GPON Traffic Analysis with TensorFlow
Václav Oujezský, Adrián Tomašov, Martin Holik, Vladislav Škorpil, Tomáš Horváth, Michal Jurčík · 2020
The paper presents the latest research results of gigabit passive optical network analysis using machine learning algorithms. TensorFlow has been used to learn the frames of GPON management traffic of the G.984.3 protocol and detect outliers to the standard. The frames are captured by FPGA programmable network card and processed with a software parser and further processed with the TensorFlow using JSON format. The proposed technique can bring a way how to test if vendors of network devices follow the standard and if the content of the frames can be treated as secure and trustworthy.