Alocação de Recursos baseada em Clustering com Aprendizado de Características e Orientação a QoS em Redes LTE-Advanced

• Ano de defesa: 2019
• Autor(a) principal: Santos, Einar César
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Engenharia Elétrica
• Palavras-chave em Português: Alocação de Recursos; Aprendizado de Máquina; LTE-A; QoS; Simulação; Clustering; Resource Allocation; Machine Learning; Aprendizado de Características; LTE-Advanced; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA::TELECOMUNICACOES::SISTEMAS DE TELECOMUNICACOES; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufu.br/handle/123456789/25825; http://dx.doi.org/10.14393/ufu.te.2019.2060

The increasing demand for wireless mobile networks access has increased operational and management costs as well as the minimum levels of Quality of Service (QoS) required for applications provisioning in Long Term Evolution Advanced (LTE-A) networks. Although the latest LTE-A specifications offer cost-effective technologies, they are not prepared to deal with Big Data, characterized by a large volume of data constantly produced at high velocity, variety, veracity and value in information and communication systems. In this sense, Machine Learning is widely used and recommended to process Big Data. The Clustering-Based Resource Allocation (CBRA) strategy, designed for autonomous resource allocation by the use of clustering (a task performed by Machine Learning), categorizes users and applications from patterns discovered in communication system data in which it is employed, providing better QoS parameters. However, to be able to classify properly, the CBRA requires a better definition of features informed as well as parameters adapted to the analyzed data during the clustering execution. This work proposes a CBRA mechanism, specifically developed for LTE-A, provided with an algorithm for autonomous feature learning and another for clustering de-parameterization. The proposal is evaluated through computational simulation and compared with other related mechanisms existing in the literature. The results show better classification quality, in comparison with the number of analyzed features, and adequate performance of the mechanism for real-time video applications.