Search Results (628)

As artificial intelligence (AI) advances, deep learning models are shifting from convolutional architectures to transformer-based structures, highlighting the importance of accurate floating-point (FP) calculations. Compute-in-memory (CIM) enhances matrix multiplication performance by breaking down the von Neumann architecture. However, many FPCIMs struggle to maintain high precision while achieving efficiency. This work proposes a high-precision hybrid floating-point compute-in-memory (Hy-FPCIM) architecture for Vision Transformer (ViT) through post-alignment with two different CIM macros: Bit-wise Exponent Macro (BEM) and Booth Mantissa Macro (BMM). The high-parallelism BEM efficiently implements exponent calculations in-memory with the Bit-Separated Exponent Summation Unit (BSESU) and the routing-efficient Bit-wise Max Finder (BMF). The high-precision BMM achieves nearly lossless mantissa computation in-memory with efficient Booth 4 encoding and the sensitivity-amplifier-free Flying Mantissa Lookup Table based on 12T Triple Port SRAM. The proposed Hy-FPCIM architecture achieves 23.7 TFLOPS/W energy efficiency and 0.754 TFLOPS/mm² area efficiency, with 617 Kb/mm² memory density in 28 nm technology. With almost lossless architectures, the proposed Hy-FPCIM achieves an accuracy of 81.04% in recognition tasks on the ImageNet dataset using ViT, representing a 0.03% decrease compared to the software baseline. This research presents significant advantages in both accuracy and energy efficiency, providing critical technology for complex deep learning applications. Full article

Air cycle systems, once largely replaced by vapour-compression technologies due to efficiency concerns, are now re-emerging as a viable and sustainable alternative for highly dynamic thermal applications and excel in ultra-low temperature. By using air as the working fluid, these systems eliminate the need for synthetic refrigerants and comply naturally with evolving environmental regulations. This study presents the conceptual design and simulation-based analysis of a novel air cycle machine developed for advanced automotive testing environments. The system is intended to replicate a wide range of climatic conditions—from deep winter to peak summer—through the use of fast-responding turbomachinery and a flexible control strategy. A central focus is placed on the radial turbine, which is designed and evaluated using a modular, open source framework that integrates geometry generation, off-design CFD simulation, and performance mapping. The study outlines a potential operating strategy based on these simulations and discusses a control architecture combining lookup tables with zone-specific PID tuning. While the results are theoretical, they demonstrate the feasibility and flexibility of the proposed approach, particularly the turbine’s role within the system. Full article

Deep Neural Networks (DNNs) are the primary approach for enhancing the real-time performance and accuracy of Keyword Spotting (KWS) systems in speech processing. However, the exceptional performance of DNN-KWS faces significant challenges related to computational intensity and storage requirements, severely limiting its deployment on resource-constrained Internet of Things (IoT) edge devices. Researchers have sought to mitigate these demands by employing Binary Neural Networks (BNNs) through single-bit quantization, albeit at the cost of reduced recognition accuracy. From an information-theoretic perspective, binarization, as a form of lossy compression, increases the uncertainty (Shannon entropy) in the model’s output, contributing to the accuracy degradation. Unfortunately, even a slight accuracy degradation can trigger frequent false wake-ups in the KWS module, leading to substantial energy consumption in IoT devices. To address this issue, this paper proposes a novel Probability Smoothing Enhanced Binarized Neural Network (PSE-BNN) model that achieves a balance between computational complexity and accuracy, enabling efficient deployment on an FPGA platform. The PSE-BNN comprises two components: a preliminary recognition extraction module for extracting initial KWS features, and a result recognition module that leverages temporal correlation to denoise and enhance the quantized model’s features, thereby improving overall recognition accuracy by reducing the conditional entropy of the output distribution. Experimental results demonstrate that the PSE-BNN achieves a recognition accuracy of 97.29% on the Google Speech Commands Dataset (GSCD). Furthermore, deployed on the Xilinx VC707 hardware platform, the PSE-BNN utilizes only 1939 Look-Up Tables (LUTs), 832 Flip-Flops (FFs), and 234 Kb of storage. Compared to state-of-the-art BNN-KWS designs, the proposed method improves accuracy by 1.93% while reducing hardware resource usage by nearly 65%. The smoothing filter effectively suppresses noise-induced entropy, enhancing the signal-to-noise ratio (SNR) in the information transmission path. This demonstrates the significant potential of the PSE-BNN-FPGA design for resource-constrained edge IoT devices. Full article

In this paper, we demonstrate a receiver-side lookup table (Rx-side LUT) equalization method for high-speed short-reach intensity modulation and direct detection (IM/DD) transmission systems, which alleviates the computational complexity of neural network-based equalization algorithms. Compared to conventional pre-equalization techniques applied at the transmitter side, which utilize distortion correction values stored in LUTs derived from the transmitted symbols and their corresponding recovered counterparts, the Rx-side LUT relies solely on receiver-side data. The received data to be equalized serves as the index of the LUT, with a nearest-neighbor algorithm employed to find the element closest to the index and then return the corresponding equalization result from the table. With a lightweight lookup process, the proposed method releases the computation complexity of neural network-based equalization algorithms by replacing the computationally intensive operations performed during the inference phase. Experimental results indicate that compared to baseline fully connected neural network (FCNN) and gated recurrent unit (GRU) equalization, the Rx-side LUT could decrease the algorithm execution time by 25.5% and 34.6% for 100 GBaud and 22.8% and 36.9% for 112 GBaud PAM4 signals, respectively, while maintaining comparable system performance. The proposed scheme provides a low-complexity solution for high-speed, low-cost IM/DD optical interconnects. Full article

This paper presents the first-of-its-kind full-crown Detached Eddy Simulation (DES) of a radial turbine designed for operation in a transcritical CO₂-based power cycle. The simulation domain contains not only the main blade passage but also the exhaust diffuser and the rotor disk cavities. To ensure accurate simulation of the turbine, two hybrid RANS/LES models, using the Improved Delayed Detached Eddy Simulation (IDDES) approach, are validated in a flow around a circular cylinder at $\mathrm{Re}=3900$, obtaining excellent agreement with other experimental and numerical studies. The turbine simulation was performed using the k-$\omega$-SST-based IDDES model, which was identified as the most appropriate approach for accurately capturing all relevant flow dynamics. Thermophysical properties of CO₂ are modeled with the Span–Wagner reference equation, which was evaluated by a highly efficient spline-based table look-up method. A preliminary assessment of the grid quality in the context of DES is performed for the full-crown simulation, and characteristic flow features of the main passage and cavity flow are highlighted and discussed. Full article

The need for unpredictable sequences of bits is common in many important security applications. These sequences can only be generated by true random number generators (TRNGs). Apart from the natural analog domain for TRNGs, this type of generator is also required as a digital-based solution, particularly leveraging field-programmable gate array (FPGA) platforms. Despite the number of existing FPGA-based implementations, new solutions that use different types of entropy sources, utilize fewer FPGA resources, or ensure higher throughput are still being sought. This paper presents an architecture of a simple TRNG targeted for implementation in FPGAs. As a source of entropy, the TRNG exploits jitter in capacitive oscillators and metastability in flip-flops. The capacitive oscillators, in turn, use the input–output cells of an FPGA chip and unconnected external pins and cyclically charge and discharge the parasitic capacitance associated with these pins. The TRNG needs a small number of FPGA resources, namely 13 look-up tables (LUTs), 12 flip-flops, and 3 unused pins. Its throughput is approximately 12.5 Mbit/s for AMD/Xilinx Artix-7 FPGA family chips. The presented TRNG passes all the NIST statistical tests for a wide range of operating conditions. Full article

Multispectral LiDAR point clouds, which integrate both geometric and spectral information, offer rich semantic content for scene understanding. However, due to data scarcity and distributional discrepancies, existing methods often struggle to balance accuracy and efficiency in complex urban environments. To address these challenges, we propose DA-GSGTNet, a novel segmentation framework that integrates Gated Stratified Graph Transformer Blocks (GSGT-Block) with Dynamic Aggregation Transition Down (DATD). The GSGT-Block employs graph convolutions to enhance the local continuity of windowed attention in sparse neighborhoods and adaptively fuses these features via a gating mechanism. The DATD module dynamically adjusts k-NN strides based on point density, while jointly aggregating coordinates and feature vectors to preserve structural integrity during downsampling. Additionally, we introduce a relative position encoding scheme using quantized lookup tables with a Euclidean distance bias to improve recognition of elongated and underrepresented classes. Experimental results on a benchmark multispectral point cloud dataset demonstrate that DA-GSGTNet achieves 86.43% mIoU, 93.74% mAcc, and 90.78% OA, outperforming current state-of-the-art methods. Moreover, by fine-tuning from source-domain pretrained weights and using only ~30% of the training samples (4 regions) and 30% of the training epochs (30 epochs), we achieve over 90% of the full-training segmentation accuracy (100 epochs). These results validate the effectiveness of transfer learning for rapid convergence and efficient adaptation in data-scarce scenarios, offering practical guidance for future multispectral LiDAR applications with limited annotation. Full article

This paper presents a multiscale histogram excess-distribution strategy addressing the structural limitations (i.e., insufficient dark-region restoration, block artifacts, ringing effects, color distortion, and saturation loss) of contrast-limited adaptive histogram equalization (CLAHE) and retinex-based image-contrast enhancement techniques. This method adjusts the ratio between the uniform and weighted distribution of the histogram excess based on the average tile brightness. At the coarsest scale, excess pixels are redistributed to histogram bins initially occupied by pixels, maximizing detail restoration in dark areas. For medium and fine scales, the contrast enhancement strength is adjusted according to tile brightness to preserve local luminance transitions. Scale-specific lookup tables are bilinearly interpolated and merged at the pixel level. Background restoration corrects unnatural tone compression by referencing the original image, ensuring visual consistency. A ratio-based chroma adjustment and color-restoration function compensate for saturation degradation in retinex-based approaches. An asymmetric Gaussian offset correction preserves structural information and expands the global dynamic range. The experimental results demonstrate that this method enhances local and global contrast while preserving fine details in low light and high brightness. Compared with various existing methods, this method reproduces more natural color with superior image enhancement. Full article

Accurate docking of Autonomous Guided Vehicles (AGVs) is a critical requirement for efficient automated production systems in Industry 4.0, particularly for collaborative tasks with robotic arms that have a limited working range. This paper introduces a cost-effective software-upgrade solution to enhance the precision of the final docking phase without requiring new hardware. Our approach is based on a two-stage strategy: first, a switch from a global dead reckoning system to a local navigation scheme, is triggered near the docking station; second, a dedicated Takagi-Sugeno Fuzzy Logic Controller (FLC), guides the AGV to its final position with high accuracy. The core novelty of our FLC is its implementation as a gain-scheduling lookup table (LUT), which synthesizes critical state variables—heading error and distance error—from limited proximity sensor data, to robustly handle positional uncertainty and environmental variations. This method directly addresses the inadequacies of traditional odometry, whose cumulative errors become unacceptable at the critical docking point. For experimental validation, we assume the global navigation delivers the AGV to a general switching point, near the assembly station with an unknown true pose. We detail the design of the fuzzy controller and present experimental results that demonstrate a significant improvement, achieving repeatable docking accuracy within industrially acceptable tolerances. Full article

In the V-cycle of distributed electric wheel loaders (DEWLs), transport accounts for about 70% of the cycle, making energy saving urgent, while shovel-stage slip limits traction stability. This paper proposes a two-module control framework: (i) a preferred-motor transport strategy that reduces parasitic losses and concentrates operation in high-efficiency regions; and (ii) a load-ratio-based front–rear torque distribution for shoveling that allocates tractive effort according to instantaneous axle vertical loads so that each axle’s torque respects its available adhesion. For observability, we deploy a pre-calibrated lookup-table (LUT) mapping from bucket cylinder pressure to the front-axle load ratio, derived offline from a back-propagation neural network (BP-NN) fit. Tests on a newly developed DEWL show that, compared with dual-motor fixed-ratio control, transport-stage mechanical and electrical power drop by 18–37%, and drive-system efficiency rises by 6–13%. During shoveling, the strategy reduces the peak inter-axle slip from 22–35% to 13–15% and lowers the mean slip to 2.6–5.9%, suppressing sawtooth-like wheel-speed oscillations without sacrificing peak capacity. The method reduces parasitic energy flow, improves traction utilization, and is readily deployable. Full article

In the context of autonomous vehicles, proper lane merging is critical as it can reduce the traffic bottleneck and lead to safer road transportation. To obtain a collision-free and efficient lane merging, advanced control algorithms need to be designed to smoothly coordinate multiple vehicles to form a platoon. Model predictive control (MPC) is such a controller capable of forecasting future states of multiple vehicles by optimizing their control inputs while satisfying the constraints. Prior MPC-based studies mostly utilized offline planning with a precomputed lookup table of feasible maneuvers to model lane merging. Although these model designs reduce the online computational load, they lack flexibility, as they rely on predefined scenarios and cannot easily adapt to dynamic or unpredictable situations. In this study, we present a centralized MPC framework capable of online trajectory tracking under dynamic constraints and disturbances, for collision-free operation in tightly spaced multi-vehicle platoons. To evaluate the flexibility of our online algorithm, we examine the role of prediction horizon—the time window over which future states are forecasted—and platoon size in determining both the feasibility and efficiency of merging maneuvers. Our results reveal that there exists an optimal prediction horizon at which braking and acceleration can be minimized, thereby reducing energy consumption by 35–40%. Additionally, we observe that increasing the prediction horizon beyond the minimum required for feasibility can alter the vehicle sequence in the platoon. Capturing the changes in vehicle sequence (e.g., who leads or yields) when prediction horizon varies, is a consequence of online trajectory optimization. This vehicle sequence change cannot be captured by offline planning that relies on precomputed look-up table maneuvers. We also found that as the number of vehicles increases, the minimum feasible prediction horizon increases significantly. Full article

A systematic and highly efficient current commands generator for an interior permanent magnet synchronous motor (IPMSM) in electric vehicles is proposed. This paper integrates maximum torque per ampere (MTPA), maximum power control (MPC), and maximum torque per voltage (MTPV) criteria for optimal efficiency, and systematically establishes an optimal current control commands workflow. A rapid current commands mapping technique and an automatic high efficiency of all speed range current command generator are proposed. The automatically generated commands table can be effectively applied in a motor controller to reduce the energy consumption of an electric vehicle for all operating speed range. A graphical user interface (GUI) tool for the generator, which can automatically produce the current command (look-up tables, LUT) in an Excel format, is designed. High-speed field-weakening and zero-torque cruising (ZTC) in electric vehicles are also thoughtfully considered. By using the proposed method, motor controller designers can more rapidly adjust required motor current command tables and speed up the development period. Both GUI simulation and experimental results show the effectiveness and feasibility of the proposed method. Full article

The exponential growth of multimedia communication necessitates advanced techniques for secure data transmission. This paper details a new data hiding method centered on a predictive voting mechanism that leverages neighboring pixels to estimate a pixel’s value. Secret data are concealed within these predictions via a purpose-built lookup table, and the retrieval process involves re-estimating the predicted pixels and applying an inverse mapping function. Experimental results demonstrate that the proposed method achieves an embedding capacity of up to 686,874 bits, significantly outperforming previous approaches while maintaining reliable data recovery. Compared with existing schemes, our approach offers improved performance in terms of both embedding capacity and extraction accuracy, making it an effective solution for robust multimedia steganography. Full article

Climate change results from disruptions in Earth’s radiation energy balance. Radiative forcing is the dominant factor of climate change. Yet, most studies have focused on radiative effects within the calculated actual albedo, usually overlooking the angle effect of regions with large-scale and highly varied terrain. This study produced the actual albedo databases by using albedo retrieval look-up tables. And then we investigated the spatiotemporal variations in land surface albedo and its corresponding radiative effects in the Yellow River Basin from 2000 to 2022 using MODIS-derived reflectance data. We employed time-series, trend, and anomaly detection analyses alongside surface downward shortwave radiation measurements to quantify the radiative forcing induced by land-cover changes. Our key findings reveal that (i) the basin’s average surface albedo was 0.171, with observed values ranging from 0.058 to 0.289; the highest variability was noted in the Loess Plateau during winter—primarily due to snowfall and low temperatures; (ii) a notable declining trend in the annual average albedo was observed in conjunction with rising temperatures, with annual values fluctuating between 0.165 and 0.184 and monthly averages spanning 0.1595 to 0.1853; (iii) land-cover transitions exerted distinct radiative forcing effects: conversions from grassland, shrubland, and wetland to water bodies produced forcings of 2.657, 2.280, and 2.007 W/m², respectively, while shifts between barren land and cropland generated forcings of 4.315 and 2.696 W/m². In contrast, transitions from cropland to shrubland and from grassland to shrubland resulted in minimal forcing, and changes from impervious surfaces and forested areas to other cover types yielded negative forcing, thereby exerting a net cooling effect. These findings not only deepen our understanding of the interplay between land-cover transitions and radiative forcing within the Yellow River Basin but also offer robust scientific support for regional climate adaptation, ecological planning, and sustainable land use management. Full article

Different optimization techniques for the SHAKE variant of SPHINCS+ are explored on an FPGA with the means to find a power-efficient model for resource-constrained devices. This work explores multiple hashing implementations, such as registering inputs and directly feeding data to hashing units, as well as different variations in hashing permutations per clock cycle. The design is evaluated based on resource requirements, the signature generation rate, and both static and active power consumption. This design shows a decrease in energy consumed per signature by 20% to 30% compared to other state-of-the-art SPHINCS+ implementations, while only using 12–14k lookup tables (LUTs), depending on the SPHINCS+ variant. Moreover, an amendment is proposed to the SPHINCS+ specification that allows for decreased processing time and memory consumption while maintaining the security level and non-deterministic properties. This is accomplished by rearranging the inputs in the random oracle model. Full article