Otimização da escolha de modelo de propagação por medição de campo e inteligência artificial

• Ano de defesa: 2019
• Autor(a) principal: Botelho, Alberto Leonardo Penteado
• Orientador(a): Akamine, Cristiano
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Presbiteriana Mackenzie
• Palavras-chave em Português: televisão digital; SBTVD; modelo de propagação; rede de frequência única; predição de cobertura; medição de campo; inteligência artificial; aprendizagem de máquina; aprendizagem por classificação
• Área do conhecimento CNPq: CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA::TELECOMUNICACOES::SISTEMAS DE TELECOMUNICACOES
• Link de acesso: http://dspace.mackenzie.br/handle/10899/24493

The propagation model to be chosen in the design of a digital terrestrial broadcast station is a critical point for predicting the coverage area. There are several models, with specific characteristics that may be better than others in certain situations. This dissertation presents a study of the choice of propagation model, through the use of artificial intelligence (AI). A brief review of the SBTVD (Brazilian System of Digital Television), the complexity operation in SFN (Single Frequency Network) and the most widely used propagation models in the literature. The comparison of propagation models was elaborated with the field measurements and simulations by the Progira coverage prediction software, which works on an ArcGis geoprocessing platform that considered the criterion of smallest average error (absolute mean deviation, standard deviation and root mean square error) between the field measurement and the software simulation. The propagation model ITUR P. 1812-3 had the best average performance. To optimize the analysis of choice of propagation models, an AI method was developed by machine learning, classification learning, so that the computer can formulate aspects of human intelligence and have the ability to choose the best propagation model for each study area, not restricted to sites measured in the field. The Support Vectors Machines and Nearest Neighbor Classifiers learning models displayed a significant improvement of the average error in comparison to the model of propagation of smallest average error