Brillouin Optical Time Domain Analyzer enhanced by Artificial/Deep Neural Networks

Abstract

We report our recent studies on the use of Neural Networks to process the measured Brillouin gain spectrum (BGS) from Brillouin Optical Time Domain Analyzer (BOTDA) and extract temperature information along fiber under test (FUT). Artificial Neural Network (ANN) is trained with ideal Lorentizian BGS before it is used for temperature extraction. Its performance is evaluated by comparison to conventional curve fitting techniques, showing better accuracy especially at large frequency scanning step during the acquisition of BGSs. We have also applied advanced hierarchical Deep Neural Network (DNN) in BOTDA for temperature extraction to improve the training and testing efficiency. We believe that ANN/DNN can be attractive tools for direct temperature or strain extraction in BOTDA system with high accuracy.