Various aspects described herein relate to a machine learning based signal recovery. In one example, a computer-implemented method of noise contaminated signal recovery includes receiving, at a server, a first signal including a first portion and a second portion, the first portion indicative of data collected by a plurality of sensors, the second portion representing noise; performing a first denoising process on the first signal to filter out the noise to yield a first denoised signal; applying a machine learning model to determine a residual signal indicative of a difference between the first signal and the first denoised signal; and determining a second signal by adding the residual signal to the first denoised signal, the second signal comprising (i) signals of the first portion with higher magnitudes than the noise in the second portion, and (ii) signals of the first portion having lower magnitudes than the noise in the second portion.

Acoustic imaging waveforms are measured utilizing a downhole acoustic tool within a wellbore, and then aligned relative to a main echo of each waveform. The aligned waveforms are then subjected to a first low-pass filter. Residuals are extracted by determining differences between the aligned waveforms and the filtered waveforms. The residuals are aligned to corresponding acoustic firing pulses of the downhole acoustic tool. The aligned residuals are subjected to a second low-pass filter. The measured waveforms are aligned to the corresponding acoustic firing pulses. Noise associated with the downhole acoustic tool is removed from the pulse-aligned, measured waveforms utilizing the filtered residuals.

A method for estimating a time variant signal representing a seismic source obtains seismic data recorded by at least one receiver and generated by the seismic source, the recorded seismic data comprising direct arrivals and derives the time variant signal using an operator that relates the time variant signal to the acquired seismic data, the operator constrained such that the time variant signal is sparse in time.

The present method predicts and separates dispersive surface waves from seismic data using dispersion estimation and is completely data-driven and computer automated and no human intervention is needed. The method is capable of predicting and suppressing surface waves from recorded seismic data without damaging the reflections. Nonlinear signal comparison (NLSC) is used to obtain a high resolution and accurate dispersion. Based on the dispersion, surface waves are predicted from the field recorded seismic data. The predicted surface waves are then subtracted from the original data.

A method for processing microseismic data, comprises: receiving the microseismic data acquired by one or more multicomponent sensors; convolving the microseismic data with an operator that is applied to all of the components of the microseismic data; and applying a multicomponent filter operator to the convolved microseismic data. The microseismic data may result from human activity or be entirely natural. The filtering preserves the polarity of the received data while improving the signal-to-noise ratio of the filtered data.

Techniques are disclosed relating to deblending of sources in multi-source geophysical survey data, including marine or land-based data. Recorded data may be aligned to a primary source. A deblending procedure may be iteratively applied to produce a residual term and deblended estimates for the primary source and one or more secondary sources. Following an iteration of the deblending procedure, the resultant data may be sorted according to a domain that renders the one or more secondary sources incoherent with respect to the primary source. The domain used for sorting may be different from a domain used to sort during an immediately prior iteration. In embodiments, the deblending procedure may use coherency filtering, and the coherency filtering may be weighted according to a signal-to-noise metric generated from the data being deblended.

Waves from cement bond logging with a sonic logging-while-drilling tool (LWD-CBL) are often contaminated with tool waves and may yield biased CBL amplitudes. The disclosed LWD-CBL wave processing corrects the first echo amplitudes of LWD-CBL before calculating the BI. The LWD-CBL wave processing calculates a tool wave amplitude and a phase angle difference as the difference of the phases between the tool waves and casing waves. The tool waves are then used to correct the LWD-CBL casing wave amplitude and remove errors introduced from tool waves. In conjunction with the sets of operations described, the LWD-CBL wave processing also include array preprocessing operations. Array preprocessing may employ variation of bandpass filtering and frequency-wavenumber (F-K) filtering operations to suppress tool wave.

Processes and systems described herein are directed to performing marine surveys with marine vibrators that emit orthogonal coded pseudo-random sweeps. In one aspect, coded pseudo-random signals are generated based on coded pseudo-random sequences. The coded pseudo-random sequences are used to activate the marine vibrators in a body of water above a subterranean formation. The activated marine vibrators generate orthogonal coded pseudo-random sweeps. A wavefield emitted from the subterranean formation in response to the orthogonal coded pseudo-random sweeps is detected at receivers located in a body of water. Seismic signals generated by the receivers may be cross-correlated with a signature of one of the orthogonal coded pseudo-random sweeps to obtain seismic data with incoherent residual noise.

A method includes receiving observed seismic data, determining an envelope or magnitude of the observed seismic data as a first observed value, generating a variable noise term based in part upon the first observed value, and utilizing the variable noise term to determine a likelihood function of a stochastic inversion operation. The method also includes utilizing the likelihood function to generate a posterior probability distribution in conjunction with the stochastic inversion operation and applying the posterior probability distribution to characterize a subsurface region of Earth.

A system for attenuating coherent noise from seismic data comprises one or more sensors configured to sense waves generated by a seismic source and a coherent noise attenuation module communicably coupled to the one or more sensors and comprising a processor and memory. The coherent noise attenuation module is operable to receive a plurality of traces of seismic data from the one or more sensors and apply a first transformation to the plurality of traces, identify one or more outlier waveforms in the transformed traces, attenuate the identified outlier waveforms, and apply a second transformation to the plurality of traces that is the inverse of the first transformation.