” and “c”, the torque ripple went down from 0.25 N·m to 0.1 N·m, and the speed ripple reduced from 2 rpm to 0.3 rpm with the new FTC. In the case of a fault in phases “a” and “d”, the torque ripple was cut from 0.6 N·m to 0.2 N·m, and the speed ripple dropped from 5 rpm to 0.5 rpm after implementing the new FTC. Furthermore, the dynamic performance of the motor during faults and during changes in speed and load was just as good as when the motor was operating normally.

Electronics doi: 10.3390/electronics15215164

Authors: Fernando Broncano Aurora Cuartero Alberto López-Trigo Jesús A. Torrecilla-Pinero Andrés Caro

Access to social network graphs is typically restricted due to data protection regulations, which prevent direct observation of their true structure. These graphs exhibit inherent inequality, measurable in sampled regions using various metrics such as the Gini coefficient. This study evaluates the effectiveness of snowball sampling followed by induced-subgraph densification in reproducing attention concentration. In subsets of crawled synthetic populations, the in-degree Gini coefficient is biased, with the direction of bias changing at a true value of 0.31. The share of the top percentile and the Gini coefficient for PageRank are distorted likewise. In synthetic populations containing up to 20 million nodes, induced-subgraph estimates are influenced primarily by local neighbourhood structure and the tail of the degree distribution, rather than by concentration. Comparisons to null models do not validate these estimates. Per-node counts obtained directly from crawled accounts closely approximate the true values, whereas no sampling method effectively corrects for the induced subgraph. This outcome aligns with predictions from binomial thinning of node degrees and demonstrates that uniform sampling performs the worst. On Bluesky, a census of 20.7 million complete profiles yields a Gini coefficient of 0.93. Snowball sampling of the crawled accounts produces an in-degree Gini coefficient of 0.63 and 0.88 for the counts. The crawl thus substantially underestimates the top-1% share by a factor of six.

Electronics doi: 10.3390/electronics15215163

Authors: Jin Hyo Jeon Kyeong Il Ko Meong Hun Lee

Environmental conditions may provide information relevant to respiratory deterioration in pig barns, but their value beyond recent acoustic observations remains unclear. This study developed a controlled semi-synthetic benchmark to examine when multimodal fusion of irregularly sampled environmental data and recent acoustic observations improves prediction of future respiratory-acoustic activity. Scores were derived from the aSwine dataset using a frozen PANNs representation and a recording-disjoint classifier, which achieved a macro-averaged held-out ROC AUC of 0.695 and average precision of 0.176. Empirical score distributions were incorporated into simulated 42-day sequences containing ammonia, temperature, relative humidity, and carbon dioxide. Eight models and controls were compared over 20 paired repetitions at coupling strengths 0, 2, 4, and 6. History-matched concatenation achieved mean ROC AUCs of 0.496, 0.518, 0.569, and 0.581, compared with 0.519, 0.494, 0.497, and 0.507 for audio-only prediction. Its advantage remained significant after Holm correction only at coupling 6. Time-aware fusion did not consistently outperform history-matched concatenation or its past-history-permuted control. Additional analyses showed that acoustic persistence, history length, generating assumptions, and alert definitions affected model rankings. Event-level evaluation using validation-selected thresholds showed no consistent operational advantage for fusion. These findings remain conditional on the empirical acoustic-score pools and the specified generator. Validation using synchronized real-farm audio, environmental, and veterinary data is needed.

Electronics doi: 10.3390/electronics15215162

Authors: Fahad Al-Otaibi

Uncertainty-aware wind and photovoltaic (PV) forecasting must account for grid-related operating constraints. A conditional diffusion framework generates joint 24 h renewable scenarios using separate German EMHIRES series (262,968 hourly values), chronological 70/15/15 partitioning, and a simplified IEEE 14-bus DC benchmark. A cosine noise schedule with a terminal cumulative alpha of 2.43 × 10−7 was adopted, and DDPM and physics-constrained denoisers were trained using three independent random seeds. Daily persistence, a hybrid wind/PV persistence baseline, a residual-bootstrap baseline, and final-only and reverse-step projection controls were evaluated. For seed 42, PC-Diffusion with physics-gradient guidance achieved an MAE of 0.07401, a CRPS of 0.05153, and 91.80% DC-benchmark feasible trajectories, compared with 0.07404, 0.05289, and 14.61% for DDPM. Hybrid wind/PV persistence achieved a lower MAE (0.06080), while the residual bootstrap obtained a lower CRPS (0.04881). Guided sampling required 22.29 ms per 24 h forecast window, compared with 2.76 ms for DDPM with final-only projection on the evaluated hardware. Raw guided prediction intervals remained undercovered (PICP90 = 71.81%). Validation-calibrated interval bounds achieved 95.25% empirical coverage without changing scenario feasibility, whereas dispersion-based scenario adjustment reduced feasibility. Physics-guided diffusion therefore improves scenario feasibility under training-derived DC constraints, although point-forecast accuracy, probabilistic calibration, and operational feasibility remain subject to important trade-offs.

Electronics doi: 10.3390/electronics15215161

Authors: Hua Wang Yue Wei Mengjie Wang Weidao Li

Optical satellite networks (OSNs) based on low-Earth-orbit (LEO) satellites and optical inter-satellite links (ISLs) are considered a promising component of future high-capacity space communications. However, the high mobility of LEO satellites leads to frequent changes in satellite–ground connectivity, which may cause interruptions in end-to-end optical paths and increase routing instability. To mitigate the impact of frequent satellite–ground handovers, in this paper we investigate routing selection problems for two categories of services with different temporal requirements, namely real-time and time–torrent services. First, a real-time and time–torrent service routing (RTSR) problem was established based on the relevant network and service models. Second, a slow-shot topology model was developed to represent the time-varying network topology through a sequence of snapshots. Third, two routing algorithms, namely slow-shot-based routing selection (SS-RS) and multi-delivery parallel SS-RS (MSS-RS), were proposed to accommodate different service requirements in the time and capacity dimensions, respectively. Extensive simulations were conducted to evaluate the proposed algorithms in terms of their service blocking probability, delivery interruption ratio, bandwidth utilization, and average hop count. The results illustrate the effectiveness of the algorithms, while exhibiting trade-offs regarding service blocking, resource utilization, and routing performance.

Electronics doi: 10.3390/electronics15215160

Authors: Somayeh Khamoushi Jeff Kilby Gonzalo Farias Lorenzo Garcia Krishnamachar Prasad

This review systematically surveys Sliding-Mode Control (SMC) strategies applied to knee and lower-limb exoskeletons with direct knee-related control or evaluation, emphasising research published over the last decade (2015–2025). The paper first outlines key control challenges arising from nonlinear joint dynamics, external disturbances, sensor feedback, and uncertainties in human–robot interaction. The review then analyses 126 unique studies, including peer-reviewed journal articles and full conference papers, covering classical SMC, higher-order formulations, observer-based architectures, intelligent adaptive schemes, and optimisation-enhanced approaches. Beyond tracking accuracy and chattering behaviour, the review considers practical aspects relevant to wearable rehabilitation systems, including sensing requirements, real-time implementation, patient-specific adaptation, and human–robot interaction. These aspects highlight current methodological trends, practical limitations, and the gap between control performance and real-world rehabilitation use. The review aims to support the development of more robust, safe, and practically deployable control frameworks for knee and lower-limb exoskeletons in rehabilitation applications.

Electronics doi: 10.3390/electronics15215159

Authors: Sangwoo Park

Large language models (LLMs) have achieved strong performance by scaling model size, but this progress has also made inference increasingly expensive. Mixture-of-Experts (MoE) and early exit are two representative approaches for reducing inference cost. However, early exit has been studied mainly for dense Transformers rather than MoE models. Existing early exit methods often base exit decisions on confidence signals derived from intermediate language-model (LM) head outputs, leading to repeated LM-head computation. We observe an opportunity to reduce this decision cost by exploiting MoE-native routing signals, which are already produced during MoE inference. In this work, we use these routing signals as informative cues for layer-wise early exit and propose a routing-aware double-gate method for MoE LLMs. Stage 1 uses MoE routing signals to screen candidate exit layers. Stage 2 invokes the LM head only for selected candidates as a final verification step. Across representative MoE backbones and six benchmark tasks, Double gate preserves accuracy while reducing executed layers, intermediate LM-head evaluations, and head invocation rate compared with early exit baselines. Measured end-to-end inference latency is reduced by 53.84%, 10.77%, and 34.24% relative to full-depth inference on Mixtral-8x7B, Qwen1.5-MoE, and DeepSeek-MoE-16B, respectively.

Electronics doi: 10.3390/electronics15215158

Authors: Minjong Cheon

Safety evaluations usually test refusal, abstention, and stated confidence separately. We treat them as one property, appropriate uncertainty expression, and audit it in text and vision across six small open-weight LLMs and five vision-language model (VLM) configurations. On a 240-item Korean benchmark, judged refusal is higher for harmful than for benign-looking requests in every model and under every labeling rule, but model-level orderings depend on the rule (primary rule: zephyr-7b-beta refuses harmful requests least, 42.5%, and benign ones most, 18.8%). Against 150 blinded labels from one annotator, the judges agree on refusal (κ=0.63) but overstate benign-prompt refusal (10.4% vs. 3.6%). Replacing sensitive words in benign prompts did not lower refusal (−1.9 pp [−5.4, 1.7]). Two VLMs abstain far more on out-of-distribution images, yet their stated confidence when they answer is equivalent within ±5 points to that on in-distribution images. On six MedMNIST v2 tasks, VLM accuracy is near chance (13.2–15.6% vs. 14.6% expected; ConvMixer reference on the same images: 67.8%) at ≈80% stated confidence, and stated evidence locations match CNN saliency only at chance level. Scaling from 2B to 8B shrinks the confidence gap without improving accuracy. At a fixed generation budget, only a step-by-step instruction to assess intent and risk raised benign-prompt refusal (+9.0 pp); generic step-by-step instructions did not. Neither scale nor chain-of-thought prompting resolves these failures.

Electronics doi: 10.3390/electronics15215157

Authors: Jinhuan Yu Xifeng Zheng Ran Zhan Yufeng Chen Yang Wang Hui Cao Yu Chen Xinyue Mao

RGB channel response differences and model prediction bias limit joint grayscale luminance and chromaticity calibration in passive-matrix (PM) Mini-LED displays. A method combining local response modeling and output residual correction is therefore proposed. A forward model constructed from single-channel and equal-RGB-code responses is used to optimize a baseline compensation mapping allowing separate RGB drive-code adjustments. At five baseline compensation operating points, residuals are calculated as the differences between simulated outputs and model predictions evaluated at the same RGB triplet. These residuals are interpolated along the gray-level axis to update luminance and chromaticity predictions before reoptimizing the mapping. Numerical evaluation based on measured base-response data shows that, over 250 gray levels excluded from residual sampling in the original scenario, joint correction reduces the mean full-scale-normalized absolute luminance error from 0.2885% to 0.1028% and the mean chromaticity deviation Δu′v′ from 0.002823 to 0.001126 relative to the baseline mapping. Compared with residual correction using existing equal-RGB-code responses, these metrics decrease by a further 20.1% and 8.1%, respectively. Both advantages persist across 30 numerical recalibration runs in the original scenario. However, in scenario S1 with stronger candidate-code-dependent deviations, joint correction produces a higher mean luminance error than the baseline, while retaining a lower mean chromaticity deviation. The results indicate that sparse residual feedback can improve calibration at unsampled gray levels when sampled residuals adequately represent prediction bias within the candidate domain. Independent measurements at non-equal RGB triplets are still needed to validate physical calibration performance.

Electronics doi: 10.3390/electronics15215155

Authors: Ghadeer Aljafen Suliman Aladhadh

The rapid development of generative artificial intelligence tools has contributed to the spread of misinformation and falsehoods, making it easy to fabricate lies, misinformation, and rumors across various sensitive fields such as health and politics. However, these tools are a double-edged sword; they can also be used to detect and limit the spread of misinformation and have become essential in the process of fact-checking. Therefore, in this paper, we review the evolution of the automated fact-checking pipeline in the context of these emerging technologies. Based on a structured literature search, we review and synthesize 81 studies across four major stages: claim detection, evidence collection, veracity prediction, and explanation generation. Moreover, we explicitly distinguish claim-level verification from fake news detection and propose two taxonomies: one organizing veracity prediction models by architectural family, from traditional machine learning to large language models, and one classifying fact-checking and fake-news datasets into four classes: general, multimodal, multilingual, and topic specific. This classification provides an extensive comparative analysis of their attributes, annotation approaches, and open-access sources, facilitating future benchmarking. Finally, we explore open pipeline challenges—including LLM hallucinations and operational constraints—and outline future research directions.

Electronics doi: 10.3390/electronics15215156

Authors: Huda Ali Alatawi

Tabular foundation models (TFMs) are pretrained once across many tables and applied to a new table by conditioning on its labelled rows, with no training on that table. They challenge gradient-boosted decision trees, the long-standing default for tabular prediction, but the published comparisons are hard to interpret, because studies differ in data regime, split design, baseline tuning, and metric. This review assesses the evidence regime by regime. From 76 peer-reviewed works selected under a documented protocol, it defines TFMs by reusable cross-table pretraining, organises them by pretraining data, context mechanism, scale, modality, and task interface, and compares them with tree ensembles, deep tabular networks, and tabular-to-image methods. Seven recurring claims are graded on peer review, evaluator independence, split match, baseline strength, and replication, with all scores published. One claim is strongly supported: on small-to-medium independent and identically distributed tables, TFMs match or exceed tuned tree ensembles without per-dataset tuning. Tuned tree-based and deep models remain as strong or stronger on temporal, feature-engineered data; above 10,000 rows the evidence is developer-reported except for one independent clinical benchmark; and no eligible study tests a frozen TFM under a grouped split or beyond its feature envelope. Tabular-to-image methods win only on small or very high-dimensional tables. Among the comparison-bearing studies, about one in ten use a deployment-matched split, none of them an independent evaluation of a TFM, and more than half of those evaluating a pretrained model report no contamination check. A reporting checklist, an evidence ladder, and seven research directions turn these findings into practice.

Electronics doi: 10.3390/electronics15215153

Authors: Tong Jia Yan Sun Xingyu Pan Peijuan Li Yiting Liu Guangming Chen Wen Zhang Lushuo Li

In orchard harvesting, complex occlusions, dense obstacles, and limited workspace make manipulator motion planning an autonomous decision-making problem in unstructured environments, requiring collision avoidance while balancing efficiency and trajectory smoothness. However, traditional Artificial Potential Field (APF) and Rapidly-Exploring Random Tree (RRT) algorithms have limitations in meeting harvesting demands. To address this, an enhanced APF-RRT algorithm is proposed via three improvements: potential field modeling, sampling expansion, and path optimization—an exponential smooth attractive field and a non-linear repulsive field to improve potential-field continuity and guidance stability; an obstacle-distance-based dynamic expansion strategy for rapid search and precise avoidance, reducing redundant nodes; and cubic B-spline smoothing to enhance path quality and motion stability. Validation was conducted under laboratory-simulated orchard obstacle conditions rather than in actual orchards. Compared with traditional RRT, the enhanced algorithm reduces planning time, path length, and sampling nodes by 83.55%, 18.04%, and 81.20% in 2D and by 89.77%, 20.10%, and 88.74% in 3D. Further manipulator experiments show reductions in sparse scenarios of 20.38%/21.22%, 19.71%/20.02%, and 18.46%/18.74% (single-/multi-category) and in dense scenarios of 21.62%/18.57%, 21.20%/19.65%, and 20.36%/18.74%. Results demonstrate improved planning efficiency, shorter paths, and enhanced motion stability, validating the algorithm’s generalization and adaptability.

Electronics doi: 10.3390/electronics15215154

Authors: Hyun-A Park

AI-based enterprise document management systems increasingly integrate semantic retrieval, Retrieval-Augmented Generation (RAG), multimodal interaction, and data analytics. In such environments, protected objects extend beyond original documents to chunks, embeddings, retrieval outputs, generated responses, and analytical results. Consequently, conventional login-based or static role-based access control is insufficient. This paper proposes a three-layer authentication protocol that sequentially performs session authentication, role/permission verification, and event-based group-membership authentication. Unlike conventional group-key approaches, the proposed method uses event-specific group keys only as authentication material, not for document re-encryption. By combining forward and reverse hash chains, the protocol supports group key secrecy, forward secrecy, and backward secrecy in dynamic group environments. Security analysis shows that the protected authentication values are computationally indistinguishable from random bit strings of the same length under the stated assumptions. In addition, the efficiency analysis and prototype-based evaluation show constant-time online authentication per request, together with practical latency, throughput, and manageable membership-change overhead. Therefore, the proposed scheme provides a practical application-layer access-control architecture for AI-based enterprise document services.

Electronics doi: 10.3390/electronics15215136

Authors: Weili Chen Yecheng Liu Wei Wei Jiangning Wei Weiliang Guo

The magnetotelluric (MT) signals are susceptible to various types of noise during acquisition, such as square waves, triangular waves, and pulse noise. Consequently, the effective signal is overwhelmed, leading to a significant degradation in the signal-to-noise ratio. Traditional wavelet threshold (WT) denoising can separate signals from noise through multi-scale analysis. However, its performance depends heavily on the selection of wavelet basis functions, decomposition levels, threshold functions, and threshold selection methods, which leads to insufficient adaptability in complex noise environments, causing issues such as signal distortion or residual noise. To address the above challenges, this paper proposes a divide-and-conquer strategy based on a ResNet-guided adaptive wavelet denoising method. The method first constructs a synthetic dataset containing multiple types of typical noise, and then establishes the optimal wavelet parameter combinations for different noise categories through systematic experiments. Subsequently, a residual neural network (ResNet) is trained on this dataset to construct a multi-class classification model capable of classifying the dominant noise type in each data window. In practical applications, the ResNet model is employed to recognize the dominant noise type in MT data, and based on the recognition results, the optimal wavelet parameter combinations are adaptively invoked for processing, thereby achieving targeted, category-specific denoising. Experimental results on simulated data and field data from two sites demonstrate that the proposed method effectively preserves the features of the desired signal under strong noise backgrounds and improves MT data quality.

Electronics doi: 10.3390/electronics15215152

Authors: Arshad Syed Muhammad Nashit Kamal Mazhar Malik Mingqi Chen Qiang Li Zhong Ming

Quadrotors attract broad military and civilian interest, yet their underactuated, highly coupled, and nonlinear dynamics make precise control difficult. While sliding-mode control (SMC) offers robust trajectory tracking, tuning its parameters remains a tedious manual task. To address this challenge, this research presents deep deterministic policy gradient (DDPG) with SMC for a DJI Tello quadrotor whose aerodynamic coefficients are estimated using blade element momentum theory. Two controller variants are evaluated separately. The offline variant uses DDPG to select bounded fixed SMC gains for two reference-trajectory comparisons against manually tuned SMC and PID/PD baselines. The online variant uses a residual DDPG actor to update position-loop gains during a separate square-trajectory simulation while the attitude loop gains remain fixed. The fixed-gain comparisons show lower reported position-tracking RMSE for the DDPG-tuned controller under the tested conditions. The online experiment demonstrates bounded gain variation and its effect on square-trajectory tracking under the specified mass-uncertainty protocol.

Electronics doi: 10.3390/electronics15215151

Authors: Kuan Li Yanqi Hou Yudun Li Kongming Sun Tongyu Liu

Synchronized fixed-frequency current control (SFFC) enables inverter-dominated islanded microgrids to regulate power without relying on frequency deviations. However, conventional SFFC may exhibit power oscillations when battery storage systems (BSSs) transition from discharging to charging, while unbalanced and nonlinear loads can degrade voltage quality. To address these challenges, this paper proposes an improved SFFC scheme integrating power decoupling, multi-frequency voltage compensation, and current-capacity coordination. First, the mechanism of power oscillations during BSS discharging-to-charging transitions is analyzed, and a voltage-vector-oriented controller is developed to decouple active- and reactive-power regulation. Second, the effects of negative-sequence and harmonic currents on voltage distortion are analyzed while considering the inverter equivalent impedance. Accordingly, a compensation method combining notch and low-pass filters is developed to extract and compensate for the fundamental negative-sequence and selected harmonic components. Third, a residual-current-capacity-based allocation strategy is developed to coordinate reactive-power support, negative-sequence voltage compensation, and harmonic suppression within inverter current limits. Furthermore, a satellite-synchronized mode-transition strategy is incorporated to facilitate transitions between grid-connected and islanded operation. Simulation and experimental results demonstrate improved power regulation and voltage quality under different operating conditions. Specifically, the experimental line-voltage unbalance index decreases from 3.83% to 1.64%, while the line-voltage total harmonic distortion decreases from 4.89% to 1.98%. Stable BSS charging and discharging operation and grid-connected-to-islanded transitions are also experimentally demonstrated.

Electronics doi: 10.3390/electronics15215150

Authors: Chin-Feng Lee Shao-Ting Huang Sue-Chen Hsueh Yung-Chen Chou

Face images are widely used in recognition, social media, and digital evidence, so tampering can undermine trust. This study proposes a blind tamper detection and self-recovery method for aligned, grayscale CelebA face images. A vector-quantized variational autoencoder (VQ-VAE) turns each image into a discrete codebook index map, which is scrambled, redundantly embedded by 2-LSB substitution, and recovered at verification by position-wise mode voting; a feature-consistency mask and a pixel-residual mask then localize tampering and guide fusion-based restoration. Watermark embedding introduces negligible distortion (whole-image average PSNR 44.19 dB, SSIM 0.985). On 23 test images, F1-scores are close to 1 for large-area contiguous tampering and image splicing at tampered-area ratios up to 80%, whereas salt-and-pepper noise lowers precision. Whole-image average recovery PSNR/SSIM is 30.96 dB/0.918 and 20.50 dB/0.699 under 40% and 80% large-area tampering (30.95 dB/0.919 and 20.36 dB/0.689 for splicing); these values are influenced by the intact pixels retained during fusion and do not measure tampered-region fidelity. The results apply to accidental corruption and uninformed tampering; benign processing such as JPEG compression or resizing is treated as tampering. VQ-VAE codebook indices thus serve as embeddable, bit-exact self-recovery information recoverable by mode-based voting.

Electronics doi: 10.3390/electronics15215149

Authors: Yangyang Xi Yong Wei Yifei Zhang Mei Wang Ruyi Gan

Conductance-based water-cut measurements are sensitive to coupled changes in temperature and salinity. This study develops a Bayesian-regularized backpropagation neural network (BR-BPNN) for compensation using the mean sensor peak voltage, temperature, and salinity as inputs. A grid of 198 measured calibration points was used to construct a trilinear response surface and generate 9000 derived samples for developing BR-BPNN, a ridge-regularized nonlinear basis-expansion baseline (MPR), and support vector regression (SVR). On 40 independently measured test conditions spanning 55–95% nominal water cut, the fitted BR-BPNN achieved a mean absolute error of 1.72 percentage points, compared with 3.52 for MPR and 2.32 for SVR under the reported model-selection procedures. Training-strategy comparisons and physical-plane holdouts further assessed interpolation-based training. The selected 3-10-5-1 network was deployed on an STM32L496RGT3 and evaluated with newly acquired sensor inputs under the same 40 nominal conditions. Embedded evaluation achieved a mean absolute error of 1.868 percentage points and a mean network forward-inference time of 52.05 μs. The results demonstrate a compact implementation of coupled compensation for embedded conductance-based water-cut measurement within the tested laboratory domain.

Electronics doi: 10.3390/electronics15215148

Authors: Shengtao Lyu Huimin Lu

Weakly supervised video anomaly detection (WS-VAD) aims to train models using only video-level labels while simultaneously achieving precise localization of anomalous segments. Existing methods universally assume that input videos are captured under favorable imaging conditions. However, in real-world deployments, adverse illumination conditions such as foggy illumination, rain, and nighttime cause severe image degradation, leading to significant feature distribution shift and substantially degrading detection performance. Illumination-robust weakly supervised video anomaly detection built upon the Information Bottleneck (IB) architecture. Our framework introduces three complementary innovations: (1) an Illumination-Conditioned Normalization layer (WCN) that dynamically adjusts illumination-adaptive affine parameters through a lightweight illumination-aware branch, effectively counteracting the contamination of feature distributions caused by illumination degradation; (2) a Motion-Aware Temporal Enhancement module (MATE) that extracts motion cues via frame differencing and injects them into the feature sequence via gated modulation, alleviating the over-compression of motion-based anomaly signals by the information bottleneck; and (3) a Temporal Reasoning Module (TRM) that re-models temporal dependencies through a lightweight cross-frame attention mechanism, restoring the model’s capability for precise temporal localization of anomalous behaviors. Comprehensive experiments on ShanghaiTech Campus, UCF-Crime, and XD-Violence benchmarks demonstrate that RobustNet significantly outperforms the DS-IB Net baseline and prior state-of-the-art methods under clear, foggy, rainy, nighttime, and mixed illumination conditions, while ablation studies verify the effectiveness and complementarity of all three innovations. RobustNet is also efficient: the three modules add only +4.3% parameters and cost −6.7% throughput relative to the DS-IB Net baseline, and an ordering ablation confirms that placing MATE before WCN is preferable to the reversed order.

Electronics doi: 10.3390/electronics15215147

Authors: Abdul Aziz Zazai Yuan Gao

Finite-control-set model predictive control (FCS-MPC) gives low-distortion voltage regulation for LC-filtered two-level three-phase inverters, but its computation grows exponentially with the prediction horizon. We trained neural networks offline to imitate it. Four imitators were formed by crossing two network types, a pattern-recognition classifier (NNPR) and a regression network (NNRG), with data from one-step and two-step FCS-MPC teacher controllers, giving NNPR1, NNPR2, NNRG1, and NNRG2. All were evaluated in MATLAB/Simulink with linear resistive loads at sixteen training and four unseen interpolated conditions. Mean output-voltage total harmonic distortion (THD) was 2.55% for NNPR1, 0.89% for NNPR2, 2.89% for NNRG1, and 1.36% for NNRG2, against 2.32% and 1.73% for one-step and two-step FCS-MPC, respectively. The training horizon mattered more than the network type, and the classifier gave lower mean THD at both horizons. NNPR2 gave lower THD than its teacher controller at all sixteen conditions while switching 26% less often. The advantage came mainly from the load current, which the teacher controller estimated but the networks received as measured; given the measured value, the teacher controller came within 0.13 percentage points of NNPR2. Network computation is horizon-independent; MATLAB execution times are indicative only. Results are simulation-only; hardware-in-the-loop validation with delay compensation must precede deployment.

Electronics doi: 10.3390/electronics15215146

Authors: Shaochuan Zhao Xiaotian Wang Jiawei Jiang Hefeng Yin Bing Liu Yong Zhou

RGB-T semantic segmentation fuses visible (RGB) and thermal infrared (TIR) images to enhance model robustness under complex lighting conditions. However, most existing methods employ unified fusion strategies in the spatial domain, limiting their ability to fully exploit the complementary characteristics of the two modalities across different frequency components. To address this issue, we propose a novel fusion model that achieves finer cross-modal collaboration through frequency-aware feature disentanglement and adaptive fusion mechanisms. Specifically, the model first enhances the multi-granularity representation of dual-modality features using attention mechanisms. The features are then transformed into the frequency domain, where a dynamic band disentanglement mechanism adaptively separates high-frequency details from low-frequency semantics based on the input content. A heterogeneous fusion pathway is further designed to differentially aggregate high- and low-frequency components, emphasizing cross-modal detail alignment in the high-frequency path and semantic consistency in the low-frequency path. The proposed model is evaluated on three widely-used RGB-T segmentation benchmarks, namely MFNet, PST900, and FMB, and achieves state-of-the-art performance, demonstrating its effectiveness and superiority.

Electronics doi: 10.3390/electronics15215145

Authors: Antanas Čenys Justinas Janulevičius

Hybrid finality-gadget consensus protocols expose many tunable parameters, that are usually fixed by convention. We present a three-stage verify-then-optimize method. Stage 1, a machine-checked consistency check of the quorum-intersection bound, covers 325 committee and quorum settings by counter abstraction, cross-validated by three independent models; the verdicts reduce to the two-thirds quorum rule, fix the quorum at 2f + 1, and show that the gadget stops finalizing at 75% participation. Stage 2 certifies that confirmed but unfinalized blocks resist reorganization. For a latest-message fork choice with proposer boost under random network delays, attacked by an adversary that splits its votes and times their release per validator, we compute exactly, as a Markov decision process, the worst-case probability that a block is confirmed and then reorganized within 256 slots of its proposal, and check it against an identity-level model. With 13 and 25 validators and an adversary at the gadget’s fault bound, the balancing attack succeeds almost surely unless the boost lies in a narrow window, and a risk of 10−6 requires 38 to 90 slots of confirmation, 1.5 to 3.5 times the depth a proof-of-work estimate prescribes. Stage 3 reports operating points: against such an adversary every slot length trades confirmation latency against message load.

Electronics doi: 10.3390/electronics15215144

Authors: Bingyan Cui Zhen Liu

Accurate pixel-level segmentation of pavement distress from unmanned aerial vehicle (UAV) imagery is important for automated road inspection. However, this task remains challenging because pavement defects often exhibit small spatial scales, low contrast, irregular boundaries, and complex topological structures. Moreover, many UAV pavement datasets provide only bounding-box annotations, which limits their direct use for fine-grained segmentation. To address these issues, this study proposes AeroPave-SAM3, a weakly supervised pavement-distress segmentation framework built on a promptable vision foundation model. First, a box–text dual-prompt mechanism converts bounding-box and class annotations into high-confidence pseudo masks through prompt-consistency evaluation and confidence-aware refinement. Second, a class-conditioned morphology-aware adapter is introduced to model the different characteristics of linear, network-like, and regional distress patterns. Boundary, topology, and consistency constraints are further incorporated to improve contour accuracy and structural continuity. Finally, a global-to-local self-prompted refinement strategy performs native-resolution local inference without manual prompts, enabling better recovery of thin cracks and complex boundaries in high-resolution UAV images. Experiments on UAV-PDD2023 using a manually verified pixel-level evaluation benchmark show that AeroPave-SAM3 achieves 84.36% mIoU, 91.38% Dice, 87.03% Boundary F1, and 90.12% clDice. Compared with the SAM3 baseline, these metrics improve by 3.09, 1.71, 4.42, and 4.41 percentage points, respectively. The results demonstrate that AeroPave-SAM3 provides an effective annotation-efficient solution for fine-grained UAV pavement-distress segmentation.

Electronics doi: 10.3390/electronics15215143

Authors: Ching-Che Chung Wei-Chia Huang Duo Sheng

Physical Unclonable Functions (PUFs) generate unique chip identifiers from manufacturing variations, but transmissions over public networks can be exploited through eavesdropping. To improve variation resilience, the proposed SRPFD-PUF employs bit self-test (BST) for stable-CRP selection, while response-bit tolerance is considered in an illustrative authentication flow. By minimizing additional circuit overhead, strong baseline PUF metrics are maintained at low hardware cost, supporting integration in Internet of Things (IoT) devices. An illustrative PUF-based authentication flow is presented to demonstrate a potential IoT application; end-to-end protocol implementation and security validation remain outside the scope of this study. The three-condition BST screen yields a minimum simulated logarithmic reliability score of 98.93% under the evaluated PVT conditions within 0.675–1.05 V and −40–150 °C. With the circuit architecture and LFSR configuration disclosed, the evaluated machine-learning attacks yield prediction accuracies of 47.68–52.46%, indicating resistance to the tested modeling methods. The TSMC 16 nm FinFET physical design occupies 0.0117 mm2, while post-layout simulation estimates 0.3089 mW at 100 MHz; no fabricated-silicon measurements are reported.

Electronics doi: 10.3390/electronics15215142

Authors: Zhicheng Cheng Qi Long Mei Wang Jigang Huang

Tiny surface defects are difficult to inspect because of their small scale, weak contrast, diverse morphology, and similarity to normal textures. To address the conflict between early candidate retention and false-positive suppression, this study proposes a coarse-to-fine inspection framework that separates candidate detection from final verification. You Only Look Once (YOLO) 26s first generates defect candidates, which are locally reconstructed and analyzed by a region-of-interest (ROI)-guided Multi-Task SAM (ROI-MT-SAM) for joint segmentation, semantic prediction, and mask-quality estimation. Multi-Evidence Verification (MEV) then integrates complementary confidence and spatial cues to refine the final decision, while residual-transfer augmentation improves defect diversity under limited-data conditions. Experiments show that ROI-MT-SAM improves mean Intersection over Union (mIoU) from 0.3918 to 0.8869 compared with whole-image SAM, with particularly substantial gains for small defects. MEV further reduces false positives from 245 to 46 and achieves a precision of 0.9565 and an F1-score of 0.9241. These results demonstrate the potential of combining candidate-centered representation with multi-evidence verification for tiny-defect inspection of automotive interior surfaces under the evaluated conditions.

Electronics doi: 10.3390/electronics15215141

Authors: Fahad AlGhamdi

We study how linear probes read Modern Standard Arabic (MSA)–Egyptian Arabic code-switching in Linguistic Code-switching Evaluation (LinCE) tweets using Qwen3-8B-Base, ALLaM-7B-Instruct-preview, Gemma-2-9B and Fanar-1-9B. Probes trained on monolingual text classify 54–56% of MSA words in Egyptian contexts as Egyptian, but only 4–8% of Egyptian words in MSA contexts as MSA. Lexical and contextual controls reduce the gap, leaving 12–20 percentage points after regression and 19–28 after matching. Probes trained on balanced code-switched data recover the inserted word’s variety with 96–98% accuracy, so the asymmetry does not imply loss of variety information. Probe readings correlate with next-token variety preference. Replacing the outputs of five patching-selected attention heads with reference means changes both measures relative to layer-matched random heads, but reduces MSA-to-Egyptian misclassification by only 1–14 percentage points. Effects change direction with a different replacement mean and also occur in monolingual text. External news comments reproduce the directional readout gap (38–69 points), although the difference between lexically matched directional shifts remains unresolved. Register remains a possible contributor, and head selection and interventions use one tweet corpus. Probe-based analyses should therefore report switching directions separately and distinguish monolingual readouts from recoverable word variety.

Electronics doi: 10.3390/electronics15215140

Authors: Mingxu Chen Xueyan Wang Bingye Zhang Peng Lin Pei Li Yiying Li Guoping Zou

Latent insulation defects complicate the ampacity assessment of high-voltage cable intermediate joints because localized electrical stress may increase without a discernible change in surface temperature. This study develops a simulation-based framework that combines electro-thermal analysis, Gaussian-kernel inversion, and current-limit calculation to investigate the potential incorporation of simulated defect information into allowable-current assessment. A 25-dimensional feature representation supports the estimation of internal hotspot temperature and the numerical discrimination between healthy and defective insulation states. The simulations reveal that micro-scale defects substantially intensify localized electric fields but produce only negligible surface-temperature variations, highlighting the difficulty of assessing internal insulation conditions from thermal measurements alone. Under noise-free simulation conditions, the model demonstrates the numerical separability of healthy and defective insulation states within the simulated dataset. A bisection solver determines the allowable current subject to the hotspot-temperature limit, while an illustrative heuristic derating rule adjusts this current to account for potential risks associated with insulation defects. The framework provides a simulation-based approach for investigating the integration of hotspot-temperature estimation and defect-related current adjustment, while highlighting the need for improved noise robustness and experimental validation before practical application.

Electronics doi: 10.3390/electronics15215139

Authors: Weibo Wang Shenwen Bai Yongkang Zheng Mingmao Gong

To address the inherent issues in traditional reactive power compensation strategies—such as compensation lag, the lack of load morphological identification capability, and bus transient shocks induced by multi-modality switching—this paper proposes a dual-modality collaborative reactive power optimization algorithm for distribution networks, integrating deep prediction and load morphological perception. First, a prediction model based on Trend-Residual decoupling and Gate-Aware factors is formulated to achieve precise load forecasting. Second, a morphological identification mechanism is designed to explicitly isolate the slowly varying intervals and the abrupt change intervals of the load. Finally, within the slowly varying intervals, the Artificial Hummingbird Algorithm is introduced for global optimization to provide steady-state anchors for the optimization of subsequent abrupt change intervals; conversely, within the abrupt change intervals, Finite Control Set Model Predictive Control integrating a terminal gravitational mechanism is employed for dynamic tracking, thereby realizing a flexible soft-start for the discrete compensation devices. Simulation results based on a modified IEEE 69-node system demonstrate that, under the strict premise of constraining the physical switching frequency of thyristor-switched capacitors, the proposed algorithm effectively curtails network losses, which is beneficial to prolonging the service lifespan of reactive power compensation equipment and improving the overall power quality.

Electronics doi: 10.3390/electronics15215138

Authors: Giacomo Cabri Mauro Leoncini

In the context of smart cities, reservation-based parking systems may accept more requests than they can ultimately honor when users stay longer than declared. This paper introduces a risk-aware admission-control framework that explicitly accounts for overstay uncertainty while preserving high facility utilization. For each future time slot affected by a request, admission is based on an upper bound on the probability that too few occupied stalls will be released. We derive three alternatives: a dependence-robust Markov bound, a sharper Chernoff bound under conditional independence, and a Cantelli bound incorporating common exogenous impediments. We also compare uniform and aggregate allocations of a reservation-level risk budget and establish an incremental per-acceptance reliability guarantee. Simulation experiments under saturated demand, different overstay regimes, correlated disruptions, and four operational contexts show that nominal admission and static buffering may accept many reservations that cannot be honored. By contrast, risk-aware policies substantially reduce or eliminate no-park events while maintaining high utilization. Uniform Chernoff admission achieves the highest service rate in most scenarios, serving more users than the deterministic baselines despite accepting fewer reservations. The experiments also reveal that uniform slot-level budget allocation implicitly favors shorter stays, an effect that can markedly increase throughput when parking durations are heterogeneous.

Electronics doi: 10.3390/electronics15215137

Authors: Bo-Kai Peng Shao-Kang Huang Chen-Chien Hsu Wei-Yen Wang

Task-conditioned robotic manipulation requires grasps that are geometrically plausible and compatible with downstream function. Although vision–language models (VLMs) have shown promise for semantic reasoning in robotic manipulation, VLM-selected grasps may not consistently satisfy both requirements. To address this issue, we propose a two-stage generate-then-reason framework for task-conditioned 6-DoF grasp selection. In Stage 1, a VLM localizes task-relevant object regions, followed by segmentation and GraspNet-based generation of plausible grasp candidates. In Stage 2, another VLM ranks grounded grasp candidates according to their compatibility with the specified downstream task. The evaluation in this journal includes a 60-trial full-system physical study and four controlled replay experiments that examine Stage 1 localization, Stage 2 cross-model ranking, repeated-inference reliability, and cross-stage candidate-pool sensitivity. The physical system achieved an overall task success rate of 63.3%. In the replay analysis, Gemini 3 Flash Preview achieved the highest Stage 1 localization accuracy of 95.0% and the lowest median Stage 1 latency of 11.09 s. For Stage 2 ranking, GPT-5.6 Sol achieved the highest Top-1 task-valid rate of 92.9% on identical frozen candidate pools and showed the strongest repeated-inference behavior among the evaluated critics. The cross-stage sensitivity experiment yielded task-valid selection rates ranging from 70.8% to 87.5% across six model combinations, indicating sensitivity of downstream selection to upstream candidate evidence within the evaluated conditions. These results indicate that physical task completion and stage-specific VLM behavior capture different aspects of system performance and should be evaluated separately when assessing VLM-based grasp-selection systems.

Electronics doi: 10.3390/electronics15215135

Authors: Hector Poblete Ania Cravero Ninoska Diaz

Large language model agents are increasingly used to analyze data, generate code, and execute operations in data-engineering environments. However, successful technical execution does not guarantee that an operation is authorized or that its results are correct and verifiable. This creates governance challenges when specialized agents modify persistent data. This study presents a Governed Human–AI–Agent Workflow that coordinates a Data Expert Agent and a Data Engineering Agent under human supervision. The objective is to establish explicit responsibilities for planning, authorization, execution, verification, and provenance. The workflow was developed using an adapted Design Science Research Methodology and demonstrated through a controlled Data Lakehouse experiment using synthetic academic data. Documented execution failures were analyzed to identify governance requirements and develop operational controls. These failures included repeated execution, inconsistent evaluation units, premature remediation, and unintended effects across data-quality rules. The resulting artifact comprises nine stages and four prompt-based contracts: Planning, Execution, Verification, and Refinement. An initial intervention increased the data-quality detection F1 score from 0.8084 to 0.9077 but introduced 49 false negatives. A subsequent authorized refinement increased F1 to 0.9988, with one remaining false negative. These results illustrate the importance of verifying the effects of technical interventions before accepting a new persistent data state. The contribution is a process-oriented governance artifact that integrates human authorization, specialized agent responsibilities, bounded execution, read-only verification, and persistent evidence. The study provides a formative demonstration within one controlled academic laboratory scenario. Independent evaluation across other datasets, platforms, and institutional environments remains necessary.

Electronics doi: 10.3390/electronics15215133

Authors: Sandyka Bala Hans Indrawan Sucipto Nico Surantha

Credential-based attendance mechanisms such as QR codes and RFID cards record who checked in but not whether that person is physically present, which makes them vulnerable to proxy attendance. Vision-based alternatives depend on line of sight and camera coverage, whereas Bluetooth Low Energy (BLE) indoor positioning can passively verify the presence of a student’s device. However, received signal strength indicator (RSSI) measurements are highly sensitive to multipath propagation, so neighboring workstations with similar signal characteristics are difficult to distinguish. This study evaluates the feasibility of workstation-level BLE localization in a classroom with 40 workstations and four beacons. RSSI was smoothed with a Kalman filter and combined with relational features, namely beacon ranks and pairwise RSSI differences, as inputs to Weighted K-Nearest Neighbor (WKNN), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM) classifiers at 40, 20, 8, and 4 location classes. Kalman filtering improved accuracy for all four models at every granularity (+3.34 to +11.89 percentage points); for the two main models, WKNN and RF, all paired cluster-bootstrap 95% confidence intervals of this gain were above zero (+4.58 to +8.71 percentage points). The relational features improved tree-based ensembles at 40 classes (RF: +1.38 percentage points, 95% CI [+0.03, +2.88]; XGBoost: +1.35 percentage points, point estimate) but reduced WKNN accuracy at the finer granularities. RF reached 40.68% accuracy at 40 classes, 61.06% at 8 classes, and 73.37% at 4 classes, and all four classifiers performed similarly at 40 classes. These results show a clear trade-off between spatial resolution and classification reliability with four beacons; identity binding and session-level attendance decisions remain to be evaluated.

Electronics doi: 10.3390/electronics15215134

Authors: Xiaobing Yu Yuqing Zhang Chao Zheng Zhihui Fan Tianming Zhan

To address video jitter and key-structure positional drift in industrial surveillance scenarios, this paper proposes a video stabilization method based on key-structure semantic constraints and grid motion fusion. First, Segment Anything Model 3 (SAM3) is employed to extract key structures and estimate semantic motion from their positional variations. The resulting semantic motion is then fused with deep unsupervised trajectory-based (DUT) grid motion to enhance key-structure positional stability while maintaining overall video stability and preserving local geometric structures. Subsequently, a geometric valid support domain is constructed using the fused motion to identify invalid boundary regions introduced by frame reprojection. The Flow-Guided Transformer (FGT) model is further employed to complete these regions and improve the visual completeness of the stabilized videos. Experiments conducted on general video sequences and industrial surveillance scenes demonstrate that the proposed method achieves favorable performance in geometric structure preservation and overall stability while reducing key-structure positional drift. Qualitative results further illustrate that the geometric support masks identify regions requiring completion and guide FGT in generating visually plausible boundary content in the presented examples. Ablation studies and fusion-weight sensitivity analysis further demonstrate the complementary roles of semantic motion and grid motion in achieving balanced stabilization performance.

Electronics doi: 10.3390/electronics15215132

Authors: Chao Zhang Weifeng Ji

To address the characteristics of persistent attack propagation, heterogeneous associations, and temporal evolution of attack behaviors in dynamic networks, this paper proposes AC-ODE, a dynamic network attack prediction model based on attack-context graph neural ordinary differential equations, which jointly predicts host node risk and attack propagation along links. Specifically, a relation-aware historical state encoding module is first designed to extract multidimensional and comprehensive cybersecurity context from multi-frame temporal graph snapshots. An attack-context-conditioned continuous propagation module is then constructed to model the continuous evolution of latent states on discretely observed graphs, while using attack context to modulate propagation intensity. Finally, a historical–future state gated fusion mechanism is introduced to enable joint node–edge prediction. Compared with existing methods based on discrete snapshot modeling, AC-ODE more accurately characterizes risk variation trends as network attacks transition from activation to attenuation. Experimental results on three public datasets—LANL, the DARPA Operationally Transparent Cyber (OpTC) dataset, and CICIDS2017—show that AC-ODE outperforms multiple baseline methods in terms of both AP and Hits@10, achieving an edge-level AP of 0.974 on OpTC, which corresponds to a 73.8% relative improvement over the second-best method. Ablation studies and temporal evolution analyses further demonstrate that the core modules play an important role in improving prediction accuracy and capturing the temporal evolution patterns of network attacks.

Electronics doi: 10.3390/electronics15215131

Authors: Liyuan Yu Guang Yang Hongwei Wang Yi Liu Sirui Yao Haotian Zhu

An equivalent-circuit model is developed to investigate the effects of fabrication-induced phase perturbations on coherent power combining in terahertz resonant tunneling diode (RTD) oscillator arrays. The resonant tunneling and residual excess currents are represented by two nonlinear voltage-controlled current sources, while the slot antenna is modeled as an equivalent RLC resonant network. Fabrication-induced phase perturbations are modeled by introducing controlled variations in the intrinsic capacitance of individual oscillator units. A GaInAs/AlAs RTD oscillator operating at 220 GHz delivers an output power of 74.56 μW, and its simulated current–voltage characteristics agree well with experimental data. The model is subsequently extended to 1 × 2, 2 × 2, and 4 × 4 arrays, yielding output powers of 140.67, 269.71, and 1053.39 μW and combining efficiencies of 94.33%, 90.43%, and 88.30%, respectively. For the 2 × 2 array, increasing the phase perturbation from 0° to 2° reduces the output power from 269.71 μW to 255.32 μW. The proposed model enables the analysis of phase perturbations arising from fabrication-induced variations in individual RTD oscillator units and provides a theoretical basis for the design and optimization of scalable, high-efficiency terahertz RTD oscillator arrays.

Electronics doi: 10.3390/electronics15215128

Authors: Shike Wang Jinyu Wen Yabing Sun Lin Cui

To address the modeling and stability analysis of large-scale renewable energy plants, this paper proposes a K-means-based multi-device clustering and equivalent-model simplification method. A plant comprising 1000 MW wind turbines, a 250 MW/1000 MWh energy storage system, and ±384 Mvar static var generators is represented by five equivalent units connected in parallel. A total terminal-impedance model of the parallel equivalent-unit system is then developed, and its impedance intersections with the grid-side impedance are analyzed to assess the grid-connected stability of the renewable energy plant. Time-domain simulation results validate the accuracy of both the theoretical model and the stability assessment.

Electronics doi: 10.3390/electronics15215130

Authors: Gibeom Park Jungsik Kim

Silicon carbide (SiC) devices used in space and nuclear environments are exposed to radiation that creates displacement defects (DDs), but the effect of DDs near the gate oxide/SiC interface has not been analyzed systematically. Using technology computer-aided design (TCAD) simulation, this study analyzes how near-interface DDs in a 4H-SiC metal-oxide-semiconductor field-effect transistor (MOSFET) affect the threshold voltage (Vth), transconductance (gm), and on-current (Ion), and builds deep neural network (DNN) models that predict these effects. The DDs were modeled as acceptor-like traps with varied energy levels, concentration, depth, lateral position, and distribution width. The degradation was largest at the channel entrance and the edge of the junction field-effect transistor (JFET) region. The center depth mattered little while the trap region reached the interface, whereas a small gap between the trap region and the interface strongly reduced the degradation. Shallow-level traps reduced Ion without shifting Vth, whereas only deep-level traps caused an actual Vth shift. The DNNs accurately predicted the degradation for new TCAD conditions in which the trap parameters simultaneously took values not used in training and outperformed grid interpolation and other regression models. The trained DNNs thus rapidly predict the degradation at the simulated lateral positions for radiation reliability assessment.

Electronics doi: 10.3390/electronics15215129

Authors: Wei Xuan Jun Zhu

Speech enhancement on edge and wearable devices operates under strict power budgets; generative enhancers buy quality through repeated on-device evaluations. On a Jetson Orin NX, we characterize solver truncation and solver-free retraining for a 65.59 M-parameter flow-matching enhancer. Same-board latency, energy, and quality measurements yield a decision rule comparing two quantities: N*, the smallest step count PESQ-equivalent to the retrained model (±0.05), and Nmax, the largest a mode affords. N* = 4 for the reference on this two-speaker corpus; Nmax (p95) is 5 at the unconstrained mode and 2 at every constrained mode. We keep the solver where affordable; otherwise, under the PESQ criterion, only retraining reaches that quality among the compared recipes. Power modes are not a performance axis; incremental energy varies across budgets (2218 vs. 2022 mJ/s); a 0.048 M reference at MAXN shows 16.3× lower sustained end-to-end RTF at approximately 1% of the energy, 0.208 PESQ lower, with no detectable SI-SDR difference. At one pass, retraining gains 0.173 PESQ but loses 0.75 dB SI-SDR. The procedure is specified, not validated beyond this configuration: re-measure Nmax on the target board and N* after a model, criterion, distribution, or arithmetic change.

Electronics doi: 10.3390/electronics15215127

Authors: Yubo Zhang Xuefeng Zhu Xueyan Hao Jiaxuan Liang Qidan Zhang Bo Zhang Xingcheng Yu

In cyber-physical power systems (CPPS), the inrush current generated during transformer reclosing can easily cause misjudgment of protection devices and thus serves as an important trigger for cascading failure propagation. Existing studies mostly rely on black-box electromagnetic transient simulations or simplified magnetization models, lacking analytical approaches that account for hysteresis and residual magnetism, and therefore fail to accurately characterize the coupling mechanism between physical transients and information-layer dynamics. To address this issue, an analytical transient model for three-phase transformer reclosing is established based on the Jiles–Atherton theory, in which the hysteresis loop and residual magnetism are embedded into the electromagnetic transient process as a differential–algebraic equation (DAE) set, and the semi-analytical dependence of the inrush peak, harmonic characteristics, and decay law on the residual flux and closing angle is derived. On this basis, an inrush-feature-driven probability model of protection misjudgment is constructed, a CPPS cascading failure propagation model incorporating physical–information state transitions is developed, and an adaptive reclosing recovery control strategy based on the node fault-state reachability is proposed. Simulations on the IEEE 39-bus system show that the proposed model accurately reproduces the quantitative influence of hysteresis residual magnetism on inrush characteristics (the peak error is within 10% compared with measured/detailed simulation results), and the proposed control strategy reduces the misjudgment rate per reclosing operation from 6.78% to 1.18% (a reduction of approximately 83%) and the post-recovery load loss from 0.182% to 0.001%, compared with the conventional reclosing scheme, significantly improving the resilience and recovery efficiency of the CPPS.

Electronics doi: 10.3390/electronics15215126

Authors: Meghal Mandot Olusola Oluwakemi Abayomi-Alli Nikolaj Goranin Adebayo Adewumi Abayomi-Alli

The growing prevalence of online abuse has increased the need for automatic cyberbullying and hate speech detection, particularly in under-resourced and morphologically complex languages with limited annotated data and language resources. This study presents a multilingual English–Lithuanian framework combining machine learning, deep learning, transformer models, three data augmentation techniques, and explainable AI to address class imbalance, limited data, and model transparency. Overall, transformer-based approaches outperformed traditional methods. For the English dataset, averaged results over five runs with different random seeds showed model-dependent performance. Syn-Aug achieved the best F1-scores for LR (0.8791 ± 0.0029), SVM (0.8810 ± 0.0028), and RoBERTa (0.9287 ± 0.0023), while contextual BERT augmentation achieved the best F1-scores for CNN (0.9124 ± 0.0020) and BERT (0.9274 ± 0.0021). The highest English performance was obtained with the soft-voting BERT-RoBERTa ensemble, with an F1-score of 0.9356 ± 0.0035. For Lithuanian hate speech classification, hold-out validation and five-fold cross-validation were employed and evaluated on the same independent test set. For the five-fold cross-validation, XLM-RoBERTa and Syn-Aug achieved the best accuracy (0.9458) and F1-score (0.9363), followed by XLM-RoBERTa with XLM-Aug (accuracy 0.9072; F1-score 0.9336). Under the hold-out protocol, XLM-RoBERTa with XLM-Aug achieved the best accuracy (0.824) and F1-score (0.8109). These results show that Lithuanian classification performance varied with augmentation and evaluation protocol. Cross-language transfer using semantically aligned binary labels further highlighted the importance of language-specific data. Overall, the proposed approach supports efficient, interpretable, and reliable multilingual cyberbullying detection.

Electronics doi: 10.3390/electronics15215125

Authors: Jolanta Chmielińska Jacek Jakubowski Kinga Karwowska Damian Wierzbicki

Nowadays, synthetic aperture radars are widely used on various satellite imaging platforms. The high availability of SAR imagery makes it very popular in various automatic object detection systems. However, the visual interpretation of SAR data is difficult even for experts, which poses the primary challenge in preparing the data required for training and evaluating deep learning models. Therefore, using data augmentation techniques to expand the original training dataset is a popular approach. In this paper, a method of synthetic SAR object image generation is proposed using Principal Component Analysis. The presented technique is based on principal component perturbations and can be used for data augmentation to train object detection models. The numerical experiments performed with the YOLO model showed that the proposed method improved AP50 on the test set for transport aircraft detection, with statistical significance confirmed by bootstrap analysis.

Electronics doi: 10.3390/electronics15215124

Authors: Zhongzhen Hu Caihui Zhu Zhaohui Jin Xin Chen Cheng Liu Zewen Dong Jian Qiu Hui Zhao Kefu Liu Tingfeng Wang

The push–pull inverter topology inherently incorporates a transformer, making it suitable for electrosurgical generators without requiring an additional transformer. However, conventional push–pull topologies exhibit limitations, including high switch stress and the absence of a dedicated path for releasing transformer leakage-inductance energy, which may increase switching losses at high frequencies. This paper presents a bipolar pulse generator based on a push–pull inverter topology, offering a new engineering approach for electrosurgical applications. The prototype achieves soft-switching operation over a load range of 30–1000 Ω, reducing design complexity, extending load adaptability, and improving system reliability while meeting the stringent pulse-width control requirements of bipolar electrosurgical applications. In addition, based on an equivalent-impedance model of tissue welding, the operating modes and soft-switching mechanisms of the push–pull inverter under different dynamic load conditions are analyzed, and strict constraints relating to dead time, load resistance, and circuit parameters are derived. A bipolar pulse-generator prototype operating at 400 kHz was developed to validate the theoretical analysis. Experimental results show a conversion efficiency of 86–94% over a dynamic load range of 25–250 Ω. At a moderate nominal load, the system maintains an efficiency above 93% over an output-power range of 4.92–183.69 W, achieving high-efficiency operation across a power range relevant to electrosurgical applications. While maintaining tissue-welding quality, the proposed system improves overall efficiency, demonstrating the feasibility and engineering value of the topology and control strategy for bipolar electrosurgical generators.

Electronics doi: 10.3390/electronics15215119

Authors: Wenzhi Chen Zihao Hua Qichao Lu Libo Cao Bo Cheng Chuanchang Liu Penghan Song

Whether edge-scale language models can call tools, and whether teams help, is understudied. We present a fine-grained diagnostic evaluation of tool calling on open-source edge models (0.6B–9B). The single-agent framework measures four dimensions and attributes every run to one of seven failure modes; parameter filling, not tool selection, separates single-agent models. The evaluation extends to multi-agent systems: an orchestrator-workers workflow over 30 MCP tools, a two-layer failure taxonomy of role delegation and worker execution, and a noise-floor protocol of 30 repeats with paired t-tests, exact permutation tests, and bootstrap intervals. Delegation rises steeply with orchestrator scale. Even a Qwen3-0.6B coordinator can delegate: a template-controlled sweep traces its zero delegation to the chat template attached to the model tag rather than to weight precision, and the same weights reach 12.5–22.1% under the template embedded in the model file. Orchestrator-workers teams do not significantly outperform a matched single model or best-of-3, an oracle-aided upper bound rather than a deployable selector, at any scale we tested: delegation overhead consumes the accuracy workers achieve. These findings hold on 4090 GPUs and reproduce on an Ascend 310P NPU. Cross-family runs reproduce the general directional pattern, and the tested orchestrator-workers teams provide no demonstrated advantage over the corresponding single-model baselines under this benchmark and deployment setup.

Electronics doi: 10.3390/electronics15215123

Authors: Yonghui Liu Xiaonan Tan Mu Zhou

In power grids, the introduction of renewable energy may affect frequency stability. Therefore, this paper considers adjustable predefined-time sliding-mode frequency control of microgrids with hybrid energy storage systems (HESSs) and cyberattacks. First, by analyzing dynamics of the ultracapacitors, batteries and cyberattacks, models of the HESS and aperiodic DoS attacks are established. To compensate the disturbances induced by renewable energy, an adjustable predefined-time observer (APTO) is designed, which combines adaptive laws such that the observation error converges to the specified boundary with a predefined time that is regulated by a flexible scalar. Based on the observer, an integral sliding surface is designed. Then, by introducing a switched system, the stability of the sliding-mode dynamics is analyzed. Moreover, to guarantee reachability of the sliding surface, a predefined-time sliding-mode controller is designed. Compared with the existing results, the proposed controller is independent of the initial states. At last, the two-area microgrids are used to illustrate the effectiveness of the designed controller. Compared with SMC and FTSMC, the frequency deviation is reduced by 6.79% and 4.78% and the convergence time is reduced by 31.6% and 13.3%. It is shown that the proposed controller achieves fast convergence and strong robustness.

Electronics doi: 10.3390/electronics15215122

Authors: Mahmoud Ahmed Sebastien Genevey Yves Audet Yvon Savaria

This paper proposes an on-chip, ultra-low-power cold-start circuit for a low-voltage, fast start-up of a boost converter for autonomous thermal energy harvesting from slight temperature differences. An improved charge transfer obtained from a Schmitt trigger (ST) clock booster enables voltage multiplication of the low input voltage during a cold start, providing low power consumption and a high boosting ratio. The start-up charge pump operates with an on-chip clock generated by an ultra-low-voltage ring oscillator that comprises modified Schmitt trigger (ST) delay cells, as previously reported by the authors. The proposed start-up architecture combines a previously reported advantageous one-shot control mechanism to start the boost converter efficiently. Theoretical analysis supports the proposed start-up scheme, and it is implemented in 22 nm FD-SOI CMOS technology. Simulation results show that, when combined with an inductive boost converter, the proposed harvesting system starts operating in 1.2 s when harvesting energy from a 48 mV DC source. To the best of the authors’ knowledge, 48 mV is the lowest reported minimum start-up voltage for a simulated fully integrated start-up circuit operating without additional sources of energy. The complete primary boost converter that was simulated requires a standard 82 μH off-chip inductor.

Electronics doi: 10.3390/electronics15215121

Authors: Jorge Díaz-Villegas Zihao Xiao Ania Cravero

Agentic Software Engineering (ASE) systems depend on decisions about context, reasoning, coordination, tool use, execution, and verification. This systematic mapping study maps problems, design decisions, operationalizations, and evidence in peer-reviewed studies of LLM- or foundation-model-based agentic systems in Software Engineering from 1 January 2024 to 13 July 2026. ACM Digital Library, IEEE Xplore, Scopus, and Springer Nature Link were searched on 13 July 2026 using the same Boolean string. From 1069 records, 94 studies were included. A frozen, single-paper-isolated extraction protocol identified 1018 design decisions, followed by deterministic multi-label synthesis; no formal study-level risk-of-bias assessment was performed. Of these decisions, 633 were linked to 656 problems or objectives, and 913 (89.7%) had empirical or comparative evaluation records. Only 19.8% of evaluation records attributed evidence to an individual decision, whereas 52.3% assessed complete systems; multi-agent configurations represented 68.7% of problem-linked decisions. Heterogeneous metrics and system-level evaluation limit inference about effectiveness. The findings expose a gap between evaluation coverage and decision-level attribution and motivate ablations and human-centered evaluations. The review received no external funding and was not registered.

Electronics doi: 10.3390/electronics15215120

Authors: Audrey Rah Yuhua Chen

This offline study examines alpha and beta spectral differences around the task annotations of the PhysioNet Electroencephalography (EEG) Motor Movement/Imagery Dataset. Motor-cortex EEG was analyzed from the PhysioNet motor-execution and motor-imagery recordings. The time origin was the dataset task annotation (T1 or T2), which marks the instructed target interval rather than independently measured movement onset. The primary one-second window, from −1.0 to 0.0 s, precedes the task annotation and lies in the preceding rest block when that block fully contains it. Spectral power features in the alpha and beta frequency bands were computed using Welch’s method. Subject-wise z-score normalization was applied prior to statistical analysis. A comprehensive statistical evaluation was performed, including non-parametric group comparisons, pairwise post hoc tests with Bonferroni correction, effect size estimation, bootstrap confidence intervals, normality assessment, and rank correlation analysis. The original 109-subject Kruskal–Wallis contrast is significant (Hα=74.79, p≪0.001; Hβ=289.01, p≪0.001). Reanalysis established that these alpha and beta power differences compare the end of rest with the end of the preceding task, rather than a measured pre-movement interval with rest. Under the Methods-faithful per-trial ERD/ERS definition, alpha ERD did not differ significantly across rest/left/right (H=3.95, p=0.139) and is therefore not treated as a primary confirmed finding. Supplementary subject-held-out classification (107 participants) yielded an AUC of 0.595 for post-cue versus pre-cue rest discrimination, while pre-cue prediction of the upcoming hand remained at chance after sequence control. These findings do not establish limb-specific intention detection before movement. These findings characterize sensorimotor EEG dynamics and provide a foundation for future research on EEG-informed neurorehabilitation.

Electronics doi: 10.3390/electronics15215117

Authors: Daniel D. Campo-Ossa G. Alejandra Feo-Cediel Fabio Andrade Rengifo

In power-quality studies of power electronic converters, total harmonic distortion (THD) summarizes voltage and current spectra separately, obscuring their joint organization across harmonic orders. We develop the Cross-Order Distortion Descriptor (CDD), a magnitude-only descriptor of paired per-order root-mean-square (RMS) voltage and current spectra. In its exact factorization, the fundamental-related part is fixed by the fundamental RMS product and the two THD values, whereas the harmonic–harmonic part additionally depends on κ∈[0,1], the complement of the normalized same-order spectral overlap, which approximately locates a record below the THD-imposed ceiling. In the pair with the largest separation in κ among 2500 simulated diode-bridge rectifier supply spectra with equal voltage THD and nearly equal current THD, κ=0.96 and 0.34 accompany different conduction intervals; reversing the third-harmonic phase, with voltage magnitudes unchanged, reverses that ordering. In induction-motor records, the harmonic–harmonic part of CDD rises by 11.8% through κ at a nearly unchanged scale under illustrative error scenarios, whereas photovoltaic-inverter tests approximately double both THD values at nearly constant κ. In capacitor records, κ nearly follows joint spectral concentration; in data-center and capacitor records, harmonic–harmonic contributions stay small at low voltage THD, as the bounds require. CDD characterizes paired harmonic spectra, separating THD-determined and order-dependent contributions.

Electronics doi: 10.3390/electronics15215118

Authors: Tao Zhou Xiangxiang Zhang Hongqiang Yu Defang Chang Jiangxiu Zhao

Image fusion is a research hotspot in the field of computer vision, which obtains complementary information from the different modal images and then presents it in a comprehensive, highly information-cohesive fused image. Encoder–decoder networks are an important technical method in the field of image fusion. This paper reviewed the research status and development of the encoder–decoder networks in the field of image fusion from the following aspects: Firstly, the 21-image fusion-related datasets and the 27 open-source codes of deep learning-based image fusion are summarized. Secondly, the model structure of encoder–decoder networks and the basic principles for image fusion based on the encoder–decoder network are described. Thirdly, encoder–decoder networks fusion models are summarized into four aspects: convolutional auto-encoder network (CAE Net), sparse auto-encoder network (SAE Net), U-shape encoder–decoder network (U-net), and variational auto-encoder network (VAE Net). In U-net, there are five types: U-net incorporating residual connection mechanism (Res U-net), U-net incorporating dense connection mechanism (Dens U-net), U-net incorporating multiscale connection mechanism (MS U-net), U-net incorporating attention mechanism (Att U-net), and U-shaped encoder–decoder networks (U-net) combined with Generative Adversarial Network (GAN U-net). Fourthly, the existing image fusion evaluation metrics are summarized from both subjective and objective aspects. Fifthly, the application of encoder–decoder network in five kinds of image fusion, such as multi-exposure image fusion, multi-focus image fusion, infrared and visible image fusion, medical image fusion, and remote sensing image fusion, are summarized. Finally, there is a discussion of the main challenges faced by encoder–decoder networks in the image fusion domain and an outlook on future directions. This paper systematically analyzed the application of encoder–decoder networks in the image fusion areas, which has positive implications for the progress of image fusion technology.

Electronics doi: 10.3390/electronics15215116

Authors: Jing Qiu Anming Dong Guijuan Wang Jiguo Yu

Intelligent reflecting surfaces (IRSs) have been recognized as a disruptive and revolutionary technology for achieving efficient and secure information transmission. This paper investigates beamforming optimization methods for an IRS-aided MIMO system with an eavesdropper. A secrecy rate maximization beamforming problem subject to both the transmit power constraints at the base station (BS) and the reflection constraints at the IRS is considered. The problem is non-convex and challenging to solve due to the coupled optimization variables. To address this issue, an iterative alternating optimization (IterAO) algorithm is proposed. The method is easy to implement, but it needs more running time to converge. We then propose a projected gradient method (PGM), which is proven to converge to a stable secrecy rate in relatively less time. It is shown that the proposed IterAO and PGM methods converge to almost the same secrecy rate ultimately, but the PGM significantly reduces the computational complexity. In addition, the secrecy rates of the proposed IterAO and PGM are higher than that of the benchmark scheme. It is also shown that the system with an IRS offers significant achievable secrecy rate gains over that without.

Electronics doi: 10.3390/electronics15215114

Authors: Alessandra Lumini Maichol Dadi

Beach volleyball video analysis is challenging because technical actions are short, the ball is small and fast, player occlusions are frequent, and broadcast footage contains camera cuts. This paper presents Vollytics a modular single-camera framework for technical-action detection, event spotting, rally reconstruction, and score inference from broadcast and fixed-camera videos. The system combines a You Only Look Once (YOLO)-based detector with court geometry estimation, geometry-aware filtering, lightweight player–ball association, and volleyball-specific temporal reasoning to transform frame-level visual observations into a structured representation of match progression. We curate 23,651 annotated frames containing player, ball, and technical-action bounding boxes, together with a video-based dataset comprising 231 action events and 44 rallies. Frame-level technical-action detection achieves a Macro Action F1-score of 0.572, while temporal consolidation yields a Macro F1-score of 0.851 for event spotting under an absolute temporal tolerance of 2 s. At rally level, Vollytics reconstructs 43 of 44 annotated rallies, achieving an F1-score of 0.966. On three complete broadcast matches, score inference produces an absolute point-count error of 8 over 215 official points (3.7%). The detector runs at 30.8 FPS on the tested GPU, supporting faster-than-real-time analysis of 25 FPS video. These results provide an initial validation of integrating frame-level visual perception with spatial and temporal domain reasoning for automatic beach-volleyball match analysis. Compared with previous single-camera beach-volleyball approaches, Vollytics provides a more complete framework in terms of functional scope and evaluation breadth, extending isolated tracking or interaction recognition to quantitatively evaluated event spotting, rally reconstruction, and progressive score inference.

Electronics doi: 10.3390/electronics15215115

Authors: Fang Cao Yaokun Ku Xiaoxue Cao Zhitao Chen Gang Li Wenyan Han Bailiang Lu Zhi Chen

Traffic risks on bridge segments exhibit spatiotemporal dependence under congestion, low visibility, and sudden traffic incidents. This study develops and evaluates a reproducible simulation framework that links short-term risk-proxy forecasting with graded low-position lighting control for a 1200 m linear bridge corridor. The framework uses six traffic and environmental scenario families, a direction-separated graph convolution and gated recurrent unit model, explicit baselines and graph ablations, and a rule-based upstream coordination controller. The model predicts a continuous conflict-exposure proxy for 20 control zones over a five-minute horizon. Across ten training seeds, the proposed DS-GCN-GRU achieved a mean RMSE of 0.1094 ± 0.0004, MAE of 0.0773 ± 0.0007, macro-F1 of 0.8339 ± 0.0082, Level 3 recall of 0.8777 ± 0.0095, and high-risk PR-AUC of 0.9292 ± 0.0010. Real-world safety benefits and driver responses to the proposed lighting commands have not been validated.

Electronics doi: 10.3390/electronics15215113

Authors: Nicola Zanfardino Marco Giovannini Lorenzo Ricciardi Celsi

Machine-learning (ML) portfolio studies often emphasize forecast accuracy, although economic value depends on how forecasts propagate through uncertainty, constraints, turnover, and transaction costs. We evaluate a leakage-controlled decision-support pipeline combining weekly Elastic Net and Random Forest excess-return forecasts with uncertainty-calibrated Black–Litterman (BL) allocation across five ETFs. Data from 2021–2023 form the training sample, 2024 is used for walk-forward selection and confidence calibration, and 2025 is a final 53-week holdout; model coefficients may update with newly observed data, but the configuration is frozen before the holdout. Neither pooled ML model beats the zero or historical-mean forecasts on the pre-specified Mean NRMSE. BL–Elastic Net obtains a net Sharpe ratio of 2.762 with 0.86% average weekly turnover and a −3.11% maximum drawdown; descriptively, its Sharpe exceeds 1/N (2.425) and mean–variance (2.324) but not GMV (3.069). However, circular block-bootstrap 95% confidence intervals for all BL–Elastic Net and BL–Random Forest Sharpe differences versus 1/N, GMV, and mean–variance include zero. Sensitivity, regime, and implementation analyses therefore support a bounded conclusion: forecast ranking and portfolio ranking can diverge, but the available holdout does not establish statistically significant risk-adjusted dominance.

Electronics doi: 10.3390/electronics15215112

Authors: Kevin Niño-Tejada Jhonathan L. Rivas-Caicedo Laura Saldaña-Aristizábal Juan F. Patarroyo-Montenegro

Wearable Inertial Measurement Units (IMUs) enable full-body motion tracking studies beyond instrumented laboratory environments, which tend to be complex and rigid. However, current IMU-based motion tracking algorithms require complex neural network models that limit online deployment on the wearable device—or devices—due to hardware constraints. This paper explores the concept of distributing the computational load of a full-body joint-angle estimation algorithm across multiple wearable devices. We present a Distributed Bidirectional Recurrent Neural Network (D-BRANN) scheme that decomposes a sparse-IMU full-body joint angle estimator into five anatomical branches. Each branch processes one local IMU reading and a shared hip/root reference across three incremental stages: leaf position, joint position, and rotation estimation. Then, the distributed inferences of all branches are merged through a residual fusion algorithm. Compared to centralized benchmarks, D-BRANN requires around 66% fewer parameters and floating-point operations (FLOPs) per prediction. Despite this reduction, D-BRANN achieves SIP errors (mean global rotation errors of the upper arms and thighs) and whole-body angular errors that are non-inferior to those of the centralized pipeline benchmark within a 2° margin, with a SIP error of 14.88° compared to 14.95°, while its SIP error under fixed-window online inference increases moderately (18.66° compared to 16.69°). These results show that sparse-IMU pose estimation can be partitioned into substantially smaller computational components while preserving comparable offline SIP accuracy and real-time processing throughput, supporting future deployment across distributed wearable devices.

Electronics doi: 10.3390/electronics15215111

Authors: Guangxiao Song Chongxiao Qu Qingjie Liu Daying Quan Baoqing Zeng

Automatic modulation recognition (AMR) from short in-phase and quadrature (I/Q) sequences remains challenging at low signal-to-noise ratios (SNRs), as informative local structures become less distinguishable and discriminative evidence may be distributed across time. We present DITE-Net, a Dendritic-Inhibitory Temporal Evidence Network for short-sequence AMR. Inspired by neural information-processing functions, the architecture combines dendritic compartmental integration (DCI) for input-dependent multiscale feature aggregation, excitation–inhibition regulation (EIR) for gated residual response control, and distributed temporal evidence integration (DTEI) for contextual processing and bidirectional recurrent aggregation. The neural-circuit-inspired terminology reflects functional design analogies rather than a biologically faithful simulation. Experiments on RADIOML 2016.10A and RML22 demonstrate that DITE-Net achieves competitive recognition accuracy. Additional ablation studies and low-SNR evaluations indicate that removing BiGRU causes the largest accuracy reduction among the evaluated components, while the independent effects of adaptive routing and EIR remain modest and dataset-dependent. Computational analyses further quantify the trade-off between recognition accuracy and inference cost for fixed-window processing.

Electronics doi: 10.3390/electronics15215110

Authors: Olabode A. Olanipekun Carlos J. Montalvo Sean G. Walker John T. Wade Bhushan Lohar

Here, the interactions between Urban Air Mobility (UAM) and general aviation (GA) as a socio-technical system (STS) are studied. Their characteristics are also outlined as they relate to the complexity usually associated with such kinds of systems, the aviation industry being an enterprise where both are described as subsets. Using agent-based modeling (ABM), the conceptualization of the socio-technical system (STS) model and its incorporation into the development of a NetLogo program are succinctly presented. Key elements within this complex framework were captured using high-level abstraction, including interfaces such as the physical infrastructure consisting of the airport, airspace (or highway), waypoints, runways, and so on; the social components, one of which was represented by the volume of passenger traffic to and from the airport which was further abstracted in terms of the number of outbound and inbound flights; and the electronic/digital components including surface radar as well as calls by air traffic controllers to the pilot which were also abstracted as separate-aircraft and dynamic-separate-aircraft procedures in the NetLogo code developed for the agent-based model (ABM). The results, discussion and conclusion are presented. Finally, after conducting a number of simulations with the developed NetLogo model by varying the number of airplanes and UAMs for each iteration, the setup with four airplanes and two UAMs resulting in 54 and 27 check-ins/outs over a duration of 291 ticks offered the best configuration with the least likelihood of mid-air collision occurring.

Electronics doi: 10.3390/electronics15215109

Authors: Sergio Villanueva López Emilio Soria-Olivas Manuel Sánchez-Montañés

In industrial visual inspection, individual components may look normal even when their arrangement or quantity is incorrect. For example, a cable may be plugged into the wrong slot of a connector, although every part in the image is intact. These logical anomalies are difficult for methods such as PatchCore, which score image regions independently, while the vision-language pipelines that can reason about them are slow and costly to deploy on a production line. In this work, we present Positional Contextual Regression (PC-Reg), which is a method that learns from defect-free images how each region relates to its surroundings, without gradient-based training. A frozen vision backbone provides the patch features, and a closed-form ridge regression at each grid position predicts every patch feature from its neighbors; the Mahalanobis distance of the prediction residual, calibrated per position, measures how far a region departs from its expected context. The attention of the backbone weights the neighbors, and a complementary score captures the content of the four image quadrants. With one fixed configuration across 32 categories of MVTec LOCO AD, MVTec AD, and VisA, the combined method, PC-RegDual, reaches 85.7%, 98.8%, and 92.6% AUROC (87.5%, 99.0%, and 93.5% with a larger backbone). On the same features, it outperforms a nearest-neighbor baseline that follows PatchCore on MVTec LOCO AD by 5.3 percentage points overall and by 14.6 percentage points on logical anomalies, and it outperforms PaDiM and a mean-subtraction baseline on all three benchmarks. Diagnostic feature-space tests show that PC-Reg responds to broken spatial relationships that independent patch scoring misses, and an evaluation on production-line images (BTAD) gives results comparable to the best baselines. On a consumer GPU, fitting takes about 5 s per category (72 s on one CPU core) and inference 55 ms per image, with deterministic output and no external services, so new products can be set up in minutes on local hardware.

Electronics doi: 10.3390/electronics15215108

Authors: Şahin Korkmaz Mustafa Alkan Aynur Koçak

Detectors of text generated by large language models can achieve high scores by exploiting shared source documents, data sources, or generator-specific artifacts rather than genuine linguistic signals. This is a serious concern for textual deepfakes written to mislead readers or impersonate human authors. We present a methodological audit of a Turkish human-versus-synthetic text classifier. The archived BERTurk model reproduced its near-perfect row-random test result (F1 = 0.9990), but 86.38% of test samples shared a root document with the training data. We then built a corrected 215,074-sample cohort from a single generator (Qwen2.5-7B-Instruct) and four development sources, using a group-disjoint split and three seeds. Within this cohort, BERTurk achieved F1 = 0.9958 ± 0.0001 (XLM-R: 0.9923 ± 0.0006), whereas cross-source testing lowered F1 to 0.8990. In a separate, prospectively frozen extension, nine new BERTurk models evaluated on held-out generator families reached a mean macro-F1 of 0.9379. However, on 500 independent Turkish Wikipedia texts, the same nine models showed a mean false-positive rate of 0.7384 and a balanced macro-F1 of 0.5500, failing all external-validity criteria. High in-cohort scores therefore do not indicate general performance beyond the evaluated generators and sources. The main contribution is a traceable evaluation protocol that separates historical, corrected, and failed-validity results.

Electronics doi: 10.3390/electronics15215107

Authors: Xin Zhang Xuewei Xiang Hui Li Gongcheng Zou Zhi Chen Cheng Yang Yifan Jian

Accurate assessment of shielding sleeve failure severity is essential for ensuring the safe and reliable operation of reactor coolant pump motors. However, local shielding sleeve failures induce only weak variations in conventional motor operating characteristics, while strict safety constraints make it difficult to obtain sufficient actual failure samples. Moreover, distribution discrepancies among different data domains further limit the assessment accuracy and generalization capability of conventional data-driven methods. To address these issues, this paper proposes a shielding sleeve failure severity assessment method based on multi-source domain-adaptive adversarial transfer learning. First, parametric finite-element models of shielding sleeve bulging and wear are established. By analyzing the sensitivity of different operating characteristics to shielding sleeve degradation, eddy current loss is selected as the failure-sensitive indicator, and mechanism-based mapping models between eddy current loss and the bulging displacement and wear depth are derived. Second, a multi-source domain-adaptive adversarial transfer network is developed. A shared–private feature extraction architecture is employed to jointly represent common degradation information and source-specific characteristics, while domain-adversarial learning is used to reduce the feature-distribution discrepancy between the source and target domains. The prediction results from multiple source domains are further adaptively fused according to their distribution relevance to the target domain. Finally, the proposed method is validated using both simulation and prototype experimental data. The prediction accuracy exceeds 96.8% for all simulation-based transfer tasks, with an overall improvement of 8.6% compared with the single-source adversarial transfer model. When experimental data are used as the target domain for simulation-to-experiment transfer, the prediction accuracy reaches 94.7%. These results demonstrate the effectiveness and cross-domain generalization capability of the proposed method for shielding sleeve failure severity assessment under limited target-domain labeling conditions.

Electronics doi: 10.3390/electronics15215106

Authors: Zhiyu Li Lulu Zhang Xiaoyang Wang Lili Hu Zhilei Zhang Cheng Zhang Changwei Zhang

As wireless communication technologies continue to evolve, malicious jamming becomes a major factor that limits system performance and communication quality, especially in emergency communication scenarios where reliable links are essential for timely information delivery and rescue coordination. Conventional anti-jamming transmission methods often suffer from two issues. In the detection stage, the jamming threshold is difficult to determine. In the data transmission stage, spectrum resources are not efficiently utilized. To address these limitations, this paper jointly considers jamming detection and data transmission, and proposes a joint jamming avoidance and suppression method based on intelligent detection threshold adaptation. Specifically, reinforcement learning is first used to iteratively optimize the detection threshold according to transmission outcomes, which improves the accuracy and robustness of jamming detection. Then, based on the detection results, the jamming status is dynamically evaluated, then jamming avoidance and jamming suppression are jointly employed during data transmission, which effectively improves spectrum utilization. Simulation results show that, under both continuous-band jamming and comb-spectrum jamming, the proposed scheme nearly doubles the system sum rate and also achieves significantly higher detection accuracy than conventional methods, which demonstrates strong robustness and adaptability for emergency communication systems.

Electronics doi: 10.3390/electronics15215105

Authors: Ioannis Kansizoglou Vasileios Sevetlidis Christos Kokkotis Panagiotis Siaperas Antonios Gasteratos

Sign Language Alphabet Recognition (SLAR) is a critical component of inclusive Human–Computer Interaction (HCI), particularly for facilitating communication with Deaf and Hard-of-Hearing (DHH) individuals. The paper at hand proposes an attention-enhanced lightweight Convolutional Neural Network (CNN) architecture for interpretable SLR, exploiting the Convolutional Block Attention Module (CBAM) to improve both classification performance and model transparency. The introduced method is assessed on the Sign Language MNIST dataset, where the integration of attention mechanisms improves classification accuracy compared with the corresponding attention-free CNN configurations. Additionally, qualitative attention maps are presented to visualize the spatial regions emphasized by the model during classification, complemented by a quantitative attention-perturbation analysis assessing the functional relevance of these regions. Furthermore, a comprehensive analysis is conducted regarding the computational cost introduced by the CBAM integration, demonstrating a favorable trade-off between accuracy, interpretability, and model complexity. A comparative analysis with representative methods evaluated on the same dataset is also presented, contextualizing the proposed architectures in terms of recognition performance and computational efficiency. This study underscores the potential of attention mechanisms for developing more transparent and inclusive AI systems using static SLAR, providing a foundation for future real-world HCI and assistive communication applications. From an assistive technology and occupational therapy perspective, such recognition systems may support the development of accessible Augmentative and Alternative Communication (AAC) interfaces aimed at reducing communication barriers and facilitating participation of DHH individuals in everyday activities.

Electronics doi: 10.3390/electronics15215104

Authors: Jangmook Kang Seongwon Bae Jongbin Park Jangwoo Park Cheolhee Yoon

Autonomous mobile robots (AMRs) that depend on remote inference introduce communication and service availability into the navigation control path. This paper presents a cloud-independent mapless-navigation architecture integrating visual features distilled offline from a vision–language model, PointNet++ light detection and ranging (LiDAR) features, and a Twin Delayed Deep Deterministic Policy Gradient (TD3) controller. An 8-bit integer (INT8)-quantized Vision Transformer student processes red–green–blue (RGB) images on an NVIDIA Jetson Orin Nano, while late fusion projects visual and geometric features into a common 512-dimensional policy input. Runtime navigation uses RGB and LiDAR observations without natural-language commands. A four-class threat model distinguishes the remote-inference interfaces removed by this arrangement from residual sensor, local-interface, and supply-chain risks. The evaluation combines public benchmark inputs with a fixed-seed synthetic profiling sequence and reports descriptive timing, power, and path-deviation summaries. Analytical checks establish the projection dimensions, nominal weight-storage reduction, and power–time calculations. The results provide a descriptive characterization of this integration, with sensor-attack mitigation reserved for future evaluation. The contribution is an explicit perception–control architecture, an analysis of its remote-inference dependencies, and a conceptual mapping from deployment objectives to security and safety frameworks.

Electronics doi: 10.3390/electronics15215103

Authors: Bailin Li Xin Wei Siyuan Ren Nan Wang

Electrical single-line diagrams contain symbols with substantial scale variation, thin-line structures, hollow contours, and dense layouts, posing challenges for automatic symbol detection. This study proposes SAFE-YOLO, a YOLOv11s-based detector that integrates local structure modeling, spatial–channel interaction, and multi-scale feature fusion. The structure–frequency-enhanced C3k2 module (C3k2_SF) combines adaptive sampling with frequency-aware channel recalibration. Spatial and channel-wise interactive attention (SICA) introduces bidirectional modulation between spatial and channel branches, while an adopted parallel asymptotic feature fusion network (PAFFN) supports cross-scale feature aggregation and recalibration. Experiments on a dataset of 543 electrical single-line diagrams covering 24 symbol categories showed that SAFE-YOLO achieved an mAP@0.5 of 93.6 ± 0.2%, compared with 88.6 ± 0.1% for YOLOv11s. Measured throughput decreased from 117.6 to 74.1 images/s, while mean latency increased from 8.4 to 13.6 ms per image. These results indicate improved detection accuracy under the adopted evaluation protocol, accompanied by increased inference time. SAFE-YOLO provides a task-oriented option for offline electrical-drawing processing when the additional processing cost is acceptable.

Electronics doi: 10.3390/electronics15215102

Authors: Khalaf Alsalem

The task of classifying clinical case reports by their specialty is challenging because of the heterogeneity of the text in the reports, overlapping terminology of specialties, class imbalance, and the use of proxy labels in the biomedical literature where the reports are located. We tested a MultiCaRe-CERGFormer multi-view classifier, which fuses word- and character-level TF–IDF features obtained via truncated SVD, precomputed biomedical document embeddings, and specialty-specific concept indicators and features learned from similar cases in the training partition. A gated fusion module is provided to fuse the encoded branches for multiclass prediction. There were 8564 case reports, assigned to 13 specialty categories created from the metadata. The models were evaluated with five-fold grouped cross-validation, where the reports in the same source article belong to the same fold. The proposed model achieved a mean accuracy of 0.8152 ± 0.0070, a macro-F1 of 0.6693 ± 0.0179, a weighted-F1 of 0.8147, a macro-AUC of 0.9706 ± 0.0049, and a macro-average precision of 0.7430 ± 0.0227. The mean macro-F1 (0.6705 ± 0.0264) of the word- and character-level TF–IDF SVM was slightly higher. Therefore, the approach proposed was competitive under the evaluated protocol but was not "leading" on all metrics. Due to the nature of the targets, which were based on metadata of the articles instead of clinician adjudication, these results provide a measure of the prediction of proxy labels and are not meant to indicate diagnostic validity, clinical utility, or external generalizability.

Electronics doi: 10.3390/electronics15215100

Authors: Lin Li Jian Zheng Jialiang Xu Zuliang Lu

Feature extraction effectively captures meaningful information from raw signals or data to improve the accuracy and real-time performance of detection, classification, and prediction. The selection of an effective discriminant projection method is crucial for extracting discriminant features from the raw signals or data. However, some existing discriminative projection methods face two major limitations. First, the fixed graph structure adopted by some of the methods during feature extraction makes them susceptible to noise or outliers. Second, the predefined zero-one label matrix used by some of the methods to guide the model in learning the discriminative projection matrix leads to the loss of some intrinsic information inherent in the data. To overcome these limitations, we propose a framework called discriminative projection with flexible marginal and local relationship learning (DP-FRL). The key innovations of DP-FRL are primarily reflected in three aspects: (1) utilizing a flexible target matrix to guide the projection matrix in learning marginally structured relationships in the data; (2) employing permutation matrices in the local relationship regularization term reduces the dependence of the projection matrix on the order of features, effectively enhancing its ability to flexibly capture local relationships in the data; and (3) introducing a rank-one quadratic penalty to constrain the projection matrix, which helps learn more discriminative projections. To effectively solve the proposed framework model, we designed an iterative optimization algorithm. Finally, the classification experiments conducted on real-world datasets demonstrate that, compared to related methods, the features extracted by the DP-FRL method are more effective for classification tasks.

Electronics doi: 10.3390/electronics15215101

Authors: Sizhe Shen Haobo Li Jinglei Zhang Longjiang Li Hong Yuan Chuntao Chang Xiaoming Wang

The performance of integrated navigation using global navigation satellite systems (GNSS) and inertial navigation systems (INS) in challenging urban environments is highly dependent on the quality of GNSS observations, particularly under multipath and non-line-of-sight conditions. Conventional weighting strategies based on predefined empirical relationships with satellite elevation angle or signal quality are limited in characterizing the nonlinear and time-varying characteristics of GNSS measurement errors. To address this issue, this study proposes a deep learning-based adaptive weighting method for tightly coupled GNSS/INS integration. Satellite elevation angle, signal-to-noise ratio (SNR), observation residuals, and normalized observation residuals are jointly utilized to characterize observation quality, and a neural network is employed to estimate satellite-specific measurement noise parameters for constructing the measurement noise covariance matrix. Since ground-truth measurement noise is difficult to obtain directly, the network is optimized using position, velocity, and attitude errors as the training objective. Vehicle-based experiments were conducted in representative environments, including open-sky areas, tree-lined roads, urban street canyons, and overpass sections. The proposed method achieves east, north, and up position root mean square errors (RMSEs) of 0.38, 0.75, and 1.14 m along route A, corresponding to reductions of 15.6%, 17.6%, and 17.4% compared with the combined elevation angle and SNR weighting strategy. Along route B, the corresponding RMSEs are 0.56, 0.71, and 1.18 m, representing reductions of 25.3%, 43.2%, and 34.1%, respectively. Consistent improvements are also achieved in velocity and attitude estimation. These results demonstrate that the proposed adaptive weighting strategy effectively improves the accuracy and stability of tightly coupled GNSS/INS integration under challenging observation conditions.

Electronics doi: 10.3390/electronics15215099

Authors: Yonghui Wang Hong Qian Wenbin Zhao

Infrared and visible image fusion, as a pivotal image processing technique, aims to integrate complementary information from heterogeneous source images into a high-quality fused representation. However, the intrinsic domain discrepancy between infrared and visible modalities, coupled with the susceptibility to low-light environments, significantly hampers the fusion process and degrades the quality of the results. To address these limitations, this paper proposes PSGAN, a triple-discriminator generative adversarial network incorporating a pseudo-sensing mechanism. Specifically, our method synergistically integrates the distributional alignment capability of generative models with a pseudo-sensing mechanism that simulates the inverse process of perceptual imaging. This approach effectively alleviates the domain discrepancy across disparate imaging physics, empowering the network to extract robust, cross-modal aligned features. Furthermore, an additional illumination discriminator is incorporated into the conventional dual-adversarial framework to guide the fused images toward achieving optimal luminance and contrast, even in extreme darkness. Finally, to comprehensively quantify the performance of fusion algorithms across multiple dimensions, this paper develops a multi-criteria evaluation framework based on Dempster–Shafer (D-S) evidence theory. Experimental results demonstrate that the PSGAN algorithm achieves superior fusion accuracy and strong generalization capability, while simultaneously providing significant performance gains for downstream detection tasks.

Electronics doi: 10.3390/electronics15215098

Authors: Suvish Josiah E. Balota Priyalakshmi Sheela Senthilarasu Sundaram

Sensor capability has outpaced the architectures available to integrate it: in near-body multimodal sensing, the binding constraint is no longer whether a variable can be measured, but whether a platform coordinates heterogeneous channels while preserving timing integrity, contextual interpretability, and extensibility. The gap is acute in ageing-focused research, which needs both side-specific comparison and heterogeneous sensing. We present the Deterministic Embedded Acquisition Platform Biosensing Hub (DeAP-BioHub), a dual-core STM32H755 architecture isolating acquisition from variable-latency subsystems by peripheral ownership. A real-time Cortex-M4 domain performs scheduled sensing, timestamping, and record creation, and a bare-metal Cortex-M7 domain carries telemetry, file-system logging, and operator interaction; the two share only a non-cacheable region and hardware semaphores, exchanging fixed 464-byte records with build-time offset assertions. With telemetry and logging active, the environmental channel held its nominal period at 1000.000 ms, at 1 ms timestamp resolution (SD 0.188 ms, n = 452), 96.5% exactly at nominal; the infrared channel’s 0.517% excess traced to that sensor’s own refresh cycle rather than the scheduler, and no storage-synchronization signature was detectable in the four sessions (combined Rayleigh test, p = 0.99). Contextual sensing shares that scheduling, record structure, and timestamp source rather than sitting in metadata: a fingertip registration was summarized descriptively by a first-order approach (τ = 154 s, R2 = 0.997), and a session reading flat on infrared alone stayed diagnosable from the environmental record. Ageing-focused research is the target application: this is laboratory verification of the acquisition substrate, not evaluation in older adults or clinical validation.

Electronics doi: 10.3390/electronics15215096

Authors: Fan Li Jishuo Qin Yuan Si Binqi Ma Shujun Xin Dan Wang

Finite-energy frequency support through multiple high-voltage direct-current (HVDC) links requires coordination of power delivery, device constraints, and support withdrawal, because maximizing the initial response can exhaust a source too early or deepen the frequency minimum after support ends. This paper develops a two-layer method that separates the energy supplied to a synchronous system by asynchronously connected sources from the power redistributed inside it by embedded links. A local scalar projection enforces the present source and capacitor constraints at every converter, with explicit conditions for discrete invariance, bounded-error sampled execution, finite-rate withdrawal, and the capacitor discharge time. A causal regional linear program allocates support over time and constrains the frequency response after the scheduled withdrawal; its feasibility condition is derived in terms of the remaining source energy, and late calculations or expired instructions revert to local frequency response. In a two-area benchmark with 900 MJ of source energy and a 20 MW internal transfer limit, 5 Hz coordination reduces the maximum frequency depression by 17.38% and 13.83% relative to independently searched static gains, and three-phase average-value electrical realization retains 16.57% and 13.25%. In 70 newly executed plant-perturbation cases with load damping, governor, secondary, inertia, and tie-line parameters varied by up to 20%, the controller improves on fixed droop in every case and on the searched static gains in 63, without device-limit violations; in a three-area extension with three stations the reductions are 6.60% to 17.55% relative to searched static gains and 38.04% to 42.05% relative to fixed droop. Measured calculation times identify the linear program as the only real-time bottleneck and quantify how its size grows with the number of areas and stations.

Electronics doi: 10.3390/electronics15215097

Authors: Jawon Koo Eunki Kim Hankyu Lee Han-Bin Lee Dongsup Song

In high-bandwidth memory (HBM) testing through the direct access (DA) port, the response-path bandwidth is far lower than the internal data bandwidth. Because most read responses are fault-free, transferring every response incurs substantial transfer overhead and test time. This paper proposes three fault-data transfer methods. Selective staggered read suppresses fault-free response transfers and overlaps faulty-response transfers with the subsequent read latency. Duplicate fault suppression replaces identical fault responses at consecutive addresses with a short duplicate code. Failing-bit indexing transmits only the positions of failing bits. The proposed architecture controls whether, when, and in which format each response is transferred without modifying the DA port interface or the March command and address sequence. Cycle-accurate simulation of an HBM2 test model shows that, across Random, Column-Correlated, and Row-Correlated Fault models with fault rates of 0.01–5%, the proposed method reduces the DA response data by 99.32–99.99% and the total test time by 44.86–45.01% compared with the conventional method. Synthesis shows that the added logic is smaller than that of on-chip fault collection and signature compaction schemes and independent of the memory capacity. The proposed architecture attains nearly the same test time as on-chip fault collection and signature compaction schemes while delivering every faulty response and its failing-bit positions to the tester, so that its diagnostic coverage is not bounded by on-chip storage.

Electronics doi: 10.3390/electronics15215095

Authors: Anna Drabczyk Paweł Uss Katarzyna Bucka Paula Mazur Robert P. Socha

Thin-film capacitors and supercapacitors are gaining increasing attention due to the rapid development of modern electronics, flexible devices, microsystems and wearable technologies. Their small size, low weight and compatibility with integrated electronic systems make them promising components for next-generation applications and energy-storage systems. In recent years, significant progress has been achieved in the development of thin-film structures with improved capacitance, higher energy density, lower leakage current and better thermal and electrical stability. Advances in thin-film engineering have also enabled the use of high-k dielectric materials, ultrathin insulating layers, oxide-based electrodes and advanced carbon materials to enhance device performance. This review presents recent progress in thin-film capacitor and supercapacitor technologies, focusing on device architectures, dielectric and electrode thin films and reliability-related challenges. Different configurations, including metal–insulator–metal (MIM), metal–oxide–semiconductor (MOS), interdigital, flexible and three-dimensional structures, are discussed. Special attention is given to oxide thin films, nanostructured electrodes, interface quality, defect-related phenomena and scaling effects influencing electrical performance. Current market trends and commercial perspectives are also discussed in the context of emerging electronic applications. Finally, key technological challenges and future development directions for thin-film capacitors and supercapacitors are highlighted.

Electronics doi: 10.3390/electronics15215094

Authors: Huy Nguyen Loc Cao Nguyen Nam Hoang Nguyen

Radio frequency (RF) for wireless systems is widely installed across applications such as mobile communications, maritime networks, smart factories, and smart homes because of its easy setup compared with wired systems. Nevertheless, potential health concerns associated with exposure to RF waves have driven interest in optical camera communication (OCC) using visible light as a safer alternative. Concurrently, artificial intelligence (AI) is transforming automation and complex problem-solving across IoT, smart farming, and smart home ecosystems. In this study, we propose a Camera On–Off Keying (C-OOK) scheme that incorporates YOLO26n for multiple users, which supports light source detection and tracking, along with a denoising autoencoder (DAE) algorithm to assist the decoding process in long-range and mobile environments. This approach improves the traditional C-OOK by supporting multiple users simultaneously, increasing distances, and lowering the bit error rate (BER). Experimental results demonstrate reliable communication up to 15 m in a mobile environment.

Electronics doi: 10.3390/electronics15215093

Authors: Zhian Long Riwei Chen Jinjiang Zhou Yong Ge Guozhi Yang Xin Che

Wind power prediction is important for utility grid energy scheduling. However, prediction accuracy suffers from poor data quality and limited feature engineering. Therefore, this paper proposes a data-driven ensemble learning framework to deal with these issues. First, the paper divides raw data into four seasonal groups. A preprocessing pipeline integrating the 3σ rule, Hermite interpolation, and a wind power curve modeling method detects and corrects outliers. Second, we develop a two-stage prediction strategy. In the first stage, a hybrid deep learning model including convolutional neural networks, gated recurrent units, and an attention mechanism jointly learns spatio-temporal features, with hyperparameters optimized using the exponential-trigonometric optimization (ETO) algorithm and the input length determined using the partial autocorrelation function. In the second stage, a temporal convolutional network with ETO-optimized hyperparameters predicts the error sequence, capturing multi-scale nonlinear relationships between input features and error dynamics. Experimental results on real-world wind farm data demonstrate that the proposed method achieves a mean RMSE of 5.89±0.11 MW, MAE of 4.05±0.09 MW, MAPE of 7.24±0.18%, and R2 of 0.928±0.008 over 50 independent runs. A representative single run achieves an RMSE of 5.89 MW, MAE of 4.02 MW, MAPE of 6.9%, and R2 of 0.93.

Electronics doi: 10.3390/electronics15215092

Authors: Tania Tanzin Hoque Cátia Silva Md Tofael Ahmed Tiago Pinto

Deep learning models are increasingly used for time-series applications; however, they introduce significant computational challenges, particularly when handling the largest workload datasets. Long Short-Term Memory (LSTM) networks are widely used in time- series applications because they can learn sequential patterns and time-dependent relationships within data. Nevertheless, LSTM training is computationally expensive and time-consuming. This work investigates the performance of LSTM models across heterogeneous computing environments, focusing on the impact of sequential and parallel execution strategies. Experiments were conducted on multiple platforms with different Central Processing Unit (CPU) and Graphics Processing Unit (GPU) hardware using a 13-year (4748 observations) original dataset synthetically expanded to 47,480 observations to evaluate computational workloads rather than additional real-world observations. The results show that task-level parallel execution reduces wall-clock execution time across the tested systems, while predictive performance remains broadly comparable across the tested configurations. Furthermore, the GPU-based implementation achieved lower execution times than the CPU-based implementation in the tested configurations. Overall, this study highlights the importance of HPC and parallel computing for accelerating deep learning workflows in time series applications. The findings confirm that computational performance is highly dependent on hardware capabilities and execution strategies, although accuracy remains stable across heterogeneous platforms.

Electronics doi: 10.3390/electronics15215091

Authors: Shensheng Tang

This paper investigates adaptive, cost-based probabilistic offloading between two coexisting wireless systems: a type-1 system and a type-2 system that opportunistically exploits idle type-1 resources. The coupled dynamics are modeled as a finite level-dependent quasi-birth–death (LD-QBD) process with explicit block structure, enabling exact stationary evaluation. From the stationary distribution, we derive teletraffic-consistent metrics including blocking, dropping, utilization, throughput, and cost-based performance measures. Building on these metrics, we develop a windowed, adaptive admission controller that combines KPI smoothing, cost-aware probabilistic mapping, PI-based regulation, and safety backoff mechanisms. Numerical results demonstrate that the proposed controller effectively balances throughput, admission opportunity, and preemption risk under changing traffic conditions while achieving the lowest overall blocking-and-drop cost among the practically useful policies considered. The framework provides an analytically grounded framework for investigating adaptive opportunistic offloading in wireless networks.

Electronics doi: 10.3390/electronics15215090

Authors: Ghita Hana Mourad Hana Abdelmjid Saka

Smart Universities increasingly rely on connected learning platforms, smart classrooms, Internet of Things (IoT) devices, cloud services, and digital administrative systems. This integration expands the attack surface and creates security requirements involving educational data, heterogeneous infrastructure, and the continuous availability of academic services. This survey examines AI-based intrusion detection through three complementary perspectives: campus-specific threats and architectural layers, detection approaches and their deployment trade-offs, and operational evaluation and governance requirements. It reviews machine learning, deep learning, ensemble and hybrid models, lightweight detection, federated learning, and explainable AI, relating their capabilities and limitations to device, edge, network, cloud, and application layers. The analysis highlights gaps between benchmark performance and evidence of suitability for operational university environments, including limited campus-specific validation, incomplete reporting of latency and resource requirements, and insufficient assessment of privacy safeguards and explanation usefulness. Conventional detection metrics remain necessary but should be complemented by false-alert workload, end-to-end latency, temporal robustness, communication costs, and analyst-centered evaluation. The survey provides a campus-oriented synthesis, an operational evaluation matrix, and a staged research and deployment roadmap prioritizing documented datasets, accountable data governance, baseline assessment, and monitored pilots before broader federation or automated response. AI-based intrusion detection offers promising capabilities for Smart Universities, but its practical value requires validation under representative campus conditions and alignment with institutional resources, privacy obligations, and incident-response processes.

Electronics doi: 10.3390/electronics15215089

Authors: Juan Lata-Garcia Luis Andrade Hover Torres

This study evaluates the impact of the integration of non-synchronous distributed generation on the frequency stability, rotor-angle response and dynamic performance of the electrical system associated with Substation 50 La Troncal (Ecuador). The network was modeled in DIgSILENT PowerFactory 2024 with six busbars, and eight renewable-penetration scenarios (15–50% of the 32.8 MW demand) were assessed under N-1 (loss of the COAZUCAR TG3 generator) and N-2 (TG3 plus the Ocaña 2 line) contingencies applied at t = 1 s over a 20 s horizon. Steady-state results showed voltage profiles within acceptable ranges and no overloads, whereas the dynamic analysis revealed non-compliances, with average ROCOF of 0.772 Hz/s, frequency nadir of 59.192 Hz and angular deviation of 8.75°. A battery energy storage system (BESS) with virtual inertia was then implemented, improving the nadir to 59.672 Hz, the ROCOF to 0.220 Hz/s and the angular deviation to 6.10°. An ANSI 51 overcurrent scheme (Siemens SIPROTEC 7SJ6001) installed on the 69 kV line at B3 was verified with operating times compatible with the protection requirements. The combination of energy storage, virtual inertia and conventional protection constitutes a technically viable alternative to increase non-synchronous distributed generation while preserving stability and operational security.

Electronics doi: 10.3390/electronics15215088

Authors: Hyeon Woo Kim Hyun Jun Kim Meong Hun Lee

Smart barns rely on vision, wireless, and environmental sensing systems for continuous livestock monitoring, but sensing reliability may vary with cattle behavioral context. This study investigates the associations between herd-level behavioral contexts and multimodal sensing degradation and proposes a context-aware recovery framework using a Sensing Reliability Degradation Index (SRDI). Data collected from 12 cattle in a single semi-open barn over 14 days were organized into 2016 non-overlapping ten-minute windows and classified into five contexts: Normal, Feeding, Moving, Resting, and Crowding. Feeding and Crowding were associated with increased visual occlusion and wireless degradation, whereas Moving exhibited greater tracking instability. Mean SRDI before recovery ranged from 0.19 in Resting to 0.69 in Crowding. In an offline evaluation, the proposed framework increased data completeness from 92.4% to 97.1%, detection continuity from 84.6% to 91.8%, and packet delivery ratio from 93.1% to 97.6%, while reducing mean transmission delay from 176 to 134 ms. Mean SRDI decreased from 0.32 to 0.23, representing a relative reduction of 28.1%. Compared with a context-independent adaptive baseline, SRDI decreased from 0.28 to 0.23. Independent manual validation of behavioral context labels yielded an accuracy of 89.3% and a macro-averaged F1-score of 0.88. Statistical and robustness analyses further supported the reported differences under the evaluated conditions. These findings suggest that behavioral context can provide useful supplementary information for sensing reliability assessment and recovery selection. However, the single-barn observational dataset, shared vision-derived inputs, and offline log-based evaluation limit causal interpretation and generalizability. Future work should validate the framework using multi-site studies and prospective real-time deployment.

Electronics doi: 10.3390/electronics15215087

Authors: Antoni Lenart Grzegorz Rzym

Software-Defined Networking (SDN) exposes fine-grained flow- and port-level telemetry that is attractive for machine learning detection of Distributed Denial-of-Service (DDoS) attacks in the programmable edge networks envisioned for 5G and 6G. A recurring difficulty is that the labels used to train such detectors are themselves produced by a signature-based Intrusion Detection System (IDS) whose coverage is uneven across attack types, so it is unclear how much of a detector’s quality is limited by the labels rather than the telemetry. We investigate this question within a virtualised 5G mobile edge testbed based on Open5GS and UERANSIM, incorporating a new Security Module integrating Open vSwitch (OVS), the Ryu controller, and the Snort IDS. We generate benign traffic and four DDoS attacks—TCP SYN, UDP and ICMP volumetric floods and the application-layer GoldenEye HTTP keep-alive exhaustion—with a scheduler that balances the number of occurrences per attack type, and we label every one-second sample twice: once from Snort alerts and once from the simulator’s ground truth. Because a quarter of the sixty-seven telemetry features are ten-second rolling means or first differences and attacks span many consecutive seconds, a sample-level split leaks temporal structure between training and test; we therefore evaluate with a window-grouped protocol that holds out entire attack occurrences, and we use a fourth campaign, collected independently and held out in full, as the principal test. Four classifiers—logistic regression, a support vector machine with a radial-basis kernel, a random forest, and XGBoost (gradient-boosted trees)—are compared. Under window-grouped cross-validation on the three development campaigns, XGBoost reaches a macro-averaged F1 of 0.92: the volumetric attacks are detected almost perfectly, while GoldenEye is markedly harder (F1 around 0.66) because at the flow level its telemetry overlaps benign traffic. Training on ground-truth rather than IDS-derived labels improves macro-F1 for every classifier on a common within-pool test set and raises GoldenEye F1 by up to 0.55, largest precisely where the IDS is blind, as Snort’s second-level coverage of GoldenEye is an order of magnitude below that of the volumetric floods; a window-level paired bootstrap over held-out attack occurrences places the macro-F1 and GoldenEye-F1 gains above zero with 95% confidence for the random forest and XGBoost detectors, a per-sample McNemar test being reported only as a complement. A detector trained on three campaigns largely transfers this performance to an independently collected fourth campaign of the same testbed, with macro-F1 slightly lower (0.90 versus 0.92) and a clear drop for GoldenEye (F1 0.56 versus 0.66), so the result reflects robustness across independent runs of one testbed rather than broad generalisation. The findings indicate that, in this setting, IDS-derived label quality is a first-order determinant of SDN-telemetry DDoS detection for the application-layer GoldenEye attack; for the volumetric floods, whose IDS coverage is near-complete (91–99.6%), the label regime matters little, and that grouped, cross-campaign evaluation is necessary to measure their real-world behaviour.

Electronics doi: 10.3390/electronics15215086

Authors: Raddad Almaayn Ashok Karukutla Henok Berhanu Tsegaye Petro Mushidi Tshakwanda Harsh Kumar Michael Devetsikiotis

In additive manufacturing (AM) supply chains, systems that record provenance and score reputation as separate transactions let lifecycle records and assessments diverge. We present a Hyperledger Fabric chaincode that commits a provenance event, the Bayesian reputation update it implies and their link in one read–write set. An on-chain policy admits ratings only from counterparty organizations; reputation gates authorize future activities but never block recording outcomes. Under client crashes in the measured partial-write window, the unified design produced no divergence in 3900 trials (2900 with version 1.0, 1000 with version 1.1), while a client-orchestrated two-chaincode deployment diverged in all 1000 trials of two control batches. A same-channel InvokeChaincode arrangement also produced none in 1000, at 4.5 ms more endorsement and 11.7% lower throughput at 100 transactions in flight. On four desktops, open-loop load saturates near 12 tx/s, and a 12-h soak at 6 tx/s committed 98.7% of submissions with no partial record found. Ledger experiments cover gate refusal, failed outcomes, recovery and disputes; with client retry, 540 contended ratings each committed once. A simulation quantifies gate errors and shows that evaluation-time decay reacts slowly when active actors’ performance drops. The scores record submitted outcomes; they are not validated measures of performance.

Electronics doi: 10.3390/electronics15215085

Authors: Mohammed M. Alammar Muneer Parayangat Mohammed Sayeeduddin Habeeb

Ultra-Reliable Low-Latency Communication (URLLC) systems require extremely reliable packet delivery under stringent latency constraints, particularly in mission-critical wireless applications such as industrial automation, autonomous transportation, and intelligent control systems. In such systems, communication reliability is jointly influenced by finite-blocklength decoding behavior, stochastic wireless channel variations, queue evolution, and packet deadline violations. This paper develops a Lyapunov-based cross-layer transmission control framework for finite-blocklength URLLC systems with deadline-constrained queues and stochastic traffic arrivals. The proposed method jointly models short-packet communication reliability, queue dynamics, packet expiration, and adaptive transmission parameter selection within a unified mathematical formulation. An end-to-end (E2E) reliability metric incorporating both physical-layer (PHY) decoding failures and queue-induced deadline violations is introduced. In order to satisfy long-term reliability and power constraints while preserving queue stability, virtual queues are constructed and integrated into a Lyapunov drift-plus-penalty optimization framework. Based on the proposed formulation, an online transmission control algorithm is developed for adaptive selection of transmission blocklength, coding rate, and transmit power according to instantaneous queue and wireless channel conditions. Unlike reinforcement-learning-based methods, the proposed framework does not require offline training or prior statistical knowledge of traffic and channel distributions. The analysis establishes bounded real-queue occupancy under capped admissions and deadline expiration, and shows that mean-rate stability of the virtual queues implies the corresponding surrogate-risk and power constraints. Classical Lyapunov scaling is discussed as a conditional benchmark rather than an established optimality guarantee for the implemented urgency-weighted controller. Simulation results show that the proposed framework reduces time-average transmit power by approximately 33.0% relative to Adaptive-FBL, 42.3% relative to EC-FBL, 46.7% relative to the simplified Lyapunov baseline, and 71.6% relative to fixed transmission, while maintaining packet-level reliability, delay, and queueing performance close to the adaptive benchmarks. Simulation results show that the proposed approach achieves a favorable balance between reliability, latency, and energy consumption across different traffic conditions.

Electronics doi: 10.3390/electronics15215084

Authors: Lijuan Wang Krassie Petrova Mee Loong Yang

Software-defined wireless sensor networks (SDWSNs) are deployed in mission-critical applications such as environmental monitoring, smart cities, and healthcare. However, existing protocol architectures such as SDN-WISE lack built-in mechanisms for the detection or mitigation of Denial-of-Service (DoS). Existing trust-based security solutions are rarely fully integrated into the protocol stack; nearly all of them rely on manually tuned attack-specific detection thresholds and metric weights, which severely undermine their effectiveness. This study integrates the adaptive trust model ATAW-TM into the SDN-WISE protocol stack; the ATAW-TM model was adjusted to the resource-constrained sensor node environment by embedding three lightweight adaptations that void the invocation of the exponential and logarithmic software library functions: an inverse-quadratic kernel of Gaussian-like cooperation probability, a quadratic form of second-order Rényi (replacing Shannon entropy) for entropy weight, and a softsign for the aging factor. To evaluate the integrated model, we developed a comprehensive evaluation framework encompassing 14 distinct DoS attack types and the data tampering attack against SDWSNs. The integrated model was evaluated through extensive Cooja-based simulations. The experimental results indicate that the proposed security solution effectively addresses the majority of the DoS attack range and the data-tampering attack through attack detection, abnormality identification, and mitigation. However, for grey hole and sinkhole attacks, the attacker’s trust value stabilizes at a level lower than that of legitimate nodes but above the detection threshold; second, the detection precision of flooding-type attacks is relatively low and may result in victim nodes being wrongly classified as malicious. These security enhancements are achieved with only negligible performance overhead, introducing less than 5.3% additional hop delay and 8.1% additional forwarding delay compared with the standard SDN-WISE. Furthermore, simulation results show that the proposed method can adapt to variations in network traffic conditions without requiring manual threshold tuning. Overall, the proposed lightweight trust-based attack recognition model offers a practically achievable security solution for resource-constrained SDWSNs, without sacrificing protocol efficiency.

Electronics doi: 10.3390/electronics15215083

Authors: Wyssem Fathallah Dhaou Bouchouicha Mongi Ben Moussa Abdelilah Mansour Hedi Sakli

Compact ultra-wideband (UWB) multiple-input multiple-output (MIMO) front-ends must stay matched over a wide continuum, reject licensed 5G interferers and remain uncorrelated at sub-wavelength spacing. This paper formulates coexistence as the inversion of a mask M = (N(f), I(f), rho-max, eta-min) and treats complementary split-ring resonators (CSRRs) and a split-ring resonator (SRR) fence as operators acting on electrically distinct regions. Sufficient conditions for a conditionally factorised inverse are derived (bounded interaction on the support of N(f), bounded cross-sensitivity and a well-conditioned rejection Jacobian), together with a first-order bound on the inversion error. The rejection factor is implemented by inverse regressors with a forward-consistency check and an out-of-domain detector (random forest and k-nearest neighbours, R-squared = 0.9825 and 0.9921 on unit-cell data); non-uniqueness is handled by selecting, among forward-consistent candidates, the geometry with the largest fabrication margin. While ten target specifications are queried across the C-band and WLAN windows, only the two fabricated unit cells have undergone dedicated full-wave and experimental validation, whereas the remaining eight are compared against their nearest training-set neighbours, leaving multi-target generalisation across the specification space only partially validated. A two-element FR4 prototype designed as an experimental testbed for the union of a C-band and a WLAN mask, with 10 mm edge-to-edge spacing (0.07 wavelength at 2.1GHz and 0.097 at the 2.9GHz lower matched edge in full wave), is measured and scored against the mask. In measurement, frequency-domain coupling stays below −20 dB from 2.1 to about 14.8GHz (full wave: 67% of the band). Rejection is the weak coordinate: the measured reflection of the two-port board peaks at about −9.5 dB in both stop-bands, against −2.9 and −2.0 dB in full wave, so the −10 dB criterion is met over about 44% of the combined mask, and the C-band mask is covered only from about 3.9GHz. This partial compliance exposes the effects of uncompensated unit-cell-to-aperture transfer offsets, stop-band bandwidth mismatch and FR4 Q-factor damping, showing that the prototype demonstrates the diagnostic scoring pipeline rather than a fully compliant hardware realisation of coexistence-mask inversion. The four-configuration test, carried out in full wave at the fabricated design, shows an interaction residual (0.61 in magnitude norm) far above the rejection-margin threshold (0.019), so condition (C1) is not met and factorised inversion remains a strictly conditional formulation that is not experimentally validated by the current prototype. Moreover, because pattern-based correlation is evaluated only from simulated patterns at six spot frequencies (ECC at most 0.015) without measured pattern validation, and because group delay and fidelity factor have not been assessed, wideband performance claims are restricted to continuous-wave frequency-domain matching and isolation. The changes needed to close this gap, the tolerance budget of both cells and the validation that remains open are stated explicitly, and modelled results are separated from measured ones.

Electronics doi: 10.3390/electronics15215082

Authors: Konstantinos Manolakis-Sempagios Dimitra Georgiou Ilias Maglogiannis

In today’s world of rapidly evolving internet communication, email remains a widely used medium for direct message exchange across ordinary users, organizations, and defense-related environments, where spam campaigns can threaten operational continuity and communication integrity. Despite significant advances in spam filtering techniques using deep learning approaches, comparatively less attention has been given to the robustness of traditional machine learning classifiers under cross-corpus distribution shift. This study benchmarks five classical classifiers—Naive Bayes, Logistic Regression, Random Forest, Linear Support Vector Machine (SVM), and XGBoost—for spam email detection. A unified framework incorporating text preprocessing, TF-IDF feature extraction, engineered metadata features, stratified 10-fold cross-validation, and cross-corpus evaluation was applied to the sampled public email corpora, including one LLM-generated synthetic corpus (Dataset C). Classifier performance was assessed using standard classification metrics, dependence-aware statistical testing, retention-based analysis, and confusion-matrix evaluation. Corpus characteristics were examined using lexical diversity, vocabulary overlap, TF-IDF sparsity, and metadata distribution analyses. Results showed that Linear SVM achieved the highest within-corpus F1-score in the sampled conventional corpora (0.9868 on Dataset A), but this ranking did not consistently carry over to the tested transfers, where performance varied by classifier, feature set, and direction. XGBoost with combined features achieved an F1-score of 0.9233 in the A-to-C transfer, while Linear SVM achieved 0.9437 in B-to-C; these results do not establish general model rankings. Metadata-only scores above the within-corpus baseline in some transfers were consistent with the target class composition, including all-spam predictions, and do not establish a transferable metadata effect. The findings are limited to the sampled corpora and experimental settings; transfer to Dataset C does not establish generalization to LLM-generated spam more broadly.

Electronics doi: 10.3390/electronics15215081

Authors: Magdalena Budnarowska Erika Stracqualursi Giuseppe Attolini Giampiero Lovat Jerzy Mizeraczyk Salvatore Celozzi

Cylindrical shields are worthy of a specific investigation because of their diffusion in practical applications and of the analytical and numerical peculiarities. Main applications are related to cables, electromedical machines, drones, and others. The general case of finite-length, penetrable shields with and without apertures is addressed. The results obtained by means of a frequency-domain integral equation model (solved by means of a Method of Moments procedure) are used to infer general considerations in the absence of apertures and to validate the predictions ensuing from a numerical commercial code. The semi-analytical formulation is also useful for a rapid tuning of shield design. Approximate expressions available in the literature are also used for comparison and validation. In this way, the external field sufficient to impair the functionality of electronic devices protected by complex three-dimensional cylindrical enclosures may be easily estimated under the assumption of linearity of the shielding materials.

Electronics doi: 10.3390/electronics15215080

Authors: Jingxi Xie Ran Ran Xiangyi Cai Shiyuan He Jiwei Wei Yang Yang

Video moment retrieval (VMR) aims to localize the temporal segment in an untrimmed video that semantically matches a natural-language query. Although recent detection transformer (DETR) approaches have improved this task, they primarily emphasize query-conditioned cross-modal alignment and often underexploit the intrinsic temporal dynamics of videos. In particular, clip-level features contain frequency-domain cues that reveal how visual content changes across time, yet such cues remain underexplored for moment localization. In this paper, we propose FE-DETR, a frequency-domain expert-driven DETR framework for dynamic-aware video moment retrieval. FE-DETR introduces Multi-span Frequency Experts (MFE), which select expert-specific span lengths from frequency-domain scores of a video and perform expert-specific frequency decomposition to capture multi-scale dynamic patterns. The resulting frequency-enhanced representation is fused with regular cross-modal features for query-video alignment. To further improve localization, we design an Expert-targeted Encoder (ETE) with a distinction loss that encourages query-relevant and background clips to be handled by different expert groups, and we incorporate a multi-expert decoder for fine-grained temporal boundary prediction. On TACoS, FE-DETR achieves a three-seed mean of 40.37% Recall@1 at an intersection-over-union (IoU) threshold of 0.5 and a mean IoU of 37.52%. Experiments demonstrate strong performance on TACoS and competitive results on QVHighlights and Charades-STA.

Electronics doi: 10.3390/electronics15215079

Authors: Qunxin Liu Jun Lin Hao Yuan Yue Liu Yuyi Wang

Function Block Diagrams (FBDs) generated by large language models (LLMs) can compile in an industrial integrated development environment (IDE) while still violating the intended control behavior. We present a runtime diagnosis and repair pipeline that connects a topology with typed pins, PLCopen XML import, CODESYS compilation, and SoftPLC execution. The pipeline returns structured compilation diagnostics and runtime state observations that guide localized repair. Across a benchmark of 20 cases, the full configuration achieves 20/20 compilations on the first pass, compared with 8/20 without retrieval and 5/20 for a naive baseline. Repair raises the setting without retrieval to 13/20 within five attempts. The ablation shows that retrieved function block interfaces are the main driver of first pass compilation, while repair provides complementary recovery. For runtime evidence, a compiled pump cascade program initially passed 2 of 6 diagnostic assertions. After two cycles of diagnosis and repair, it passed all 18 assertions in an expanded final verification suite. Across the benchmark suite, 118/118 executed assertions derived from requirements were satisfied after repair. A targeted defect injection study detected 7/8 defects not visible at compilation across four mechanisms. These results demonstrate a practical path for compiling, testing, and repairing control logic generated through FBD workflows.

Electronics doi: 10.3390/electronics15215078

Authors: Ali Nauman Tahir Khurshaid Sang Bong Rhee

The increasing deployment of low-power Internet of Things (IoT) devices in dense urban and rural environments is significantly constrained by Non-Line-of-Sight (NLoS) propagation, where buildings and terrain obstruct direct communication with terrestrial Base Stations (BSs). This paper investigates a Reconfigurable Intelligent Surface (RIS)-assisted Unmanned Aerial Vehicle (UAV) acting as an aerial relay to establish a virtual Line-of-Sight (LoS) communication link between blocked IoT devices and the BS through passive beam reflection. Since IoT devices operate under stringent energy constraints, the system is evaluated in terms of bandwidth-normalized Energy Efficiency (EE). For fixed numbers of devices and RIS elements and fixed power parameters, maximizing EE over the discrete RIS phase shifts is equivalent to maximizing the sum rate under a fixed power budget, a non-convex combinatorial problem where its search space grows exponentially with the RIS size. To address it without explicit channel modeling, a Q-learning-inspired phase-optimization heuristic is proposed, in which discrete RIS phase configurations are learned for each channel realization from scalar reward feedback only. Simulation results show that, for varying numbers of RIS elements, IoT transmit powers, and active devices, the proposed method achieves higher EE than random phase-shift selection and fixed passive reflection.

Electronics doi: 10.3390/electronics15215075

Authors: Hao Zhang Haiwei Wang Lu Chen Yang Zhang Yu Liu Yong Shi

Ultra-fast charging stations can impose substantial short-term power fluctuations and high transformer capacity requirements on distribution networks. To address the coupled planning and operational challenges of such stations, this study proposes a planning-operation coordinated sizing framework for a photovoltaic-integrated ultra-fast charging station. Unlike many existing approaches that address transformer sizing, energy storage configuration, and battery degradation separately or only partially couple them, the proposed framework jointly determines transformer capacity, battery capacity, and supercapacitor capacity while incorporating battery cycling degradation into the long-term storage cost evaluation. Representative charging scenarios are extracted from historical charging-load data and used for capacity planning. At the planning level, the average daily comprehensive cost is minimized by coordinating transformer investment, hybrid energy storage investment, operating cost, and degradation-dependent battery lifetime. At the operation level, battery and supercapacitor dispatch is optimized considering both operating economy and grid-connected power fluctuation. The upper-level sizing problem is solved using quantum-behaved particle swarm optimization. Sobol-sequence initialization and a nonlinear contraction-expansion coefficient are adopted in the implementation to enhance the distribution of the initial population and adjust the search range during the iterative process. The lower-level mixed-integer linear programming problem is solved using Gurobi. In the studied commercial-area case, the proposed hybrid energy storage configuration reduces the transformer capacity from 326 kVA to 201 kVA. Compared with the no-storage scheme, total costs are reduced by 14.9% and 14.3%, while the absolute deviations of grid-connected power are reduced by 93.4% and 94.8% for the two representative days. Compared with battery-only storage, daily battery cycle counts are further reduced by 60.5% and 65.2%. The results demonstrate that the proposed coordinated sizing framework can simultaneously reduce infrastructure capacity requirements and operating costs, improve grid integration performance, and alleviate battery cycling degradation in PV-integrated ultra-fast charging stations.

Electronics doi: 10.3390/electronics15215077

Authors: Lin Bie Haibo Zhang

The traditional dynamic window approach (DWA) cannot perceive semantic risk levels in substation environments. We propose a lightweight semantic navigation scheme that takes a prior semantic map as input, applies differentiated safety inflation to semantic regions according to category (danger zones dsafety × 1.5, secondary forbidden zones dsafety × 0.5), and fuses them into the Nav2 costmap via per-cell maximum fusion. The standard DWA thereby gains hierarchical avoidance capability at the costmap representation level, without modifying the DWA planner core. Gazebo simulations were performed with 80 runs per group, each collected under an independent simulation restart. Compared with the conventional baseline, the proposed scheme increases the minimum clearance from the robot silhouette to the physical equipment boundaries from 0.42 m to 0.66 m and the minimum signed clearance to the boundary of the pure semantic forbidden zone—a physically empty region marked only in the semantic layer—from −0.36 m (penetration in every run) to +0.46 m. Center-point entries into semantic forbidden zones are reduced to zero (80/80 → 0/80), at the cost of an 11.8% increase in path length. Statistical significance is confirmed by Welch’s t-tests with Holm–Bonferroni adjustment and by non-parametric Mann–Whitney U tests. Comparison with the Nav2 native KeepoutFilter shows that hard boundaries without soft buffering degrade DWA traversability: the right-angle expanded mask failed in all 50 attempts, and the rounded expanded mask succeeded in only 48% (24/50), whereas the proposed scheme maintained 100% (80/80). Ablation, noise, and offset–delay experiments with 30 runs per condition further validate the robustness of the scheme. In the exploratory new-start-point comparison (n = 10 per group), differentiated inflation matched uniform large inflation at the highest-risk device (0.764 m vs. 0.771 m) while keeping the path 2.6% shorter. The proposed scheme provides a loosely coupled, easily deployable semantic navigation safety enhancement for substation inspection robots.

Electronics doi: 10.3390/electronics15215076

Authors: Zhiqi Li Zhenyu Liu

Ultra-wideband (UWB) technology has been widely applied to indoor positioning due to its high-precision ranging. In practical indoor scenarios, inertial measurement units (IMUs) are commonly fused with UWB to suppress measurement errors caused by non-line-of-sight (NLOS) propagation. Frequent switching between line-of-sight (LOS) and NLOS conditions easily generates severe outliers in UWB ranging data. Furthermore, the conventional extended Kalman filter (EKF) lacks an adaptive adjustment mechanism for innovation covariance, which significantly reduces the positioning robustness of the fusion system. To address the above issues, this paper establishes a tightly coupled UWB/IMU positioning model for complex indoor scenarios and proposes an improved robust adaptive extended Kalman filter (IR-AEKF) algorithm. After calibrating UWB ranging errors, IMU bias, and scale factor errors, the system state equation and UWB ranging observation model are constructed under the tightly coupled fusion framework. On this basis, innovation statistics are leveraged to incorporate Huber-type weights and adaptive scaling factors that dynamically tune the observation noise covariance matrix. Meanwhile, a ranging outlier elimination threshold is set to suppress gross errors induced by NLOS interference. Experimental results demonstrate that the IR-AEKF can effectively suppress disturbances arising from abnormal ranging outliers. In dynamic positioning tests, the mean positioning errors along the X-axis and Y-axis are 0.032 m and 0.035 m, respectively. The IR-AEKF provides an effective technical solution for UWB/IMU fusion positioning in complex indoor environments.

Electronics doi: 10.3390/electronics15215073

Authors: Jhonatan Barroso-García Lisbeth Rodríguez-Mazahua Felipe Castro-Medina Asdrúbal López-Chau Giner Alor-Hernández José Antonio Palet-Guzmán

The accelerated growth of medical databases, particularly those storing images in DICOM (Digital Imaging and Communications in Medicine) format, has increased the need for efficient strategies to optimize the retrieval and querying of medical information. This work presents KÁIXAR, a horizontal fragmentation method for medical databases that integrates workload analysis, DICOM images, and content-based retrieval. The proposed method employs operation records (logs) to identify access patterns and, through the Pareto principle (80/20), selects the most representative queries for generating the fragmentation scheme. Subsequently, KÁIXAR proposes and allocates fragments by considering the hardware characteristics of servers located at the sites where the recorded queries originated. As a verification mechanism, the system enables the visualization of fragmented DICOM images and their associated metadata. Experimental evaluation showed scenario-dependent performance benefits. In an additional physical-site validation under a high-workload scenario, KÁIXAR reduced mean query latency from 108.735 ms to 15.062 ms relative to XAMANA, corresponding to an 86.15% reduction and a 7.22× speedup; the paired latency difference was statistically significant (p<0.001). These results support KÁIXAR as an efficient alternative for horizontal fragmentation in medical databases.

Electronics doi: 10.3390/electronics15215074

Authors: Ruixian Wang Rui Ma Qin Yan

With the continuous increase in electric vehicle (EV) charging power and power density, timely and accurate diagnosis of open-circuit faults in power devices is essential for the reliable operation of EV chargers. To address the strong dependence of existing open-circuit fault diagnosis methods on electrical features and their limited generalization across operating conditions, an open-circuit fault diagnosis method for EV chargers employing a power factor correction (PFC) stage and an inductor-inductor-capacitor (LLC) resonant converter is proposed based on real-world charging conditions and multi-device dynamic thermal responses. First, an electrothermal model incorporating device losses, Foster thermal networks, and a state-of-charge (SOC)-dependent equivalent load is established to obtain multi-device junction-to-case temperature-difference responses. Then, representative SOC operating conditions are determined from real-world charging data, and fault simulations are performed to construct a thermal-response dataset. Finally, based on this dataset, a convolutional block attention module (CBAM)-temporal convolutional network (TCN)-gated recurrent unit (GRU)-Attention model is developed for fault identification and localization. Compared with single-SOC training, representative-SOC training improves the accuracy under noise-free unseen-SOC conditions from 64.62% to 99.74%, while maintaining 91.38% accuracy at a signal-to-noise ratio of 3 dB. These results demonstrate that the proposed method provides strong cross-SOC generalization capability and noise robustness.

Electronics doi: 10.3390/electronics15215072

Authors: Guangjiao Chen Mingyang Deng Wensheng Zhang Tianxiang Xu Xingchen Li

Geographic routing in intelligent connected vehicle convoys is challenged by rapid topology changes, intermittent connectivity, and frequent route interruption. This paper presents IV-GPSR, a link-lifetime-aware extension of GPSR for intelligent connected vehicle convoys. It replaces the one-dimensional lifetime approximation with a geometric link-expiration estimate, normalizes forwarding progress, current-link lifetime, and local neighbor availability, and integrates the resulting score into the standard greedy perimeter store carry forward sequence while remaining local and stateless with respect to global topology. Independent MATLAB packet-level experiments show statistically significant packet delivery ratio (PDR) gains over GPSR of 2.48 percentage points at 160 vehicles and 15 m/s and 4.11 percentage points at 25 m/s. Aggregate throughput is approximately 1.63–1.74 Mbit/s, consistent with 512-byte packets at 3.5 packets/s, while control overhead remains near 31%. IV-GPSR does not significantly outperform optimized AFB-GPSR, AK-GPSR, or SD-GPSR proxies and incurs a small delay penalty relative to GPSR. A weight study shows that PDR is most sensitive to the lifetime coefficient; the best grid point is α = 0.10, β = 0.80, γ = 0.10, whereas the main experiments use α = 0.35, β = 0.50, γ = 0.15 for balanced contributions.

Electronics doi: 10.3390/electronics15215071

Authors: Shixuan Lyu Wei Wang Yudong Xu Yinzhang Cheng Xiang Jing Yichen Yu Yin Yao

To identify the growing number of new entities connected to low-voltage distribution networks when field measurements are limited, this study proposes an intelligent identification method based on tabular in-context learning (TabICL). An IEEE 33-bus distribution-network model was constructed in PowerFactory to simulate five scenarios: load only, load with EV charging piles, load with PV, load with wind turbines, and load with both PV and EV charging piles. The resulting dataset comprised 96-point daily active-power curves. TabICL captured dependencies across time points, and its performance was directly compared with that of SVM, random forest, XGBoost, and MLP using identical training and test sets and evaluation metrics. TabICL achieved an accuracy of 92.50% and a macro-averaged F1 score of 0.9386, exceeding the four baselines by 15.00–17.50 percentage points in accuracy and by 0.1990–0.3075 in macro-averaged F1. Its accuracy remained 0.8888 at 40 dB; however, the pronounced degradation at lower signal-to-noise ratios identifies noise robustness as a limitation of the current method. These simulation results demonstrate the feasibility of identifying new grid-connected entities under limited measurement conditions, while validation with measurements from operating distribution networks remains necessary.