We present a load-independent class-Φ₃ oscillator and its design method based on particle swarm optimization (PSO). Instead of explicitly deriving analytical design equations, the proposed method evaluates zero-voltage switching (ZVS) and constant current (CC) output directly from time-domain circuit simulations and incorporates them into an objective function for optimization. By exploiting the high convergence stability and flexibility of PSO, load-independent ZVS and CC can be systematically achieved even for complex class-Φ₃ circuits. The validity of the circuit topology and design method is verified through experimental results.

In this study, we proposed four van der Pol oscillators coupled in a ring structure using linear memristors. It was found that this network exhibits various stable states depending on the initial values of the oscillators and memristor parameters, and that the synchronous state and spatiotemporal patterns change because of the linear memristors. These results suggest the potential for using memristors in the physical implementation of brain-inspired networks that mimic the mechanisms of the human brain, such as in physical reservoir computing and brain-inspired computing.

In this study, a structural design is investigated to improve the stability of the autonomous prediction of chaotic time series using a spiking neural network (SNN) as a reservoir. A two-layer SNN reservoir composed of excitatory and inhibitory (E/I) neurons is constructed, and the prediction output is learned by updating only the readout weights online using FORCE. Using the Lorenz system as a benchmark, normalized root mean squared error (NRMSE) is evaluated by separating the training (teacher-forcing) and closed-loop autonomous segments; the results indicate that the two-layer architecture reproduces key waveform features and that the autonomous-segment prediction accuracy varies with the E/I ratio.

This paper presents a quantitative analysis of analog-to-digital conversion (ADC) architectures for enabling high-speed physical reservoir computing (PRC) through spatial parallelization. In PRC systems that process broadband signals, such as optical reservoirs, the limited temporal resolution of individual conversion interfaces constitutes a critical hardware bottleneck. To overcome this limitation, this study investigates a strategy that compensates for restricted sampling rates by spatially distributing multiple low-speed conversion units and examines its impact on reservoir dynamics using 1-bit ΔΣ modulators. First, the influence of the oversampling ratio (OSR) on computational accuracy is quantified to determine the OSR required to approach the performance of an ideal analog baseline. Furthermore, the concept of time-space equivalence is demonstrated, wherein increasing the spatial density of parallel modulators effectively substitutes for high temporal resolution. Simulation results show that massive parallelization of low-cost, low-speed conversion units can achieve signal restoration accuracy comparable to that of expensive high-speed serial ADCs. These findings provide practical design guidelines for overcoming electronic bottlenecks in PRC through strategic utilization of spatial resources, thereby enabling high-performance processing under realistic hardware constraints.

The surge in computational demand for deep learning systems necessitates energy-efficient hardware. Spiking neural networks (SNNs) implemented with resistive memory devices offer a low-power solution but have constraints regarding weight precision. This study evaluated the trade-off between recognition performance and power dissipation in weight-quantized unsupervised SNNs. Our analysis revealed that performance degradation caused by reduced bit-widths can be effectively compensated for by scaling the network size. Notably, large-scale networks utilizing binary weights demonstrated robust inference capabilities comparable to those of higher-precision models. Furthermore, circuit-level simulations confirmed that minimizing weight precision results in a substantial reduction in power consumption. These findings suggest that large-scale, low-precision architectures provide a compelling design strategy for energy-efficient edge neuromorphic systems.

The hippocampal CA1 region is important for memory formation. However, how memory is encoded in hippocampal CA1 is still unclear. A previous study showed that emotional experiences in rats induce the experience-specific diversification of synaptic currents and ripple-firing events in hippocampal CA1. Ripple firing is a neural activity event observed in extracellular potentials and is generated by the synchronized firing of multiple neurons in hippocampal CA1 within approximately 50 ms. On the basis of this observation, we hypothesize that information related to episodic memory is encoded in ripple firing and that synaptic coupling strengths in hippocampal CA1 shape the diversity of ripple firing events. To verify this hypothesis, we construct a hippocampal CA1 neural circuit model and analyze how synaptic coupling strengths affect the diversity of ripple-firing events.

Speech detection under heavy noise conditions remains a significant challenge, even for modern voice activity detection (VAD) methods. In this study, we propose a framework based on spectro-temporal modulation (STM) features that demonstrates superior performance compared with conventional VAD-based methods. Furthermore, comprehensive analyses across various metrics and signal-to-noise ratios (SNRs) provide deeper insights into the efficacy of detectionmethods in heavy noise environments.

This study investigated the influence of blowing pressure on the sound production mechanism of brass instruments, focusing on changes in the geometric stiffness of the lips. We employed a numerical analysis method, integrating pressuredependent eigenfrequencies estimated via finite element analysis into one-dimensional time-domain simulations. The analysis suggested that increased blowing pressure may raise the lips’ eigenfrequencies through enhanced geometric stiffness associated with quasi-static deformation. By enabling independent adjustment of physical parameters in the numerical analysis--a challenge in experiments using physical artificial lips--we observed the potential contributions of factors like Young’s modulus and density to the oscillation threshold and mode transitions. Although based on specific numerical assumptions, these findings present a preliminary framework for examining the role of geometric stiffness in oscillation. This research provides fundamental insights for analyzing the dynamic characteristics of the lips and designing artificial blowing devices.

We present a computationally efficient approach to speech separation based on spatially regularized switching independent vector analysis (SR-SwIVA) incorporating an iterative source steering (ISS) update. By introducing a switching mechanism, the proposed framework models the timevarying characteristics of multichannel mixtures using multiple separation matrices, while spatial regularization based on direction-of-arrival information solves the inter-state permutation problem. To further improve efficiency, the conventional matrix-inversion-based update is replaced with an ISSbased rank-one update, significantly reducing computational cost while preserving separation performance. Experimental results in noisy and reverberant environment demonstrate that the proposed method achieves improved separation quality and faster convergence compared with conventional SRSwIVA, making it suitable for practical scenarios with limited microphone arrays.

Switching independent vector analysis (SwIVA) is a blind source separation algorithm that employs a switching mechanism over multiple separation matrices to enhance separation accuracy. However, SwIVA suffers from the global permutation problem when prior information about the sources is unavailable. Furthermore, it exhibits poor performance in reverberant environments. In this paper, we extend SwIVA to a convolutional beamforming algorithm with a spatially regularized separation method (SR-SwCIVA). We conduct experiments to confirm that SR-SwCIVA achieves excellent separation performance under reverberant conditions while maintaining strong robustness against the global permutation problem.

This paper focuses on a variable N-parallel code shift keying (VN-CodeSK) system using a (2, 2) threshold secret sharing (TSS) scheme with a frame structure. The relationship between the LED’s half-power angle and the transmission efficiency is evaluated in overlapping illumination areas and throughout the entire room. As a result, increasing the LED’s half-power angle improves the transmission efficiency throughout the entire room. This is considered to contribute to the uniformity of the lighting function.

We propose a novel convolutional neural network specialized for generating foreground/background seeds for Graph Cut segmentation. While seed generation was automated in prior work using deep models (e.g., TransUNet), such models were originally designed for general-purpose semantic segmentation and are not optimized for Graph Cut seed generation. On the basis of insights gained from our previous study, we introduce a lightweight architecture that combines a convolutional neural network (CNN) encoder with a low-frequency vision Transformer (LF-ViT) at the bottleneck to produce reliable seeds. The model is trained using 400 images generated by data augmentation from a single image of a lantana flower and evaluated on ten test images. Experimental results show that the proposed method improves Graph Cut performance over TransUNet-generated seeds (average mIoU from 0.8946 to 0.9699).

Firefighters are required to work in hazardous fire scenes. They work in collapsed buildings and smoke-filled areas. Firefighters face a considerable risk of injury because of this environment. In particular, a floor collapse is a matter of life and death for firefighters. Therefore, a supplemental safety device can monitor such conditions and lead to improved safety of firefighters. This paper proposes a firefighter walking condition monitoring system to detect floor collapse. The system attaches pressure sensors to the soles of the firefighter’s shoes. Dangerous walking conditions are detected using pressure values measured from the foot pressure sensors. Our research confirms that the voltage of the foot pressure sensor shows cyclicity during the stair-climbing condition. The voltage of the pressure sensor increases once during the floor-collapse condition, then rapidly decreases. After that, it increases again. It is surmised that the firefighter’s foot cannot contact the ground during a floor-collapse condition. The time variation of voltage is trained and predicted using an LSTM-based neural network. The MAE between the measured and predicted values is calculated. The MAE ranged from 0.30 to 0.55 during stair climbing, while it ranged from 0.67 to 1.50 during the floor-collapse condition. For all subjects, the MAE in the floor-collapse condition is at least 1.56 times larger than that in the stair-climbing condition. As a result, this difference can be used to reliably detect dangerous conditions and alert operators and firefighters.

Aspiration caused by an impaired swallowing can cause choking and aspiration pneumonia. Conventional swallowing assessments are often invasive and lack quantitative evaluation. This study investigated the feasibility of a noninvasive and quantitative swallowing evaluation system based on the simultaneous measurement of electromyography, swallowing sounds, and esophageal Doppler waveforms.
Characteristic events were extracted from each signal and their occurrence times were analyzed under different swallowing conditions. The experimental results demonstrate that changes in the physical properties of the swallowed samples affect the temporal relationships among the detected events. These findings suggest that the proposed multimodal measurement system can capture tissue behavior and bolus movement during swallowing and has the potential to quantitatively assess the swallowing function.

We propose a YOLO11-based segmentation model for detecting crosswalks and pedestrian signals. In addition, it is confirmed that the nighttime detection accuracy is improved by adding a sufficient number of nighttime images to the training dataset. The model performance is evaluated using daytime and nighttime test images by calculating precision, recall, and AP50.

In recent years, labor shortages caused by the aging and declining workforce have become a serious issue in Japanese agriculture. To address this problem, agricultural robots for automating labor-intensive tasks have been developed. This paper is focused on automating the leaf-cutting task in tomato cultivation, which is essential for promoting growth and preventing disease, and is proposed an image-processing-based method to detect leaf-cutting points. The effectiveness of the proposed algorithm is evaluated through the quantitative analysis of detection accuracy.

Low-level quadrotor control is challenging due to nonlinear coupled dynamics, sensor noise, and sensitivity to unmodeled effects such as battery voltage sag or ground effect [4, 5, 6]. We investigate reservoir computing (RC) as a computationally lightweight learning-based alternative that preserves real-time feasibility by training only a linear readout layer [1, 2]. We develop a Betaflight-based simulation and logging pipeline that uses Betaflight [3] as a teacher controller to generate demonstration trajectories for reservoirbased policies mapping onboard-like observations to raw motor commands. We evaluate controllers using both offline imitation error and a closed-loop stabilization protocol in simulation. Our results show that explicit short-horizon representations (via buffered reservoir states) improve imitation accuracy and can enable stable closed-loop flight in multiple trials, highlighting RC as a practical candidate for embedded lowlevel control. Our results demonstrate that reservoir-based controllers can achieve stable closed-loop flight in simulation while maintaining extremely low computational complexity, suggesting RC as a practical lightweight alternative for embedded quadrotor control.