|
<< Click to Display Table of Contents >> Navigation: »No topics above this level« Filters |
The Filters category groups together all trace-level and gather-level signal conditioning modules in g-Platform. These tools modify the frequency content, phase, or amplitude of seismic traces without changing the geometry or time-domain structure of the data. They are applied at various stages of a seismic processing flow, from initial noise suppression and wavelet shaping after demultiplexing, through to final spectral balancing before stack or inversion.
Filters in this category operate in the frequency domain, the time domain, or a hybrid of both, depending on the specific module. Many are designed to preserve the broadband signal while attenuating noise or shaping the wavelet to a desired form. Some modules alter the phase of the data (for example, zero-phase conversion and minimum-phase conversion), while others work purely on the amplitude spectrum (band-pass filters, notch filters, clip). A number of modules apply adaptive or data-driven approaches (Cadzow, SVD, Orthogonal Prediction, Decision Based Median) to separate coherent signal from random or harmonic noise.
These modules restrict the frequency bandwidth of the data, removing low-frequency and high-frequency noise while retaining the usable seismic signal.
Band-pass Butterworth applies a Butterworth filter with maximally flat amplitude response within the pass band. The roll-off steepness is controlled by the filter order. Use this module when a smooth, ripple-free frequency response is required, for example before AVO analysis.
Band-pass filter applies a trapezoidal (Ormsby-style) band-pass filter defined by four corner frequencies. The transition bands between pass and reject regions have a linear slope in the frequency domain. This is the most commonly used band-pass module in everyday processing.
Band-pass filter (Distributed) is the same trapezoidal band-pass algorithm prepared for distributed (multi-node) processing of large 3D datasets.
Time-varying band-pass filter applies a band-pass filter whose corner frequencies change with two-way travel time. This is particularly useful for land data where the dominant frequency decreases with depth due to absorption, allowing tighter high-cut corners at shallow times and more relaxed corners at deeper times without global compromise.
Notch removes a narrow band of frequencies centered on a specific frequency. Use this to suppress powerline interference (50 Hz or 60 Hz harmonics) or other monochromatic noise sources.
Notch ext is the extended notch filter supporting multiple notch bands and additional control over the notch width and shape, enabling simultaneous removal of several harmonic interference frequencies.
These modules alter the phase character of the seismic wavelet, converting between minimum-phase, zero-phase, and mixed-phase wavelets, or reshaping the data to match a reference.
Zero phase conversion converts data from a mixed- or minimum-phase wavelet to a zero-phase wavelet by applying an all-pass phase rotation. Zero-phase data is symmetric around reflection events, which is the preferred form for interpretation and inversion.
Min Phase Conversion converts data to minimum-phase character. Minimum-phase wavelets concentrate their energy at the earliest possible time, which is a prerequisite for predictive deconvolution and some NMO-based processes.
Matched filter - Calculate computes a Wiener shaping filter that transforms one seismic dataset so that its wavelet matches the wavelet of a reference dataset. The two-step workflow (Calculate then Apply) is used for cross-equalization in 4D seismic processing and for matching vintages or datasets from different acquisition systems.
Matched filter - Apply applies the shaping filter computed by the Matched filter - Calculate module. The filter is convolved with each trace to bring the wavelet character of the current dataset into agreement with the reference.
Wavelet detection estimates the embedded seismic wavelet from the data, either statistically or using well ties. The estimated wavelet can be inspected for quality control and used as input to wavelet shaping operators or inversion workflows.
Cepstrum Deconvolution estimates and removes the embedded wavelet using cepstral analysis in the log-frequency domain. It separates the slowly-varying wavelet spectrum from the rapidly-varying reflectivity spectrum without requiring an explicit autocorrelation window, making it a useful alternative to Wiener deconvolution when the wavelet is non-stationary.
These modules control trace amplitude without spectral shaping, used for clipping protection, amplitude conditioning, and DC level correction.
Clip limits all trace amplitude values to a specified maximum level, preventing extremely large spikes from distorting downstream processing steps such as deconvolution or stack normalization.
Clip selector extends the basic Clip module with additional options for selecting which traces or time zones are subject to clipping, allowing spatially or temporally adaptive clipping strategies.
DC bias removal removes a static DC offset or low-frequency trend from each trace independently. Three modes are available: constant mean removal (using sine-weighted averaging), linear trend removal, and non-linear polynomial removal. It is typically applied to raw field data before any other processing to ensure zero mean traces.
These modules detect and suppress incoherent or coherent noise, including random spikes, ground roll, and organized harmonic noise. They range from simple median-based spike detectors to advanced matrix decomposition methods.
Cadzow De-Noise Filter reduces random noise by constructing a Hankel matrix from the data in a sliding window and applying singular value decomposition (SVD), retaining only the largest singular values that correspond to coherent signal. It is effective on prestack CMP or shot gathers with moderate random noise levels.
SVD De-Noise Filter applies singular value decomposition directly to a 2D window of traces, projecting the data onto the leading singular vectors to suppress incoherent energy. This is suitable for post-stack or pre-stack data where the dominant reflection energy is well organized.
Diff filter applies a second-order discrete differentiator, which emphasizes high-frequency energy and sharpens reflections. It is equivalent to multiplying the amplitude spectrum by the square of frequency, and can be used as a simple high-frequency enhancement prior to interpretation.
Decision Based Median De-Noise Filter detects and replaces spike noise using a 2D sliding-window median test. A sample is flagged as a spike if its amplitude exceeds the local median by more than a user-specified threshold factor, and is then replaced by the median value or zeroed. It is fast and effective for impulsive noise such as that caused by nearby explosions or electrical interference.
Despike GCube performs spectral despiking on a 3D seismic volume. It operates directly on SEG-Y files in a batch mode and is designed for large post-stack datasets where the Despike module cannot be used in a standard gather-by-gather flow.
Orthogonal Prediction Random Noise Attenuation suppresses random noise by predicting the signal component using orthogonal polynomial regression in a 2D time-offset window and subtracting the residual noise. It is suited to pre-stack gathers where the reflection signal follows a predictable moveout pattern.
Fast-Wavelet applies a fast wavelet transform-based noise attenuation. The data is decomposed into wavelet subbands; coefficients below a noise threshold are zeroed before reconstruction, resulting in a smoothly denoised output.
Spike Decon is a spiking deconvolution module that compresses the seismic wavelet towards a spike by computing and applying the inverse of the autocorrelation-based Wiener filter. It is used to maximize temporal resolution before band-pass filtering.
FK and spatial filters operate on the combined frequency-wavenumber content of the data to suppress spatially coherent noise that cannot be removed by frequency filtering alone.
FK filter applies a fan filter in the frequency-wavenumber domain to suppress dipping noise such as ground roll, air waves, or guided waves that plot in a distinct linear fan of dips in the F-K spectrum. A polygon or velocity-based fan can be defined to reject the noise zone.
FK 3D PostStack extends the FK filter to post-stack 3D volumes, operating simultaneously in the inline and crossline wavenumber directions to attenuate acquisition footprint and other spatially periodic noise patterns.
Several modules in the Filters category are specifically intended for post-imaging conditioning of migrated volumes.
RTM Imaging runs a Reverse Time Migration imaging condition, propagating wavefields forward and backward in time through a velocity model and computing the cross-correlation imaging condition. RTM is the highest-fidelity depth migration method, handling turning waves and steep dips that Kirchhoff migration cannot image correctly.
RTM imaging post processing applies conditioning filters to RTM output, including low-pass spatial filtering and amplitude normalization, to suppress low-frequency artifacts and migration noise that are characteristic of the RTM imaging condition.
Impedance Post PSDM processing conditions depth-migrated angle or offset gathers for acoustic impedance inversion, applying wavelet-consistent scaling and phase correction so that the migrated gathers are consistent with the convolutional model required by inversion.
Volume Loop iterates a selected filter module over an entire 3D seismic volume stored on disk, providing a mechanism to apply any gather-level filter at volume scale without loading the full dataset into memory at once.
GWellLogViewer displays well log curves alongside seismic data in a shared time or depth axis. It is used during wavelet estimation and filter design to verify that the processed seismic is consistent with the subsurface model defined by well data.