DAS-N2N: Machine learning Distributed Acoustic Sensing (DAS) signal denoising without clean data

TL;DR

This work tackles denoising of DAS data without clean training labels by introducing DAS-N2N, a weakly supervised Noise2Noise approach that uses two spliced fibres to generate two noisy copies of the same underlying signal. A lightweight 3-layer U-Net is trained to map one noisy copy to the other, effectively mapping noise to a distribution statistic while preserving the true signal, enabling rapid denoising of data from a single fibre after training. Evaluated on Antarctic Rutford Ice Stream data, DAS-N2N substantially improves SNR of microseismic icequake signals and outperforms conventional bandpass and Wiener filtering as well as a self-supervised baseline (jDAS), with processing times under 1 second for 30 seconds of data on a single GPU. The method generalizes to other DAS datasets (e.g., offshore Oregon cables) with minimal retraining, offering a practical, real-time denoising solution for DAS-based earthquake detection and monitoring, albeit with some signal leakage that may require calibration for quantitative amplitude studies.

Abstract

This article presents a weakly supervised machine learning method, which we call DAS-N2N, for suppressing strong random noise in distributed acoustic sensing (DAS) recordings. DAS-N2N requires no manually produced labels (i.e., pre-determined examples of clean event signals or sections of noise) for training and aims to map random noise processes to a chosen summary statistic, such as the distribution mean, median or mode, whilst retaining the true underlying signal. This is achieved by splicing (joining together) two fibres hosted within a single optical cable, recording two noisy copies of the same underlying signal corrupted by different independent realizations of random observational noise. A deep learning model can then be trained using only these two noisy copies of the data to produce a near fully-denoised copy. Once the model is trained, only noisy data from a single fibre is required. Using a dataset from a DAS array deployed on the surface of the Rutford Ice Stream in Antarctica, we demonstrate that DAS-N2N greatly suppresses incoherent noise and enhances the signal-to-noise ratios (SNR) of natural microseismic icequake events. We further show that this approach is inherently more efficient and effective than standard stop/pass band and white noise (e.g., Wiener) filtering routines, as well as a comparable self-supervised learning method based on masking individual DAS channels. Our preferred model for this task is lightweight, processing 30 seconds of data recorded at a sampling frequency of 1000 Hz over 985 channels (approx. 1 km of fiber) in $<$1 s. Due to the high noise levels in DAS recordings, efficient data-driven denoising methods, such as DAS-N2N, will prove essential to time-critical DAS earthquake detection, particularly in the case of microseismic monitoring.

Figures (9)

• Figure 1: Map and schematic illustration showing the DAS experiment setup. Top left: Map showing geographic location of DAS array (gold triangle) on Rutford Ice Stream, Antarctica. Main: Schematic illustration of the DAS experiment. DAS fibre array was deployed in triangular configuration on the surface of Rutford Ice Stream, with two single-mode fibres hosted within a single cable jacket spliced at cable end. See Hudson_Baird_Kendall_Kufner_Brisbourne_Smith_Butcher_2021 and Supporting Information in Hudson_Baird_Kendall_Kufner_Brisbourne_Smith_Butcher_Chalari_Clarke_2021 for further details.
• Figure 2: Implementing DAS-N2N. A) Raw data is split into input (Fibre 1) and target (Fibre 2) training data. B) Data is divided into smaller sections (128 samples x 96 channels) for model training, with augmentation (vertical / horizontal flipping) randomly applied to each training sample pair. C) Once the model is trained, only the input data (Fibre 1) is required for denoising.
• Figure 3: In-sample example of two icequakes (S-wave arrivals only) recorded by DAS deployment (time in seconds after 2020-01-17 01:30:19.232 UTC). A) Raw DAS data. B) Butterworth (2-pass, 4th order) 10 – 100 Hz bandpass filtered DAS data. C) Wiener filtered (7x7 window size) DAS data. D) jDAS filtered DAS data. E) DAS-N2N filtered DAS data. Icequake S-waves arrive at DAS channel 0 at time 0.4 s and 0.55 s, respectively. Strain rate is recorded in units of strain/s (counts).
• Figure 4: Local signal-to-noise ratio (SNR) estimates for each example in Figure \ref{['fig:fig3']}. SNR is calculated using semblance (Equation \ref{['eq:7']}) and a 13-channel x 19-sample 2D moving window
• Figure 5: Individual DAS trace (top) and corresponding spectrogram (bottom) for DAS channel 255 in each example in Figure \ref{['fig:fig3']}. Strain rate is recorded in units of strain/s (counts).