Electronics Letters最新文献

Chiplet-based systems leveraging a domain-specific reusable interposer (DSRI) technology achieve superior performance, power, and cost (PPC) advantages. This paper proposes AutoDSRI, an automated framework for chiplet generation and topology synthesis (TS) in DSRI design. AutoDSRI introduces a hierarchical graph partitioning method, first partitioning domain-generic esults show an average of 29.18% improvement in PPC over traditional methods.

This study proposes a structure that replaces the stepped metallic ridge of an inline waveguide-to-coaxial adapter operating in the 40–60-GHz band with a ridge structure based on a printed circuit board (PCB). The PCB-based ridge is designed using stepped via holes and patterned copper foil to provide an impedance-matching function without the need for metal processing. A prototype device was manufactured and measurements were made using a vector network analyser (VNA). Based on simulations and measurements, reflection coefficients of less than –15 dB were achieved across the entire band. Therefore, the proposed adaptor performs well in the U-band at low cost.

The Recursive Pseudo Random Pursuit Strategy (RPRPS) is proposed to predict the frequency difference of atomic clocks relative to reference. It further improves the computational efficiency and reduces the time complexity. The core of RPRPS is to replace the refitting of the updated predictor and recalculating of the sum of square residuals of the unupdated predictors in a recursive process. In experiments, it is employed to predict the readings of the cesium clock and hydrogen maser relative to UTC (NIM). Compared with PRPS, the experimental results show that RPRPS reduces running time by about 70% without reducing predictive accuracy.

This paper proposes a deep learning-based network for blood collection tube label edge detection, named TLED_UNet. The proposed network integrates a multi-scale feature extraction module, a residual shrinkage module based on a mixed-domain attention thresholding mechanism, and a bidirectional LSTM temporal perception unit. In addition, a task-specific composite loss function is designed to significantly improve the model's accuracy and robustness when handling diverse types of blood collection tubes. Experimental results validate the effectiveness of the proposed method in real-world engineering scenarios: the probability of the label completely obscuring the visual window of the blood collection tube is close to 0, and the incidence rate of labelling deviation exceeding 1 mm is below 0.5%, demonstrating a clear performance advantage over traditional detection methods.

Text-based person search (TBPS) aims to retrieve pedestrian images from a database based on natural language descriptions. However, existing TBPS methods often assume that image-text pairs in training datasets are perfectly aligned, neglecting noisy annotations characterized by coarse-grained or mismatched descriptions. This letter presents a robust TBPS framework that addresses these challenges through two key innovations. First, we design a dual-channel Gaussian mixture model (GMM) to identify noisy image-text pairs by leveraging both global and local feature-level alignment losses. Second, for the detected noisy samples, we generate pseudo-texts using a multimodal large language model (MLLM) and filter them via a dynamic semantic consistency scoring mechanism to ensure high-quality supervision. Extensive experiments on ICFG-PEDES and RSTPReid demonstrate that our method consistently improves top-k retrieval metrics.

Convolutional neural networks (CNNs) have widely been adopted in object detection systems to achieve high performance. The CNN-based object detection systems consist of three stages: preprocessing, neural network processing and postprocessing. The preprocessing stage becomes a critical bottleneck as lightweight CNNs have been developed and the neural network processing stage is generally accelerated in hardware. As multiple operations are performed in the preprocessing stage, large memory access is required for intermediate data between the operations. In addition, a number of computations are required to resize an input image using image interpolation where interpolation ratios are calculated for each pixel. This paper proposes software optimization techniques to reduce the memory access and computations in the preprocessing stage. The operations performed in separate nested loops are efficiently fused using local registers to remove memory access. In addition, redundant computations are removed by caching the interpolation ratios across frames. To evaluate the proposed techniques, they were applied to an object detection system implemented on the Xilinx Zynq-7000 FPGA platform. The experimental results show that the proposed techniques reduce the preprocessing time by 2.19$�\times$ to 2.67$�\times$ and increase the overall inference throughput by 1.61$�\times$ to 1.79$�\times$ compared to the baseline.

WiFi-based gesture recognition is a promising privacy-preserving sensing modality. Existing research has predominantly relied on fixed equipment configurations from third-person perspective, inherently limiting deployment flexibility and interaction possibilities. Gesture recognition from first-person perspective, using wearable devices, offers a superior alternative. However, its deployment is critically hindered by the high computational overhead of current deep learning models on embedded or mobile devices. To address the problems, this paper designed joint channel dimensionality reduction and temporal convolutional network (JCDR-TCN), a lightweight model that integrates channel dimensionality reduction and a TCN for first-person gesture recognition. Raw data are collected using a Raspberry Pi 4B, processed and then classified by JCDR-TCN. Evaluated on a self-collected first-person perspective dataset, our JCDR-TCN achieves a recognition accuracy of 95.63%. Simultaneously, the model only contains 0.05 million parameters and requires 8.38 million floating-point operations (FLOPs), demonstrating a favorable trade-off between accuracy and computational efficiency, thereby making it well-suited for embedded and mobile deployment.

Millimetre-wave multiple-input multiple-output (MIMO) radar has many advantages such as flexible adaptability to complex environments and high detection accuracy. Among the numerous waveforms of MIMO radar, DDMA has a wide application prospect benefiting from its technical superiority. In this paper, we use empty-band based Doppler division multiple access (DDMA) to resolve the ambiguous velocity and then for further signal separation. But there remains a problem of separating multiple targets at the same distance. Thus, we propose a frame-interleaved binary mask method for the separation of empty-band DDMA signals. Simulation results demonstrate that our proposed method can effectively separate empty-band DDMA signals in the same condition.

This letter addresses the challenges of complex fault classification in a high-penetration new energy power grid, proposing a novel intelligent fault classifier based on a rule-embedded parallel deep convolutional neural network (DCNN). Traditional fault classification methods are hindered from recognizing the complex faults since they rely solely on the data in terms of a single time-frame. The classic Artificial Intelligence (AI) fault classifiers, mainly based on raw fault signals, also have a relatively limited ability to recognize complex faults. To address this, this study proposes a novel fault classification method by combining the model-driven method with the data-driven method. By designing the feature vector of parallel DCNN with characterized indexes, the proposed method inherits the ability of existing model-driven methods to locate fault types. By employing the panoramic data and an advanced CNN, the hidden features of the transition fault, developing fault are revealed. Simulation results and quantitative analyses validated the superior performance of the proposed method in complex target scenarios, attaining 97.22% accuracy in complex fault classification.