Aplicações de aprendizagem de máquina na estimação de qualidade de enlace em comunicações sem fio

Abstract

The popularization of embedded devices and their direct wireless communication, i.e., machine-to-machine (M2M) communication, is a growing trend in today’s networks. An increasing number of embedded devices support different types of applications, such as healthcare, surveillance, gas monitoring, among others, which require a high level of communication reliability due to the critical information they carry. However, the use of many of these devices increases competition for the frequency spectrum, making it difficult to achieve reliable communication. To overcome these limitations, the use of link quality estimators (LQE - Link Quality Estimators) is crucial to provide efficient communication. In order to enable a robust and fast communication under adverse conditions, such as noise, this work proposes two new LQE, called PRR2 and LQM, which use multiple link quality metrics and different strategies for calculating the estimate. The PRR2 estimator captures variations in link quality in the short term, but also considers long-term variations. The LQM estimatotors are compared with the state-of-the-art in a prototype using USRPs (Universal Software Radio Peripheral) radios, and the results show that the proposals increase the reliability of the links by reducing the number of retransmissions by up to 61% and by increase the delivery rate by up to 43%.