Seismogram Fingerprints As a Tool for Automatic Filtering of Low-Frequency Noise

Abstract

The article begins with a review of publications on low-frequency noise suppression techniques. Denoising seismograms of earthquakes, explosions, and other seismic events is the goal of the study. It demonstrates that a branch in the theory and practice of seismogram processing is currently being actively developed, in which they are analyzed in a two-dimensional time–frequency plane. Additional second- and third-level add-ons appear in addition to the existing methods, which makes it difficult both to understand the essence of the methods and interpret their results. In our article, we attempted to order things in them. As an alternative to numerous add-ons to time–frequency analysis, we proposed our own approach. We believe that it will not only make the analysis clearer, but also increase its accuracy. Our method is based on the application of fingerprint technology to the results of continuous wavelet transform of a seismogram. In difficult cases, we recommend using a more advanced version of it: the redundant fingerprint method. It provides a convenient opportunity to objectively assess the frequency responses of all components of the seismogram. Based on the results of analysis, the automatic information system can select the optimal cutoff frequency for the filter in order to clear the seismogram of low-frequency noise and minimally distort the signal shape. This is especially important if the spectra of both partially overlap and if the noise intensity is high. The method may find itself in demand for automatic classification of seismic events by the nature of their source using machine learning technologies.