Improving the Brillouin frequency shift measurement resolution in the Brillouin optical time domain reflectometry (BOTDR) fiber sensor by artificial neural network (ANN)

We propose to use an artificial neural network (ANN) model to improve the Brillouin frequency shift (BFS) resolution of the sub-meter spatial resolution differential cross-spectrum Brillouin optical time domain reflectometry (DCS-BOTDR) fiber sensor. ANN technique was chosen due its nonlinear mapping characteristic that makes it suitable to effectively extract the BFS distribution along the fiber optic. Importantly, the ANN model offers more flexibility in estimating the BFS because one does not need to set the initial parameters like the conventional polynomial function based on Lorentzian curve fitting (LCF) model. In our ANN model, we used the BFS distribution data measured by the pulse duration TL = 14 ns at 56,000 times averaging as the ground truth in the training phase, because they provided the highest BFS resolution compared to that of obtained by other pulse duration cases, as reported previously. After training the data, we tested them with the BFS distribution results for TL = 2, 4 and 60 ns cases, which they showed poor BFS resolution when processed by the conventional model. For TL = 2 and 4 ns cases, the deployment of ANN has significantly improved the BFS resolution at least two times higher than that by the conventional polynomial fitting. Surprisingly, for TL = 60 ns case, despite the reduced number of averaging from 56,000 to 7000 times (i.e., eight times faster measurement), the BFS resolution has improved about seven times from 21.13 MHz to 2.88 MHz. This indicates the significant contribution of the ANN model in reducing the measurement time in the DCS-BOTDR

Automated summary

We propose using an artificial neural network (ANN) model to improve the Brillouin frequency shift (BFS) resolution of the differential cross-spectrum Brillouin optical time domain reflectometry (DCS-BOTDR) fiber sensor. The ANN model offers more flexibility in estimating BFS and significantly improves the resolution.