de convolution
DESCRIPTION
Deconvolution attempts to undo unwanted convolution processesapplied to the seismic wavelet as it passes through the seismic systemUltimate purpose:to extract the reflectivity seriesDeconvolution improves temporal resolution by compressing thewaveletThis can include attenuation of multiplesTRANSCRIPT
1
Presentation title - Page 1 Geoscience Training Centre
Cefoga
BASIC SEISMIC PROCESSING BASIC SEISMIC PROCESSING
DECONVOLUTION DECONVOLUTION
PART 06PART 06
Presentation title - Page 2 Geoscience Training Centre
Cefoga
Purpose of DeconvolutionPurpose of Deconvolution
Deconvolution attempts to undo unwanted convolution processes
applied to the seismic wavelet as it passes through the seismic system
Ultimate purpose:
to extract the reflectivity seriesto extract the reflectivity series
Deconvolution improves temporal resolution by compressing the
wavelet
This can include attenuation of multiplesThis can include attenuation of multiples
2
Presentation title - Page 3 Geoscience Training Centre
Cefoga
Convolutional Model of the EarthConvolutional Model of the Earth
Deconvolution makes some assumptions about the seismic system:
Vertically propagating downVertically propagating down--going plane wavegoing plane wave
earth made of horizontal layers, constant Vearth made of horizontal layers, constant V
compression (P) wave at normal incidence, no shear wavescompression (P) wave at normal incidence, no shear waves
Source signature does not change: stationarity Source signature does not change: stationarity
Despite the fact that non of these criteria are not satisfied,
deconvolution is a remarkably robust process
Presentation title - Page 4 Geoscience Training Centre
Cefoga
Reflection coefficient Wavelet Result
Principle of Superposition Principle of Superposition -- 11
Reflection coefficient for a single boundary represented by a spike:size indicates degree of impedance contrastsize indicates degree of impedance contrastcan be positive or negativecan be positive or negative
Source wavelet replicates itself:
3
Presentation title - Page 5 Geoscience Training Centre
Cefoga
Principle of Superposition Principle of Superposition -- 22
Earth reflectivity series is a series of reflection coefficients
Convolution of reflection coefficients with wavelet Convolution of reflection coefficients with wavelet
*
=
Presentation title - Page 6 Geoscience Training Centre
Cefoga
ResolutionResolution
To properly identify all boundaries wavelet must be ‘removed’
Reverse the effect of the convolutionReverse the effect of the convolution
Inverse FilterInverse Filter
DECONVOLUTIONDECONVOLUTION
3 boundaries How many boundaries
If reflections are closer (in time) than the wavelet length then
interference occurs and these events cannot be uniquely identified
4
Presentation title - Page 7 Geoscience Training Centre
Cefoga
Convolution in the Time Domain ...Convolution in the Time Domain ...
The convolution of the reflectivity series with the source wavelet can be
expressed as:
x(t) = w(t) * e(t)x(t) = w(t) * e(t)
where:where:
xx is recorded traceis recorded trace
ee is reflectivity time seriesis reflectivity time series
ww is seismic wavelet (source signature in this model)is seismic wavelet (source signature in this model)
noise - free model
Presentation title - Page 8 Geoscience Training Centre
Cefoga
… Multiplication in the Frequency Domain… Multiplication in the Frequency Domain
assume reflectivity is randomamplitude spectra multiplied
Amplitude spectrumof wavelet
X
Amplitude spectrum of reflectivity
(white)
Amplitude spectrumof trace
=
5
Presentation title - Page 9 Geoscience Training Centre
Cefoga
Inverse FilteringInverse Filtering
From
xx((tt) = ) = ww((tt) * ) * ee((tt))
to remove w (t)
derive operator derive operator gg((tt) such that ) such that gg((tt) * ) * ww((tt) = spike) = spike
gg((tt) is mathematical inverse of ) is mathematical inverse of ww((tt) = ) = w w --11((tt))
Inverse filterInverse filter = = w w --11((tt))
Presentation title - Page 10 Geoscience Training Centre
Cefoga
DeconvolutionDeconvolution TypesTypes
Deconvolution falls into two main categories:
Statistical Statistical DeconvolutionDeconvolution• Operators are derived statistically from the seismic data
Deterministic Deterministic DeconvolutionDeconvolution• Operators are derived from known or modelled functions
6
Presentation title - Page 11 Geoscience Training Centre
Cefoga
Statistical Deconvolution: Application CasesStatistical Deconvolution: Application Cases
Wavelet estimated directly from the seismic data trace(s)
zerozero--lag spike ( ‘spiking deconvolution’ )lag spike ( ‘spiking deconvolution’ )
spike at arbitrary lagspike at arbitrary lag
time advanced form of the input (‘predictive’)time advanced form of the input (‘predictive’)
zerozero--phasephase
any desired arbitrary shape (‘shaping’)any desired arbitrary shape (‘shaping’)
Presentation title - Page 12 Geoscience Training Centre
Cefoga
Spiking Deconvolution Spiking Deconvolution -- 11
White Noise (pre-whitening)
constant added to zeroconstant added to zero--lag of autocorrelationlag of autocorrelation
prevents amplification of noise at edges of frequency spectrumprevents amplification of noise at edges of frequency spectrum
Autocorrelation of trace is that of the wavelet
Assumes stationarity
7
Presentation title - Page 13 Geoscience Training Centre
Cefoga
Spiking Deconvolution Spiking Deconvolution -- 22
X
=Wavelet Operator
Spiked output
prewhitening
amplitude
frequency
amplitude
frequency
amplitude
frequency
Presentation title - Page 14 Geoscience Training Centre
Cefoga
PrePre--Whitening Whitening -- 11
Amplitude Spectrum - No White Noise
0
1
2
3
4
5
6
7
8
9
10
0 25 50 75 100 125
Inverse of Amplitude Spectrum - No White Noise
0
2
4
6
8
10
12
0 25 50 75 100 125
Amplitude Spectrum with White Noise
0
2
4
6
8
10
12
0 25 50 75 100 125
Inverse of Amplitude Spectrum with White Noise
0
1
2
3
4
5
6
7
8
9
10
0 25 50 75 100 125
8
Presentation title - Page 15 Geoscience Training Centre
Cefoga
Effect of White Noise in Spiking Effect of White Noise in Spiking DeconvolutionDeconvolution
Added white noise
(% of zero lag)
0.1
No deconvolution
1.0
50.0
100.0
(Operator length 80ms)
Presentation title - Page 16 Geoscience Training Centre
Cefoga
Effect of Operator Length in Spiking Effect of Operator Length in Spiking DeconvolutionDeconvolution
40
80
160
No deconvolution
(Prewhitening 1%)
Total Operator Length
(ms)
9
Presentation title - Page 17 Geoscience Training Centre
Cefoga
Operator Length Operator Length -- 11
Inspect autocorrelation of input trace
Use first transient zone
Input Wavelet Autocorrelation
Presentation title - Page 18 Geoscience Training Centre
Cefoga
Predictive Deconvolution Predictive Deconvolution -- Principles Principles -- 11
Predictive deconvolution presupposes that if an event can be predicted, it can not be a primary - the reflection series we have already defined as being random, i.e. non-predictable
aim to remove the predictable components
Unpredictable:Reflection Series
t 2t3t
4t
Predictable:Reverberations
10
Presentation title - Page 19 Geoscience Training Centre
Cefoga
Predictive Deconvolution Predictive Deconvolution -- Principles Principles -- 22
Two methods:
PREDICTION filterPREDICTION filter• Predicts time advanced form of input series• Subtraction from input
PREDICTION ERROR filterPREDICTION ERROR filter• Convolve with input
Presentation title - Page 20 Geoscience Training Centre
Cefoga
Effect of Gap in Effect of Gap in PredicitvePredicitve DeconvolutionDeconvolutionTotal Operator Length
(ms)
Gap
(ms)
88 8
96 16
104 24
120 40
No deconvolution
(Prewhitening 1%)
80 0(4)
11
Presentation title - Page 21 Geoscience Training Centre
Cefoga
MultiMulti--channel Deconvolution channel Deconvolution -- 11
Calculate an operator from a group of traces
Preserves continuityPreserves continuity
Avoids deconvolving the geologyAvoids deconvolving the geology
signal well estimated but the noise is not well attenuatedsignal well estimated but the noise is not well attenuated
Presentation title - Page 22 Geoscience Training Centre
Cefoga
DECMC ExampleDECMC Example
spiking DECON
multi -channel DECON
12
Presentation title - Page 23 Geoscience Training Centre
Cefoga
Frequency Domain Frequency Domain DeconvolutionDeconvolution
Input Trace
Compute Autocorrelation
Compute W-L Operator
Fourier TransformAmplitude Spectrum Phase Spectrum
Amplitude Spectrum Phase Spectrum
Multiply AddInverse Fourier
Transform to get output trace
Fourier Transform
Presentation title - Page 24 Geoscience Training Centre
Cefoga
Zero Phase Zero Phase DeconvolutionDeconvolution
Input Trace
Compute Autocorrelation
Compute W-L Operator
Fourier TransformAmplitude Spectrum Set Phase = 0
Amplitude Spectrum Phase Spectrum
Multiply AddInverse Fourier
Transform to get output trace
Fourier Transform