To address the problem of low signal-to-noise ratio (SNR) in long-distance sensing using a single-pulse Brillouin optical time-domain analyzer (BOTDA), a fusion method of pulse coding and denoising convolutional neural network (DnCNN) is proposed to improve the BOTDA SNR over long distances. This method can be widely used in long-distance engineering fields, such as long-distance oil and gas pipeline leakages, optical fiber composite overhead ground wire (OPGW) cable safety warnings, and submarine cable monitoring. When traditional single-pulse BOTDA performs long-distance sensing, it is generally necessary to increase the peak pulse power or the cumulative average number of measurements to obtain higher SNR. However, an excessively high peak input power causes a modulation instability effect, resulting in a decrease in the measurement accuracy of the system. An excessive cumulative average number significantly increases the measurement time of the system. Therefore, a fusion method of pulse code and denoising convolutional neural network (DnCNN) is used to improve the SNR of BOTDA. This method can effectively improve the SNR, extend the sensing distance, and accelerate the measurement speed, while maintaining the spatial resolution of the system.

First, the signal strength enhancement principle of Golay coding BOTDA and the noise characteristics of the Brillouin gain spectrum (BGS) are analyzed, and the SNR enhancement scheme of the Golay coding fusion DnCNN is constructed. Under similar experimental conditions, the BGS along the fiber is acquired using single-pulse and Golay coding, and the BGS acquired using Golay coding is denoised using a trained DnCNN. Subsequently, the peak pulse power is increased to 110 mW, and single-pulse measurement signals averaged 2000 times are collected and compared with the signals obtained by the fusion method averaged 100 times. The results are compared at 5 m spatial resolution and root mean square error (RMSE) of the temperature change area of less than 0.2 MHz. The block-matching 3D filtering algorithm (BM3D), complete ensemble empirical mode decomposition with adaptive noise combined with wavelet threshold (CEEMDAN-WT), and DnCNN are used to reduce the noise of the data collected using the Golay coding, and the effects and running times of different noise reduction methods are compared. Finally, a gradient-temperature experiment is conducted to verify the effectiveness of the fusion method at different temperatures.

The results of the experiments show that compared with the single-pulse modulation mode, at the same pulse peak power, the fusion method can increase the system sensing distance from 10.8 km to 100 km, and the SNR at 10.8 km is increased by 18.92 dB. Compared with the Golay coding modulation mode, the fusion method increases the SNR by 9.17 dB at the 100 km end (Fig. 9). It is further verified that at 5 m spatial resolution and RMSE of temperature change area of less than 0.2 MHz, the cumulative average times required by the fusion method decreases from 2000 times to 100 times, and the measurement time is shortened from 1056 s to 194 s compared with single-pulse modulation (Fig.12). Comparing the time required by BM3D, CEEMDAN-WT, and DnCNN to reduce the noise of the experimental data, DnCNN only took 4.62 s, whereas BM3D and CEEMDAN-WT required approximately 8 h and 27 h, respectively (Fig. 13). The Brillouin frequency shift (BFS) and temperature change in the BFS curve obtained by the fusion method maintains a good linear relationship, and the temperature information along the fiber can be accurately restored at different temperatures (Fig. 14).

In this study, a fusion method of pulse coding and denoising convolutional neural network (DnCNN) is proposed. This method can increase the sensing distance of the BOTDA system from 10.8 km to 100 km under a cumulative average of 100 times and a pulse peak power of 18 mW, and the SNR along the fiber is improved. At a measurement accuracy of 100 km sensing distance, 5 m spatial resolution, and RMSE of temperature change area of less than 0.2 MHz, the measurement time of the fusion method is shortened from 1056 s to 194 s compared with the single-pulse method. This fusion method can be used to measure the temperature in long-distance oil and gas pipeline leakages, OPGW cable safety warnings, submarine cable monitoring, and other engineering fields.