Evaluating measurement uncertainty in Brillouin distributed optical fibre sensors using image denoising

In 2016, our research team proposed in an issue of Nature Communications1 the use of multidimensional signal processing, especially image denoising techniques, to improve the signal-to-noise ratio (SNR) of distributed optical fibre sensors. The benefits of the method were demonstrated for distributed Raman and Brillouin sensors, both proving a significant performance improvement. Here we show that, while the SNR enhancement for the case of Brillouin distributed sensing was correctly estimated in our publication1, an overestimation of the Brillouin frequency uncertainty reduction was reported as a result of the inadvertent use of a conventional methodology for performance evaluation. Based on a better understanding of the impact of image denoising applied to Brillouin distributed sensors, here we report a revised estimation of the Brillouin-frequency shift (BFS) uncertainty obtained after image denoising, verifying that although 2D image denoising can significantly improve the measurement SNR, this cannot be fully transferred to the overall Brillouin sensing performance. Based on novel findings and a deeper understanding of the method, and for the sake of clarity, we start by clearly stating the fundamental limitations and real benefits of image denoising applied to Brillouin distributed sensing, by partially reusing the analysis and conclusions drawn in ref. 2. In the following, the effects of 2D filtering in the 2 dimensions of the matrix containing the measured data (i.e. in the Brillouin-spatial domain along the fibre and in the Brillouin-frequency domain describing each local Brillouin gain spectrum) are separately addressed.