Aplicações de aprendizagem de máquinas às comunicações móveis: gerenciamento de recursos e avaliação de desempenho

Abstract

In order to bear the growth in the traffic volume expected for the coming years, the next generation of mobile communication systems relies on advanced technologies such as multiple orthogonal subcarriers and coordination of multiple transmission points. The radio resources are organized in a higher number of dimensions, making resource allocation and assessment of link-level performance more complex tasks. Based on techniques from machine learning, we investigated novel ways of addressing these tasks in order to perform them efficiently. This thesis provides two proposals: (i) clustering of transmission points and (ii) design of a link-to-system interface. In proposal (i), the k-means algorithm is used to identify the strength signal vectors that are similar to each other, leading to a decrease on the cooperation complexity. In proposal (ii), we make use of artificial neural networks to learn the behavior of link-level simulations, resulting in a link-to-system interface more reliable for certain situations on the cell-edge region. The results obtained in both proposals confirm machine learning as an appropriate approach to the problems addressed, being able to lead to alternative interpretations and efficient solutions.