Probabilistic deep neural network based signal processing for Brillouin gain and phase spectrums of vector BOTDA system

Abstract

We demonstrate a novel probabilistic Brillouin frequency shift (BFS) estimation framework for both Brillouin gain and phase spectrums of vector Brillouin optical time-domain analysis (BOTDA). The BFS profile is retrieved along the fiber distance by processing the measured gain and phase spectrums using a probabilistic deep neural network (PDNN). The PDNN enables the prediction of the BFS along with its confidence intervals. We compare the predictions obtained from the proposed PDNN with the conventional curve fitting and evaluate the BFS uncertainty and data processing time for both methods. The Brillouin phase spectrum generally provides a better measurement accuracy with reduced measurement time in comparison to the Brillouin gain spectrum-based measurement, for an equal signal-to-noise ratio and linewidth. The proposed method is demonstrated using a 25 km sensing fiber with 1 m spatial resolution. The PDNN based signal processing of the vector BOTDA system provides a pathway to enhance the BOTDA system performance.