Parametriza??o das distribui??es da estat?stica de teste GID sob as hip?teses Ho e H1 via redes neurais aritificiais

• Ano de defesa: 2019
• Autor(a) principal: Lemes, Alan Lima
• Orientador(a): Guimar?es, Dayan Adionel
• Banca de defesa: Guimar?es, Dayan Adionel , Ynoguti, Carlos Alberto , Rodrigues Leite, Jo?o Paulo Reus
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Instituto Nacional de Telecomunica??es
• Programa de Pós-Graduação: Mestrado em Engenharia de Telecomunica??es
• Departamento: Instituto Nacional de Telecomunica??es
• País: Brasil
• Palavras-chave em Português: R?dio Cognitivo; GID; Sensoriamento Espectral Cooperativo; Teste de Ader?ncia; Redes Neurais Artificiais
• Área do conhecimento CNPq: Engenharia de Telecomunica??es
• Link de acesso: https://tede.inatel.br:8080/tede/handle/tede/187

Spectral shortage is a major constraint to the advancement of wireless communication systems especially when such systems must provide a high data rate and support high connection density, which is expected from the fifth generation of telecommunications networks. Cognitive radio technology allows opportunistic and efficient use of bands that may be underutilized in the electromagnetic spectrum and, therefore, may be a solution to the aforementioned problem. To determine free spectral bands, cognitive radios use a technique called spectral sensing. Many sensing techniques have been proposed in the literature, but performing the performance evaluation of such techniques and relating them to the systemic parameters is not a trivial task. Recently the GID (Gini index detector) test was proposed for centrelized cooperative spectrum sensing on cognitive radio systems. Its main features are the low computational complexity, the robustness against unequal and dynamical noise and received signal powers. In this dissertation the procedures and the results of the goodness-of-fit of the GID test statistic are presented to diverse distributions of probability. It is demonstrated that the Stable distribution adequately characterizes the statistic under hipotese H0, while the Generalized Extreme Value distribution best applies to H1. Two artificial neural networks are then developed to establish the mapping between the systemic parameters and the parameters that characterize these distributions, allowing theoretical calculations of the performance and the decision threshold of spectral sensing are performed.