Fast and accurate Brillouin frequency shift extraction in Brillouin optical time domain reflectometry (BOTDR) distributed fiber sensor by using ensemble machine learning algorithm

Abstract

To improve the Brillouin frequency shift (BFS) resolution measurement and processing time of the differential cross-spectrum Brillouin optical time domain reflectometry (DCS-BOTDR) fiber sensor, our team suggests employing the ensemble machine learning (EML) technique. Because it gave the best BFS resolution compared to the other TL cases, we used the BFS distribution data recorded by the pulse duration TL =14 ns case as ground truth to train the EML model in this work. After that, we tested the EML model for TL =4, 60, and 90 ns cases. We improved the BFS resolution for all TL situations by approximately 2.85 MHz, comparable to our resolution when TL was equal to 14 ns. This result demonstrates that the EML algorithm is reliable, efficient, and highly accurate in its predictive capabilities. Additionally, we have documented a rapid processing time of approximately one second. In addition, we have successfully demonstrated 20 cm spatial resolution measurement for TL =60 and 90 ns, which was not previously possible with the usual DCS-BOTDR signal processing method.