IEEE Latin America Transactions

Nonlinear Granger Causality Analysis Using Neural Network Architectures for Sequential Data

Diego N. Chacón Wilches, A. Orjuela-Cañón (Author) — Wed, 15 Apr 2026 00:00:00 +0000

This work presents an analysis of nonlinear Granger causality (NNGC) computation based on artificial neural networks (ANN) architectures. The study evaluates the impact of using computational intelligence models as ANN, and how the training parameters can modify the causality estimation. For this, the employing from three chaotic maps (Hénon, Ikeda, and Tinkerbell) and one neuron-like map (Rulkov) in bivariate scenarios were implemented. Three architectures from the ANN were used such as multilayer perceptron in a mode of the nonlinear autoregressive models, long-short memory term, and gated recurrent unit architectures were used to compute the NNGC, applying a forecasting on sequential data techniques. Results demonstrated that NNGC is highly sensitive to neural network parameters, such as the number of neurons, lag length, and batch size, with an optimal configuration by varying across chaotic maps. Comparisons with classical Granger causality were tested, revealing that neural networks effectively discover nonlinear relationships missed by linear methods, particularly in the Hénon and Rulkov maps.

Detecting Obstetric Violence Tweets from Mexico: Annotation Guidelines and Classification with LLMs

Wed, 15 Apr 2026 00:00:00 +0000

This paper presents the construction and analysis of a manually annotated corpus of tweets related to Obstetric Violence (OV) shared on Twitter (now X). The study aims to identify different types of violence experienced by women during the perinatal period, as well as activism efforts that seek to raise awareness about OV. The methodology includes data collection through keyword filtering, manual annotation guided by typologies of OV, and a descriptive analysis using BERTopic to identify themes in the data. The tweets were classified into categories such as OV, Non-OV, and Activism, and further annotated based on narrator type and type of OV violence. The study also evaluates the performance of large language models (LLMs) — including ChatGPT, Copilot, and Meta’s LLaMA — for zero-shot classification of tweets, highlighting their limitations in accurately identifying nuanced cases of OV. The research contributes a labeled dataset, a detailed annotation guide, and insights into the challenges of detecting OV in social media texts. It underscores the importance of addressing the invisibility and normalization of OV in both healthcare and NLP research.

The Hybrid Spectral - Gradient Saliency Pruning: A combination of multiple filter scoring criteria

Wed, 15 Apr 2026 00:00:00 +0000

Deep convolutional neural networks (CNNs) have achieved remarkable performance in visual recognition but remain computationally expensive for deployment on embedded or edge devices. This paper introduces Hybrid Spectral–Gradient Saliency Pruning (HSGSP), a structured pruning framework that unifies spectral analysis and data-driven gradient saliency to achieve efficient CNN compression. The proposed method incorporates a lightweight Frequency Relevance Network (FRN) that learns to estimate the spectral importance of convolutional filters through frequency-band energy ratios, enabling fast, task-driven scoring. A hybrid saliency metric fuses the FRN’s spectral relevance with gradient-based Taylor sensitivity, ensuring filters are preserved only when important both spectrally and task-wise. An adaptive iterative schedule dynamically adjusts pruning intensity based on validation feedback, preventing over-pruning and maintaining stability. Experiments on CIFAR-10 and CIFAR-100 using VGG-16BN demonstrate up to 90% parameter reduction with negligible accuracy loss, outperforming recent structured pruning methods. Furthermore, on a Raspberry Pi 5, our pruned model delivers a 3.4x inference speedup while slightly improving accuracy, and when permitting only a 1% accuracy trade-off, the speedup increases dramatically to 7.5x. The results confirm that combining spectral cues with gradient saliency offers a robust and interpretable path toward efficient CNN deployment. The official implementation code of our method is available at https://github.com/locth/HSGSP.git.

Evaluation of Imputation Techniques using Reanalysis Data for Meteorological Variables in Northern Chile

Wed, 15 Apr 2026 00:00:00 +0000

The article explores a study on meteorological data imputation in Northern Chile, an arid region with complex geomorphology. Obtaining complete and high-quality time series poses a challenge due to data loss at meteorological stations, hindering climate change analysis in the area. Six imputation techniques were evaluated using reanalysis data from the CFSR and CFSv2 models, integrated into a single data set as an alternative to the use of neighboring meteorological stations. These models are valuable for the study of climate, especially when meteorological stations are not available or have data problems. For the research work, the six stages of the CRISP-DM methodology were developed, providing a robust framework. The results show that the Direct Imputation, Hot-Deck, Weighted K-Nearest Neighbor Imputation and Inverse Distance Weighting techniques obtain the lowest residual errors according to meteorological variables, while the NR technique is consistently inferior compared to the other techniques evaluated. The study concludes that it is essential to evaluate imputation techniques and reanalysis models based on the specific geographic area where they will be applied. Reanalysis data represents the study area’s behavior and meteorological variables with varying degrees of accuracy. As a result, the best imputation technique differs depending on the geographic region, reanalysis model, and meteorological variable.

Unimodal Model for Emotion Detection in Immersive Virtual Learning Environments Using Spatial Analysis of Hands and Head

Wed, 15 Apr 2026 00:00:00 +0000

This study introduces a unimodal model for emotion detection in Virtual Reality (VR) environments that depends only on the spatial information from the user's head and hands during interactions within an Immersive Virtual Learning Environment (IVLE). The goal is to eliminate the need for additional sensors, offering an emotional recognition method that can be scaled for multiuser settings. Data on rotation and position from VR devices were collected along with self-reported valence and arousal ratings from 65 participants. Primary and spatial features were extracted, generating mean and median vectors. Random forest regression techniques were then used to predict valence and arousal values in SMOTE augmented data. A paired random pre-augmentation data was used to further test the models in a closer-to-final-implementation scenario. The models achieved accuracies of 70% and 76% for valence prediction using the mean and median vectors, respectively. For arousal, the accuracies were 83% (mean vector) and 87% (median vector). The findings suggest that the median-based approach improves performance, although it involves higher feature dimensionality. This model enables the non-invasive inference of a user's emotional state in VR environments, without cables or extra sensors. This advancement enhances user experience and lowers implementation costs. These results provide a foundation for integrating affective tutors in IVLEs, with potential applications in education and training involving large groups.

Computed torque for restricted Lagrange-Euler systems via DAEs and LMIs

Julián León, David Vázquez, Miguel Bernal (Author) — Wed, 15 Apr 2026 00:00:00 +0000

An extension of the well-known computed-torque technique for restricted Lagrange-Euler systems is presented in this work. It is shown that adding reaction forces to the feedforward term as well as designing the feedback term via linear matrix inequalities allows computed torque to deal with trajectory tracking problems in closed kinematic chains. Implementation issues via index-1 differential algebraic equations and recently appeared toolboxes are discussed. A fully reproducible case study is included to illustrate the effectiveness of the proposal as well as a real-time implementation which point out that non-restricted plants are also amenable to the novel technique.

Comparison of Two Harmonic Load-Pull Systems Using Frequency-Domain and Time-Domain Measurements for Evaluating GaN Transistor Output Power and Efficiency

Wed, 15 Apr 2026 00:00:00 +0000

Two active real-time harmonic load-pull systems were developed: one operating in the time domain (TD-AHLP) and the other one in the frequency domain (FD-AHLP). The main distinction between them lies in their receivers: the first utilizes a four-channel oscilloscope, while the second uses a vector network analyzer. Measurements of output power and drain efficiency were performed using both load-pull systems on a MACOM CGH40010F GaN-HEMT transistor, biased as a Class-B power amplifier (V_GS = -2.92 V, V_DS = 28 V, I_DS = 200 mA), at a fundamental frequency of 1 GHz. The performance of both systems was compared to evaluate the harmonic loads up to their third harmonic, output power, drain efficiency, and power gain of a Harmonic Tuned Power Amplifier (HTPA). Results demonstrate high consistency between the two systems, with mean relative deviations below 3% for both output power and drain efficiency. Both systems identify specific second-harmonic impedance regions along the periphery of the Smith chart where performance metrics remain nearly constant and largely independent of the second-harmonic reactance.

A Simple Patch Antenna Design with Multiple Linear Polarizations for RFID Readers

Trang Hoang-Thu, Hung Pham-Duy, Hung Tran-Huy, Hyun Chang Park (Author) — Wed, 15 Apr 2026 00:00:00 +0000

This paper proposed a compact antenna with multiple linear polarization realization for radio frequency identification (RFID) operating at ultra-high frequency (UHF) band. The main radiator is a crossed patch, which is excited by different feeding positions to adjust the current flowing on the patch. To minimize the unwanted effect of the biasing circuit on the antenna radiation performance, the antenna is excited through a reconfigurable feeding network with one input and three outputs. By controlling the ON/OFF states of three PIN diodes integrated into the feeding network, four different modes including horizontal, vertical, and both diagonal polarizations can be realized. The final antenna design with overall dimensions of 0.67 λ × 0.67 λ × 0.04 λ at 915 MHz shows good matching performance from 910 to 925 MHz. Within this band, the broadside gain values for all polarization states are better than 6.2 dBi.

Analysis of Ferroresonance in Photovoltaic Park Distribution Systems

Wed, 15 Apr 2026 00:00:00 +0000

Ferroresonance is a phenomenon of particular interest, characterized by causing temporary overvoltages in electrical systems, posing a risk to system integrity. Currently the increasing integration of renewable energy sources, such as wind and photovoltaic, generally implement distribution systems with underground cables interconnected trough transformers with different configurations, which increases the possibility of scenarios susceptible to the ferroresonance phenomenon. Since these overvoltage cause deterioration in the power quality and can damage to equipment. In this regard, this paper presents the modeling of photovoltaic park distribution system which connects the photovoltaic system through three winding transformers (Dy1y1) collecting the energy to transmit it through a power transformer connected to the 115 kV system, cable lengths are analyzed to determine the potential occurrence of the ferroresonance phenomenon under phase closing and opening conditions, in addition to evaluating the impact of maximum and minimum photovoltaic generation under system open conditions.

A High Step-up Single-switch based Cubic-gain DC–DC Converter for PV Applications

Jayasri Boda, Giridhar A.V., Narasimharaju B.L. (Author) — Wed, 15 Apr 2026 00:00:00 +0000

The growing use of renewable energy sources (RES’s), particularly photovoltaic(PV) systems, has increased the need for efficient high step-up DC-DC converters. This paper proposes a single switch cubic-gain (SSCG) converter capable of achieving a high voltage gain with minimal circuit complexity. The proposed converter achieves cubic voltage gain through the integration of a quadratic boost converter(QBC) combined with a voltage-lift technique, enabling substantial voltage boosting using only a single switch. Detailed steady-state analysis are carried out under both continuous and discontinuous conduction modes, along with an evaluation of the influence of non-ideal components on practical voltage gain and efficiency. The comparison results with the existing converter topologies highlight the distinctive advantages of the SSCG converter, including the simple structure, minimum total component count, continuous input current, common ground configuration, and high step-up capability. A 48/400 V, 50 kHz, 200 W hardware prototype is developed and experimentally verified, demonstrating close agreement with theoretical analysis. The results confirm that the proposed SSCG converter is a promising solution for high-gain photovoltaic energy conversion applications.

Table of Contents June 2026

Daniel Ulises Campos Delgado — Wed, 15 Apr 2026 00:00:00 +0000