Deconvolution — g-Platform
g-Platform module

Deconvolution

Geomage's implementation of the surface-consistent modules works in a multi-frequency mode, which allows accurate amplitude estimation, equalization and enhancement across the frequency spectrum. One operator is solved for every surface point — source, receiver, CDP and offset — and the wavelet family beside it covers everything from a stationary Wiener inverse to cepstrum liftering.

Download free trial How the operator is solved Watch Demos
One per surface point
Source, receiver, CDP and offset
Band by band
Filters solved per frequency band
Calculate → apply
Operators stored, then applied
Surface-consistent deconvolution workspace in g-Platform before the operator is applied, with the NMO gather, its autocorrelation, the amplitude spectrum, the receiver amplitude correction map and the solver convergence curves
The same workspace after surface-consistent deconvolution in g-Platform, with a compressed autocorrelation and a broadened amplitude spectrum
g-Platform workspace with amplitude spectra, source and receiver operator panels and gathers before and after amplitude and deconvolution operators
In view
Before deconvolution
A seismic trace is a wavelet convolved with reflectivity — and then everything the survey did to it. The source signature, the coupling under each geophone, the terrain, the offset and the near-surface all leave their own mark, and none of them is the geology you are after. Surface-consistent processing takes that literally: it treats the trace as the product of a shot term, a receiver term, a CDP term and an offset term, averages the log spectra of each, and separates the components with a Gauss-Seidel iteration until they stop changing. What comes out is an operator per surface point — calculated once, stored, and applied as deconvolution alone or together with the amplitude recovery.
Surface-consistent processing

One operator for every trace that shares a surface point.

Two module pairs run on the same principle: surface-consistent deconvolution for the wavelet, surface-consistent amplitude correction for the amplitudes. Each is a Calculate stage that solves the operators and an Apply stage that puts them on the data.

The idea in one line
A single operator is calculated for all the traces that have the same surface point in common — one for the shot gather, a separate one for the receiver gather — rather than a fresh operator for every trace.
Why the data need it
A survey is designed for ideal conditions and acquired across real terrain. Instrument response, near-surface variation and changing subsurface properties distort the recorded wavelet from one surface location to the next.
Four contributions, separated
Each shot, each receiver position, each CDP and each offset contributes its own spectral modification. The solver averages the log spectra of each set and isolates the four components from one another.
Multi-frequency by construction
The filters are computed for each frequency band, with a minimum and a maximum frequency to restore, so the operator equalizes the spectrum band by band instead of applying one average correction across the whole of it.
Iterate to convergence
Gauss-Seidel iterations continue until the difference between successive passes falls under a threshold you set. The solver convergence per frequency band is plotted, so a run that will not settle tells you so.
Calculate, then apply
The Calculate stage writes the operators, which can be saved to file and reloaded later; the Apply stage puts them on the data in the shot, receiver or offset domain — deconvolution only, or deconvolution together with amplitude recovery.
The amplitude twin
Surface-consistent amplitude correction runs the same machinery on amplitudes — scale factors computed per trace and distributed across source, receiver, CMP and offset until they stabilise — and an advanced variant lets you define your own domains and trace headers.
Everything you need to judge it
Common-shot and common-receiver gathers before and after, source and receiver amplitude maps for each frequency band, the frequency response spectrum and the autocorrelation that tells you what operator length the data want.
Wavelet shaping

Compress the wavelet, whatever shape it arrived in.

The wavelet that reaches the recorder is not the one that left the source: attenuation, dispersion and lithology have all reshaped it on the way. Which deconvolution you reach for depends on whether that reshaping is stationary, time-varying or periodic.

Stationary
Decon - Wiener

The classical least-squares inverse. The trace is modelled as the source wavelet convolved with reflectivity plus noise, and the filter is the one that minimises the mean square error between estimated and true reflectivity — broadening the band and sharpening the wavelet.

Non-stationary
Time-variant Wiener Decon

The same reasoning with the stationarity assumption dropped: the wavelet is estimated at each time and the inverse filter varies down the trace, so shallow and deep reflectors are not sharpened by the same operator.

Time-frequency
Decon Gabor

Built for wavelets that are not stable at all. A Gaussian window decomposes the trace into frequency content and its variation over time, and attenuation is part of the model rather than something the operator has to fight.

Periodic events
Cepstrum Deconvolution

In the cepstrum, convolution becomes addition and periodicity collapses to isolated peaks — so short-period multiples, bubble pulses, ghosts and reverberations can be removed with a lifter in the quefrency domain, and the minimum-phase wavelet extracted, where a spiking or predictive filter would struggle.

Marine source
Designature

Airgun arrays fire a peak and then a bubble. Given the far-field signature — measured, derived from near-field hydrophones or modelled — designature builds the operator that removes the residual bubble and converts the wavelet to minimum phase, with the receiver ghost added if wanted.

Demultiple
Predictive (gap) route

Deconvolution doubles as a demultiple technique on land and at sea. The surface-consistent operator can be solved in the time domain as a predictive deconvolution, with a prediction interval and a percentage of white noise for stability, and cepstrum liftering takes the strongly periodic reverberations.

Workflow

Six steps from a distorted wavelet to a balanced gather.

1
Come in clean
The input the Calculate stage expects has already been through the denoise sequence and the statics, so the spectra it averages describe the earth and the acquisition rather than the noise.
2
Mute what sits above the breaks
A mute picking file removes anything ahead of the first arrivals, with a taper so the edge does not become a feature of the operator.
3
Choose the domains
Common source and receiver, plus offset, plus a bin term — and an azimuth domain where the distortion follows direction. Offset minimum, maximum and step bound what takes part.
4
Solve the operators
Gauss-Seidel iterations run to the threshold you set, with the autocorrelation length and a spectral smoothing filter shaping the design. The result can be saved to file and reloaded on the next pass.
5
Apply them
Apply in the same domain you solved in, choosing deconvolution alone or deconvolution with amplitude recovery, and adding white noise as the stabiliser.
6
Read the QC
Shot and receiver gathers before and after, source and receiver amplitude maps per frequency band, convergence curves and the autocorrelation and spectrum that show the wavelet actually compressed.
Operators & scale

Parameters that decide the operator, and a run that finishes.

The Calculate stage is a solver, and it is exposed as one: the domains it decomposes over, the window it designs in, the frequency range it restores and the iteration budget it is allowed — alongside the execution controls a survey-sized job needs.

Operator settings
What shapes the deconvolution filter
Domains
Common source and receiver, optionally with offset and a bin term, plus an azimuth domain with a chosen number of azimuths for direction-dependent distortion.
Solver
A threshold factor that stops the iteration when successive passes stop changing, and an explicit iteration count — with a fast-solving path when the traces are read from disk rather than RAM.
Design window
Autocorrelation length and a smoothing filter over the spectra; the autocorrelation of a gather before and after decon is the display that sets them.
Frequency range
Minimum and maximum frequency to restore, defaulting to the whole band, so the operator can be kept off the parts of the spectrum with no signal in them.
Time or frequency
Solve in the frequency domain, choosing minimum or zero phase, or in the time domain as a predictive deconvolution with a prediction interval.
Stabilisation
A percentage of white noise added to the process, plus a desired amplitude level that the corrected data are brought to.
Running it on a survey
Compute, memory and what comes back out
CPU or GPU
Selected per module, with distributed execution across the processing server and a per-machine bulk size in megabytes.
Cluster controls
Thread limits on the nodes and on the main machine, a job suffix and an affinity tag to identify your workflow in the server QC interface.
Scripts
Your own scripts can run before and after the workflow — switching cloud nodes on and off around a long job, for instance.
Big-data path
Trace headers can be read straight from disk instead of loaded into RAM, and traces are read per chunk, for datasets that will not fit in memory.
Operator files
Calculated operators save to their own file and load back, so a solved survey does not have to be re-solved to be re-applied.
Run information
Counts of sources, receivers, bins and frequency bands, and the frequency step, reported from the run itself.
Modules in this group
The deconvolution and surface-consistent modules named on this page — not the full list
SC Deconvolution - calculate SC Deconvolution - apply SC Amplitude correction - Calculate SC Amplitude correction - Apply SC Amplitude correction advanced Decon - Wiener Time-variant Wiener Decon Decon Gabor Cepstrum Deconvolution Designature
More g-Platform capabilities

Resolution earned here is resolution the image keeps.

The same project, the same batch processing and the same cluster carry a survey from field tape through to depth. These are the other topics in the g-Platform workflow.

g-Platform overview Signal processing & denoise Multiple attenuation Velocity analysis Stack imaging Time imaging Depth imaging FWI & RTM VSP processing Refraction statics Residual statics First-break picking Geometry assignment & QC Data & geometry formats
FAQ

Questions, answered.

What does surface-consistent actually mean?
That a single operator is calculated for every trace sharing the same surface point, rather than one operator per trace. The reasoning is physical: a survey is designed for ideal conditions and then acquired across real terrain, so each shot point, each receiver position, each CDP and each offset imposes its own spectral modification on the recorded trace. Surface-consistent processing assumes exactly that decomposition and reverses it — averaging the log spectra for each shot, each receiver, each CDP and each offset, then isolating the individual components with a Gauss-Seidel iteration.
Which deconvolution suits which wavelet?
Wiener deconvolution assumes a stationary wavelet and designs a least-squares inverse filter that minimises the difference between estimated and true reflectivity. When the wavelet changes with time — which it does, because attenuation is depth-dependent — the time-variant Wiener form designs a filter that varies down the trace, and Gabor deconvolution works in the time-frequency domain with attenuation built into its assumptions. Periodic components such as reverberations, bubble pulses and short-period multiples are better handled in the cepstrum, where periodicity collapses to isolated peaks that a lifter can remove.
Which operator domain should I calculate in?
Common source and common receiver are the base pair: the source domain absorbs what the dynamite, vibroseis or airgun source itself imposed, the receiver domain what receiver sensitivity and local noise imposed. Adding the offset domain accounts for amplitude and spectral variation with offset, and adding the bin term accounts for what varies spatially. An azimuth domain is available as well, for distortions that follow direction, with the number of azimuths under your control.
Can deconvolution attenuate multiples?
Yes — it is a standard demultiple technique on both onshore and offshore data, and predictive or gap deconvolution is the form usually meant. In g-Platform the surface-consistent operator can be solved in the time domain as a predictive deconvolution, with a prediction interval and a percentage of white noise for stability, and cepstrum deconvolution removes strongly periodic reverberations and short-period multiples in the quefrency domain. For water-bottom and interbed multiples in their own right, the demultiple toolkit — SRME, SRMP and the Radon family — is on the multiple attenuation page.
Get started

Solve the operators on your own survey.

Surface-consistent deconvolution and amplitude recovery, band by band, with the convergence and the maps to prove it. Talk to Geomage about a demo, or take g-Platform for a trial run.

Download free trial Request pricing