Aplicação de redes neurais na identificação de multipercursos para a obtenção de parâmetros de dispersão temporal

• Ano de defesa: 2019
• Autor(a) principal: Nogueira, Sandra Eloi Ferreira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual do Maranhão; Brasil; Campus São Luis Centro de Ciências Tecnológicas – CCT; PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA DA COMPUTAÇÃO E SISTEMAS - PECS; UEMA
• Palavras-chave em Português: Redes Neurais Artificiais; CFAR; Mean Excess Delay; RMS Delay Spread; Retardo RMS; Retardo de potência; Power delay; Engenharias
• Link de acesso: https://repositorio.uema.br/jspui/handle/123456789/3386

The temporal dispersion is one of the intrinsic problems of the multipath that affects the integrity of the signal and one of the ways to combat it is through the knowledge of the medium delay parameters and RMS delay spread, which are extracted from the power delay profile .These parameters need to be estimated with a certain accuracy, since other data depend on them, such as, equalizers, cyclic prefix, coherence band, bit error rate (BER), data transmission rate, among others, that are used in the project of wireless systems. This accuracy dependents directly on the number of valid multipaths found in the power delay profile. Therefore, the cleaning of the delay profile is an important step in the channel characterization, so that the number of studies using several filtering techniques is increasing: Constant False Alarm (CFAR) [1], [2] and [3 ], CLEAN [4] [5] and WAVELET [6], in order to identify the valid multiperpaths of impulsive noises. Therefore, this work proposes the use of an alternative power-delay profile cleaning technique using artificial neural networks to identify and eliminate impulsive noise. The technique of filtering using RNA with feed-forward architecture through the back-propagation algorithm was tested and compared with the results produced by the CFAR technique. RNA tests were carried out in five different propagation environments, including urban, semi-urban, rural and predominantly vegetated regions at different transmission frequencies. The neural technique showed a better efficacy with a mean accuracy of 91.38% and, consequently, more satisfactory values than the CFAR for the identification of valid multiperceptions. The results of the Mean Excess Delay and RMS Delay Spread parameters obtained by the new technique presented adequate values when compared with the CFAR and with the values described in the ITU-R norm P.1411-9