Search Results (384)

This paper proposes a method of fault detection in analog circuits which involves various steps, including selection of benchmark circuits, dataset preparation, signal decomposition, model training, and performance analysis. The main aim of this work is to provide solid performance even in noisy environments. Monte Carlo analysis is used to generate a synthetic dataset with 200 runs per fault class by introducing component tolerances and realistic faults. A multi-stage pipeline is proposed; it begins with resampling the signals and normalizing them, and then noise is added at different levels: 5 dB, 10 dB and 20 dB. Feature fusion is performed by combining time-, frequency-, and statistical-domain features. Statistical-domain features are extracted by applying Variational Mode Decomposition (VMD) to split them into four IMF levels, followed by the application of Continuous Wavelet Transform (CWT) for time–frequency-domain analysis. Support Vector Machine (SVM), Random Forest, and Gradient Boosting are used as base-level classification models. A stacking ensemble model is developed which uses Random Forest, Gradient Boosting, and Extra Trees as base learners and Logistic Regression as the meta-learner. Full article

The exponential growth of digital video traffic over satellite networks demands innovative approaches to optimize spectral efficiency while ensuring high quality of experience (QoE). Conventional Adaptive Coding and Modulation (ACM) schemes respond solely to Channel State Information (CSI), neglecting the perceptual importance of video content. This paper proposes a comprehensive Quality-Driven Adaptive Coding and Modulation (QACM) framework that dynamically allocates physical-layer resources based on joint channel conditions and content-aware quality metrics. The framework introduces a Quality Significance Factor (QSF) that quantifies video complexity and priority, enabling intelligent trade-offs between spectral efficiency and quality robustness. We implement a complete simulation testbed incorporating DVB-S2X-compliant ModCods with multiple code rates (1/2, 3/4, and 5/6) and higher-order constellations (up to 64QAM) over Additive White Gaussian Noise (AWGN) channels. Extensive experimental results using H.264/AVC sequences demonstrate that, while standard ACM achieves 30.14 dB PSNR for high-motion football sequences at 16 dB SNR with 64QAM-1/2, QACM improves this to 41.40 dB by switching to QPSK-1/2, representing an 11.26 dB gain. We provide comprehensive BER analyses across the 0–20 dB SNR range, statistical significance validation (p < 0.01 for quality improvements), computational complexity analysis showing 15.2% overhead, and detailed comparisons with prior arts. The framework demonstrates scalability to higher-order modulations while maintaining 23% weighted QoE improvement over conventional ACM. This work provides a validated, implementable cross-layer solution for next-generation satellite broadcasting systems. Full article

Fibre Bragg Grating (FBG) demodulation technology is central to structural health monitoring. However, spectral distortion and noise caused by complex environments, along with the challenge of balancing accuracy and real-time performance in existing deep learning algorithms, severely limit its application in large-scale dynamic sensing networks. To address this challenge, this paper proposes a hybrid learning framework named TimeMixer-LightGBM. The framework first employs a pure MLP-based TimeMixer model to efficiently extract multi-scale spectral sequence features from FBG reflection spectra, and then uses a LightGBM model (gradient boosted decision trees) to perform fast regression on these features for centre wavelength shift prediction. Experiments on synthetically distorted FBG spectra (including asymmetric shape variations and additive white noise) show that the method achieves picometre-level accuracy (RMSE = 2.128 pm) with an average processing time of only 0.08 ms per spectrum, representing a speedup of about 4.5 times over the latest deep learning models for FBG demodulation. It also exhibits excellent noise robustness, maintaining an average absolute error of 1.5 pm. Ablation experiments confirm the necessity and synergy of the hybrid architecture. The framework was further applied to demodulate double-peaked overlapping spectra, outperforming existing methods under mild, moderate and severe overlap conditions while keeping inference times in the sub-millisecond range. This study provides a novel and effective technical solution for real-time high-precision FBG demodulation, validates the effectiveness of pure MLP architectures in spectral analysis, and lays a theoretical foundation for deploying such demodulation on embedded edge devices, thereby achieving a favourable balance of accuracy, speed and scalability. Full article

Deep learning models have achieved remarkable performance in radar jamming recognition, yet they remain highly vulnerable to adversarial attacks—small, carefully crafted perturbations that cause misclassification—posing a critical threat to intelligent electronic countermeasure systems. Existing adversarial defenses suffer from an inherent accuracy–robustness tradeoff, limited defensive knowledge sources, and poor generalization to unseen attacks, while the additional challenge of model lightweighting for resource-constrained radar platforms remains largely unaddressed. This paper proposes Adversarial Distillation Defense (ADD), a training framework that synergistically integrates adversarial training with knowledge distillation to produce lightweight yet robust jamming recognition models. In ADD, an adversarially pre-trained teacher model simultaneously transfers its classification knowledge on clean samples and defensive knowledge on adversarial samples to a compact student model through four complementary loss terms. Extensive experiments on a simulated dataset comprising seven radar jamming types demonstrate that ADD achieves the strongest defensive performance among the compared defenses under both white-box and black-box attacks across varying perturbation strengths and jamming-to-noise ratios. Feature-space visualization further confirms that ADD enables the student model to maintain well-separated class clusters even under strong adversarial perturbations. These results indicate that ADD offers an effective strategy for building secure and lightweight deep learning models for radar jamming recognition. Full article

Automatic modulation classification (AMC) becomes challenging in multipath fading channels, particularly for high-order quadrature amplitude modulation (QAM) signals whose constellation points are strongly distorted by inter-symbol interference, phase rotation, and fading. This paper presents a channel-type-guided hierarchical AMC framework that combines blind preprocessing with deep learning. In the first stage, the received in-phase and quadrature (IQ) signal is downsampled and preprocessed using blind signal processing techniques. Blind source separation (BSS) is used for additive white Gaussian noise (AWGN) and flat fading channels, whereas the constant modulus algorithm (CMA) followed by BSS is used for multipath fading channels. A convolutional neural network (CNN) then performs first-stage modulation classification and generates a QAM-family flag. If the first-stage output corresponds to a QAM-family signal, a second-stage refinement path is activated. In this path, a convolutional denoising autoencoder (CDAE) is applied to the original received signal to mitigate multipath-induced distortion, followed by BSS preprocessing and a dedicated CNN classifier for 16-QAM, 64-QAM, and 256-QAM. Simulation results over AWGN, flat fading, and multipath Rician fading channels show that the proposed hierarchy improves high-order QAM classification in the considered settings, especially for 64-QAM and 256-QAM under multipath fading with stronger time variation. Multi-frame Softmax averaging further improves decision stability. The results support the use of classical blind preprocessing and selective CDAE-based refinement as a practical, complementary front end for AMC in controlled multipath simulation scenarios, while real over-the-air validation and automatic channel-category detection remain future work. Full article

Deploying medical visual question answering (VQA) systems over wireless networks introduces a fundamental challenge: channel-induced image degradation may corrupt the visual representations extracted by large vision-language models (VLMs), leading to unreliable diagnostic decisions. We investigate the robustness of frozen LLaVA-1.6, BLIP-2, and BioViL-T hidden-state features under additive white Gaussian noise (AWGN), Rayleigh fading, and six combined JPEG-compression-plus-channel conditions (quality factors $q\in \{20,50,70\}$) across signal-to-noise ratios (SNRs) from $-5$ to $+20$ dB. A lightweight MLP classifier is trained exclusively on clean features and evaluated on channel-degraded features, enabling controlled analysis of representation robustness without retraining. We introduce the Feature Robustness Score (FRS), defined as the difference between cosine similarity and normalized L2 drift of clean versus degraded features, together with a validation-set FRS threshold analysis as a label-free retraining criterion. A wavelet sub-band energy analysis further characterizes the spectral distribution of channel-induced feature drift. Experiments on PathVQA and VQA-RAD reveal four key findings: (1) LLaVA-1.6 features maintain cosine similarity above $0.98$ across all eight channel conditions and all SNR levels, with statistically significant MLP gains at every tested point ($p<0.05$, McNemar’s test); (2) BLIP-2 and BioViL-T features are less stable but still support consistent MLP improvements, with BioViL-T performing competitively on VQA-RAD, suggesting domain alignment matters; (3) JPEG compression quality ($q=20,50,70$) has negligible impact on feature drift, establishing VLM features as JPEG quality-invariant; and (4) wavelet analysis confirms that channel noise primarily affects high-frequency detail bands while preserving low-frequency semantic content. Full article

Accurate waveform classification in noisy environments is an important task in modern communications, radar signal analysis, biomedical signal interpretation, industrial monitoring and other signal processing systems. This paper investigates the performance of three neural network architectures: Multilayer Perceptron (MLP), one-dimensional Convolutional Neural Network (1D-CNN), and Long Short-Term Memory (LSTM) for multiclass waveform classification in the presence of Additive White Gaussian Noise (AWGN). A time series dataset consisting of multiple waveform classes is generated and corrupted with AWGN across a wide range of signal-to-noise ratio (SNR) levels to simulate noisy signal distortion conditions. The three models are trained and evaluated under identical conditions to ensure a fair comparison. Their classification performance is evaluated in terms of accuracy, Confusion Matrix (CM), Receiver Operating Characteristic (ROC) curve and the Area Under the ROC curve (AUC) across varying SNR values. Simulation results demonstrate that the 1D-CNN effectively captures local temporal patterns and achieves superior robustness in classification at moderate and high SNR levels. The LSTM model demonstrates the ability to capture temporal dependencies in sequential waveform data but exhibits sensitivity to waveform variations due to amplitude, phase and frequency changes and noise at lower SNR values. The MLP, although computationally simpler, shows comparatively limited performance in low-SNR conditions due to its lack of temporal feature extraction capability. For the case of multiclass deterministic waveforms, the accuracy of classification for the 1D-CNN and LSTM is nearly 100% at SNR = 5 dB showing their robustness in classification, whereas the accuracy of MLP is approximately 70% that shows poor classification in noisy conditions. When there is random amplitude, frequency and phase variations in the waveforms, the accuracy of the 1D-CNN and MLP increases with SNR, and 1D-CNN superior to MLP. However, the LSTM accuracy fails to improve with SNR, resulting in poor classification performance in such a scenario. The results provide an insight into the suitability of different neural architectures for waveform classification tasks in noisy communication or other time series applications and highlight the advantages of convolutional feature extraction for robust signal recognition. Full article

Herein, p-NbSe₂ thin films were deposited onto n-Si substrates to fabricate an n-Si/p-NbSe₂ (SNS) heterojunction for visible light communication (VLC) applications. Structural analysis revealed that the NbSe₂ films possess a trigonal phase and are composed of slightly elongated and irregularly shaped grains with an average size of 0.131 μm. Electrical characterization showed that the SNS heterojunction exhibits pronounced rectifying behavior, with a bias-dependent asymmetry factor reaching 6.6 × ${10}^3$. The photodetection performance of the device was evaluated under illumination from white, blue, red, tungsten, and infrared LEDs. The device exhibited excellent photodetection characteristics across the visible region, achieving a maximum responsivity of 3.79/3.68 AW⁻¹, external quantum efficiency of 1160/809%, noise equivalent power of 4.43 × 10⁻¹⁴ /4.57 × 10⁻¹⁴ WHz^−1/2, and specific detectivity of 3.91 × 10¹²/3.79 × 10¹² Jones under blue/white light illumination, confirming its practical relevance for VLC systems. In addition, frequency-dependent photocurrent measurements under modulated blue and white LED illumination revealed −3 dB bandwidths of approximately 775 Hz and 716 Hz, respectively, supporting the potential of the n-Si/p-NbSe₂ photodiode for low-frequency VLC-related visible-light detection. Compared with previously reported photodiodes used in VLC and IR technologies, the present device demonstrated superior responsivity and EQE%, together with competitive NEP and detectivity. The enhanced performance is attributed to efficient photocarrier generation and collection across the Si/NbSe2 heterojunction. These results confirm that the fabricated SNS photodiode is a promising candidate for high-sensitivity and efficient visible light communication applications. Full article

The high peak-to-average power ratio (PAPR) of orthogonal frequency division multiplexing (OFDM) signals poses a significant challenge for power-limited satellite transponders, leading to power amplifier nonlinearity and reduced system efficiency. This paper proposes a novel joint algorithm named Polar-SLM-CPM for efficient PAPR suppression in satellite coded OFDM (COFDM) systems. The core of this scheme is a deeply integrated design that synergistically combines polar coding, intelligent selective mapping (SLM), and adaptive continuous phase modulation (CPM). Unlike conventional approaches that treat these components separately, our method leverages the constant-envelope property of CPM for inherent PAPR limitation, employs a gradient-learning-optimized intelligent SLM mechanism for efficient low-PAPR sequence search, and utilizes capacity-approaching polar codes to guarantee transmission reliability. The synergistic operation is mathematically modeled and extensively evaluated via MATLAB simulations. Results demonstrate that the proposed algorithm achieves a substantial PAPR reduction of approximately 4.2 dB at a complementary cumulative distribution function (CCDF) of ${10}^{-3}$ while maintaining bit error rate (BER) performance comparable to conventional polar-coded OFDM under additive white Gaussian noise (AWGN) channels. Further analyses on synchronization, computational complexity (Big-O), parameter sensitivity, spectral efficiency trade-offs, and robustness in realistic nonlinear/phase-noise channels are provided, confirming the scheme’s practical viability. This work presents a balanced and effective solution for enhancing the power efficiency and signal integrity of next-generation integrated satellite communication and navigation systems employing COFDM-CPM waveforms. Full article

As the offshore wind industry scales toward 15 MW capacity, the reliability of Direct-Drive Permanent Magnet Synchronous Generators (DD-PMSGs) becomes critical. However, real-world run-to-failure data for these massive, multi-pole machines is virtually non-existent, creating a barrier for developing effective data-driven diagnostic systems. This study proposes a high-fidelity framework for detecting permanent magnet faults in the International Energy Agency (IEA) 15 MW Reference Wind Turbine. Using Finite Element Analysis (FEA), a dataset (magnetic flux and back electromotive-force (EMF)) capturing the electromagnetic signatures of healthy and faulty states of a PMSG under varying severities is generated. To improve the power of computer vision, 1D time-series signals were transformed into 2D images. Specifically, Gramian Angular Fields (GAFs) and Recurrence Plots (RPs) were applied to magnetic flux density signals, while Markov Transition Fields (MTFs) were applied to back-EMF signals. These representations were then fused into multi-channel Red-Green-Blue (RGB) images and processed via a ResNet-18 Deep Transfer Learning model using a strictly non-overlapping, leakage-free dataset partitioning strategy. The proposed framework achieved a classification accuracy of 99.45% on noise-free data. Furthermore, robustness testing under varying levels of Additive White Gaussian Noise (AWGN) (30 dB, 40 dB, and 50 dB Signal-to-Noise Ratio (SNR)) demonstrated sustained high performance, maintaining over 90% accuracy even under severe 30 dB noise conditions. Comparative analysis proved that this multi-channel fusion significantly outperforms single-channel encoding methods, which collapse under heavy noise, validating the scalability of the framework and applicability for next-generation condition monitoring in harsh offshore environments. Full article

Image enhancement is of pivotal importance in the detection of defects in wood veneers. However, acquired images frequently exhibit signs of blurring, uneven illumination, and insufficient contrast, which can lead to a reduction in the accuracy of defect recognition. In this study, an algorithm based on Grünwald–Letnikov (G–L) fractional-order differentiation is proposed for the enhancement of wood veneer defect images. Initially, the gain characteristics of differential amplitude-frequency responses on high- and low-frequency image components are analyzed, and the feasibility of the method is demonstrated by linking these characteristics with the frequency-domain distributions of live knot, dead knot, and crack defects. Secondly, an eight-direction mask operator is constructed based on the G–L definition, and a DC component preservation factor is introduced to eliminate the luminance drift caused by mask truncation. The application of the mask is performed independently on the R, G, and B channels, and a dynamic blending mechanism is designed to achieve a balance between texture enhancement and structural fidelity. Finally, a set of six evaluation metrics (AG, E, PSNR, RMSE, SSIM, and VIF) is employed to assess the quality of enhanced images. The proposed algorithm is then compared with five existing algorithms (SSR, MSR, MSRCR, CLAHE, and AGC) under both noise-free and additive white Gaussian noise conditions. The findings indicate that the G–L fractional-order differentiation algorithm facilitates a more balanced representation of image features, thereby enhancing contrast, brightness, and textural contours. This approach results in more authentic color reproduction and superior visual quality. Full article

Hybrid brain–computer interfaces (BCIs) have attracted growing research attention because they combine the millisecond-level temporal resolution of electroencephalography (EEG) with the spatially informative hemodynamic responses of functional near-infrared spectroscopy (fNIRS). However, most existing deep fusion methods rely on static late-fusion strategies, which tend to underexploit latent cross-modal dependencies and are vulnerable to modality-specific signal degradation. To address these limitations, we propose MGFNet, a multi-granularity fusion network for hybrid BCI decoding. MGFNet contains three components: (1) intra-modal encoders that learn modality-specific spatiotemporal representations from EEG, oxygenated hemoglobin (HbO), and deoxygenated hemoglobin (HbR) signals; (2) cross-modal interaction encoders that temporally align paired modalities and use dilated convolutions to capture long-range EEG-fNIRS dependencies; and (3) a Coupling-Guided Sparse Component Routing (CGSCR) module that estimates sample-specific cross-modal coupling and performs adaptive discrete routing. We further introduce a deep supervision strategy to stabilize optimization and improve branch-level discriminability. Under a within-subject held-out evaluation protocol on a public benchmark dataset, MGFNet achieved classification accuracies of 99.40% on the n-back task and 99.03% on the word generation (WG) task, outperforming representative comparison methods evaluated under a matched protocol. Ablation studies further confirmed the contributions of the intra-modal encoders, the cross-modal interaction encoders, and the CGSCR module. Under controlled EEG corruption with additive white Gaussian noise at −10 dB, MGFNet outperformed a static-fusion variant by 9.23 percentage points on the n-back task and 6.31 percentage points on the WG task. These results support the effectiveness of MGFNet in the present offline within-subject setting and indicate improved robustness under controlled single-modality degradation. Full article

Background/Objectives: In this three-stage study, we aimed to adapt an Auditory-Visual Stroop test (AV-Stroop test) for tinnitus subjects, evaluate the correlation between performance in the conventional Stroop test (C-Stroop test) and the AV-Stroop test; assess the effect of cognitive screening test performance on the AV-Stroop test’s results; and apply the AV-Stroop test in participants with tinnitus and controls. Methods: At the First Stage, the AV-Stroop test was adapted using white noise (WN), pure tone (PT), and narrow band (NB) sound stimuli. At the Second Stage, results of the AV-Stroop test, the C-Stroop test, and the Montreal Cognitive Assessment (MOCA) were compared (n = 45). At the Third Stage, the AV-Stroop test was applied to participants with and without tinnitus (n = 70). The tinnitus group was assessed with an additional test track (stimuli matched to tinnitus spectral characteristics, Tinnitus Pitch). Results: We adapted 34 training and evaluation tracks for the AV-Stroop test. AV-Stroop test’s results were correlated with C-Stroop test’s total task time (WN, p-value = 0.002; NB and PT, p-value < 0.001 comparing C-Stroop word reading task; and WN, NB, and PT, p-value < 0.001 for C-Stroop color naming task), and number of errors (NB, p-value < 0.001 comparing C-Stroop word reading task, and p-value = 0.012 for C-Stroop color naming task). Participants’ MOCA scores were not associated with AV-Stroop test performance. Participants with tinnitus required more time and made more errors in the AV-Stroop test. Additionally, the tinnitus group made more errors in the Tinnitus Pitch track. Conclusions: The AV-Stroop test proved to be an accessible, easy-to-administer tool for evaluating attentional and inhibitory control in participants with tinnitus. The stimulus with spectral characteristics similar to tinnitus perception was more effective in assessing top-down executive control in participants with the symptom. Full article

Rapid and reliable DC fault detection is critical to the safe operation of Voltage Source Converter High Voltage Direct Current (VSC-HVDC) multi-terminal grids, where low system impedance causes fault currents to rise within milliseconds, demanding detection within 1 ms. Discrete Wavelet Transform (DWT) has emerged as a leading signal processing technique for this purpose. However, no comprehensive performance study exists comparing the principal mother wavelets Daubechies (db), Symlets (sym), and Coiflets (coif) across the key operational variables of noise environment, cable length, and grid topology. This paper presents a systematic comparative evaluation of six wavelets (db4, db8, sym3, sym5, coif3, coif5) for DC fault detection in both three-terminal and four-terminal VSC-HVDC grids, assessing performance against four metrics: detection delay, accuracy, noise tolerance, and computational efficiency. Internal close-up and internal remote DC faults were simulated under no-noise conditions and white Gaussian noise levels of 30 dB, 20 dB, and 10 dB, with additional tests at cable lengths of 50 km and 400 km. Results demonstrate that db4 consistently achieves the lowest detection delay with high accuracy for four-terminal configurations under varying noise conditions, while sym3 proves most adaptable across both topologies for multiple cable lengths owing to its consistent detection delay. Real-time validation using an OPAL-RT hardware-in-the-loop (HIL) platform confirms the simulation findings, reinforcing the suitability of sym3 for multi-terminal grid deployment. These results provide actionable guidance for the selection of mother wavelets in DWT-based protection algorithms for modern VSC-HVDC systems. Full article

This study investigates the robustness of Frenet–Serret curvature ($\kappa$) and torsion ($\tau$) estimates derived from noisy discretely-sampled three-dimensional space curves, with emphasis on the comparative performance of cubic spline and cubic Hermite interpolation methods. Accurate estimation of these geometric invariants is essential for reliable analysis of curves arising in signal processing and shape reconstruction; yet, the higher-order derivatives required for their computation exhibit pronounced sensitivity to measurement noise. We examine curves constructed through a Hilbert transform-based parameterization of the form $\mathrm{r}\left(t\right)=\left(X\left(t\right),A\left(t\right)sin\varphi \left(t\right),g\left(t\right)\right),$ where discrete samples are contaminated with additive white Gaussian noise at varying signal-to-noise ratios. Reconstruction is performed using cubic spline interpolation, which ensures global $C^2$ continuity, as well as cubic Hermite spline interpolation, which provides $C^1$ continuity with local tangent control. Frenet frame computations are then applied via regularized finite difference schemes. To characterize noise amplification theoretically, we derive the Curvature Stability Index (CSI) and Torsion Stability Index (TSI) as first-order variance bounds under the delta method. While these indices formalize the derivative-order dependence of noise sensitivity, Monte Carlo simulations reveal that empirical variance exceeds theoretical predictions by factors of ${10}^4$ to ${10}^6$, indicating dominance of nonlinear error propagation. Nevertheless, the indices establish that torsion instability arises fundamentally from third-order derivative structure rather than ground-truth magnitude. Numerical experiments across three geometric regimes constant-invariant helices, variable-curvature helices, and planar curves with identically zero torsion demonstrate that the ratio of the torsion root mean square error to curvature root mean square error consistently ranges from $6.5$ to $9.8$. This disparity persists even in the degenerate planar case, where $\tau \equiv 0$ analytically, confirming that torsion sensitivity is an intrinsic property of the Frenet–Serret formulation. Across all configurations, cubic spline reconstruction yields lower Monte Carlo mean RMSE and reduced empirical variance compared to Hermite spline, providing superior stability for derivative-based invariant estimation. Full article