High-Precision Recognition of Jump Event in Brillouin Optical Time-Domain Sensors

We propose a novel post-processing method based on long and short-term memory artificial neural networks (ANNs) to improve the recognition accuracy of temperature/strain jump events in Brillouin optical time domain sensors (BOTDSs). The method achieves centimeter-level accuracy in extracting edge locations of temperature/strain jump events, without spatial resolution (SR) deterioration caused by pulsewidth or sampling rate. In a proof-of-concept experiment, we use a Brillouin optical time domain reflectometry sensor to measure the Brillouin gain spectrum (BGS) slope of a 1 km sensing fiber. With an averaging number of over 1000 and a data sampling rate of 200 MSample/s, the method achieves a recognition uncertainty of temperature jump event edges of about 5.2 cm, which is only 10% of that of the conventional methods and other ANN-based methods. Moreover, the proposed method can be used to realize fast sensing with a measurement time of 0.08 s and a processing time of 0.03 s.