부산대학교 도서관

We propose and investigate a method to estimate Brillouin frequency shift (BFS) for Brillouin optical time domain analysis (BOTDA) using Brillouin gain and loss spectra (BGS and BLS) along with support vector machine classifier (SVC) that is one of the typical machine learning (ML) models. BGS and BLS are simultaneously swept by dual-frequency probe light, where a curve with double peaks is obtained as a function of sweep frequency. We call the obtained curve the virtual gain spectrum (VGS), from which BFS is estimated. We conducted simulation to investigate the effect of the signal-to-noise ratio (SNR) and the number of sweep points in terms of BFS estimation. Besides, the accuracy of the proposed BFS estimation method was evaluated by the experiments. Both simulation and experimental results showed that the proposed method using VGS exhibits less error in BFS estimation compared to the conventional method using BGS. In the experiment, when the frequency of probe light was swept with a step of 1 MHz, the average standard deviation of estimated BFS was 0.713 MHz in VGS SVC, while it was 1.434 MHz in BGS SVC. Even when the frequency sweep step was 10 MHz, the average standard deviation was 0.930 MHz in VGS SVC, whereas it was 3.301 MHz in BGS SVC.