We present a real-time, fully automatic vehicle traffic monitoring system based on phase-sensitive Optical Time Domain Reflectometry ( -OTDR) using distributed fiber optic sensors. The solution is designed for embedded on-the-edge deployment, leveraging a compact system composed of an FPGA with an integrated processor. It enables direct, in-stream processing of -OTDR signals without requiring data storage or offline analysis. The system emulates, among others facilities, the functionality of a simplified Hough Transform (HT) through a lightweight hardware architecture based on an array of binary shift registers, optimized for detecting near-vertical line segments that correspond to the signature of moving vehicles. The proposed solution demonstrates that existing optical fiber networks can be effectively repurposed for low-power, real-time traffic monitoring using dedicated embedded hardware. (Supplementary material available: https://github.com/marsistefano/ApplePie25 .)
This paper presents a collaborative GitHub framework we are developing to enhance project-driven learning in microcontroller-based embedded systems education. While common teaching methods provide foundational knowledge, students often struggle with system-level integration, real-world constraints, and professional workflows. Our framework addresses these gaps by structuring project development around version control, peer collaboration, and instructor-guided code reviews, simulating industry practices. Key features include scaffolded progression (modeling, implementation, and verification), and enforced documentation standards. A case study involving an IoT alarm system - developed using STM32 and ESP32 microcontrollers - demonstrates the framework’s effectiveness in teaching hardware/software co-design, debugging, and teamwork. Results highlight improved student engagement, motivation, and proficiency in industrial-grade development. The methodology emphasizes personalized mentorship in small groups, balancing autonomy with structured feedback. All materials are openly available to foster community adaptation.
Quantized neural network kernels using very-low-precision arithmetic such as 4-bit integer weights are gaining popularity for their reduced memory requirements and enhanced computational efficiency. However, standard processor architectures are often not optimized for such fine-grained computations. In this paper, we investigate the utilization of the SIMD Within A Register (SWAR) technique to efficiently execute quantized DNN kernels with 4-bit integer weights. Leveraging bit-level parallelism through SWAR, we achieve significant speedups (up to 7 × ) compared to standard implementations. We discuss the key SWAR-based implementation strategies and demonstrate their efficacy through experimental results. C++ source code available at https://github.com/lorenzograssi01/swaruint4 .
This paper presents the practical demonstration of a wireless sensor network (WSN) based on commercial nano-drones for structural integrity assessment of critical infrastructure. Leveraging the “Crazyflie 2.1” platform, each drone operates as a self-sustained sensor node equipped with vibration-sensing capabilities and an energy harvesting system. The proposed architecture supports dual-mode operation - flight and sensing - through a dual-MCU configuration that enables efficient power management and network integration. The system was validated on a full-scale damaged concrete bridge, where three drones were deployed to collect vibration data under operational conditions. Experimental results show strong agreement between the nano-drone measurements and those of benchmark commercial accelerometers, achieving a worst-case frequency identification error below 0.3
Accurate classification of Alzheimer’s Disease (AD) from brain MRI scans remains a critical yet challenging task due to limited annotated datasets and variability in image quality. This study presents a deep learning pipeline that enhances model robustness by first combining filtered images, denoised using the Non-Local Means (NLM) filter, with their original unfiltered counterparts, creating a hybrid dataset that captures a wide range of image conditions. To further improve training diversity, two augmentation techniques were applied: (1) Generative Adversarial Networks (GANs) for high-quality image generation and resizing, and (2) real-time conventional transformations, including rotation and flipping. The proposed framework was evaluated using three state-of-the-art convolutional neural network architectures—ResNet50V2, DenseNet169, and EfficientNetV2-B0. Among them, EfficientNetV2-B0 achieved the highest classification accuracy of 99
