Aplicação de algoritmos de aprendizado de máquina para a demodulação de sinais de comunicação por luz visível

Abstract

Due to the exponencial technology advance, the amount of data and it’s complexity are becoming higher. Thus, the use of Machine Learning become interesting in telecommunications area to signal demodulation. This work purpose a method made by three steps to develop a machine learning based OOK demodulator to visible light communication. These three steps are: data aquisition, pre-processing, building and training machine learning algorithms. Data was acquired by a public dataset developed by Ma et al. (2019), having multiple modulations in a visible light communication system varying the distance between emitter and receptor by 0 cm to 180 cm. Data will be filtered by a level three gaussian filter and splitted by two approachs: A machine learning model for each distance, creating multiple specific models, and an unique machine learning model for all dataset, creating a general purpose approach. The used machine learning algorithms are: Decision trees, bagging and boosting ensemble methods. The metric used is model accuracy. Random Forest and Gradient Boosting had the best accuracies in both approachs, furthermore, filter also presented an accuracy gain in both approachs. Finally, specific models approach get better results than the general purpose approach.