<?xml version="1.0" encoding="UTF-8"?>
<feed xmlns="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
<title>Trabalhos de Graduação</title>
<link href="https://hdl.handle.net/1884/30071" rel="alternate"/>
<subtitle>Biblioteca Digital de Trabalhos de Graduação da UFPR.</subtitle>
<id>https://hdl.handle.net/1884/30071</id>
<updated>2026-09-03T23:29:26Z</updated>
<dc:date>2026-09-03T23:29:26Z</dc:date>
<entry>
<title>Pré-distorção digital de amplificadores de potência multimodos</title>
<link href="https://hdl.handle.net/1884/107161" rel="alternate"/>
<author>
<name/>
</author>
<id>https://hdl.handle.net/1884/107161</id>
<updated>2026-08-21T16:31:20Z</updated>
<published>2025-01-01T00:00:00Z</published>
<summary type="text">Pré-distorção digital de amplificadores de potência multimodos
Resumo: Este trabalho apresenta um estudo detalhado sobre a modelagem comportamental de amplificadores de potência multimodos utilizados em sistemas modernos de comunicação sem fio. Considerando que esses dispositivos operam em condições de elevada demanda espectral, ampla faixa dinâmica e modos distintos de funcionamento, torna-se essencial compreender seus efeitos de não linearidade e memória, que degradam diretamente a qualidade do sinal transmitido. Nesse contexto, este trabalho compara duas abordagens de modelagem: o modelo de Polinômios com Memória e uma Rede Neural Perceptron de Múltiplas Camadas, representando uma técnica baseada em aprendizado de máquina. A metodologia adotada envolveu o uso de dados reais de entrada e saída de um amplificador de potência multimodo, submetidos a etapas de pré-processamento, como a organização e normalização das amostras. O modelo polinomial foi ajustado por meio de técnicas de otimização não linear, buscando estimar coeficientes capazes de representar tanto o comportamento estático quanto o dinâmico do dispositivo. Também foi projetada e treinada uma rede neural perceptron utilizando aprendizado supervisionado, avaliando diferentes funções de ativação. A análise comparativa entre os modelos foi conduzida por meio de métricas quantitativas, como erro quadrático médio e erro absoluto médio, e pela avaliação visual das curvas que relacionam a amplitude de entrada e saída do amplificador. Os resultados obtidos evidenciaram as limitações do modelo de Polinômios com Memória diante das características mais acentuadas de não linearidade e das transições abruptas entre modos de operação. Embora apresente baixa complexidade computacional, o modelo polinomial não conseguiu reproduzir com precisão regiões de compressão e descontinuidades típicas de amplificadores multimodos. Em contraste, a Rede Neural Perceptron demonstrou maior capacidade de ajuste, reduzindo significativamente o erro em todas as métricas avaliadas e capturando com fidelidade o comportamento não linear do dispositivo. Além disso, o uso de normalização adequada e funções de ativação não saturantes contribuiu para uma convergência estável e eficiente durante o treinamento. Conclui-se que técnicas baseadas em aprendizado de máquina representam uma alternativa robusta para a modelagem de amplificadores de potência modernos. Os resultados obtidos estabelecem uma base sólida para o desenvolvimento futuro de técnicas de pré-distorção digital baseadas em redes neurais, as quais têm potencial para melhorar de forma significativa a linearidade, a eficiência energética e o desempenho global de transmissores utilizados em sistemas de comunicação.; Abstract: This work presents a detailed study on the behavioral modeling of multimode power amplifiers used in modern wireless communication systems. Considering that these devices operate under high spectral demand, wide dynamic range and distinct operating modes, it becomes essential to understand their nonlinear and memory effects, which directly degrade the quality of the transmitted signal. In this context, this work compares two modeling approaches: the Memory Polynomial model and a Multilayer Perceptron Neural Network, representing a machine-learningbased technique. The methodology adopted involved the use of real input and output data from a multimode power amplifier, subjected to preprocessing steps such as sample organization and normalization. The polynomial model was adjusted using nonlinear optimization techniques to estimate coefficients capable of representing both the static and dynamic behavior of the device. Additionally, a perceptron neural network was designed and trained using supervised learning while evaluating different activation functions. The comparative analysis between the models was conducted using quantitative metrics, such as mean squared error and mean absolute error, and the visual assessment of curves relating the input and output amplitudes of the amplifier.The results highlighted the limitations of the Memory Polynomial model when confronted with stronger nonlinear behavior and abrupt transitions between operating modes. Although it presents low computational complexity, the polynomial model was unable to accurately reproduce compression regions and discontinuities typically observed in multimode amplifiers. In contrast, the Perceptron Neural Network demonstrated superior fitting capability, significantly reducing error across all evaluated metrics and faithfully capturing the nonlinear behavior of the device. Furthermore, the use of proper normalization and non-saturating activation functions contributed to stable and efficient convergence during training. It is concluded that machine-learning-based techniques represent a robust alternative for the modeling of modern power amplifiers. The results obtained provide a solid foundation for the future development of neural-network-based digital predistortion techniques, which have the potential to significantly improve linearity, energy efficiency and overall performance of transmitters used in broadband communication systems.
Orientador: Prof. Ph.D. Eduardo Gonçalves de Lima; Trabalho de Conclusão de Curso (graduação) - Universidade Federal do Paraná, Setor de Tecnologia, Curso de Graduação em Engenharia Elétrica; Inclui referências
</summary>
<dc:date>2025-01-01T00:00:00Z</dc:date>
</entry>
<entry>
<title>Development of the Drive Selector in MATLAB with object-oriented data management for motor selection in construction power tools</title>
<link href="https://hdl.handle.net/1884/107160" rel="alternate"/>
<author>
<name/>
</author>
<id>https://hdl.handle.net/1884/107160</id>
<updated>2026-08-21T14:39:09Z</updated>
<published>2025-01-01T00:00:00Z</published>
<summary type="text">Development of the Drive Selector in MATLAB with object-oriented data management for motor selection in construction power tools
Resumo: O desenvolvimento de ferramentas elétricas modernas requer a integração entre medições experimentais do motor e requisitos operacionais para determinar configurações viáveis do drive. Esta tese apresenta a migração e a reformulação da arquitetura do "Drive Selector", uma ferramenta interna de seleção, originalmente desenvolvida em uma linguagem de baixo nível para um caso industrial, para uma implementação baseada em MATLAB. O sistema reprojetado reproduz as funcionalidades centrais da ferramenta original, incluindo o processamento de dados reais de motores dentro de uma estrutura orientada a objetos e a disponibilização de uma interface para avaliar pontos de operação (OPs) do motor, apoiando assim a seleção de configurações adequadas de motor. Os dados foram reorganizados em  uma estrutura hierárquica orientada a objetos utilizando scripts em MATLAB, e um novo pipeline  baseado em arquivos .mat substituiu o processo anterior composto por múltiplas etapas. Essa reestruturação reduziu dependências e melhorou a clareza, o desempenho e a manutenibilidade.  A interface gráfica foi desenvolvida no MATLAB App Designer, oferecendo um ambiente para configurar OPs e visualizar diferentes configurações de motor a fim de identificar a opção mais adequada. A validação, realizada por comparação com a implementação anterior, confirmou resultados consistentes e maior usabilidade. A versão resultante do Drive Selector em MATLAB fornece uma base sustentável e extensível, apoiada por um design modular, estrutura de código mais clara e documentação completa. Funcionalidades que permanecem fora do escopo desta versão são identificadas como extensões futuras para continuidade do desenvolvimento; Abstract: Modern power-tool development requires integrating experimental motor  measurements with operating requirements to determine feasible drive configurations. This thesis presents the migration and redesign of the "Drive Selector," an internal selection tool originally developed in a low-level language for an industrial case, into a MATLAB-based implementation. The redesigned system reproduces the core functionalities of the original tool, including parsing real motor data within an object-oriented structure and providing an interface for evaluating Operating Points (OPs) of motors, thereby supporting the selection of appropriate motor configurations. The data were reorganized into a hierarchical object-oriented structure using MATLAB scripts, and a new data pipeline based on .mat files replaced the previous multi stage process. This redesign reduced dependencies and improved clarity, performance, and maintainability. The graphical interface was developed in MATLAB App Designer, offering an environment for configuring OPs and visualizing different motor configurations to identify the most suitable option. Validation through comparison with the earlier implementation confirmed consistent results and improved usability. The resulting MATLAB Drive Selector provides a maintainable and extensible foundation, supported by modular design, clearer code structure and complete documentation. Features that remain outside the scope of this version are identified as future extensions for continued development
Orientador: Eduardo Gonçalves de Lima; Trabalho de Conclusão de Curso  (graduação) - Universidade Federal do Paraná, Setor de Tecnologia, Curso de Graduação em Engenharia Elétrica; Inclui referências
</summary>
<dc:date>2025-01-01T00:00:00Z</dc:date>
</entry>
<entry>
<title>Análise em regime permanente senoidal de circuitos contendo elementos com derivada de ordem fracionária</title>
<link href="https://hdl.handle.net/1884/107159" rel="alternate"/>
<author>
<name/>
</author>
<id>https://hdl.handle.net/1884/107159</id>
<updated>2026-08-21T14:37:04Z</updated>
<published>2025-01-01T00:00:00Z</published>
<summary type="text">Análise em regime permanente senoidal de circuitos contendo elementos com derivada de ordem fracionária
Resumo: No contexto educacional, ferramentas para simulações de circuitos são essenciais para o aprendizado prático de alunos e pesquisadores, pois facilitam a  análise do comportamento dos circuitos em diferentes condições, promovem iterações rápidas de modificações, previnem falhas e diminuem custos. Neste contexto, existem diferentes tipos de análises: a análise em corrente contínua (CC), a análise em corrente alternada (CA), e a análise transitória. Tradicionalmente, dispositivos como capacitores e indutores, de natureza passiva dinâmica, presentes em circuitos elétricos possuem derivada inteira. Outros dispositivos de comportamento de sistemas dinâmicos, como amplificadores, também podem ser modelados, mas por meio de derivadas de ordem fracionária. Contudo, os simuladores de circuitos comerciais só permitem a análise de circuitos com derivada inteira. O objetivo deste trabalho é desenvolver e implementar análises de CA e de equilíbrio harmônico em circuitos contendo elementos passivos de ordem fracionária. Para isso, foram criados dois circuitos: um circuito linear, para o qual foram realizadas análises de CA (manualmente e utilizando o software QUCS) em suas variações com capacitor e com elemento de derivada fracionária; e um circuito não linear, também com variações com capacitor e com elemento de derivada fracionária, para o qual foi implementada a análise em equilíbrio harmônico. Obtivemos que as curvas de tensão e correntes do capacitor formadas são senoidais e possuem defasagem de 90°, de forma que, quando plotamos o gráfico da corrente pela tensão obtivemos uma curva em forma de elipse. As curvas de tensão e correntes do elemento de derivada fracionária formadas também são senoidais para o circuito linear, mas com diferença de fase diferente de 90°, de forma que, quando plotamos o gráfico da corrente pela tensão obtivemos uma curva em forma de uma elipse com seu eixo maior  deslocado. No caso do circuito não linear com elemento de derivada fracionária, as curvas de tensão e corrente obtidas apresentaram formas mais complexas, com  defasagens não triviais e amplitudes distintas. O gráfico da corrente pela tensão revelou uma trajetória alongada e com laços, evidenciando a presença de efeitos de memória e a interação entre múltiplas frequências, características típicas desse tipo de elemento em regime permanente. Além dessas análises, também foi feita a modelagem de uma antena real a partir das medições de impedância fornecidas no arquivo experimental. Para isso, foram ajustados dois modelos equivalentes: um RLC paralelo convencional e outro composto por elementos fracionários. Comparando as curvas medidas com as calculadas, nota-se que o modelo RLC consegue seguir muito bem a região da ressonância, enquanto o modelo fracionário se aproxima    melhor da resposta após a ressonância, onde o comportamento da antena passa a ser mais dispersivo. A diferença entre os modelos também é observada na análise dos erros quadráticos médios (MSE, do inglês, Mean Squared Error) mostrando que o uso de elementos fracionários resulta em uma modelagem mais fiel da impedância quando comparado ao modelo RLC simples. Esse resultado sugere que os elementos de ordem fracionária capturam com mais facilidade os efeitos de perda, memória e variação dinâmica presentes nesse tipo de estrutura; Abstract: In the educational context, circuit simulation tools are essential for practical learning, as they allow students and researchers to analyze circuit behavior under different conditions, enable rapid iteration of modifications, prevent failures, and reduce  costs. In this setting, different types of analysis can be performed, such as direct current  (DC), alternating current (AC), and transient analysis. Traditionally, passive dynamic devices such as capacitors and inductors present integer-order behavior. Other systems, such as amplifiers, may also be modeled using fractional-order derivatives. However, commercial circuit simulators only support integer-order modeling. The goal of this research project is to develop and implement AC and harmonic balance (HB) analyses for circuits containing fractional-order passive elements. Two test circuits were created: a linear circuit, for which AC analyses were carried out manually and using QUCS, considering both a capacitor and a fractional-order element; and a nonlinear circuit, also evaluated in two variations (with a capacitor and with a fractional-order element), for which harmonic balance analysis was implemented. The voltage and current curves obtained for the capacitor were sinusoidal with a 90° phase shift, producing an elliptical I–V relationship. For the fractional-order element in the linear
circuit, the waveforms were also sinusoidal, but with a phase difference other than 90°, 
resulting in an ellipse whose major axis was displaced. In the nonlinear circuit with a fractional-order element, the voltage and current curves became more complex, showing nontrivial phase shifts and distinct amplitudes. The I–V curve produced an elongated trajectory with loops, evidencing memory effects and the interaction of multiple frequencies, which are characteristic features of fractional-order elements in steady state. In addition to these analyses, a real antenna was modeled using impedance measurements from an experimental dataset. Two equivalent models were fitted: a conventional parallel RLC and a model composed of fractional-order elements. Comparing measured and calculated curves, the RLC model closely matched the resonance region, while the fractional-order model provided a better fit above the resonance, where the antenna behavior becomes more dispersive. This difference is also reflected in the analysis of the mean squared errors (MSE), indicating that the use of fractional-order elements results in a more accurate representation of the impedance when compared to the simple RLC model. These results suggest that fractional-order elements are more capable of capturing the loss mechanisms, memory effects, and dynamic variations present in such structures
Orientador: Prof. Ph.D. Eduardo Gonçalves de Lima; Trabalho de Conclusão de Curso  (graduação) - Universidade Federal do Paraná, Setor de Tecnologia, Curso de Graduação em Engenharia Eletrica; Inclui referências
</summary>
<dc:date>2025-01-01T00:00:00Z</dc:date>
</entry>
<entry>
<title>Circuitos lineares e não lineares em regime transitório descritos por modelos comportamentais</title>
<link href="https://hdl.handle.net/1884/107158" rel="alternate"/>
<author>
<name/>
</author>
<id>https://hdl.handle.net/1884/107158</id>
<updated>2026-08-21T14:30:52Z</updated>
<published>2025-01-01T00:00:00Z</published>
<summary type="text">Circuitos lineares e não lineares em regime transitório descritos por modelos comportamentais
Resumo: O presente trabalho teve como objetivo principal realizar a modelagem e a análise do comportamento de circuitos eletrônicos, lineares e não-lineares. Tentando capturar os efeitos dinâmicos obtidos pela não-linearidade, utilizou-se dados obtidos de amplificadores de potência. Para isso, foi conduzido um processo sistemático envolvendo simulação de circuito, extração de dados, identificação dos modelos, além da avaliação do desempenho de diferentes técnicas de predição, buscando compreender como certos modelos simulam a resposta real de um sistema físico e quais as limitações associadas a cada abordagem. Inicialmente, desenvolveu-se uma simulação completa de um circuito linear, determinando-se tensões e correntes e se aplicou filtros digitais para realizar a modelagem do mesmo. Aanálise revelou uma dificuldade no processamento do regime transitório do modelo, o que compromete a fidelidade de modelos lineares tradicionais. Esse comportamento reforçou a necessidade de empregar técnicas mais sofisticadas de identificação capazes de capturar  memórias dinâmicas e relações não-lineares entre entrada e saída. Na etapa seguinte, foi aplicada a modelagem auto-regressiva com entradas exógenas (do inglês, AutoRegressive with eXogenous inputs) (ARX), uma técnica clássica de identificação de sistemas lineares. Embora útil para representar dinâmicas simples e regiões aproximadamente lineares do circuito, o modelo ARX mostrou desempenho limitado quando confrontado com as não-linearidades intrínsecas do amplificador. As simulações destacaram erros consideráveis, sobretudo em regiões onde a resposta física do circuito mudava abruptamente. Métricas como erro quadrático médio (do inglês, Mean Square Error) (MSE) e coeficiente de determinação, R2, confirmaram que, apesar de capturar parcialmente a tendência da saída, o ARX não era capaz de reproduzir com precisão as transições dinâmicas mais críticas. Emseguida, a técnica auto-regressiva não-linear com entradas exógenas (do inglês, Non linear AutoRegressive with eXogenous inputs) (NARX) foi aplicada com o objetivo de melhorar a capacidade de modelagem. Ao incluir termos não-lineares e realimentação das próprias saídas, o NARX demonstrou desempenho superior ao ARX, reduzindo significativamente o erro de predição. O modelo foi capaz de representar de maneira mais precisa as regiões críticas e capturar as relações dinâmicas entre entrada e saída de forma mais fiel ao comportamento físico do circuito. A última etapa do projeto consistiu na aplicação de redes neurais artificiais em estrutura NARX. Essa abordagem combina a flexibilidade dos modelos baseados em aprendizado de máquina com a capacidade de memória dinâmica característica dos modelos autoregressivos. Os resultados obtidos foram significativamente superiores aos demais métodos, apresentando valores de MSE extremamente baixos e R2 próximos de 1. Isso evidencia a capacidade do modelo de redes neurais auto-regressivas com entradas exógenas (do inglês, Neural Network AutoRegressive with eXogenous inputs) (NNARX) em aproximar funções altamente não-lineares, reproduzindo com grande fidelidade a resposta do amplificador. No conjunto, os resultados permitiram comparar diferentes abordagens de modelagem e concluir que, embora modelos lineares sejam úteis como referência, técnicas não-lineares avançadas são mais adequadas para descrever sistemas eletrônicos reais com comportamento complexo. Assim, o projeto demonstra o potencial e a relevância de métodos inteligentes na análise e predição de circuitos não-lineares.; Abstract: The present work had as its main objective the modeling and analysis of the behavior of both linear and nonlinear electronic circuits. In an effort to capture the dynamic effects resulting from nonlinearity, data obtained from power amplifiers were used. To this end, a systematic process was carried out involving circuit simulation, data extraction, model identification, and performance to evaluate the different prediction techniques, aiming to understand how certain models simulate real physical system response and what limitations are associated with each approach. Initially, a complete simulation of a linear circuit was developed, determining voltage and current, then digital filters were applied to model it. The analysis revealed a difficulty in processing the transient response of the model, which compromises the fidelity of traditional linear models. This behavior reinforced the need to employ more sophisticated identification techniques capable of capturing dynamic memory and nonlinear relationships between input and output. In the following stage, AutoRegressive with eXogenous inputs (ARX) modeling, a classical technique for the identification of linear systems, was applied. Although useful for representing simple dynamics and approximately linear regions of the circuit, the ARX model demonstrated limited performance when confronted with the intrinsic nonlinearities of the amplifier. The simulations highlighted considerable errors, especially in regions where the physical response of the circuit changed abruptly. Metrics such as Mean Square Error (MSE) and The present work had as its main objective the modeling and analysis of the behavior of both linear and nonlinear electronic circuits. In an effort to capture the dynamic effects resulting from nonlinearity, data obtained from power amplifiers were used. To this end, a systematic process was carried out involving circuit simulation, data extraction, model identification, and performance to evaluate the different prediction techniques, aiming to understand how certain models simulate real physical system response and what limitations are associated with each approach. Initially, a complete simulation of a linear circuit was developed, determining voltage and current, then digital filters were applied to model it. The analysis revealed a difficulty in processing the transient response of the model, which compromises the fidelity of traditional linear models. This behavior reinforced the need to employ more sophisticated identification techniques capable of capturing dynamic memory and nonlinear relationships between input and output. In the following stage, AutoRegressive with eXogenous inputs (ARX) modeling, a classical technique for the identification of linear systems, was applied. Although useful for representing simple dynamics and approximately linear regions of the circuit, the ARX model demonstrated limited performance when confronted with the intrinsic nonlinearities of the amplifier. The simulations highlighted considerable errors, especially in regions where the physical response of the circuit changed abruptly. Metrics such as Mean Square Error (MSE) and R2 confirmed that, despite partially capturing the output trend, ARX was not capable of accurately reproducing the most critical dynamic transitions. Next, the Nonlinear AutoRegressive with eXogenous inputs (NARX) model was applied with the objective of improving modeling capability. By incorporating nonlinear terms and feedback of the output itself, the NARX model demonstrated superior performance compared to ARX, significantly reducing prediction error. The model was able to more accurately represent critical regions and capture the dynamic relationships between input and output in a way that was more faithful to the physical behavior of the circuit. The final stage of the project consisted of applying artificial neural networks using the NARX structure. This approach combines the flexibility of machine-learning-based models with the dynamic memory capabilities characteristic of autoregressive models. The results obtained were significantly superior to the other methods, presenting extremely low MSE values and The present work had as its main objective the modeling and analysis of the behavior of both linear and nonlinear electronic circuits. In an effort to capture the dynamic effects resulting from nonlinearity, data obtained from power amplifiers were used. To this end, a systematic process was carried out involving circuit simulation, data extraction, model identification, and performance to evaluate the different prediction techniques, aiming to understand how certain models simulate real physical system response and what limitations are associated with each approach. Initially, a complete simulation of a linear circuit was developed, determining voltage and current, then digital filters were applied to model it. The analysis revealed a difficulty in processing the transient response of the model, which compromises the fidelity of traditional linear models. This behavior reinforced the need to employ more sophisticated identification techniques capable of capturing dynamic memory and nonlinear relationships between input and output. In the following stage, AutoRegressive with eXogenous inputs (ARX) modeling, a classical technique for the identification of linear systems, was applied. Although useful for representing simple dynamics and approximately linear regions of the circuit, the ARX model demonstrated limited performance when confronted with the intrinsic nonlinearities of the amplifier. The simulations highlighted considerable errors, especially in regions where the physical response of the circuit changed abruptly. Metrics such as Mean Square Error (MSE) and R2 confirmed that, despite partially capturing the output trend, ARX was not capable of accurately reproducing the most critical dynamic transitions. Next, the Nonlinear AutoRegressive with eXogenous inputs (NARX) model was applied with the objective of improving modeling capability. By incorporating nonlinear terms and feedback of the output itself, the NARX model demonstrated superior performance compared to ARX, significantly reducing prediction error. The model was able to more accurately represent critical regions and capture the dynamic relationships between input and output in a way that was more faithful to the physical behavior of the circuit. The final stage of the project consisted of applying artificial neural networks using the NARX structure. This approach combines the flexibility of machine-learning-based models with the dynamic memory capabilities characteristic of autoregressive models. The results obtained were significantly superior to the other methods, presenting extremely low MSE values and R2 coefficients close to 1. This demonstrates the ability of NNARX (Neural Networks AutoRegressive with eXogenous inputs) to approximate highly nonlinear functions, reproducing the amplifier’s response with great fidelity. Overall, the results made it possible to compare different modeling approaches and conclude that, although linear models are useful as a reference, advanced nonlinear techniques are more adequate for describing real electronic systems with complex behavior. Thus, the project demonstrates the potential and relevance of intelligent methods in the analysis and prediction of nonlinear circuits. coefficients close to 1. This demonstrates the ability of NNARX (Neural Networks AutoRegressive with eXogenous inputs) to approximate highly nonlinear functions, reproducing the amplifier’s response with great fidelity. Overall, the results made it possible to compare different modeling approaches and conclude that, although linear models are useful as a reference, advanced nonlinear techniques are more adequate for describing real electronic systems with complex behavior. Thus, the project demonstrates the potential and relevance of intelligent methods in the analysis and prediction of nonlinear circuits. confirmed that, despite partially capturing the output trend, ARX was not capable of accurately reproducing the most critical dynamic transitions. Next, the Nonlinear AutoRegressive with eXogenous inputs (NARX) model was applied with the objective of improving modeling capability. By incorporating nonlinear terms and feedback of the output itself, the NARX model demonstrated superior performance compared to ARX, significantly reducing prediction error. The model was able to more accurately represent critical regions and capture the dynamic relationships between input and output in a way that was more faithful to the physical behavior of the circuit. The final stage of the project consisted of applying artificial neural networks using the NARX structure. This approach combines the flexibility of machine-learning-based models with the dynamic memory capabilities characteristic of autoregressive models. The results obtained were significantly superior to the other methods, presenting extremely low MSE values and R2 coefficients close to 1. This demonstrates the ability of NNARX (Neural Networks AutoRegressive with eXogenous inputs) to approximate highly nonlinear functions, reproducing the amplifier’s response with great fidelity. Overall, the results made it possible to compare different modeling approaches and conclude that, although linear models are useful as a reference, advanced nonlinear techniques are more adequate for describing real electronic systems with complex behavior. Thus, the project demonstrates the potential and relevance of intelligent methods in the analysis and prediction of nonlinear circuits
Orientador: Prof. PhD. Eduardo Gonçalves de Lima; Trabalho de Conclusão de Curso  (graduação) - Universidade Federal do Paraná, Setor de Tecnologia, Curso de Graduação em Engenharia Eletrica; Inclui referências
</summary>
<dc:date>2025-01-01T00:00:00Z</dc:date>
</entry>
</feed>
