ABSTRACT. Electric mobility has gained more attention within the transportation sector, where the development of innovative EV chargers must accomplish with the requirement of reduced charging times. Therefore, the implementation of more efficient and faster EV chargers is of utmost importance. This paper presents a comprehensive comparison between two topologies of three-phase AC-DC power converters: a full-bridge and a T-type. The AC-DC power converter must operate with sinusoidal AC current, and since it is linked to a back-end DC-DC isolated power stage, it is convenient the operation with variable DC-link voltage for enhancing the efficiency of the DC-DC stage. Along with the paper, it is presented a comparison regarding the switching power devices, sensors, gate-drivers, control complexity, as well as voltage and current stress in the switching power devices. It also presents an analysis in terms of efficiency and estimated loss distribution for different operating power levels and switching frequencies. In addition, it analyzes the influence of varying power levels and DC-link voltages on the total harmonic distortion of the AC currents. The results were obtained for different conditions of operation and considering balanced power and balanced current from the power grid point of view.

ABSTRACT. This work focuses on studying and designing a capacitive charge pump DC-DC converter for voltage elevation in radio frequency energy harvesting (RFEH) systems. The study aims to design a Dickson charge pump circuit that meets certain constraints and performance targets using components from the GPDK45. The article begins with a brief overview of RFEH systems, discussing their advantages and challenges. It then explores the operational principles of the Dickson charge pump, detailing its functionality and the key design considerations. The design process includes selecting and sizing MOSFETs and capacitors to optimize voltage gain and efficiency while accounting for the limitations imposed by the manufacturing process. Through simulations conducted, the performance of different transistor types and component dimensions is analyzed, allowing for an optimal balance between output voltage, efficiency, and power capability. Results indicate that the designed charge pump can achieve, for a constant input voltage of 0.5 V, an output voltage of approximately 1.48V with an efficiency of 34.8% for low-power applications around 1µW. The peak efficiency of 42.3% is attained at an output power of 2.2µW and an output voltage of 1.24 V.

ABSTRACT. DC-DC converters are fundamental to electric vehicle (EV) fast charging infrastructures, where the LLC DC DC resonant converter is widely adopted due to its soft-switching capability, high efficiency, galvanic isolation, and wide output voltage regulation. The high-frequency transformer in the LLC converter plays a key role in achieving such features, as well as for ensuring high power density. This study presents the design of a planar transformer that incorporates detailed modeling of critical parameters, including magnetizing and leakage inductance, flux density, losses, and parasitic capacitances. A custom ferrite core structure is proposed, consisting of ferrite tiles, ferrite bars, and a central ferrite block for a 25 kW LLC converter power module operating with an input voltage range of 650 V to 850 V and an output voltage varying between 200 V and 500 V. The proposed design was validated through electromagnetic simulations in ANSYS Maxwell, employing the finite element method (FEM), while the performance of the LLC converter was analyzed using PLECS simulations.

ABSTRACT. This work presents the design and analysis of a fully integrated, inductorless DC-DC step-up converter based on switched-capacitor techniques, targeting ultra-low-power intraocular applications. The proposed architecture is adapted from the work of Shih and Otis and introduces an auxiliary charge pump to improve overall efficiency. The main converter generates a regulated 1.2 V output from a low-voltage source (400 mV), using high-voltage clock signals provided by the auxiliary charge pump to enhance charge transfer. The entire control circuitry is powered by the converter itself and includes a start-up sequence based on a voltage reference and a comparator, enabling autonomous operation from ambient light. Key building blocks such as the reference current source, ring oscillator, and four-phase clock generator were implemented and validated through simulation. The converter can deliver up to 10 µA and demonstrates suitability for self-powered biomedical systems.

ABSTRACT. The dependence of power electronics converters is each more unquestionable, where a vast majority of topologies can be used supported by digital control algorithms. Normally, control is implemented using sensors, however, although less considered, sensorless techniques can also be applied, offering the possibility of reducing the hardware necessities, while increasing the control requirements. This paper proposes a sensorless current control applied to a DC-DC boost converter. The mathematical formulation of the sensorless technique is presented for contextualization with the DC-DC boost converter but is also illustrated how it can be derived for other voltage-source converters, controlled by current. The validation is performed by computer validation and corroborated by experimental results of a fully developed laboratorial prototype of a DC-DC boost converter. The results confirm the applicability of the proposed sensorless current control for different requirements of load operation.

ABSTRACT. This work explores the use of satellite imagery and open-source data to predict the operational status of floating offshore wind turbines, using the WindFloat Atlantic project as a case study. By integrating Sentinel-2 satellite images, metocean data, and offshore wind production data from ENTSOE, a machine learning model was developed to estimate the operational state of offshore wind turbines. The approach relies on using already available platform offsets from satellite imagery and correlating them with wind intensity and direction. Results show a clear relationship between turbine movement and operational status, particularly under higher wind conditions. Two classification models—Logistic Regression and Support Vector Machines—were evaluated, with Logistic Regression providing more stable performance across cross-validation scenarios. This study demonstrates the potential of remote sensing and open data to complement traditional monitoring systems, offering a scalable and cost-effective solution for offshore wind farm operation and maintenance strategies.

ABSTRACT. Water availability faces difficulties in balancing supply and demand. Factors such as climate change are causing a shortage of this resource, and the growing population is increasing the demand for it. Urban landscapes represent around 70% of urban water use and since they are privately owned, they are often difficult to access for data collection. This gap drives the need for new methods of estimating outdoor water consumption, since it can be a crucial input for decision-making in water distribution systems. This paper proposes the use of Remote Sensing (RS) and Artificial Intelligence (AI) models to collect data on the consumer’s outdoor water usage and tackle the problem of non-revenue water. In partnership with SMAS Almada, the aim is to study an area where apparent water losses are very high in order to help identify the specific location in the network where they occur and who may be responsible for them. To reach this goal, two YOLO models were created to detect and segment properties in satellite images. Using these models as a basis, a Water Consumption Pipeline (WCP) that could associate the coordinate of each property to its estimated water consumption, was developed. Employing the WCP, a database regarding the study area and a platform where users can visualize the estimated water consumption of the properties in a specific geographic coordinate were also created. This research provides a solution designed to help reducing water losses in critical areas and enhances water resource management by providing SMAS Almada with valuable data to improve asset supervision. Additionally, combining the information on billed water with the predicted usage, it is possible to perform a discrepancy analysis and gain insight on how the consumers are using water in residential areas.

ABSTRACT. This paper proposes an innovative, open-source and low-cost system, for data acquisition, processing and analysis, applied to the practice of technical swimming. This system allows not only to obtain swimming metrics, but also to detect technical errors in an athlete’s stroke execution through the usage of prototype wristbands, that allow the collection of inertial signals, a mobile application that allows the control of the wristbands, and a complete data processing and analysis system that determines the swimming style. In addition, the full stack system detects possible execution errors of strokes and evaluates the swimmer’s overall performance. The results obtained are based on real swimming data collected over several months from a high number of swimmers of varying skill levels. The proposed system achieves accuracy in the recognition of the different styles of 96.95%, 98.63%, 100.0% and 94.0% for crawl style, backstroke, breaststroke and butterfly, plus a stroke technique execution accuracy of 92.2%, using a Support Vector Machine-based approach in confined waters, i.e., an indoor swimming pool, real-time performance, surpassing state-of-the-art commercial solutions. A K-Nearest Neighbors distance-based instance classification algorithm approach is also explored in this work, with inline successful outcome.

ABSTRACT. As deep learning models continue to grow in complexity and size, their deployment on resource-constrained environments such as embedded systems and edge devices becomes increasingly challenging. Quantization has emerged as a key technique to reduce model size and improve inference efficiency by reducing the numerical precision of weights and activations. However, quantizing models without compromising accuracy remains non-trivial due to issues such as outlier sensitivity, activation range variability, and layer-specific behaviors. This paper presents a systematic and statistically grounded methodology for post-training quantization of deep learning vision models. Using AlexNet as a case study, the proposed workflow involves analyzing hyperparameter distributions, applying non-parametric statistical tests to guide quantization granularity, and visually inspecting parameter behavior across layers. The approach helps identify optimal clipping strategies and quantization configurations tailored to model architecture, enabling effective deployment with minimal performance degradation.

ABSTRACT. Semantic segmentation of medical images is an important task in healthcare, helping in the identification and localization of anatomical structures and pathological regions. Image segmentation classifies image regions at the pixel level, a powerful tool for rapid and accurate diagnosis. The deployment of these systems in real-world clinical settings is hindered by the computational demands of traditional software implementations, leading to slow processing times which discourages its daily utilization by professionals. To promote the utilization of mobile intelligent devices for medical analysis, accurate deep learning models with low complexity and energy-efficient and high-performance architectures are necessary. This work analyses several medical image segmentation models in terms of computing efficiency and performance/accuracy ratio. The most efficient model is chosen to be further optimized and executed in a low-power embedded computing platform. By optimizing the model and considering the computational characteristics of the target architecture, this work implements a mobile solution for medical image segmentation with high accuracy and real-time capabilities.

ABSTRACT. This paper presents a comparative study of a lightweight Convolutional Neural Network's (CNN) performance, for speech recognition, deployed in three microcontroller platforms with resource constraints. The devices used were the STM32F411CEU6 (ARM Cortex-M4), the Seeed Studio XIAO ESP32-C6 (dual-core RISC-V) and the RP2350 (dual-core Cortex-M33 and dual-core RISC-V). This study evaluates model time inference and energy consumption, with both quantized (Int8) and non-quantized (Float32) variants. To the best of the author's knowledge, this is the first implementation of an iterative spectrogram computation technique specifically optimized for low-memory microcontroller environments performing real-time inference. Additionally, the Pico TensorFlow Lite port was extended to support inference on the RISC-V cores of the Pico 2 W (RP2350), enabling performance comparisons between ARM and RISC-V architectures on a shared platform. Results reveal the trade-offs between ARM and RISC-V based microcontrollers regarding speed, power consumption, implementation complexity and overall performance, offering practical insights for the deployment of speech based AI on the edge.

ABSTRACT. This work aims to develop a platform for managing Augmented Reality (AR) and Virtual Reality (VR) devices, enabling multiple users to engage in collaborative and immersive experiences occurring both in the physical and digital world.

The proposed solution addresses challenges in synchronization, data transmission, and real-time rendering, so that the multiple participants - either co-located or geographically dispersed - can share a unified, high-fidelity experience.

This research intends to advance mixed reality technologies for education, entertainment, and business purposes.

To demonstrate this cross-device approach, an application using the MagicLeap2 and HTC Vive devices is implemented. This application allows various users to collaboratively interact within the same mixed reality environment.

Unity3D is the core framework of this system, and OpenXR is the layer that ensures hardware compatibility.

ABSTRACT. E-commerce has seen considerable growth, mainly driven by the COVID pandemic. With the return to normality, the problems of scheduling home delivery have increased. This problem has led to a boost in the use of PUDO (Pick-Up Drop-Off) solutions. Smart lockers stand out among these, as they are unattended and allow for extended hours of use. Existing locker solutions are usually based on pins and an iteration with a physical smart locker. The use of a mobile application makes the interaction with lockers more personalised and intuitive for both customers and couriers, making it the most efficient way to manage system flows and their data more homogeneously. This work explores different communication architectures between smartphones, local locker and central server to mitigate connectivity failures that may arise due to diverse installation environments. By implementing and testing various approaches, this research aims to identify the pros and cons of each approach in terms of reliability and security. The conclusions of this work contribute to the advancement of IoT-based cyber-physical systems in the smart locker ecosystem. To validate and test these approaches, a mobile application was developed and the existing backend server was modified to support the different communication architectures. This work was carried out in contribution with CTT.

ABSTRACT. This paper presents a methodology for modeling and importing a fixed-wing drone into a simulated environment with ease. The proposed approach improves physics approximations over current methods and optimizes parameter definitions between Gazebo and PX4. A custom automation script is developed to integrate aerodynamic properties using an Extended Vortex-Lattice Model (using AVL). Additionally, a parameter parsing mechanism is implemented to ensure consistency between simulation and real-world flight conditions.

ABSTRACT. The development of smart cities is increasingly driven by the Internet of Things (IoT), which introduces significant data flows and scalability challenges for urban networks. Integrating new sensors into large-scale, cost-effective, and sustainable infrastructures can disrupt existing network architectures. This paper presents a generic IoT sensor node metamodel that supports the description and emulation of diverse sensor node types, forming the foundation for a flexible emulation platform tailored to urban environments. The proposed solution includes a hardware and software architecture optimized for star topologies, employing low-cost, energy-efficient components and a Bluetooth-enabled configuration framework. The approach addresses key challenges such as accurate device behavior replication, scalability, and security, and is adaptable to emerging connectivity standards including LoRaWAN and NB-IoT. By providing a robust platform for testing and emulation in real-world conditions, this work addresses a critical gap in the development and validation of IoT systems for smart cities. A prototype implementation demonstrates the feasibility and effectiveness of the solution, supporting the rapid deployment of reliable urban IoT applications.

ABSTRACT. Efficient energy management in hybrid thermo-electrical microgrids is crucial for balancing thermal and electrical loads, particularly under uncertain demand and dynamic pricing conditions. Traditional deterministic and probabilistic methods often fail to capture such uncertainties, leading to suboptimal dispatch strategies and higher operational costs. This paper presents a robust energy management framework based on the fuzzy Monte Carlo method to model uncertainties in both thermal and electrical loads. Unlike conventional approaches, the proposed method enhances robustness and flexibility in responding to real-time price fluctuations by integrating a value-based pricing mechanism. The core contribution is the development of a coordinated energy dispatch strategy that simultaneously manages fuel cells, battery storage, and utility grid interaction, while incorporating demand response programs. The fuzzy Monte Carlo method enables a more comprehensive representation of both aleatory and epistemic uncertainties, supporting adaptive decision-making under volatile conditions. A realistic microgrid case study, including fuel cell, battery storage, and grid connection, is used to evaluate the proposed system. Simulation results demonstrate improved cost-effectiveness, operational stability, and load-balancing performance compared to conventional methods. The results validate the proposed framework as a viable and resilient solution for next-generation microgrid energy management systems under uncertainty.

ABSTRACT. The significant power fluctuations associated with wind farms pose a major challenge to grid power quality. One solution to mitigate these fluctuations is the use of energy storage systems. However, in battery-based storage systems, frequent and rapid charge-discharge cycles can accelerate battery degradation. To address this challenge, this work presents a hybrid solution that combines superconducting magnetic energy storage (SMES) and batteries. This choice relies on the fact that a key characteristic of SMES is its high power density compared to other energy storage technologies like batteries, along with a virtually illimited number of cycles. Hybridizing the storage system will prevent very fast or extreme charging rates, thereby reducing the risk of accelerated battery degradation. The control strategies for both storage systems are presented. The overall system is evaluated through simulation studies, which confirm the expected outcomes.

ABSTRACT. In recent years, non-residential buildings have increasingly adopted renewable energy generation systems to align with the European Union’s goal of achieving carbon neutrality by 2050. However, energy storage systems play a fundamental role in maximising the use of the generated renewable energy. Due to their high acquisition costs, adequately sizing these systems is essential. Moreover, applying an optimal scheduling strategy for energy storage operation can significantly improve the economic viability of such systems by reducing energy-related costs. In this paper, a MILP-based optimisation algorithm—incorporating battery lifespan constraints—is applied to a reference commercial building to schedule the operation of the storage system. A sensitivity analysis on the installed photovoltaic power and energy storage capacity is performed to evaluate their impact on the economic and operational performance of the optimisation algorithm under different sizing configurations.

ABSTRACT. With the growing interest in replacing fossil fuels with renewable energy, grid operators need to look at the possibilities of renewables being ancillary service providers. Due to their quicker power up times and smaller power requirements to restart after a complete blackout, photovoltaic power plants are looked on favorably to be black start units in modern grids. As testing of black start control strategies is not feasible in real scenarios and real time simulators are often used as a means to validate new control strategies and train operators. In this work, a simple yet effective control strategy is implemented and validated through real time simulation. Demonstrated results indicate that the designed control method can be used to enable black start from photovoltaic power plants.

ABSTRACT. This paper proposes a new hybrid Maximum Power Point Tracking (MPPT) algorithm combining the Particle Swarm Optimization (PSO) and Crow Search Algorithm (CSA) to enhance the performance of Maximum Power Point (MPP) tracking. The adopted approach starts with a global exploration of the search space, where 10 agents are initially uniformly distributed to identify the most promising region containing the Maximum Power Point (MPP). Once this region is determined, CSA is used to refocus all search agents to this restricted area. Then, PSO takes over for a more precise local optimization, thus ensuring a balance between efficient exploration and fast convergence. The performance of the hybrid PSO-CSA is evaluated by comparing it to the CSA-only and PSO-only methods, according to three criteria: accuracy, convergence time and reliability. The tests are performed under four irradiance profiles, including standard conditions (1000 W/m², 25°C) and three partial shading scenarios. Each test is repeated 100 times, and the standard deviation is used to measure the reliability of the results. The simulations are performed under MATLAB/Simulink on a photovoltaic array composed of six MSX-60 modules connected in series. The results show that the developed hybrid algorithm achieves higher accuracy, accelerated convergence, and improved stability compared to methods based solely on CSA or PSO.

ABSTRACT. This study presents the development of a wearable sensor designed to assess gait asymmetry. The device uses the TF mini S LiDAR sensor to detect mid-stance and calculates the gai cycle duration to assess symmetry. The Arduino Nano RP2040 Connect microcontroller was chosen for its compact size, real-time data processing capabilities, and seamless integration with the Arduino Cloud, facilitating remote monitoring and data storage. The TF mini S LiDAR was found to be the most suitable sensor due to its field of view and high frame rate, allowing accurate mid-stance detection. The final device combines portability, high data update rates, and precise detection, making it an effective tool for monitoring gait rehabilitation, including LLA. This wearable sensor contributes to improving prosthetic adaptation and offers healthcare systems a reliable means of monitoring patient progress.

ABSTRACT. Visually impaired individuals face many challenges, from identifying common objects and reading to social problems such as face recognition or picking up social cues. Most of these problems require assistance from other people or tools, such as white canes or guide dogs. However, despite being useful, these tools have limitations that can be addressed with modern technology with the recent advancements in artificial intelligence and machine learning. This work proposes a Smart Object Detector System that aims to increase the autonomy and safety of visually impaired individuals while implementing modern deep learning models without relying on external servers on the edge or in the cloud. The system optimizes a lightweight version of the most recent version of the YOLO (You Only Look Once) object detector model, YOLOv11n, for fast execution, low resource requirements, and low energy consumption. A set of hardware-oriented optimizations, such as These include quantization, pruning and knowledge distillation, are explored to maximize the model's efficiency when executed in a low cost, low energy Tensor Processing Units. The system focuses on aiding the visually impaired in navigating the outside world by identifying marks such as crosswalks or vehicles while also keeping in mind comfort, reliability, and ease of use with the goal of making it as easy to adopt as possible. The results demonstrate the viability of using a highly accurate object detection model in a low-cost device with marginal accuracy reduction (less than 5% compared to the original model) that runs locally and guarantees real-time response in less than 100 ms.

ABSTRACT. Photonic Integrated Circuits (PICs) have emerged as a highly promising technology for biomedical sensing applications. They offer interesting characteristics such as cost-effectiveness, rapid response, high sensitivity, and label-free detection. Typically, their operating principle relies on detecting variations in the refractive index at the device's surface upon interaction with a biological sample. PIC-based biosensors have become one of the most appropriate technology for lab-on-chip (LOC) due to their real-time diagnosis, extreme sensitivity, robustness, reliability, and their potential for multiplexing and low-cost mass production. These applications are redefining the boundaries of diagnostics and research since they distinguish themselves by their capability to detect multiple interactions within a single sample simultaneously. The multimode interferometers (MMI) are fundamental devices, as they can combine or split the optical power between different input and output channels, which can provide reference channels and enables parallel sensing of multiple biomarkers. This work presents a new sensor design for a multichannel refractive index sensor based on photoresist polymer waveguides fabricated on a SiO2 substrate, enabling a straightforward fabrication process using spin coating. The proposed architecture consists of a 1×2 MMI splitter followed by a three-waveguide directional coupler. We demonstrate using numerical simulations that the output power of the waveguides is sensitive to changes in the surface refractive index, providing the foundation for an effective sensing mechanism.

ABSTRACT. This paper presents a real-time seismic monitoring system based on low-power Internet of Things (IoT) nodes equipped with Micro-Electromechanical Systems (MEMS) accelerometers and Machine Learning (ML) algorithms for earthquake detection. The system employs edge computing to process data locally to lower latency and reduce cloud infrastructure dependence. Designed for energy-efficient operation, this approach enables continuous monitoring in remote or resource-constrained environments. The addition of ML enhances the detection of seismic events, offering a cost-effective and scalable solution for early warning applications. Performance metrics were used to evaluate how effectively the ML model performs, achieving high accuracy (92%) and precision (94%).

ABSTRACT. pH is a physiological parameter of great importance in biomedical research. For example, during cell growth process, pH is usually lowered due to the acidic metabolites released by the cells. This decrease on the pH value can affect biological processes and, consequently, affect the accuracy of the research. To monitor the pH changes along cell engineering experiments, cell culture media is usually complemented with a pH colorimetric indicator. However, this does not quantify the pH value, giving only an indication of pH ranges. To overcome this limitation, especially for biomimetic microsystems, where homeostasis has to be secured for long periods of time, an optical sensor that quantifies the pH values, through optical transmittance reading, was developed with features that allows integration. The successful integration of the pH sensing module in a microfluidic bioreactor was achieved with a pH-detecting microchamber of only 1 mm x 1 mm (depth x width), showing the potentiality of this sensing technology to be integrated with a broad bioplatforms used for biomedical research. Overall, the integrated beam-splitter setup allowed to obtain a sensor with an optical path of 2 mm, in a 1 mm microchamber, with a sensitivity of 8.5%/pH in terms of optical transmittance response.

ABSTRACT. Using a Design-Oriented 5 DC-Parameter MOSFET model along with simulation results derived from advanced 16 nm technology, we demonstrate that a zero temperature coefficient (ZTC) bias point does not exist in the intrinsic gain of the transistor, represented by $g_m r_{ds}$. Instead, it was found that a ZTC zone is present when the transistor operates in a well-saturated condition. Although some temperature dependence persists within this ZTC zone, it is characterized by a low complementary to absolute temperature (CTAT) behavior, as indicated by an effective temperature coefficient ($TC_{eff}$) under 100 ppm/$^\circ$C, i.e., 0.01\% per each 1 $^\circ$C. Furthermore, both theoretical analysis and simulation results reveal that the ZTC zone in both strong and weak inversion does not manifest in triode operation due to the pronounced CTAT behavior of $r_{ds}$, which is not adequately compensated by the proportional to absolute temperature (PTAT) behavior of $g_m$. This research highlights the complexities of temperature dependence in MOSFET operations and introduces significant insights into transistor behavior at the nanoscale.

ABSTRACT. The Mach-Zehnder Interferometer (MZI) has been extensively utilised in a variety of modern photonic circuits. With the integration of optical switches, this component serves as a fundamental element in the development of more sophisticated structures in the contemporary era of the photonics industry. In this paper, we propose the design of a 2x2 programmable optical matrix using MZI interferometers,directionals couplers and the thermo-optical effect (TOE) as a control technique to obtain the change in the optical signal in the Bar, Cross, Partial states, resulting in a thermally reconfigurable photonic integrated circuit (PIC). The forward-only waveguide network architecture, in which light flows through each element in only one direction, will be employed. Synopsys Rsoft software will be utilised for simulation and results. The objective of this study is to analyse the design of a basic structure, which can be conceptualised as a block within a matrix or a more advanced solution for photonic applications, including, reconfigurable photonic networks, Wavelength Division Multiplexing (WDM) and artificial neural networks.

ABSTRACT. Energy harvesting (EH) is a surprisingly promising solution for sustainable IoT systems, enabling self-powered operation by converting ambient energy into usable electrical energy. This study investigates five classical DC-DC converter topologies - Boost, SEPIC, Cùk, Super Lift Luo and Flyback - by analyzing their behavior when supplied by ultra-low voltage (hundreds of mV) EH sources. The simulations were carried out to evaluate the voltage step-up capabilities, efficiency, sensitivity to load resistance, and electrical performance with different duty cycles. According to the comparative study with similar simulation conditions, the Super Lift Luo converter showed the higher voltage conversion ratio, and the Boost converter was found to be the most efficient (69.4%). The results highlight the importance of selecting suitable converter topologies based on input voltage, efficiency and load requirements. This preliminary study concludes that the Super Lift Luo and the Boost converters are the most promising for ultra-low voltage applications.

ABSTRACT. This paper presents a fully automatic modeling and parameter extraction strategy for the above-threshold and sub-threshold characteristics of Indium Gallium Zinc Oxide (IGZO) thin-film transistors (TFTs) in both linear and saturation regions of operation. A semi-physical analytical description of the current-voltage characteristics, which is captured using a single unified expression applicable to both operational regions is considered.

ABSTRACT. This paper offers a groundbreaking evaluation of the mismatch robustness of Hot Spot Bias Points (HSBP), approached through a simulation-based methodology. The HSBP consists of three distinct bias points: the Zero Temperature Coefficient (ZTC), the Transconductance Zero Temperature Coefficient (GZTC), and the Zero Distortion Bias Point (ZDBP). The ZTC and GZTC are characterized as the MOSFET drain current and transconductance that exhibit insensitivity to temperature variations, while the ZDBP represents the point at which the transconductance achieves its maximum derivative, thus resulting in a zero third-order distortion. Understanding the variability of these bias points under mismatch variations is crucial for guiding Integrated Circuit (IC) designers. Such insights enable designers to leverage the advantages of HSBPs while also recognizing their limitations in terms of robustness. Utilizing a commercial 16 nm FinFET Process Design Kit (PDK), this study investigates several devices. Monte Carlo simulations reveal that, for the minimum (maximum) transistor area of the core device, the following standard deviations and mean ratios are σ(ZTC)/µ(ZTC)=1.78%(0.070%), σ(GZTC)/µ(GZTC)=2.31%(0.066%), and σ(ZDBP)/µ(ZDBP)=6.91%(1.87%) for the ZTC, GZTC, and ZDBP, respectively. These metrics are always normalized with respect to the threshold voltage (V_T0).of the core device. Furthermore, the robustness of the HSBPs across low voltage, ultra-low voltage, and high voltage scenarios is also demonstrated.

ABSTRACT. In current mobile networks, the amplification stage is the factor which limits the Radio Access Network (RAN) energy efficiency, with the most impactful blocks being the Power Amplifiers (PAs) and power combiners. This is because available solutions have to accommodate strict linearity requirements, leading to sacrifices in efficiency. A survey was performed for the energy efficiency of common PA solutions, such as standalone, Doherty and outphasing PAs, highlighting how they either fail to offer efficient operation or do so while posing significant trade-offs, which includes addressing the issues with circuit-based power combiners. In alternative, parallel switched-mode PA and over-the-air (OTA) power combination structures are highlighted, which consider the opposite design approach, by maximizing efficiency through simultaneous PA and architecture-level optimizations, then fulfilling linearity requirements through signal processing techniques, meeting the necessary requirements for next-generation mobile networks. Optimizations for further energy-efficiency improvement as well as currently open problems are presented.

ABSTRACT. Vehicular Delay-Tolerant Networks (VDTN) address the communication challenges inherent in vehicular environments characterized by intermittent connectivity and dynamic mobility patterns. This paper investigates the simultaneous use of social and geographical routing mechanisms in VDTNs. Social information can be learned automatically from previous contact history, while geographical information is assumed to be received on each node from a GPS system and ex-changed with neighboring nodes. The routing protocol variants studied are variants of the spray and wait protocol, where both the spray and wait phases use either social or geographical in-formation. The simulation results show that the mechanism used in the search phase has more impact on the routing performance. GreedySocial, a combination of Greedy spraying of messages followed by Social (dLife-based) search of the destination had the best (higher) delivery proba-bility for higher node density situations, although for lower density scenarios its spray mechanism is not aggressive enough.

ABSTRACT. Ultra-wideband (UWB) systems are critical for indoor positioning in robotics, industrial tracking, and asset management due to their accuracy in multipath-prone environments. Like GPS satellites requiring precise orbital data, UWB systems depend on well-calibrated anchors—fixed reference points whose positional accuracy directly impacts location estimates. We systematically evaluate and compare computational calibration methods, such as Genetic Algorithms, Maximum Likelihood, and Extended Kalman Filter, using synthetic data, assessing both efficiency and error reduction in calibration and location. Nonlinear Least Squares (NLS) outperformed other approaches from this review as well as state-of-the art methods, reducing anchor calibration errors to 10.7cm (86.03\% improvement from 1-meter initial uncertainty) and tag localization errors to 5.6cm (88.35\% reduction). NLS maintained computational efficiency (mean execution time of 0.011s, proving ideal for real-world deployments where efficiency and accuracy are critical.

ABSTRACT. Sign language (SL) translation aims to facilitate communication between deaf and hearing individuals. The complex, and yet important, task has encouraged the development of innovative tools, but the literature lacks in examples of solutions that represent the naturalness, continuity and multimodality required for sign language sentences.

This paper explores an innovative approach to the development of a continuous sign language recognition (CSLR) and translation system, by applying language learning principles. The proposed system is divided into two parts: the first explores a new approach, based on children’s language learning process, to the training of a deep learning (DL) model, while the second focuses on the development of a multimodal DL architecture capable of recognising and translating continuous SL, using the implemented approach. The paper enphasis is put on the first part of the system, highlighting the key design choices aimed at mimicking human language acquisition in the model training process, the main differences between this and traditional state-of-the-art (SOTA) methodologies, and the advantages compared to other continuous sign language translation (CSLT) solutions.

Although this work will be directed at Portuguese Sign Language, Lingua Gestual Portuguesa (LGP) translation, if successful, the methodology could be extended to other sign language translation systems or even broader natural language processing (NLP) tasks, contributing to a more inclusive and accessible communication framework.

ABSTRACT. Sign language is the primary means of communication for deaf individuals or those with hearing impairments. However, as the population does not widely use it, less development has been carried out, such as in verbal language translation. This work proposes a recognition method that enables the translation of sign language into verbal language in real-time and as accessible as translations between verbal languages. The proposed system uses Google MediaPipe framework to extract landmarks from video frames, an adapted Transformer model to translate these landmarks into glosses, and a fine-tuned GPT-4o mini model to convert the glosses into verbal language. With the proposed methodology two models were trained. The first model was trained with available data from the SIGNUM dataset, where 92.93% of accuracy was achieved, with a signer that the train never visualized. The second model was trained with a database augmented with new sentences, concatenating isolated signs taken from the original. This model is used for real-time continuous translation and obtained an accuracy of 84.14%.

ABSTRACT. Analysing user-generated content from social media can potentially provide significant insight for organizations into how their customers perceive the provided services, and what potential pitfalls they fall into. However, the appropriate collection, preparation, and analysis of content is challenging. The main objective of this study is to explore methods for analysing social media content, specifically user messages from X (ex. Twitter), on airports, with the aim of discovering information that can be used to improve airport service quality (ASQ). Another objective is to find out if accessibility is discussed, what aspects of it are discussed, and how broad the discussion is at different airports.

The study uses a dataset of over 1 million tweets containing mentions of international airports. First, collected data is cleaned and preprocessed. Next, the messages that discuss airport services are filtered using a rule-based technique based on the Airport Council International (ACI) Airport Services Quality (ASQ) measurement framework and a custom dictionary of key words and phrases assigned to each service category. The presence of various airport services in the data is further studied, with special attention to accessibility. Sentiment analysis is performed to understand the polarity of customer perception on various airport services. To calculate polarity scores, the VADER lexicon is used. With polarity scores, each tweet is further assigned to a certain sentiment category. Finally, polarity scores are visualized for each airport service category, presenting a convenient tool for assessing the overall ASQ. The final part of the study is dedicated to finding AI-based alternatives to the rule-based technique for detecting user messages related to airport services. For this task, five off-the-shelf machine-learning algorithms from the popular Python library Scikit-learn are trained and tested for multi-label classification.

The custom rule-based technique for detecting the discussion on airport services evidently achieved a high accuracy of about 93%, although its capability is limited. Applying this technique showed that more than 60% of messages do not touch airport services, even if all tweets were collected by a criterion of mentioning a popular international airport. Around 0,5% of all messages concern accessibility and are mostly focused on the topic of reaching the airport. The results of the study show that the frequency of different ASQ related topics varies considerably. Access, Waiting, Facilities and Passport, all appearing in over 6% of the tweets, were relatively frequent. On the other hand, ASQ topics such as wayfinding and check-in appear only in less than 3% of the tweets. Even though accessibility was dealt by only 0,5% of the tweets in our data set of over 1 million tweets, we were able to find interesting differences between specific airports as well as the salient subtopics related to accessibility. The sentiment analysis showed a slightly positive overall sentiment in the tweets. However, interesting differences can be observed between the different services, with Passport having a positive overall sentiment and Waiting and Arrival a negative overall sentiment

ABSTRACT. Synchronous reluctance motors (SynRM) have very promising characteristics in the field of energy efficiency, because compared to induction motors, they do not have rotor currents, and compared to synchronous motors they do not need permanent magnets. Thus, the use of SynRM in variable speed applications, through electronic frequency converters, can achieve IE4 (Super Premium) efficiency classes, while induction motors are in IE1 (Standard) and IE2 (High) classes and synchronous permanent magnet motors achieve IE3 (Premium) and IE4 (Super Premium Efficiency) classes, according to international energy directives. Due to the current design techniques of electrical machines using the finite element method, frequency electronic converters, microprocessors, and control techniques, SynRM has emerged in recent years in the field of high-performance variable speed drives. The aim of this paper is to obtain the efficiency map of the SynRM through computer simulation using two-dimensional finite element analysis software and through experimental tests.

ABSTRACT. The objective of this paper is to delineate the ongoing doctoral research work that is focused on the development of a digital model intended to emulate the real-time operation of an electrolyzer that is powered by a DC/DC converter. The digital model of the converter and the proton exchange membrane (PEM) electrolyzer (EL) is presented, and it is based on an electrical equivalent model. A primary contribution of this study is the analysis of the errors resulting from the discretization process. Furthermore, the implementation and development of the digital model requires a comprehensive study of the errors and key affecting factors. Additionally, the formulation of a mechanism to reduce these errors is essential for advancing this topic. Preliminary results obtained using the digital emulator developed demonstrated its capacity to reproduce the voltage and current response applied to the electrolyzer with an error of 5% in relation to the continuous-time model.

ABSTRACT. This paper implements and analyzes the Synchronous Reluctance Machine (SynRM) and the Permanent Magnet Assisted Synchronous Reluctance Machine (PMASR). First, tests are carried out to assess the behavior of the SynRM machine without permanent magnets. Subsequently, the PMASR is analyzed in two stages: first, with the addition of Ferrite permanent magnets, and then with the addition of Neodymium-Iron-Boron (NdFeB) magnets. The purpose is to improve SynRM's performance, making it suitable for electric traction systems. The work developed interconnects the analytical models of the SynRM and PMASR, with their finite element models, where their operational analysis is verified through temporal numerical simulations related to the speed control of these motors. This approach makes it possible to observe and study the magnetic behaviour of the machine in detail by analysing the magnetic field lines, the magnetic flux and the value of the inductances along the d-axis and the q-axis. Based on the results obtained, this work intends to provide a comprehensive understanding of the impact of different magnetic materials on the performance of the PMASR, therefore contributing to the development of more higher power density electric machines for applications in electric traction systems, particularly for applications in electric vehicles.

ABSTRACT. This work explores optical communications using Visible Light Communication (VLC) technology. Utilizing RGBV LEDs, VLC is applied to urban traffic scenarios, such as intersections, to collect real-time traffic data, including vehicle position, speed, queue lengths, and waiting times. VLC transmitters installed on streetlights, traffic signals, and vehicle headlights enable vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I), infrastructure-to-vehicle (I2V), and lamp-to-vehicle (L2V) communications. Each communication type employs a 64-bit VLC frame carrying various traffic-related data, following a predefined VLC protocol.

The proposed work involves developing a simulated traffic scenario that replicates vehicle movements and active VLC communications, generating frames automatically. This requires integrating the SUMO traffic simulator with LED control hardware to automate the entire process. On the reception side, decoding algorithms will be implemented to identify and decode the optical signals transmitted by each VLC source.

These algorithms for demultiplexing and decoding the transmitted optical signals enables data recognition and mobile device localization within the surrounding environment. The study aims to evaluate the optimal topology (orthogonal, hexagonal, or mixed) for deployment within buildings to enhance system performance in terms of information transmission, indoor positioning, and user movement direction. Signal decoding techniques, such as parity control bits, will be optimized to minimize the system’s Bit Error Rate (BER).

ABSTRACT. This paper presents a novel method for automatic detection of red light running violations based on video analysis and advanced artificial intelligence techniques, specifically designed for embedded platforms. The system integrates a CNN-based traffic light classifier, the YOLOv8 object detector, and the ByteTrack tracking algorithm to identify vehicles and analyze their trajectories relative to traffic signals within user-defined regions of interest. Unlike traditional approaches relying on costly infrastructure or cloud processing, the proposed method runs entirely on-device, enabling real-time inference with reduced latency and increased data security. A prototype was developed on an NVIDIA Jetson Orin Nano and tested using real-world traffic videos under varying lighting and weather conditions. Experimental results demonstrate high detection accuracy and reliable operation, achieving frame rates of up to 29 fps. These findings confirm the method’s effectiveness and efficiency, offering a scalable, low-cost solution for intelligent traffic enforcement in resource-constrained environments.

ABSTRACT. This study integrates Visible Light Communication (VLC) and a Deep Reinforcement Learning (DRL) system to enhance traffic signal control, reduce congestion, and improve safety through real-time data-driven management. VLC leverages existing infrastructure to transmit real-time data on vehicle and pedestrian dynamics, while trained agents optimize traffic signals and vehicle trajectories across intersections. A centralized system trains a unified DRL model to coordinate local agents managing individual intersections, enabling real-time signal adjustments via a queue/request/response methodology. Simulations and realworld trials validate the approach, showing significant reductions in waiting and travel times, particularly under rerouting scenarios. Scalable to diverse intersection types, the system adapts dynamically to changing traffic conditions, improving efficiency and safety.