REV Journal on Electronics and Communications
Not a member yet
230 research outputs found
Sort by
Partial Distance Correlation-Based Motion Pattern Detection in Pangasius Fish
No full textIn computer vision, behavioral recognition of aquatic organisms plays an important role, particularly for Pangasius catfish, a fish species commonly cultured in Vietnam. This study presents a method for detecting Pangasius motion patterns comprising 38 catfish videos with a total of 236,133 extracted frames, from which 4,593 motion windows are extracted and classified into six behavioral categories (Cruising, Burst–Coast, Escape, Schooling, Milling, and Swarming) based on Partial Distance Correlation (PDC) integrated with video processing techniques and feature extraction methods. Experimental results show that Distance Correlation (dCor) on raw data yields high correlation values (0.826–0.989) but with substantial scatter. PDC with heading angle control maintains elevated values (0.804–0.979) with tighter residual clustering. When denoising is combined with heading angle control, pdCor achieves optimal efficacy (0.852–0.973). Compared with dCor, pdCor provides consistent improvements, especially for complex behaviors (Escape: 5.1%; Swarm: 3.2%). The combined strategy detects 40% of patterns better and 60% similarly, indicating pdCor does not reduce performance for simple behaviors but substantially improves detection for nonlinear, high noise patterns
SHA-RV: A RISC-V Accelerator for SHA-224/256 with Cycle Reduced ISA Extensions for Blockchain Applications
Get PDFThe Secure Hash Algorithm SHA-256 and SHA-224 are widely used for software integrity, digital signatures, and blockchain across embedded and edge platforms. Prior RISC-V accelerators still struggle to achieve low cycle counts and high system throughput on long message streams. This paper proposes a hardware-efficient RISC-V accelerator with low-latency SHA instruction extensions, named SHA-RV, to reduce cycles and improve end-to-end performance. SHA-RV integrates three optimizations: a high-bandwidth BufferSet for continuous data supply, a four-stage pipelined SHA core, a system-level double-buffering pipeline, and an FSM-orchestrated BufferSet mapping. Implemented on a Xilinx ZCU102 system on a chip, SHA-RV operates at up to 300 MHz and uses 3,146 flip-flops, 5,175 lookup tables, and 15 block RAMs. On 64-byte blocks, SHA-RV completes a block in 257 cycles, improving over related RISC-V designs by between 9.7 and 134.9 times, while reducing logic resources versus the ISOCC 2024 design by 89.4 percent in flip-flops and 85.2 percent in lookup tables. At the system level, SHA-RV achieves a throughput of 599 megabits per second and an energy efficiency of 798.7 megabits per second per watt under a real-time dynamic power assumption of 0.75 watts, outperforming representative CPUs by between 61 and 454 times in energy efficiency. These results show lower latency and superior hardware efficiency relative to prior work
Impact of Optical Crosstalk on OIRS-Assisted HAP-Based Multiuser FSO Systems over Turbulence Channels
Get PDFFree-space optical communication (FSO) utilizes laser beams to transmit data through the atmosphere. However, FSO faces significant challenges, including the strict requirement for line-of-sight (LoS) communication and terrestrial obstacles, which limit its scalability to connect multiple users in diverse environments. To address these limitations and enable reliable multi-user connectivity, the integration of high-altitude platforms (HAP) and optical intelligent reflecting surfaces (OIRS) has emerged as a critical solution. To serve multiple users simultaneously, an OIRS is equipped at the HAP to dynamically control the reflected beam from a ground station to the terminals. This study analyzes the proposed FSO system performance through the outage probability. During the analysis, practically influencing factors such as optical crosstalk, i.e., interference between OIRS regions, and atmospheric turbulence, are considered. The numerical results show the feasibility of deploying OIRS on HAP to support multiuser FSO systems. In addition, properly designing the OIRS coverage could improve the overall performance of the multiuser FSO system
E2DSR: Edge-Enhanced Representation for Deep Super-Resolution in Machine Vision Applications
Get PDFWhile deep-based super-resolution (SR) has achieved remarkable progress, state-of-the-art models like EDSR often rely solely on pixel-level information, resulting in overly smooth outputs that often fail to reconstruct the fine-grained edge details essential for downstream machine vision tasks. To address this challenge, we propose the Edge-Enhanced Deep Super-Resolution (E2DSR) model, a task-aware framework that leverages explicit edge guidance to enhance the reconstruction process with high-frequency edge information. E2DSR integrates a novel Edge Feature Enhancement Block (EFE) into a deep residual architecture, which learns to extract and fuse salient edge features from the low-resolution input. We demonstrate the effectiveness of our approach within a gesture recognition, where E2DSR significantly enhances input quality for a state-of-the-art YOLOv10 detector. Experimental results show that our method substantially outperforms the original EDSR and other approaches, improving the mean average precision (mAP) from 0.776 to 0.822 on average across four representative gesture action types. Our work demonstrates that explicit edge guidance is a crucial component for developing super-resolution models that excel in practical machine vision applications
Protograph LDPC-Coded Superposition Modulation for MIMO Channels with Triple Mixed-ADC Architectures
Get PDFThis paper investigates an energy-efficient large-scale MIMO (LS-MIMO) receiver that integrates a triple mixedADC architecture with superposition modulation (SM) and protograph-based LDPC coding. The considered receiver employs groups of one-, two-, and five-bit ADCs to reduce front-end cost compared with full-resolution designs, while offering improved performance over a uniform 1-bit baseline in the tested simulations. A double-layer factor-graph detector isdeveloped, and analytical log-likelihood expressions are derived under an additive-quantization-noise model to account for attenuation and quantization effects across ADC tiers. A PEXIT-based analysis is adapted to study iterative-decoding thresholds and examine design parameters such as SM weights, protograph matrices, and resolution allocations across various MIMO configurations. Simulation results for representative setups align with the PEXIT predictions and show performancegains for the triple mixed-ADC configuration relative to a uniform 1-bit system. In line with previous findings, equalweight SM demonstrated competitive performance and, in the tested settings, yielded the lowest thresholds. The proposed analysis framework may be useful for guiding protograph design in systems that combine SM, mixed-resolution ADCs, and LS-MIMO
Hybrid Architectures Combining Cellular and Convolutional Neural Networks for Fish Classification and Disease Detection
Get PDFFish classification and disease detection are crucial for (play crucial role in) sustainable aquaculture management, requiring high-accuracy, real-time computer vision models. This study introduces FISH-YOLOv8, an enhanced deep learning model built on YOLOv8, replacing all convolutional layers with Cellular Neural Networks (CeCNNs) to leverage their superior dynamics and noise tolerance for improved feature extraction in turbid, occluded underwater conditions. BiFormer Attention and Non-Maximum Suppression (NMS) further optimize detection accuracy and speed (enhance detection accuracy and processing speed). Evaluated on a Roboflow dataset of 1,800 images across 14 classes (10 fish species, 4 diseases), FISHYOLOv8 achieves a Mean Average Precision (mAP) mAP@50 of 0.9936 ± 0.0012 (p < 0.05) and 98.89% accuracy after 50 epochs, outperforming YOLOv8 and peers. With 52 ± 2 Frames Per Second (FPS), it offers a robust, real-time solution for aquaculture monitoring
Semi-Blind Timing Skew Calibration in TIADCs Using Second-Order Taylor Approximation and LMS Algorithm
Get PDFThis paper proposes a technique to calibrate timing mismatches in the digital domain for Time-Interleaved Analog-to-Digital Converters (TIADCs) based on second-order Taylor series approximation for application in high-speed communication and digital signal processing systems. By analyzing the error signal using second-order Taylor series approximation, the proposed technique estimates timing mismatches through the Least Mean Squares (LMS) algorithm to accelerate computation speed and reduce hardware resources. Subsequently, the timing mismatches are corrected based on the second-order Taylor series approximation. The effectiveness of the proposed technique is demonstrated through simulation results in MATLAB software. The simulation results show a significant improvement in the performance of the TIADC. Specifically, the Signal-to-Noise and Distortion Ratio (SNDR) and Spurious-Free Dynamic Range (SFDR) for a 4-channel TIADC are enhanced from 28.29 dB and 33.05 dB to 60.67 dB and 93.66 dB, respectively
HyPoNet: Fine-Grained Sleep Posture Recognition from a Single Abdominal Accelerometer
No full textFine-grained sleep posture recognition is essential for the non-invasive management of conditions such as gastroesophageal reflux disease (GERD) and obstructive sleep apnea (OSA). Traditional systems typically recognize only a limited set of coarse sleep positions, thereby restricting their clinical effectiveness in real-world scenarios. This study presents HyPoNet, a lightweight deep learning model designed to classify twelve distinct sleep postures using data from a single wearable sensor system. The proposed hardware platform consists of a tri-axial accelerometer (ADXL345) positioned on the abdomen, interfaced with a low-power microcontroller unit (ESP32) for real-time signal acquisition and wireless data transmission. Acceleration signals along the x, y, and z axes were collected from ten healthy participants performing twelve predefined sleep positions under controlled conditions. The collected data were segmented using a sliding window method, and a subject-independent evaluation strategy was applied: data from eight participants were used for training and validation (in an 80:20 split), while data from the remaining two participants were reserved for testing. HyPoNet employs a hybrid neural network architecture combining one-dimensional convolutional layers for spatial feature extraction with bidirectional long short-term memory (BiLSTM) units to model temporal dependencies in the acceleration signals. The model achieved a mean accuracy of 97.29% and an average F1-score of 90.72%, outperforming baseline models including CNN, GRU, and Transformer-based approaches. With its low computational footprint and high classification performance, HyPoNet offers a promising solution for embedded sleep posture monitoring in home-based and clinical settings
