JOURNAL OF GEOPHYSICS AND ENGINEERING 2013, vol. 10, article no.

065002 (10pp)
doi:10.1088/1742-2132/10/6/065002

High-resolution Seismic
Processing by Gabor
Deconvolution
Zengbao Chen1, 2, Yanghua Wang 2, Xiaohong Chen1 and Jingye Li1

1National Engineering Laboratory for Offshore Oil Exploration, China University of

Petroleum, Beijing, 102249, China.
2Centre for Reservoir Geophysics, Department of Earth Science and Engineering, Imperial

College London, SW7 2BP, UK.

TRESNA HANJANI KULSUM

22316003
Magister Teknik Geofisika
Fakultas Teknologi Pertambangan dan Perminyakan
Instititut Teknologi Bandung
 OUTLINE

INTRODUCTION

A NONSTATIONARY TRACE MODEL

GABOR DECONVOLUTION

SMOOTHING

APPLICATION EXAMPLE

CONCLUSION
 INTRODUCTION
• High-resolution seismic data are pre-requisite for reservoir
characterization.
• To remove the attenuation effect, inverse Q filtering has been designed
as a type of deconvolution.
• As an alternative, Margrave (1998) presented a non-stationary
convolution model which addresses the earth’s attenuation. Then
Margrave and Lamoureux (2002) developed a non-stationary
deconvolution using a Gabor transform.
 A NONSTATIONARY TRACE MODEL
• The stationary convolution model of a seismic trace, 𝑆𝑠𝑡𝑎𝑡 (𝑡), is
expressed in the time domain as :
∞

𝑆𝑠𝑡𝑎𝑡 𝑡 = න 𝑤 𝑡 − 𝜏 𝑟 𝜏 𝑑𝜏, (1
−∞

𝑤 Seismic Wavelet
𝑟 Time domain Reflectivity series

Figure 1. A depiction of stationary

convolution: a Toeplitz matrix
formed from a source wavelet
multiplying a vector of reflectivity
series. The resultant trace is a
stationary superposition of a source
wavelet scaled by reflectivity
samples at different times.
The non-stationary wavelet is a propagating wavelet, which includes the source
signature and the non-stationary effect of dissipation described by a Q model. The
impulsive effect of the attenuation can be modeled by:
∞ ∞

𝑆𝑄 𝑡 = න න 𝛼 𝜏, 𝑓 𝑟 𝜏 𝑒 2𝜋𝑖𝑓(𝑡−𝜏) 𝑑𝑓𝑑𝜏 (2
−∞ −∞

Inner integral
∞
, so we obtained
𝑎 𝜏, 𝑡 − 𝜏 = න 𝛼(𝜏, 𝑓)𝑒 2𝜋𝑖𝑓(𝑡−𝜏) 𝑑𝑓 (3
−∞
∞ , In the matrix form, we can written as below:
𝑆𝑄 𝑡 = න 𝛼 𝜏, 𝑡 − 𝜏 𝑟(𝜏)𝑑𝜏 (4
−∞

𝑆𝑄 = 𝐴𝑟 (5

𝑆𝑄 𝑡 The non-stationary
𝑟 Time domain Reflectivity series
𝐴 Non-Toeplitz matrix
𝛼 𝜏, 𝑓 The constant-Q attenuation function
Applying a general source signature with a stationary convolution to 𝑆𝑄 ,
seismic trace can be written as below:
𝑠 = 𝑊𝐴𝑟 (6
If 𝑊𝑄 = 𝑊𝐴 a seismic trace is the product of this nonstationary matrix 𝑊𝑄
with the reflectivity vector 𝑟:
𝑠 = 𝑊𝑄 𝑟 (7
In the non-absorptive limit 𝑄 → ∞, (on Figure 2).

Figure 2. A depiction of non-stationary convolution. In the matrix-vector product, the matrix

does not possess Toeplitz symmetry. Each column of the matrix contains the source
waveform modified by the attenuation process for a travel time equal to the column time.
 A NONSTATIONARY TRACE MODEL

The constant-𝑄 attenution function (Kjartansson, 1979) is

−𝜋𝑓𝜏 𝜋𝑓𝜏
𝛼 𝜏, 𝑓 = 𝑒𝑥𝑝 + 𝑖𝐻 (8
𝑄(𝜏) 𝑄(𝜏)
For a layered medium, where Q takes a different value in each layer, the
average Q is defined (Wang, 2004) by
𝑛
1 1 ∆𝜏𝑘
= ෍ (9
𝑄(𝜏) 𝜏 𝑄𝑘
𝑘=1

𝑄 Variable of time 𝜏
𝐻(. ) Hilbert transform
∆𝜏𝑘 Interval time
𝑄𝑘 Interval 𝑄
𝑘 Th-layer
𝜏 Total travel time
 GABOR DECONVOLUTION
Smoothing and
division
Gabor Spectrum

Non-Stationary Forward Gabor Gabor

Signal Transform Deconvolution

Spectrum
Deconvolution is one of the seismic data Reflectivity in freq-
processing method to enhancement vertical time domain
resolution with compressing the seismic
wavelet. It is part of sesmic processing for Invers Gabor
sparating seimic trace into two components Transform
both the seismic waveform and the
reflectivity.
Reflectivity in
time domain
 GABOR DECONVOLUTION
∞

𝑆𝑔 𝜏, 𝑓 = න 𝑠 𝑡 ℎ(𝑡 − 𝜏)𝑒 −2𝜋𝑖𝑓𝑡 𝑑𝑡 (10

−∞

1 (𝑡−𝜏)2
−
ℎ 𝑡−𝜏 = 𝑒 𝑇2 (11
𝑇 𝜋

𝑆𝑔 𝜏, 𝑓 The forward Gabor transform of a

signal
ℎ(𝑡) The Gabor analysis window

𝑓 Frequency

𝜏 The location 𝑓 the window centre

𝑇 Referred to as the half width of the

Gaussian window
Figure 3. Transformasi Gabor
(http://ensiklopediseismik.blogspot.co.id/2008
/02/transformasi-gabor.html)
 GABOR DECONVOLUTION
∞ ∞

𝑡 = 𝛾 න න 𝑆𝑔 (𝜏, 𝑓)𝑒 −2𝜋𝑖𝑓𝑡 𝑑𝑓𝑑𝜏 (12

−∞ −∞

∞ −1

𝛾 𝑡 = න ℎ 𝑡 − 𝜏 𝑑𝜏 (13
−∞

𝑆 𝑡 The Inverse Gabor transform of a

signal
ℎ(𝑡) The Gabor analysis window

𝑓 Frequency

𝜏 The location 𝑓 the window centre

𝑇 Referred to as the half width of the

Gaussian window
𝛾 𝑡 the Gabor synthesis window

Figure 4. Transformasi Gabor

(http://ensiklopediseismik.blogspot.co.id/2008/02/transformasi-gabor.html)
Figure 5. The result of a forward and inverse Gabor transform is compared with
the original signal. The differences are small enough for Gabor deconvolution.
 GABOR DECONVOLUTION
Performing Gabor transform to the non-stationary trace model (equation
6) produces Gabor spectrum as (the Forward Gabor transform):
𝑆𝑔 (𝜏, 𝑓) ≈ 𝑊 𝑓 𝐴(𝜏, 𝑓)𝑅𝑔 (𝜏, 𝑓) (14

𝑊 𝑓 The fourier transform of the source wavelet

𝐴(𝜏, 𝑓) The attenuation function

𝑅𝑔 (𝜏, 𝑓) The forward Gabor transform of the reflectivity series

𝐷(𝜏, 𝑓) The Gabor deconvolution operator presented in time-frequency

domain
Gabor Deconvolution (time-frequency domain) is written as:
𝑅𝑔 𝜏, 𝑓 = 𝑆𝑔 𝜏, 𝑓 𝐷(𝜏, 𝑓) (15
And by comparing for
𝐷 𝜏, 𝑓 = 𝑊 𝑓 𝐴(𝜏, 𝑓) −1 (16
 GABOR DECONVOLUTION
𝐷 𝜏, 𝑓 can be determined as below:
• First, preforming a smoothing process over the Gabor spectrum of the non-stationary
seismic trace to estimate the magnitude spectrum, 𝑊𝑄 𝑒𝑠𝑡 of the propagating wavelet.
• By assuming both the source wavelet and the attenuation process be minimum phased,
the phase spectrum ∅ 𝜏, 𝑓 can be determined by the Hilbert transform that followed
below equation:
∞
ln( 𝑊𝑄 𝑒𝑠𝑡 + 𝜇 𝑊𝑄 𝑚𝑎𝑥 )
∅ 𝜏, 𝑓 = න ′
𝑑𝑓 ′ (17
𝑓−𝑓
−∞
The Gabor deconvolution operator is given as (Margrave et al., 2011)
1
𝐷 𝜏, 𝑓 = 𝑒 −𝑖∅(𝜏,𝑓) (18
𝑊𝑄 𝑒𝑠𝑡 + 𝜇 𝑊𝑄 𝑚𝑎𝑥
 GABOR DECONVOLUTION

Reflectivity


Data

Figure 6. Ilustration for Gabor deconvolution

(Margrave, 2010)
Model of attenuation and
source sig.
 Figure 9. A comparison between the real
wavelet spectrum (red curve) and the
smoothed spectrum(blue curve) at time 1
s, which is marked as a dashed yellow line
in figure 8(b).

Figure 7. A depiction of the Gabor deconvolution algorithm. In each part, the image is a Gabor magnitude
spectrum: (a) An attenuated seismic trace. (b) The propagating wavelet estimated by smoothing the spectrum (a).
(c) Estimated spectrum of the reflectivity series and obtained from Gabor deconvolution. (d) The actual spectrum
of the reflectivity series.
 Smoothing
In the Gabor transform domain, the spectrum of the reflectivity series is much
more rapidly varying than that of a propagating wavelet. Hence, the spectrum of
the propagating wavelet can be estimated by smoothing over the spectrum of a
seismic trace.
The Gabor deconvolution operator can be determined alternatively by a
regularized smoothing method. That is, smoothing is implemented as Tikhonov’s
regularization. The forward operator is set simply to be the identity matrix. The
least-squares solution has the form (Fomel, 2007)
𝑚 = (𝐼 + 𝜀 2 𝐷𝑇 𝐷)−1 𝑑 (19

𝑚 Smooth model
𝑑 Data
𝐷 Regularization operator
𝜀 the scaling parameter
 Application examples

Figure 8. Constant-Q synthetic data example. (a) Band-limited reflectivity and a minimum phase source wavelet
(red) overlapped. (b) Non-stationary trace generated by convolving the reflectivity with the source wavelet, and
then applying a forward Q filter (Q = 30). (c) Stationary trace generated by only convolving the reflectivity with the
source wavelet. (d) The result of inverse Q filtering by Gabor transform. (e) The result of Gabor deconvolution using
hyperbolic smoothing. ( f ) The result of Gabor deconvolution using regularized smoothing.
Figure 9. A comparison between the results of hyperbolic smoothing (a) and regularized
smoothing(b), applied to the Gabor magnitude spectrum of the nonstationary trace
shown in figure 6(b).
Figure 10. Variable-Q synthetic data example. (a) Band-limited reflectivity and a minimum phase source wavelet (red) overlapped
shown the same as in figure 8(b). (b) Nonstationary trace generated by convolving the reflectivity with the source wavelet, and then
applying a forward Q filter determined by the variable-Q model in table 1. (c) Stationary trace generated by only convolving the
reflectivity with the source wavelet shown the same as in figures 8(c). (d) The result of inverse Q filtering by Gabor transform. (e) The
result of Gabor deconvolution using hyperbolic smoothing. ( f ) The result of Gabor deconvolution using regularized smoothing.
Figure 11. A comparison between the results of hyperbolic smoothing (a) and regularized
smoothing (b), applying to the Gabor magnitude spectrum of the nonstationary trace shown in
figure 8(b).
 Figure 12. Gabor deconvolution applied to prestack
seismic data. (b) The result of Gabor deconvolution using
hyperbolic smoothing. (c) The result of Gabor
deconvolution using regularized smoothing.

Figure 13. A comparison among the amplitude spectra of

the same trace shown yellow in Figure 10 before (red)
and after applying Gabor deconvolution using hyperbolic
smoothing (green) and regularized smoothing (blue).
Figure 14. Gabor deconvolution applied to
poststack seismic data. (a) Migration section
before Gabor deconvolution. (b) The result of
Gabor deconvolution using hyperbolic smoothing.
(c) The result of Gabor deconvolution using
regularized smoothing. All the figures are plotted
in the same scale.
Figure 15. Zoomed-in sections of the rectangular
regions in figure 12. (a) Migration section before
Gabor deconvolution. (b) The result of Gabor
deconvolution using hyperbolic smoothing. (c)
The result of Gabor deconvolution using
regularized smoothing. The ovals indicate the
areas of comparison.
Figure 16. A comparison among the average amplitude spectra of the seismic
sections shown in figure 9 before (red) and after applying Gabor deconvolution
using hyperbolic smoothing (green) and regularized smoothing (blue).
 Conclusions
• The Gabor deconvolution is a fully data-driven method for
reflectivity estimation. It yields a high-resolution estimate
of the reflectivity, even for a strong attenuation, but
without knowing or estimating the Q values.
• Compared with the hyperbolic smoothing method, the
regularized smoothing method can be an alternative or
even better to determine the Gabor deconvolution
operator.