Aprendizagem profunda aplicada a telecomunicações: classificação de modulação e controle de congestionamento

Abstract

The goal of this dissertation is to explore Deep Learning (DL) techniques applied to Telecommunications. DL has achieved success in areas such as computer vision and object detection and it is timely to investigate DL in communication problems. Then, two distinct problems are investigated. First, DL is applied to Automatic Modulation Classification (AMC) to detect the adopted modulation scheme automatically. AMC is important, for instance, in Cognitive Radios and military applications. In this dissertation, we discuss the benefits and drawbacks of using DL as an alternative for AMC and show its efficiency in comparison to other machine learning methods applied to AMC. Other DL application in communication is Congestion Control. The context is related to Fronthaul in C-RAN architecture using congestion control in order to attend the strict requirements of the 5G system. Specifically, DL is investigated in conjuction with Deep Reinforcement Learning (DRL) techniques. In this topic, this dissertation presents a framework for investigations in congestion control for Fronthaul, and the implementation of a model and environment using NS-3 and Gym API of the OpenAI group for simulation. The developed framework is validated with preliminary experiments that compare Deep Reinforcement Learning methods with traditional congestion control techniques, using as figu