IEEE Latin America Transactions

LiteML-Edge: A Contract-Centered Framework and Methodology for Replay-Driven 1:1 Python--Firmware Validation in TinyML

Tue, 20 Oct 2026 00:00:00 +0000

Tiny Machine Learning (TinyML) enables datadriven inference on resource-constrained microcontrollers; however, offline evaluation and firmware execution are often treated as loosely coupled stages, leaving reproducibility gaps between reference experiments and on-device behavior. This paper presents LiteML-Edge, a contract-centered TinyML framework and methodology for replay-driven 1:1 Python–Firmware Validation. LiteML-Edge integrates training, optimization, quantization, artifact export, firmware deployment, and per-stage validation while exporting replay artifacts, preprocessing rules, tensor I/O metadata, output mapping, post-processing, and the Rolling-24 metric policy as a shared validation contract. The framework supports fully quantized, fully floating-point, and hybrid tensor interfaces within a single workflow. The evaluation covers one environmental sensing scenario, a fixed W = 24 policy, and the TensorFlow-to-TFLM deployment path. On a WEMOS LOLIN32 (ESP32) and a NUCLEO-F411RE (ARM Cortex-M4), case studies based on a multilayer perceptron (MLP), a long shortterm memory network (LSTM), and a compact one-dimensional convolutional model (Conv1D Tiny) showed how the framework localizes agreement or mismatch across preprocessing inputs, model payloads, raw outputs, final predictions, and Rolling-24 metrics. Under replay on both targets, MLP and Conv1D Tiny achieved contract agreement across the critical path, whereas LSTM preserved input-stage agreement but showed its first reproducible divergence at the raw-output stage. Field execution is retained as operational evidence collected under the same contract, complementing replay without replacing it for equivalence claims. The resulting workflow makes validation failures more attributable, reproducible, and auditable before deployment decisions are made, while distinguishing contract conformance across platforms from hardware-dependent deployment costs.

Systematic Analysis of Critical Factors for Value Management in Cybersecurity

Vicente Merchán Rodríguez, Danny Zambrano Vera (Author) — Tue, 20 Oct 2026 00:00:00 +0000

Existing studies indicate that cybersecurity ranks among the top priorities for IT spending in telecommunications organizations, and that value management is critical for defining effective cybersecurity policies and strategies that enable efficient resource allocation. This study aims to analyze and synthesize existing research on the critical factors underpinning the strategic relevance of value management in the cybersecurity of telecommunications organizations. A systematic literature review (SLR) following the PRISMA methodology was conducted across three phases: planning, execution, and reporting. The Consensus tool was used to search the Semantic Scholar, IEEE Digital Library, Scopus, and ScienceDirect databases, yielding 950 articles. After rigorous screening and eligibility assessment, 49 key articles were selected for qualitative synthesis. Python-based tools were used for bibliometric analysis. Results show that 2023 recorded the highest publication output, with the journal Sustainability leading in article distribution. Inter-institutional collaboration and organizational culture emerged as the themes generating the greatest research interest. Six recurring categories were identified, consolidating the critical success factors (CSF) of value management in cybersecurity within telecommunications companies. The findings suggest that effective cybersecurity value management requires a holistic, risk-based approach that integrates human, technical, and strategic dimensions. Organizations must prioritize continuous adaptation and resilience against evolving threats by leveraging collaborative ecosystems and information-sharing mechanisms. Furthermore, embedding cybersecurity within broader business, innovation, and value-creation agendas is essential for sustaining competitive advantage and stakeholder trust. Compliance with standards and the application of economic models for risk quantification ensure regulatory alignment and long-term financial sustainability.

Robust Moving Horizon Estimation with Huber Loss for GNSS-Based Vehicle Localization under Outliers

Tue, 20 Oct 2026 00:00:00 +0000

This work presents a robust state estimation strategy based on the Moving Horizon Estimation (MHE) framework, incorporating the Huber loss function (MHE-H) to improve its robustness to outliers. Traditional MHE often degrades its accuracy when measurements are disturbed by large deviations. To address this limitation, the MHE-H method replaces the quadratic penalty on measurement residuals with the Huber loss function, which preserves sensitivity to small errors while limiting the impact of outliers. The robustness and stability of the resulting estimator are analyzed, and guidelines for selecting the Huber parameter are presented. The approach is validated through two case studies: a nonlinear simulation example and an experimental dataset obtained from GNSS/IMU measurements with artificially injected outliers. Results demonstrate that MHE-H significantly reduces the estimation error compared to EKF and standard MHE, while maintaining an acceptable computational cost for real-time applications, making MHE-H a promising alternative for robust estimation in uncertain and noisy environments.

Cross-Branch Contrastive Learning Algorithm in Heterogeneous Networks

Runfeng Wang (Author); Xingyang Zhao; Bingyu Zhang, Zengxin Kang (Author) — Tue, 20 Oct 2026 00:00:00 +0000

Heterogeneous information networks (HINs) provide a flexible graph representation for relational data with different entity and relation types. Such heterogeneity enriches semantic modeling, but also increases the difficulty of representation learning. Existing heterogeneous graph neural networks have achieved promising results on HIN data; however, many of them rely on supervised or semi-supervised training and require sufficient labeled nodes. Since reliable annotations are costly in large-scale networks, reducing label dependence remains an important issue. To address this problem, this paper develops Cross-Branch Contrastive Learning (CBCL), a self-supervised framework for HIN representation learning. CBCL learns node embeddings through two complementary branches. The structural-view branch models heterogeneous neighborhood relations and path-induced structural dependencies, while the path-semantic branch captures meta-path-level semantic information. A cross-branch contrastive objective aligns the two representations and transfers complementary information between them. Moreover, meta-path-based relevance is used to expand positive samples, allowing semantically related nodes to participate in contrastive optimization. By integrating structural modeling, semantic fusion, and cross-branch supervision, CBCL learns discriminative node representations without extensive labeled data.

Error Rate Analysis of Sparse Vector Coding with Diversity Techniques in the Presence of Transceiver Hardware Impairments

Tue, 20 Oct 2026 00:00:00 +0000

Ultra-reliable and low-latency communication (URLLC) is essential for mission-critical applications in 6G, necessitating stringent reliability and latency requirements. Sparse vector coding (SVC) seems to be a suitable candidate for ultra-reliable short-packet communications in future wireless systems. In this paper, we propose an integration of SVC with receiver diversity schemes, namely maximal ratio combining (MRC), equal gain combining (EGC), and selection combining (SC), in the presence of non-ideal transceiver hardware impairments. The key idea is to weight the received signal with weighting coefficients such that the reliability of the SVC transmission is further enhanced. The packet error rate (PER) is assessed and compared with that of the conventional SVC scheme without diversity under the Extended Pedestrian A (EPA) and Extended Vehicular A (EVA) standard channel models. The findings demonstrate that all diversity schemes substantially outperform the conventional method, with MRC providing the highest performance due to its optimal combining strategy. Hardware impairments (HWI) are incorporated using a signal-dependent distortion model at both the transmitter and receiver. Results suggest that the improved reliability of the proposed system model and adverse HWI effects can be traded off against the number of receiving antennas.

A Unified Namespace-Based Edge-to-Cloud Framework for Industrial Data Analytics

Tue, 20 Oct 2026 00:00:00 +0000

Industry 4.0 has intensified the demand for Industrial Internet of Things (IIoT) platforms supporting scalable and interoperable industrial analytics. However, many existing architectures treat data acquisition, processing, and visualization as loosely connected components within fragmented designs, limiting scalability, cross-domain generalizability, and accessibility for organizations lacking specialized infrastructure. This paper presents a flexible edge-to-cloud framework for industrial analytics based on a Unified Namespace (UNS) architecture. Unlike existing IIoT architectures, which rely on point-to-point or gateway-mediated integration and recover semantic context from external mappings, the proposed framework makes the namespace itself the integration contract: every message is self-describing, and integration interfaces grow linearly rather than multiplicatively with the number of connected components. The framework is evaluated on physically independent, heterogeneous hardware against three design pillars: scalability, generalizability, and accessibility. Its practical viability is demonstrated through a case study involving induction motor fault diagnosis. Per-device latency remains near 0.5s and is unaffected as independent edge nodes join. Substituting the diagnostic model, and subsequently the data type, sampling regime, and analytic paradigm, required no modification outside the analytics component, while classification accuracy remained above 98% after cross-dataset adaptation. The complete pipeline runs on single-board hardware within the free tiers of open-source and managed services, making the framework viable for resource-constrained industrial deployments.

A SAT-Based Exact Approach for the Minimum Order Frequency Assignment Problem

Nghia Dao Xuan, Phuong Dang Anh, Khanh To Van (Author) — Tue, 20 Oct 2026 00:00:00 +0000

The Minimum Order Frequency Assignment Problem (MO-FAP) is a fundamental NP-hard problem in spectrum management, where the objective is to minimize the number of assigned frequencies while satisfying interference constraints between transmitters. Current metaheuristic methods typically do not provide optimality certificates, while traditional exact methods can struggle on heavily constrained instances. This study proposes an exact framework based on Boolean Satisfiability (SAT) for solving MO-FAP. The key idea is to use standard order variables to reformulate pairwise interference constraints as interval-boundary constraints in a proposed encoding called POSE, thereby reducing the number of explicit conflict clauses relative to a direct SAT encoding. In addition, the framework includes an incremental SAT solving strategy (INCSC) based on Sequential Counter encoding, where the clause construction is designed to support incremental solving and clause reuse across iterations without rebuilding the formula. Experiments on 35 CALMA and DUTtest benchmark instances from the FAP database show that the proposed approach certifies optimality or infeasibility for every tested instance within a 600-second time limit. On the tested benchmark set, it certifies more instances than the considered exact baselines and reduces solving time on several instances where the baselines reach the time limit. These results show that the proposed SAT-based framework provides exact certification with competitive runtime on the tested MO-FAP benchmark instances.

Wavelet-Driven GAN with Attention Mechanism for Improved Raindrop Removal

Muthukumar Balamurugan, Shivarama Holla K, Varun P Gopi (Author) — Tue, 20 Oct 2026 00:00:00 +0000

The adhesion of raindrops to camera lenses creates serious distortions and image aberrations, posing a major problem for computer vision systems used in autonomous driving, surveillance, and environmental monitoring. This paper presents a powerful multi-scale restoration architecture, D2W-GAN, that combines the Dual-Tree Complex Wavelet Transform (DT-CWT) with a U-shaped network, an attention scheme, and Generative Adversarial Networks (GANs) to remove raindrops. Current techniques tend to struggle to remove artifacts introduced by raindrops and maintain overall image fidelity. The suggested method minimally distorts raindrop distortions by using DT-CWT, which simultaneously provides fine high-frequency and large-scale low-frequency components, thereby improving the separation of the distortion. The addition of attention mechanisms enables the network to focus more on the regions of the raindrops, thereby improving the restoration accuracy. Moreover, refinements to the adversarial learning framework yield more realistic, consistent results. Experimental tests show that D2W-GAN is superior to many contemporary methods and produces higher-quality restorations, as measured by PSNR and SSIM. In general, wavelet-based feature extraction, feature attention, and adaptive optimization have been successfully integrated to achieve effective image restoration under unstable, demanding conditions.

IoT-Based Monitoring and Machine Learning Classification of Urban Water Consumption Patterns

Tue, 20 Oct 2026 00:00:00 +0000

Efficient urban water management requires real-time monitoring and intelligent analytics to support both day-to-day operation and long-term resource optimization. This paper reports the design, deployment, and evaluation of an energy-aware LPWAN monitoring system implemented at a university facility in Mexico. Over five months, the system recorded hourly volumetric flow and transmitted a single daily burst, revealing clear circadian and weekly consumption patterns influenced by academic schedules and external events. To characterize consumption behavior, a percentile-based labeling strategy was employed to categorize daily water usage into four operational states: Low, Normal, High, and Anomalous. Using these labels, supervised machine learning models were evaluated for classification of consumption patterns, while unsupervised methods were used to explore natural groupings in the data. Among the tested models, Random Forest, Decision Tree, and XGBoost achieved
the highest classification performance under stratified 10-fold cross-validation. From an energy efficiency standpoint, once-per-day bursting delivered 3,816 hourly readings through 159 daily bursts (six 12-byte frames each; 954 frames total), reducing the number of daily transmissions by 75% relative to hourly uploads while preserving full hourly resolution. These results show that minimal sensing coupled with lightweight LPWAN telemetry can support scalable, pattern-driven water management and provide a practical foundation for context-aware decision support in smart urban distribution networks.

IoT-Driven Inclusive Education: Mapping the Evidence and Research Gaps

Tue, 20 Oct 2026 00:00:00 +0000

The Internet of Things (IoT) has emerged as a transformative force in education by enabling personalized and accessible learning environments. These environments are particularly effective in addressing students’ functional and cognitive diversity, thereby advancing the goals of inclusive education. Although scholarly interest in educational IoT has increased, a cohesive understanding of its role in fostering inclusion remains fragmented. This study presents a Systematic Literature Review (SLR) of publications from 2019 to 2025, providing a comprehensive analysis of IoT applications, benefits, and challenges in inclusive education. The review maps existing technological solutions, evaluates their impact on pedagogical practices, and identifies systemic barriers to sustainable integration. The findings highlight the potential of IoT to democratize access to education, especially for students with physical disabilities, while revealing its predominant focus on higher education contexts. Finally, the study outlines critical research gaps that must be addressed to promote equitable and sustainable technology adoption.

Intelligent Energy and Carbon Management in 6G: A Hybrid DRL-PPO Optimization Framework

Harshita Gangadhar Patil, Dr. Kaushlendra Sharma, Dr. Shishupal Kumar (Author) — Tue, 20 Oct 2026 00:00:00 +0000

6th generation (6G) networks are anticipated to deliver highly-reliable, low-latency communications (URLLC); therefore, the increased use of Terahertz (THz) transceivers and Holographic MIMO surfaces will result in significant sustainability challenges as a result of the excessive amount of energy these devices will require for their operations. As part of this research, we present a framework using deep reinforcement learning (DRL) techniques known as Proximal Policy Optimization (PPO) that can effectively address two goals (conflicting) - minimizing energy consumption through intelligent channel sensing techniques and minimizing carbon footprints by using grid-aware
resource allocation techniques. We demonstrate how a Carbon-Aware PPO agent can decouple network performance from carbon emissions by collecting operational energy usage records from live 5G networks and dynamically scaling them based on 6G power restrictions. Overall, we found that this approach could reduce carbon emissions approximately 84.5% versus traditional maximum performance baselines without causing any quality of service (QoS) violations; thus, effectively circumventing all limitations of static sleep-scheduling heuristic approaches.

BSR-NBV: Breaking Stagnation with Random Pose Refinement in Probabilistic Grids for Latent Space Next-Best-View Prediction

Saulo Abraham Gante-Díaz, Juan Irving Vasquez, Erik Zamora (Author) — Tue, 20 Oct 2026 00:00:00 +0000

The Next-Best-View (NBV) problem is fundamental to autonomous 3D object reconstruction, where the goal is to generate a complete 3D model by sequentially sensing a target’s surface. This paper proposes a novel latent space-based method to address this challenge. Our architecture is trained in two distinct stages: the first stage learns to encode a probabilistic grid into a compact latent representation, while the second stage performs continuous NBV regression. Furthermore, we incorporate a pose refinement mechanism to mitigate stagnation—a phenomenon where the network repeatedly predicts views with low information gain that fail to update the probabilistic grid. Experimental evaluations on the Google Scanned Objects dataset demonstrate that the proposed method achieves superior surface
coverage compared to current state-of-the-art approaches.

Dual-Band MIMO Antenna with Enhanced Isolation for Sub-6 GHz 5G and Wi-Fi Applications

Yem Vu-Van, Thao Hoang-Thi-Phuong, Quyen Nguyen-Xuan, Cuong Do-Manh (Author) — Tue, 20 Oct 2026 00:00:00 +0000

This paper presents a single-layer dual-band multiple-input-multiple-output (MIMO) antenna with high isolation characteristics for the Sub-6 GHz n79 5G NR band (4.4–5.0 GHz) and 5-GHz Wi-Fi applications. The dual-band MIMO element consists of four-square patches with embedded slots, which are capacitively excited by a probe-fed patch at the center. To achieve high isolation, shorting vias are employed to produce a transmission zero for mutual coupling reduction at both operating frequency bands. The proposed two-port antenna achieves an overall size of 0.94λ×0.47λ×0.02λ at the lowest band of 4.6 GHz. Experimental results show measured bandwidths of 2.8% (4.62 - 4.75 GHz) and 3.9% (5.50 - 5.72 GHz) for the lower and upper bands, respectively. The antenna consistently provides more than 25 dB isolation, together with gains of 4.8–5.4 dBi for the 4.6 GHz band and 5.2–6.8 dBi for the 5.6 GHz band at both individual ports. The proposed design effectively covers the 4.62–4.75 GHz portion of the n79 band and the 5.50–5.72 GHz range for 5-GHz Wi-Fi, accommodating channel bandwidths up to 100 and 160 MHz, respectively.

Single-Layer, Compact, High Isolation Dual-Band MIMO Antenna for WiFi 6/6E Applications

Cuong Do-Manh, Thao Hoang-Thi-Phuong, Quyen Nguyen-Xuan, Yem Vu-Van (Author) — Tue, 20 Oct 2026 00:00:00 +0000

This paper presents a compact single-layer dualband MIMO antenna with high isolation characteristics for WiFi 6/6E applications. The dual-band response is realized by two closely spaced capacitively excited patch resonators operating in their fundamental modes, enabling independent control of the two operating bands within a single-layer architecture. In the MIMO arrangement, two identical dual-band elements are placed in an inverted configuration and effectively decoupled by grounded stubs. The proposed two-port antenna achieves an overall size of 0.61λ × 0.61λ × 0.03λ at 5.2 GHz. Experimental results show measured bandwidths of 3.8% (5.2–5.4 GHz) and 3.6% (6.60–6.84 GHz) for the lower and upper bands, respectively. The antenna consistently provides more than 20 dB isolation, together with a stable gain of about 5.0 dBi across both bands. Moreover, its favorable diversity performance highlights its strong potential for compact WiFi 6/6E MIMO systems.

Machine Learning-Based Wi-Fi RSS Fingerprinting for Indoor Zone Classification of Biomedical Assets in Hospitals

Tue, 20 Oct 2026 00:00:00 +0000

The frequent relocation of biomedical equipment in healthcare facilities delays maintenance and reduces operational efficiency. While dedicated positioning infrastructure can mitigate these delays, its deployment often incurs high financial costs. This work proposes an indoor zone classification system that leverages existing Wi-Fi received signal strength (RSS) fingerprints and machine learning. The methodology evaluates five supervised classifiers (Naive Bayes, K-Nearest Neighbors, Support Vector Machine, Random Forest, and Multilayer Perceptron) using an empirical dataset of 360 observations. These RSS measurements were collected by an ESP8266-based sensing module across six functional zones covered by four access points. Under a 5-fold stratified cross-validation protocol, the Random Forest model achieved a mean validation accuracy of 97.57%. These results demonstrate that existing Wi-Fi infrastructure combined with low-cost edge sensing provides a viable solution for biomedical asset tracking in hospitals.

A Miniaturized Single-Feed Dual-Band Dual-Orthogonally Polarized Antenna for U2V Communication Systems

Tue, 20 Oct 2026 00:00:00 +0000

This paper presents a compact single-fed dual-band dual-orthogonally polarized patch antenna for Unmanned Aerial Vehicle (UAV) to Vehicle (U2V) communication systems. The proposed square patch is loaded with multiple gaps and shorting vias to achieve significant size reduction. Dual-orthogonal polarizations are realized by properly exciting the radiator along the horizontal and vertical directions through a single input transmission line. By adjusting the gap dimensions, the effective capacitance of the radiator is conveniently controlled, thereby enabling two distinct resonant modes. The proposed antenna operates at 5.5 and 5.9 GHz with an electrical size of 0.18 λ × 0.18 λ × 0.04 λ at the lower band. Notably, the frequency ratio between the two operating bands is only 1.07, which is smaller than that of most reported designs. In addition, the antenna exhibits symmetric radiation characteristics, with measured peak broadside gains of approximately 3.8 dBi and 3.2 dBi at the lower and upper bands, respectively. The proposed antenna is suitable for U2V systems, supporting high data rate transmission at 5.5 GHz and reliable low-latency communication at 5.9 GHz.

Sum-rate Maximization and Outage Analysis of RIS assisted RSMA System under Electromagnetic Interference

Nishandhi Soudarsanane, Surendar Maruthu (Author) — Tue, 20 Oct 2026 00:00:00 +0000

Rate-splitting multiple access (RSMA) and reconfigurable intelligent surfaces (RIS) are key enablers for 6G communications. This study examines a downlink multiple-input single-output system using RIS-assisted RSMA (RIS-RSMA) in the presence of electromagnetic interference (EMI). Although RIS improves wireless propagation through passive beamforming, it can also reflect EMI from ambient sources, which may disrupt the decoding order and limit system performance. To overcome this, we jointly optimize the transmit beamformer and phase shifts of the RIS using an alternating optimization approach. We derive and validate the analytical expressions for the outage probability (OP) and sum rate of the system via Monte Carlo simulations. The obtained results validate that RIS-RSMA reduces the OP by $60\%$ and achieves a higher sum rate of $3.22 \times 10^{5}$ bit/s compared with RIS-assisted NOMA, and mitigates the impact of EMI.

A Machine Learning Approach for Transformer Load Allocation in Power Distribution Networks

Tue, 20 Oct 2026 00:00:00 +0000

Accurate transformer load allocation is essential for distribution network planning and operation, particularly in systems without advanced metering infrastructure. This paper proposes a bottom-up data-driven methodology that integrates clustering of measured load curves, supervised classification, and energy-based scaling to estimate transformer-level demand while preserving temporal characteristics and energy consistency. The methodology was validated using real data from an Ecuadorian distribution feeder with measurements available at the feeder head and at an intermediate recloser. Power flow results were compared with the conventional load allocation methods available in CYMDIST in order to evaluate the accuracy of the proposed method against standard approaches. At the intermediate recloser, where downstream allocation accuracy is critical, the proposed method achieved the lowest error, with a mean absolute error of 32.23 kVA and a mean absolute percentage error of 6.60%, compared with the range of errors obtained with the CYMDIST methods, which reached up to 186.77 kVA and 33.05%, respectively. These results suggest that the proposed methodology provides a practical approach for improving load allocation in minimally monitored distribution networks.

A Methodology for Analyzing the Impacts of Charging Station Demand on the Distribution Network

Eutran de Jesus Ramalho, Juan Carlos Galvis Manso (Author) — Tue, 20 Oct 2026 00:00:00 +0000

This article proposes a methodology to evaluate the impacts of integrating electric vehicle charging stations into real distribution networks, combining urban mobility simulation in Eclipse SUMO and electrical modeling in OpenDSS. Charging demand is represented by probabilistic models, with Monte Carlo simulations, allowing for the estimation of load profiles and the identification of critical scenarios. An iterative criterion for optimal sizing of fast charging stations based on occupancy rate is also proposed, balancing technical performance and economic viability. The results show that the system supports initial levels of electrification, but high penetration rates cause voltage drops in the low-voltage network, transformer overload, and increased losses, especially during peak hours.

Comparative Analysis, Characterization and Simulation of a Flyback DC-DC Converter Planar Topology Transformer

Tue, 20 Oct 2026 00:00:00 +0000

This work compares two analytical methods (A and B) for sizing planar topology transformers (PTT) in 20W DC-DC flyback converters, aiming to minimise losses and core volume. In the analytical comparison, method B showed a 6.4% lower volume compared to method A. Structure B, with better performance in the analytical step, was modelled by a multiphysics analysis in finite element software, which indicated a slight increase in temperature in the core and windings, with electrical potentials consistent with the number of turns. The magnetostatic analysis showed that the predominant flux density (130-160 mT) was within the adopted reference value (160 mT). The AC resistance calculation showed that the skin and proximity effects contribute about 50% each. A prototype was built and evaluated experimentally, with measurements of DC/AC resistance, frequency response, short-circuit and no-load tests, and transformation ratio. Finally, the closed-loop circuit was simulated with equivalent parameters extracted from the tests.