This article describes the application of the optimal transfer function neural network methodology for fast and accurate function approximation of Q‐integrals, that appears in directivity expressions of circular‐shape antennas, such as loop, microstrip patch, and aperture in an infinite ground plane. These directivity expressions were revisited and linked in terms of Q‐type power‐radiation integrals. For each directivity approach using optimal neural networks–closed‐form expressions, series of Bessel functions, and adaptive Lobatto integration, the numerical results are presented and discussed in terms of accuracy and computational cost. © 2007 Wiley Periodicals, Inc. Microwave Opt Technol Lett 49: 2721–2726, 2007; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/mop.22839

This paper presents a Radial Basis Function/Multilayer Perceptron (RBF/MLP) modular neural network, training with the Resilient Backpropagation (Rprop) algorithm which has been used for nonlinear device modeling in microwave band. The proposed modular configuration employs three or more neural networks, each one with a hidden layer of neurons, and aim to take advantage of the MLP and RBF networks specific characteristics to improve learning aspects, such as: ability to learn, speed of training and learning with consistency, or generalization. Simulations through the proposed neural network models for microstrip line with anisotropic PBG (Photonic Bandgap) structure and a metallic enclosure microstrip with PBG gave responses in good agreement with accurate results (measured or simulated) available in the literature.

<p>Este trabalho trata do estudo e implementação de técnicas de processamento digital de sinais que são utilizadas em sistemas de resposta vocal como: sistemas de reconhecimento de voz, reconhecimento de locutor e sistemas de síntese de voz. Foi utilizada a linguagem de programação C na elaboração de funções que caracterizam os sinais de voz. Foram implementados algoritmos para conversão amostras-texto, detecção de início e fim, pré-ênfase, janelamento de Hamming e extração de parâmetros temporais. Estes parâmetros incluem energia, taxa de cruzamento por zero, número total de picos, diferença no número de picos e coeficiente de correlação normalizado. Foram propostos algoritmos diferenciados para a taxa de cruzamento por zero e para o detector de início e fim. Com o intuito de criar um ambiente didático, construiu-se uma interface gráfica amigável utilizando uma plataforma de desenvolvimento orientada a eventos.</p>