Abstract

Distributed Acoustic Sensing (DAS) offer unique advantages to extensive monitoring initiatives due to its long-distance, high-density arrays of real-time acoustic sensors. However, high-resolution and long-term applications generate a large amount of data, on the order of hundreds of terabytes per year, presenting challenges for processing and storage. As a result, leveraging innovative data compression techniques is essential for scaling up DAS monitoring. DAS arrays are composed of multiple channels carrying highly correlated and coherent signals. These redundancies allow to exploit inter-channel compression, which use predictions from consecutive channel signals and achieve a higher compression ratio compared to compressing each channel separately.

 

In this work, we propose a novel coding scheme for DAS data that improves state-of-the-art compression. Encoding follows a pipeline composed of intra-prediction, inter-prediction, transform and entropy coding. For lowly correlated channels, intra-channel prediction is realized by means of Linear Prediction Coding. For inter-channel prediction, we propose three different methods: (1) prediction by warping and scaling the signal from the consecutive channel, (2) a linear predictor is learned over many channels and (3) wavelet transforms are used to disentangle data signals of different frequency, allowing fine-grained prediction. Entropy coding is done by a combination of adaptive arithmetic coding and run-length coding. The implementation is divided into an encoder and a decoder. The encoder uses bitrate optimization for decisions on the prediction methods and parameter values and can be tuned for either speed or high-compression modes. In addition, we explore lossy quantization by adding tunable quantization of the transformed signal, which we observe to achieve significantly higher compression at the expense of quantization noise. The designed algorithms and the provided implementation facilitate the deployment of long-term DAS monitoring.