Abstract This paper describes the realization of an integrated filter bank using a single multi-layered surface acoustic wave (SAW) substrate and a common electrode thickness. Through finite element method simulations, we have confirmed that the loss and the electromechanical coupling coefficient of multi-layered SAW devices are relatively insensitive to the ratio of electrode thickness to electrode pitch compared to those of leaky-SAW devices. Based on this, we designed and fabricated a time division duplex filter bank comprising four filters that cover the 1.8–2.7 GHz range on a single die using a common electrode thickness. The measured performance demonstrates a low insertion loss less than 1 dB and a low temperature coefficient of frequency in the range from −24.5 to +15.5 ppm °C −1 for all bands. These results indicate that multi-layered SAW devices are highly suitable for integrated filter design.

Macroscopic substrate surface errors and microscopic groove parameters influence the optical performance of curved diffractive microstructures. However, existing profile measurement techniques face a trade-off between large-area coverage and high resolution, which limits the ability of conventional two-dimensional (2D) line-profile methods to capture the global grating morphology. To address existing limitations, this study proposes a three-dimensional (3D) profile characterization method for curved gratings across macro- and micro-scales. Seamless reconstruction of full-aperture 3D topography with submicron-scale features was achieved using laser scanning confocal microscopy-based stitching measurements. Preprocessing for feature extraction was then performed using frequency-domain separation and the iterative closest point algorithm. The 2D Gabor filter bank, traditionally used for image texture feature extraction, was extended to 3D space to precisely characterize the period distribution of the microstructures. When combined with local planar least-squares fitting, the method enables precise characterization of the 3D spatial distribution of the grating blaze angle. Experimental results demonstrate close agreement between 3D and 2D characterization, with deviations below 0.01µm in mean period and 0.05° in mean blaze angle, confirming the accuracy and reliability of the method. This study overcomes the limitations of conventional 2D line-profile analysis by enabling high-precision, cross-scale 3D global characterization of curved diffractive microstructures, supporting process optimization and quality control in advanced optical manufacturing.

IntroductionBrain-computer interfaces (BCIs) provide a movement-free means of communication and control, typically based on motor imagery (MI) tasks of hand, foot, or tongue movements. Most BCI studies focus on classifying up to four such tasks, which limits the number of available commands and restricts overall system functionality. Expanding the range of reliable mental tasks would directly increase the number of possible commands and thereby enhance the practical utility of BCIs. Singing imagery (SI) may offer an intuitive alternative or additional task to complement conventional MI paradigms.MethodsEEG data were recorded from 14 participants performing right-hand, left-hand, foot, and tongue MI, SI, and rest. Features were extracted using filter bank common spatial patterns (FBCSP), and tasks were classified with a random forest algorithm across 2-, 4-, 5-, and 6-class scenarios. Subjective data regarding participants' perceived task difficulty and general task preferences was also collected.ResultsClassification accuracies with SI included were comparable to subsets of conventional MI tasks in 2-, 4-, and 5-class scenarios. In the 6-class scenario, average accuracy was approximately 60%, with six participants exceeding 70%, the level often cited as being necessary for effective BCI control. It is reasonable to expect performance to improve further with more advanced analysis methods and participant training.ConclusionThese promising results suggest that singing imagery can serve as both an additional and an alternative task in MI-BCIs. In lower-class systems, SI may provide a valuable option for generating commands, particularly for users who may find some conventional MI tasks less intuitive. When combined with the established MI tasks, SI could increase the number of possible commands, thereby extending the functional capacity of BCI systems. Overall, this work demonstrates the potential of SI to broaden the repertoire of mental tasks available for BCI control and to advance the development of more flexible, powerful, and user-centered BCI applications.

<sec>To meet the growing demand for high-frequency broadband signal processing in complex electromagnetic environments and to overcome the limitations of traditional electronic systems such as restricted bandwidth, limited response speed, and low integration density, this paper presents a reconfigurable microwave photonic channelized receiver chip implemented on a silicon photonic platform. The proposed architecture adopts a two-stage optical filtering strategy that circumvents the typical strict wavelength alignment requirements in traditional designs, thereby greatly alleviating the challenges of system integration. In the first stage, the cascaded Mach-Zehnder interferometer (MZI)-based wavelength division multiplexers (WDMs) are used to perform Gaussian-shaped filtering of the input optical spectrum with a channel spacing of approximately 200 GHz. The second stage combines an array of coupled resonator optical waveguide (CROW) filters functioning as finely tunable bandpass elements. These CROW filters utilize curved waveguide directional couplers, which are specifically designed to address the issues found in traditional multimode interference (MMI) couplers such as high insertion loss—and in straight directional couplers, which encounter significant coupling dispersion. The optimized curved coupler exhibits an insertion loss below 0.03 dB and a coupling ratio variation of less than 10% across the 1500–1600 nm wavelength band. Filter bandwidth reconfigurability is achieved via thermo-optic tuning of the balanced MZI embedded within each CROW filter, enabling dynamic adjustment of the coupling coefficients. Each filter exhibits a continuously adjustable 3 dB bandwidth ranging from 2.25 GHz to 3.12 GHz, with an excellent 20 dB/3 dB shape factor of 3.08. This performance indicates significantly improved roll-off characteristics compared with the performance of traditional filter designs, leading to enhanced suppression of image frequency components and improved signal separation fidelity.</sec><sec>A complete microwave photon channelized receiving link is constructed using an integrated WDM-CROW filter bank. System-level simulations confirm that the architecture provides excellent broadband adaptability, supporting the channelization of radio frequency (RF) signals in two operational bands: 8–28 GHz and 8–36 GHz. The system efficiently decomposes the input wideband RF signal into eight independent intermediate frequency (IF) sub-bands. Within each sub-band, an image rejection ratio (IRR) exceeding 22 dB is maintained. The corresponding IF ranges are 1.4–3.6 GHz when configured for 8–28 GHz RF input, and 2–5 GHz for 8–36 GHz input, covering critical communication and detection bands from X-band to K-band and satisfying the requirements of multi-scenario signal processing. Furthermore, we simulate the reception and reconstruction of a 5 GHz bandwidth linear frequency-modulated (LFM) signal, successfully verifying the chip’s capability in handling wideband waveforms. These results underscore the feasibility of the proposed chip as a high-performance solution for advanced applications such as radar detection and broadband electronic warfare systems, offering a novel, integrated photonic alternative to traditional channelized reception architectures.</sec>

This study presents an automated 3D breast cancer detection framework utilizing transrectal ultrasound imaging. The proposed approach combines multi-atlas registration with statistical texture priors for accurate segmentation. The atlas database includes annotated breast images from previous cases with segmented breast surfaces. Texture features are extracted using orthogonal Gabor filter banks to enhance the robustness of feature detection. Stage-specific tumor features are utilized to train a hybrid CNN-ResNet model, which ensures precise detection and segmentation of breast tumours in new patient images. Superpixel segmentation is then applied to refine tumor boundaries, enabling detailed and accurate tumor localization. The proposed method provides an efficient and reliable tool for early breast cancer detection, aiming to support improved diagnostic outcomes and clinical decision-making.

Seismic fault identification remains critical for resource exploration and geohazard prevention, yet conventional methods suffer from subjective interpretation bias and computational inefficiency. While convolutional neural networks (CNNs) enhance automation, their neglect of multiscale frequency features limits accuracy. Here, propose a novel Wavelet-Convolutional Neural Network (W-CNN) and its variants (W-CNN R1, W-CNN R2 and W-CNN R3) that architecturally fuses discrete wavelet transforms (DWT) with CNNs, establishing a spatial-frequency learning paradigm. By embedding Haar wavelet filter banks with cross-scale residual connections, W-CNN achieves explicit decoupling of high-frequency fault details from low-frequency structural contexts, reducing parameters by 21% versus conventional CNNs. Evaluated on coal mine datasets, W-CNN R3 achieves 90.0% accuracy (F1-score 90.3%), surpassing mainstream CNNs (LeNet-5, AlexNet, VGG16) by 0.6-12.3%, with the highest recall (95.5%) and faster convergence. The model successfully resolves 30 out of 32 exposed complex micro-faults (93.8% detection rate), demonstrating strong consistency with roadway-exposed faults in geologically complex zones, which significantly enhances its predictive capability for small-scale discontinuities. The frequency selection mechanism effectively suppresses noise interference, while the optimized architecture enables orders-of-magnitude acceleration in 3D processing. This framework provides an extensible solution for intelligent geological interpretation, with critical applications in mine safety monitoring.

Location-based services (LBSs) and positioning systems have spread worldwide due to the emergence of Internet of Things (IoT) and other application domains that require real-time estimation of the position of a person, tag, or asset in general in order to provide users with services and apps with added value. Whereas Global Navigation Satellite Systems (GNSSs) are well-established solutions outdoors, positioning is still an open challenge indoors, where different sensory technologies may be considered for that purpose, such as radio frequency, infrared, or ultrasounds, among others. With regard to ultrasonic systems, previous works have already developed indoor positioning systems capable of achieving accuracies in the range of centimeters but limited to a few square meters of coverage and severely affected by the Doppler effect coming from moving targets, which significantly degrades the overall positioning performance. Furthermore, the actual bandwidth available in commercial transducers often constrains the ultrasonic transmission, thus reducing the position accuracy as well. In this context, this work proposes a novel excitation and processing method for an ultrasonic positioning system, which significantly improves the transmission capabilities between an emitter and a receiver. The proposal employs a superheterodyne approach, enabling simultaneous transmission and reception of signals across multiple channels. It also adapts the bandwidths and central frequencies of the transmitted signals to the specific bandwidth characteristics of available transducers, thus optimizing the system performance. Binary spread spectrum sequences are utilized within a multicarrier modulation framework to ensure robust signal transmission. The ultrasonic signals received are then processed using filter banks and matched filtering techniques to determine the Time Differences of Arrival (TDoA) for every transmission, which are subsequently used to estimate the target position. The proposal has been modeled and successfully validated using a digital twin. Furthermore, experimental tests on the prototype have also been conducted to evaluate the system's performance in real scenarios, comparing it against classical approaches in terms of ranging distance, signal-to-noise ratio (SNR), or multipath effects. Experimental validation demonstrates that the proposed narrowband scheme reliably operates at distances up to 40 m, compared to the 34 m limit of conventional wideband approaches. Ranging errors remain below 3 cm at 40 m, whereas the wideband scheme exhibits errors exceeding 8 cm. Furthermore, simulation results show that the narrowband scheme maintains stable operation at SNR as low as -32 dB, whereas the wideband one only achieves up to -17 dB, highlighting the significant performance advantages of the proposed approach in both experimental and simulated scenarios.

Quickly identifying anomalies in rotating machinery is crucial for safety and profitability in contemporary industry (Industry 5.0). Unidentified failures can cause costly malfunctions and production interruptions. This research proposes an innovative strategy based on Transformer for the analysis of multidimensional vibration events (VIBT), with a view to early and accurate detection of anomalies in rotating machinery. The goal is to minimize production interruptions in Industry 5.0. The study highlights the limitations of conventional vibration analysis approaches and traditional deep learning techniques, emphasizing the need for innovative solutions. VIBT incorporates transformers and a filter bank convolution (FBC) module for effective denoising, as well as an adaptive wavelet transformation (WTA) mechanism for dynamic feature fusion at various scales, thereby addressing the challenges posed by non-stationary and noisy signals. Extensive testing on the Mafaulda dataset reveals that VIBT achieves 98.1% precision and 98.8% accuracy, significantly outperforming existing standard models. The results suggest that VIBT not only improves fault detection capabilities but also optimizes maintenance strategies in industrial applications, paving the way for future research on semi-supervised learning based on transformers and the integration of intermodal data.

Electroencephalography (EEG)-based brain–computer interface (BCI) mimics the brain’s intrinsic information-processing mechanisms by translating neural oscillations into actionable commands. In motor imagery (MI) BCI, imagined movements evoke characteristic patterns over the sensorimotor cortex, forming a biomimetic channel through which internal motor intentions are decoded. However, this biomimetic interaction is highly vulnerable to signal degradation, particularly in mobile or low-resource environments where low sampling frequencies obscure these MI-related oscillations. To address this limitation, we propose a robust MI classification framework that integrates spatial, spectral, and temporal dynamics through a filter bank common spatial pattern with time segmentation (FBCSP-TS). This framework classifies motor imagery tasks into four classes (left hand, right hand, foot, and tongue), segments EEG signals into overlapping time domains, and extracts frequency-specific spatial features across multiple subbands. Segment-level predictions are combined via soft voting, reflecting the brain’s distributed integration of information and enhancing resilience to transient noise and localized artifacts. Experiments performed on BCI Competition IV datasets 2a (250 Hz) and 1 (100 Hz) demonstrate that FBCSP-TS outperforms CSP and FBCSP. A paired t-test confirms that accuracy at 110 Hz is not significantly different from that at 250 Hz (p < 0.05), supporting the robustness of the proposed framework. Optimal temporal parameters (window length = 3.5 s, moving length = 0.5 s) further stabilize transient-signal capture and improve SNR. External validation yielded a mean accuracy of 0.809 ± 0.092 and Cohen’s kappa of 0.619 ± 0.184, confirming strong generalizability. By preserving MI-relevant neural patterns under degraded conditions, this framework advances practical, biomimetic BCI suitable for wearable and real-world deployment.

Abstract While style transfer has been extensively studied, most existing approaches fail to account for the defocus effects inherent in content images, thereby compromising the photographer's intended focus cues. To overcome this shortcoming, we introduce an optimisation‐based post‐processing framework that restores defocus characteristics to stylised images, regardless of the style transfer technique used. Our method initiates by estimating a blur map through a data‐driven model that predicts pixel‐level blur magnitudes. This blur map subsequently guides a layer‐based defocus rendering framework, which effectively simulates depth‐of‐field (DoF) effects using a Gaussian filter bank. To map the blur values to appropriate kernel sizes in the filter bank, we introduce a neural network that determines the optimal maximum filter size, ensuring both content integrity and stylistic fidelity. Experimental results, both quantitative and qualitative, show that our method significantly improves stylised images by preserving the original depth cues and defocus details.

This paper presents the design of a Fixed-point Kalman Filter Bank Architecture and its implementation on the PYNQ-Z2 FPGA. The proposed architecture comprises Q1.15 fixed-point arithmetic, saturation logic and reciprocal-based safe division, which are utilized to ensure numerical stability and hardware efficiency. Initially, the architecture for a single filter is designed and simulated for a sinusoidal input, and then replicated 8 times to create a single Kalman filter bank module with a span of angular frequencies from 0.001 to 0.029 rad/sample. The constructed Kalman filter bank is created as an IP and implemented on the PYNQ Z2 FPGA board. The design communicates via AXI4-DMA interface between the Processing System (PS) and Programmable Logic (PL). Experimental results demonstrate effective denoising of sinusoidal signals under varying noise levels- low, medium and high noise. The obtained results show an average RMSE below 0.15 and a correlation coefficient above 0.95. Post place-and-route results on the device indicate a resource utilization of 37,596 LUTs (70.67%), 124 DSP slices (56.36%), and 2BRAMs (1.79%), with a maximum operating frequency of 21.4 MHz and total power consumption of 1.48 W.

The mental state of the users can significantly affect the performance of active brain-computer interfaces (aBCIs). In this work, we aim to adopt passive BCIs (pBCIs) to measure a typical mental state, frustration, which is much relevant to aBCIs. A novel paradigm has been developed that combines both aBCIs and pBCIs under different frustration levels of users. The aBCI in this work is based on classic binary motor imagery (MI). In experiments, a new strategy was implemented that uses visual feedback to induce different levels of frustration. The electroencephalography (EEG) data collected were used for both aBCIs and pBCIs. The pBCI was utilized to assess the frustration level during the aBCI tasks, and the aBCI classification models for different levels of frustration were trained. For pBCI, the filter bank common spatial pattern (FBCSP) feature extraction and support vector machine (SVM) classification were utilized to classify three (i.e., low, moderate, high) frustration levels. For aBCI, the same method (FBCSP+SVM) was used to classify left versus right MI. We also aim to improve the performance of aBCIs in such conditions, so we developed two new methods to incorporate the pBCI results to adapt three MI classifiers to the varying states of frustration. Compared to the conventional approach of directly classifying MI tasks without considering frustration, the two proposed methods increased the mean classification accuracy by 7.40% and 8.62%, respectively. (Compared with the commonly used non-emotional discrimination data, the results are improved by 4.56% and 5.87% respectively.) Within the scope of non-invasive EEG and MI-based aBCI, this study provides, to our knowledge, an initial integrated demonstration in which a frustration-level classifier (pBCI) is trained and then used to adapt MI decoding (aBCI). It should not be taken as a claim of originality beyond this context. Starting from “user subjective perception”, this paper rises to the engineering level of “objective frustration recognition and classification model adaptation”, and makes a contribution to the depth of EEG data analysis and methodological integrity.

The probability of traffic accidents has increased highly nowadays. So, accident detection is very important. Most existing vehicle accident detection systems are based on convolutional neural networks (CNNs) and variants such as YOLO, Faster R-CNN, RetinaNet, etc. The above methods have issues such as less diversity in the models, accuracy to be improved, and heavy dependence on a large amount of labeled data. This research primarily focused on developing a video-audio multimodal unsupervised vehicle collision detection system based on the YOLOv8 algorithm. The study first employs YOLOv8 to extract spatio-temporal features from video frames, combining them with a convolutional neural network to identify and localize key objects within accident scenes. Subsequently, short-time Fourier transform (STFT) and Mel filter banks extract time-frequency features from accident audio, yielding an acoustic representation aligned with the video embedding space. Two strategies are explored during the feature fusion stage: (1) Post-processing probability-weighted fusion;(2) Attention mechanism fusion. A sliding average probability curve is introduced to smooth instantaneous predictions to address temporal instability in accident detection, generating continuous accident risk estimates that better align with human perception. Results demonstrate that the proposed multimodal fusion method significantly outperforms unimodal models in accident detection tasks.

Feature Mode Decomposition (FMD) can effectively extract bearing fault features even in the case of strong interference noise by means of adaptive finite impulse response filter banks along with correlated kurtosis. Nevertheless, the filter length L and the number of decomposition modes K need to be predefined carefully in a manual way. Otherwise, mismatched parameters could lead to redundant components or even missed detection of fault information. To mitigate the reliance on manual parameter setting, recent studies have introduced optimization algorithms such as the Whale Optimization Algorithm and the Crested Porcupine Optimizer to find the optimal parameters for FMD. However, such methods usually suffer from the dilemma of easily premature convergence in global search and long-time consumption in local fine adjustment, rendering them with difficulty in meeting the requirements of real-time and accurate diagnosis. Therefore, this paper proposes an improved Crested Porcupine Optimizer (ICPO), which can dynamically balance global and local exploitation. Furthermore, a bearing fault diagnosis method named ICPO-FMD is constructed, wherein the optimal parameter combination of K and L obtained using ICPO is provided to FMD in order to decompose bearing signals into a family of intrinsic mode functions (IMFs), and then fault sensitive components are extracted according to the proposed IMF screening principle. Finally, a reconstructed signal is obtained, followed by an envelope demodulation analysis. Experiments on simulation, laboratory and engineering signals demonstrate that the proposed method can accurately extract the fault characteristic frequency and its harmonics.

Filter bank convolutional network with dual channel attention for multi-class fine joint motor imagery decoding

Joint Design of Narrowband Transmit Waveforms and Receive Filter Banks for Range Super-Resolution and Target Classification

MSAttNet: Multi-scale attention convolutional neural network for motor imagery classification.

Abstract In industrial operational environments, rolling bearing vibration signals not only contain fault related periodic impulse components, but also interference noises. The coupling of these signals attenuates the fault characteristics, particularly in cases of compound faults, significantly affecting the diagnostic accuracy of bearings. This paper proposes a rolling bearing fault separation and compound diagnosis method using a sparse Bayesian framework with adaptive prior knowledge. Firstly, the proposed method segments signals into a finite number of modes, employing filter banks to adaptively select decomposition modes with a new compound fault indicator, which reduces noise interference and eliminates irrelevant components while ensuring the preservation of fault-related information. Secondly, by employing an interpretable strategy enables fault frequency estimation, eliminating errors from empirical parameter settings and avoiding the computational complexity of parameter optimization. The feature frequencies are estimated based on the envelope harmonic product spectrum and used as a prior knowledge of sparse Bayesian learning. Finally, to avoid the restriction of current feature extraction methods that primarily focus on the selection and extraction of a single demodulation frequency band, a sparse Bayesian probabilistic model is particularly designed for compound fault diagnosis. Simulation and experimental results show that the proposed method can effectively extract and separate the fault features of each individual modes and realize the diagnosis of rolling bearings.

Respiratory distress syndrome (RDS) is one of the most serious neonatal conditions, frequently leading to respiratory failure and death in low-resource settings. Early detection is therefore critical, particularly where access to advanced diagnostic tools is limited. Recent advances in machine learning have enabled non-invasive neonatal cry diagnostic systems (NCDSs) for early screening. To the best of our knowledge, this is the first cry-based RDS detection study to include both preterm and full-term infants in a subject-balanced design, using 76 neonates (38 RDS, 38 healthy; 19 per subgroup) and 8534 expiratory cry segments (4267 per class). Cry waveforms were converted to mono, high-pass-filtered, and segmented to isolate expiratory units. Mel-Frequency Cepstral Coefficients (MFCCs) and Filterbank (FBANK) features were extracted and transformed into fixed-dimensional embeddings using a lightweight X-vector model with mean-SDor attention-based pooling, followed by a binary classifier. Model parameters were optimized via grid search. Performance was evaluated using accuracy, precision, recall, F1-score, and ROC–AUC under stratified 10-fold cross-validation. MFCC + mean–SD achieved 93.59 ± 0.48% accuracy, while MFCC + attention reached 93.53 ± 0.52% accuracy with slightly higher precision, reducing false RDS alarms and improving clinical reliability. To enhance interpretability, Integrated Gradients were applied to MFCC and FBANK features to reveal the spectral regions contributing most to the decision. Overall, the proposed NCDS reliably distinguishes RDS from healthy cries and generalizes across neonatal subgroups despite the greater variability in preterm vocalizations.

Considering that multi-band interference often leads to a significant increase in the bit error rate at the system receiver end in actual underwater acoustic communication environments, this paper proposes a subcarrier silence anti-interference technology scheme based on filter bank multi-carrier (FBMC) with index modulation (IM). First, it is analyzed that, under three different underwater acoustic channels and without added interference, the underwater acoustic filter bank multi-carrier with index modulation (FBMC-IM) communication system outperforms traditional FBMC systems in terms of bit error rate performance. Subsequently, targeting the frequency distribution characteristics of multi-band interference, this paper designs an adaptive subcarrier silence mechanism. Through notch detection, interference band information is fed back to the transmitter, and subcarriers within the communication band that overlap with the interference signal spectrum are silenced, while unaffected subcarriers continue to carry communication information, thereby achieving multi-band partitioning to avoid interference effects. Additionally, to further enhance system performance, the paper integrates Virtual Time Reversal Mirror (VTRM) channel equalization technology, which leverages the time-focusing characteristics of multipath signals to effectively suppress multipath interference and delay spread in the acoustic channel. Simulation and field test results demonstrate that the proposed subcarrier-silence-based FBMC-IM anti-interference scheme significantly improves system reliability under multi-narrowband interference conditions. In the simulated underwater acoustic channel, the BER is reduced by approximately 65–80% at a signal-to-noise ratio of 0 dB; in the 5 km test channel in the Bohai Sea, the BER is reduced by 70–85% compared to the traditional FBMC system; in the test channel near Dalian with strong multipath spread, the BER is improved by more than one order of magnitude at a signal-to-noise ratio of 30 dB, with a BER reduction exceeding 90% under the configuration of Q = 4, k = 1. These results fully validate the superior anti-interference capability and communication robustness of the proposed scheme in interfering underwater acoustic environments.