Smart Systems and Devices

An Enhanced Multi-Stream Spatial Modulation Scheme for High-Rate MIMO Systems

Minh-Tuan Le, Trung-Hieu Nguyen — Tue, 10 Mar 2026 00:00:00 +0000

This paper proposes an Enhanced Multi-Stream Spatial Modulation (En-MSM) system employing Nt transmit antennas, where two antennas are activated simultaneously, for high-rate MIMO systems. En-MSM conveys information bits through both the spatial domain, i.e., the indexes of the active antennas, and PAM/QAM symbols. Each transmit signal vector, or transmit codeword, includes one M-PAM and one conventional N-QAM symbol, with the M-PAM symbols generated as even multiples. To further improve spectral efficiency, the M-PAM symbols are combined to form new M-APSK symbols, which replace the PAM symbols in the transmit vector. This design expands the available transmit codeword set and increases the minimum distance in the signal space. A sub-optimal detector is also proposed for signal recovery in the En-MSM scheme to support scenarios requiring low detection complexity, at the expense of some performance degradation. Simulation results demonstrate that En-MSM achieves noticeable performance gains compared with conventional MSM, ESM, and GSM-MIM across various scenarios, while maintaining comparable hardware complexity. Moreover, compared with SM, En-MSM not only achieves significant performance gains but also requires substantially fewer transmit antennas.

An FPGA-Based SoC Implementation of the CRYSTALS-Dilithium Post-Quantum Digital Signature Scheme

Tuan-Anh Dang, Nhu-Quynh Luc — Wed, 15 Apr 2026 00:00:00 +0000

The rapid advancement of quantum computing threatens the long-term security of conventional public-key cryptosystems, motivating the development of practical post-quantum cryptographic solutions. Among the algorithms standardized by the National Institute of Standards and Technology, CRYSTALS-Dilithium has emerged as a leading lattice-based digital signature scheme due to its strong security guarantees and implementation efficiency. However, efficient realization of CRYSTALS-Dilithium on resource-constrained embedded platforms remains challenging. This paper presents an FPGA-based System-on-Chip architecture for the CRYSTALS-Dilithium post-quantum digital signature scheme. The proposed design adopts a hardware–software co-design approach in which computationally intensive modules, including matrix expansion, SHAKE-256 hashing, and polynomial vector operations, are implemented in hardware to exploit FPGA parallelism while maintaining reasonable resource utilization. The architecture is implemented on a Xilinx Artix-7 (Basys-3) platform and evaluated using the NIST security level 5 parameter set at an operating frequency of 100 MHz. Experimental results show that the average key generation time is 2653.63 ms, while signing and verification require 2.695 ms and 1.105 ms, respectively. Performance analysis indicates that key generation dominates the total execution time, approximately 99.85%, primarily due to the complexity of the key generation process and the current non-optimized UART data transfer mechanism. These results demonstrate the functional feasibility of the proposed low-cost FPGA SoC architecture and provide a practical baseline for future performance and resource optimization of CRYSTALS-Dilithium accelerators in embedded post-quantum cryptographic systems.

Blockchain-Enhanced Chain of Custody with Deep Learning Tampering Detection for Digital Forensics

Xuan Hung Truong, The Dung Luong, Anh Tu Tran — Fri, 30 Jan 2026 00:00:00 +0000

The preservation of digital evidence integrity currently faces critical challenges, notably manual chain-of-custody errors ranging from 15-20%, the proliferation of sophisticated tampering techniques, and inherent scalability limitations. To address these issues, this paper presents an integrated framework that synergistically combines blockchain-based immutable custody chains, dual-branch convolutional neural networks for tampering detection, and hybrid consensus mechanisms. Through systematic ablation studies, we demonstrate that the proposed smart contract automation effectively mitigates manual custody errors, ensuring a tamper-evident and immutable custody record. Furthermore, our dual-branch architecture, enhanced with adaptive fusion (α =0.6), attains a 98.5% tampering detection accuracy—representing a 5.3% improvement over single-branch baselines. Additionally, the hybrid Proof-of-Authority and Byzantine Fault Tolerant consensus mechanism delivers a throughput of 10,000 transactions per second, marking a 1,429-fold improvement over traditional blockchain implementations. A comprehensive evaluation on the NIST CFReDS, supported by statistical validation (p < 0.001), demonstrates the superiority of our approach over six baseline methods. We further provide a detailed failure analysis, a computational cost breakdown, and validation through simulated forensic scenarios, alongside proposed integration pathways for commercial forensic tools such as EnCase and FTK to facilitate practical adoption.

GradNorm Physics-Informed Neural Networks for Linear Elasticity: Adaptive Loss Balancing and Comparative Finite Element Analysis

Phuong Cuc Hoang, Thi Thanh Mai Ta, Nam Nguyen Canh — Tue, 24 Feb 2026 00:00:00 +0000

This study presents a comparative analysis between the classical Finite Element Method (FEM) and Physics-Informed Neural Networks (PINNs) for linear elasticity. A well-known challenge in PINN formulations stems from imbalances among loss components associated with governing equations and boundary conditions, which often induce training instabilities and ill-conditioned optimization dynamics. To address this issue, we employ GradNorm, a gradient-based adaptive loss-balancing strategy, to dynamically adjust the weighting parameters during training, thereby mitigating optimization stiffness and improving convergence. The proposed PINN approach is systematically evaluated against high-fidelity FEM benchmarks across various geometries and loading conditions. Numerical results demonstrate that, while FEM remains a highly computationally efficient method for linear elastic problems, PINNs equipped with GradNorm-based adaptive weighting constitute a robust, mesh-free alternative with comparable accuracy. These findings underscore the efficacy of adaptive loss-balancing strategies for enhancing the reliability of PINNs in computational mechanics.

Determination of the Energized State Operating Point of an AC Contactor Coil Using a Bayesian Neural Network

Son T. Nguyen, Tu M. Pham, Anh Hoang, Tu A. Nguyen — Sat, 20 Dec 2025 00:00:00 +0000

The paper presents a method for determining the energized operating point of an AC contactor coil using the finite element method (FEM) combined with a Bayesian neural network (BNN) model. Under actual operating conditions, the magnetic flux density and air-gap length of the electromagnet directly influence the electromagnetic force within the contactor and, consequently, the overall device performance. However, these quantities are difficult to measure accurately in practice. The objective of this study is to propose a computational approach for estimating the magnetic flux density and air-gap length of the contactor by integrating machine learning techniques with FEM simulations. The proposed method establishes an electromagnetic model of the contactor coil, in which the magnetic flux density varies from 1.0 T to 1.5 T and the air-gap length ranges from 0.05 mm to 0.30 mm, generating the corresponding voltage drops across the coil. The simulated dataset is then used to train a BNN in an inverse inference direction, enabling prediction of the magnetic flux density and air-gap length from a known operating voltage. Based on these estimated quantities, the electromagnetic attraction force is calculated using FEM, facilitating analysis of the contactor’s operating characteristics and providing a foundation for design optimization in industrial applications.

Centralized Optimal Control with a Shared Signal for Identical Subsystems: Application to Multiple Four-Tank Systems

Thu, 29 Jan 2026 00:00:00 +0000

This study proposes a centralized control solution aimed at reducing implementation costs for industrial systems comprising multiple similar subsystems. Instead of equipping each actuator to execute separate control signals, the proposed method utilizes a shared control signal to simultaneously operate multiple independent systems that share identical setpoints and technical specifications. Analytical results demonstrate that this control architecture simplifies the hardware structure and reduces the required control resources, thereby lowering investment and maintenance costs. However, the use of a common control signal introduces certain trade-offs in control performance, such as increased delay and greater liquid level oscillation compared to independently controlled systems. To validate the approach, we conducted simulations on multiple clusters of four-tank experimental models under various initial conditions. The simulation results confirm that the proposed method ensures system stability and effective setpoint tracking. These findings suggest a promising direction for the development of centralized control architectures in large-scale or multi-agent systems.

IA-RATD: Industry-Aware Retrieval-Augmented Diffusion Models for Stock Price Forecasting

Nhat Hai Nguyen, VILAYVANH KENMANY — Sun, 26 Apr 2026 00:00:00 +0000

Stock price movements are inherently influenced by complex interdependencies among companies within and across industries. To effectively capture these relationships, we propose IA-RATD, an Industry- Aware Retrieval-Augmented Diffusion model for stock price forecasting. IA-RATD extends retrieval-augmented diffusion by incorporating industry level and interstock relationships to guide the denoising process in time series prediction. Specifically, our framework retrieves relevant historical stock sequences not only based on temporal similarity but also by considering structural connections in the market, enabling the model to leverage contextual information from related companies. Experiments on two major S&P 500 stocks, GOOG and AMZN, demonstrate that IA-RATD consistently outperforms baseline diffusion models, achieving up to 17.6% lower MSE and 28.8% lower MAE compared to state-of-the-art baselines. Our findings highlight the importance of integrating market structure awareness into diffusion-based time series models for financial forecasting. The implementation is available at: https://github.com/AppliedAI-Lab/RATD stock Diffusion, Industry-aware diffusion, Retrieval-augmented generations, Stock price

Control Design for Overhead Crane with Variable Cable Length

Trong Hieu Do, Vu Dai Trinh — Wed, 14 Jan 2026 00:00:00 +0000

Although overhead crane systems play a vital role in industrial operations, their control remains challenging due to underactuated, strong dynamic coupling, and payload oscillations. Existing approaches have shown effectiveness in reducing oscillations but often struggle with external disturbances, parameter variations, and the complexities of tuning. Moreover, many previous studies assume constant payload vibration frequencies, whereas real-world operations frequently involve varying rope lengths, leading to frequency changes and reduced controller performance. To overcome these limitations, this paper proposes a dual control framework: a proportional-derivative and sliding mode control (PD-SMC) strategy for trolley positioning and payload swing suppression, combined with an active disturbance rejection control (ADRC) scheme for cable length regulation. The PD-SMC ensures accurate and robust motion control under disturbances, while ADRC provides fast tracking performance with reduced modeling dependency. Simulation results validate that the proposed approach achieves precise positioning and effective swing suppression, even under varying rope lengths.

Fast-Convergent Stabilization and Trajectory Tracking of a Two-Wheel Self-Balancing Robot on Sloped Terrain

Dinh Hieu Pham, Manh-Linh Nguyen — Fri, 16 Jan 2026 00:00:00 +0000

The Two-wheeled Balancing Robot system is increasingly demonstrating its importance in both research and practical applications. Ensuring stability and flexible mobility on complex terrains, particularly sloped surfaces, remains a significant challenge. To address this issue, the research presented in this paper first establishes a precise mathematical model for the robot system operating on a slope. Building upon this model, the paper proposes a novel control strategy based on an improved Hierarchical Sliding Mode Control (HSMC) technique incorporating a Terminal Sliding Surface. The primary objective of this controller is to achieve extremely fast convergence speed, thereby simultaneously solving two key problems: maintaining stability at a fixed position on the slope and safely navigating the robot across the sloped area to reach a target destination in a 2D model. The research also provides an in-depth analysis of the system’s operating point on the sloped terrain and offers a rigorous mathematical proof of the overall system’s stability using Lyapunov stability theory. To validate the effectiveness, simulation results on the MATLAB/Simulink platform were conducted and directly compared with those of a conventional HSMC. The obtained results demonstrate that the proposed controller not only ensures higher stability but also exhibits superior responsiveness and performance in both assigned tasks.

Development of Unmanned Aerial Vehicle and IoT System for Water Quality Monitoring and Water Sampling

Pham Duc Dai, La Phu Hien, Uong Huy Hiep — Mon, 09 Mar 2026 00:00:00 +0000

Regular monitoring of water quality has become increasingly important due to rising pollution levels, which seriously harm human health, aquatic ecosystems, and decrease the performance of water treatment plants. Nowadays, monitoring water quality can be accomplished by traditional methods (e.g., taking samples and analysing them in a laboratory) and by water quality monitoring stations. These methods can only detect pollution as it occurs and spreads to the monitored areas. It is essential to develop a system capable of detecting pollution events early, allowing authorities to respond in a timely manner. Unmanned Aerial Vehicle (UAV) has emerged as an alternative approach for water monitoring for a large scale of reservoirs. The reason lies in the fact that, UAV equipped with remote sensing techniques and sensor nodes can be flexibly deployed to different places to collect water quality data in both spatial and temporal variations and are suitable for early detection of water pollution before it widely spreads. This paper proposed an efficient UAV platform integrated with IoT system to enhance efficacy of water quality monitoring and water sampling. In particular, an effective IoT framework combining LoRa and 4G communication networks improves data acquisition and facilitates control over long distances. Meanwhile, the UAV, with a high payload capacity, ensures the collection of sufficient water samples for laboratory analysis. The water quality data is also transmitted to a web server for storage, real-time visualization, and analysis. To demonstrate the efficacy of the UAV-assisted water quality monitoring system, it is applied to measure pH, total suspended solids (TSS), and temperature parameters, and to collect water samples from an area of the lake. The data collected by the UAV system is compared with the results obtained from laboratory analysis of water samples, revealing that the developed UAV system, while capable of being deployed flexibly over large areas, provides relatively accurate results and significantly reduces labor costs associated with water sampling.

A Study on the Proposed Structure, Materials, and Manufacturing Method for Smart Airbag Shells Protecting the Head of Elderly Vietnamese Users

Thao Phan Thanh, Nhung Nguyen Thi Trang, Anh Le Thi Mai, Thuy Dang Thi Thanh — Tue, 10 Feb 2026 00:00:00 +0000

Amid the global trend of population aging and the increasing risk of falls among older adults, the development of active protective assistive devices is of critical importance. Falls not only cause severe physical injuries but also exert significant sychological and financial burdens on the elderly and their families. Smart airbags, which are widely used as safety devices in the automotive industry, should be further investigated for potential applications in healthcare, particularly in protecting the head and other vulnerable regions of the elderly body during falls. This paper presents a comprehensive synthesis and analysis of published research related to the structural design, materials, and fabrication technologies of smart airbags. Based on this analysis, it proposes research directions that are tailored to the anthropometric characteristics of the Vietnamese elderly population and aligned with domestic manufacturing capabilities. The findings provide an essential foundation for the design and development of smart airbag systems for human body protection in general, and head protection in particular, aiming to reduce fall-related injuries among older adults.

A Multiple Channel Biometric Recognition Model Using Palm Images

Fri, 03 Apr 2026 00:00:00 +0000

Biometric identification technologies are playing an increasingly important role in securing authentication. Palm-based recognition is garnering significant attention due to the unique and reliable patterns, lines, and texture structures found on the human palm. This study proposes a novel palm-based biometric recognition system that utilizes state-of-the-art computer vision and deep learning techniques. Contrast limited adaptive histogram equalization (CLAHE) and histogram equalization (HE) are applied to enhance the visibility of features in the preprocessed images under varying lighting conditions. To improve recognition accuracy, a hybrid deep learning architecture is designed by integrating a pretrained ResNet-based backbone with a multi-channel framework. This approach effectively merges multiple distinguishing properties of the palm, including lines and texture patterns. The model is trained and evaluated on a dataset of 25,600 palm images from 128 individuals, captured in different locations and from various angles. Experimental results demonstrate strong performance, with high accuracy, sensitivity, specificity, and recall, reflecting the robustness and reliability of the system. This work contributes to a scalable and efficient solution for palm-based biometric authentication, offering a promising approach for secure identity verification.

Development of Multi-Stage Data Consistency Integrated U-Net Model for Undersampled MRI k-Space Reconstruction

Thi Hoa Bui, Trong Tuyen Nguyen, Duc Khanh Pham, Duc-Tan Tran, Anh Quang Tran — Mon, 20 Apr 2026 00:00:00 +0000

Reducing magnetic resonance imaging (MRI) acquisition time is critical to minimize patient discomfort, motion artifacts, and overall scanning costs. Compressed sensing (CS) has been widely investigated as a promising solution for MRI acceleration by reconstructing images from undersampled k-space data. However, traditional CS-based methods often suffer from instability, long reconstruction times, and limited performance due to their reliance on handcrafted regularization and iterative optimization schemes. Recent advances in deep learning have opened new possibilities for improving MRI reconstruction by learning data-driven priors directly from large datasets. Nevertheless, purely data-driven models may generate visually plausible images that lack strict consistency with the acquired measurements, potentially compromising diagnostic reliability. To address this limitation, we propose a deep learning-based MRI reconstruction framework that explicitly incorporates a data consistency (DC) layer. This physics-guided constraint enforces agreement between the network-updated k-space and the originally sampled measurements, thereby enhancing reconstruction fidelity. Experimental results demonstrate that the proposed model achieves a noticeable improvement in structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) compared with conventional reconstruction methods. These findings highlight the effectiveness of integrating learned priors with physics-based constraints, paving the way for faster MRI acquisition while preserving high diagnostic image quality in clinical practice.

Ovarian Ultrasound Image Segmentation with Limited Training Data

Fri, 17 Apr 2026 00:00:00 +0000

Ultrasound imaging is pivotal for ovarian tumor diagnosis, yet it poses significant segmentation challenges due to severe speckle noise, low contrast, and high inter-patient morphological variability. These challenges are further exacerbated by the limited availability of annotated medical data, making few-shot segmentation an appealing solution. Existing few-shot segmentation models like UniverSeg offer a promising direction for such limited-data scenarios but suffer from performance instability caused by stochastic support set selection. To address this, we propose a novel CLIP-guided support selection strategy that leverages the semantic embedding space of the Contrastive Language–Image Pre-training (CLIP) model to retrieve morphologically consistent support samples for each query. By replacing random sampling with a similarity-based retrieval mechanism, our method ensures better structural alignment between support and query images. Extensive experiments on two ovarian ultrasound datasets, OvaTUS and OTU_2D, demonstrate that our approach consistently outperforms the baseline UniverSeg and other few-shot methods. Specifically, on the OvaTUS dataset, our method achieves a Dice Similarity Coefficient (DSC) of 75.8% and Intersection over Union (IoU) of 64.9%, surpassing the random selection baseline by 2.1% and 2.7%, respectively. Furthermore, our approach shows superior robustness in extreme few-shot settings (N = 1), improving the Dice score by over 8% compared to the baseline. Code will be publicly released upon acceptance.

Bridging Vision and Language in Medical Imaging: Towards Robust Automated Report Generation from PET/CT Scans

Thanh Trung Nguyen — Tue, 24 Feb 2026 00:00:00 +0000

Vision-Language Models (VLMs) have achieved significant success in multimodal reasoning in general domains, yet their application to medical imaging remains limited, especially in specialized data domains such as PET/CT. In this study, we introduce Vietnamese Positron Emission Tomography - Vision-Language Model (ViPET-VLM), a novel pipeline specifically designed for medical report generation and visual question answering tasks on PET/CT data. ViPET-VLM integrates a fusion module to combine morphological information from CT with functional signals from PET, thereby forming a richer multimodal representation. To enhance clinical reliability, we propose a regularization mechanism with specialized loss functions that not only ensure diagnostic accuracy but also guide the model's attention to critical regions of interest in the images. ViPET-VLM was evaluated on a comprehensive, expert-validated PET/CT dataset and demonstrated marked improvements over current state-of-the-art methods in both report generation and medical question answering. The model shows potential for enhancing the accuracy and clinical applicability of VLMs for medical imaging.