Support Vector Machine based Differential Pulse-width Pair Brillouin Optical Time Domain Analyzer

Abstract

Support vector machine (SVM) based differential pulse-width pair Brillouin optical time domain analyzer (DPP-BOTDA) has been proposed and experimentally demonstrated. With only one SVM model, temperature distribution along 5 km fiber under test has been successfully extracted from differential Brillouin gain spectrum (BGS) measured under different spatial resolution in DPP-BOTDA. The temperature accuracy by SVM is better than that by Lorentzian curve fitting (LCF), especially when the pump pulse width difference and the number of trace averaging used in the measurement are small, indicating larger tolerance of SVM to high spatial resolution and low signal-to-noise ratio. SVM is also more robust to a wide range of frequency scanning steps and has less accuracy degradation under large frequency scanning step. To extract temperature from 50 000 differential BGSs, 133.17 and 1.12 s are consumed by SVM-0.1 and SVM-1 °C, respectively, both of which are much shorter than that by LCF. The data processing time of SVM is further shortened with the help of principle component analysis for data dimension reduction. SVM for measurand extraction would be especially helpful in the scenario of DPP-BOTDA where high data sampling rate is required to resolve plenty of submeter scale sensing points.