NAME
stel - module to stabilize seismic traces by reducing varia-
tion in the amplitude spectrum caused by line-spectral
noise.
SYNOPSIS
stel [ -Nntap ] [ -Ootap ] [ -opiopt ] [ -Bibeta ] [ -Tntarg
] [ -usp ] [ -nfnf ] [ -nlnl ] [ -ttjtapt ] [ -Kktm ] [ -P ]
[ -R ] [ -V ] [ -? ]
DESCRIPTION
stel is essentially the same as an earlier program by Don
Wagner (also called STEL) that was used mainly to modify or
whiten amplitude spectra of seismic traces. Our motivations
and justifications are the essential differences, e.g. to
stabilize seismic traces by reducing variation in the ampli-
tude spectrum caused by line-spectral noise (typically gen-
erated by varying geometry and elastic parameters of high-
contrast near-surface layers). The stel process (described
next) may still be used as before. The basic process con-
sists of multiplication of the complex spectrum of a trace
by a real (not complex) valued function of frequency. The
real valued function is generated from the difference
between the trace and target spectra in such a way that
after multiplication the trace spectrum will better resemble
the target's. Stel is therefore mathematically equivalent to
a time domain convolution. We are stealing an amplitude
spectrum (by way of the real valued function of frequency).
Thus the original selection of name stel. The phase of
Fourier trace components is invariant under this process. So
perhaps this is only a borrowing, not a steal. Honor and
appearances will be (mostly) preserved. If you did not
notice crude puns in the preceding, please read more care-
fully!
The process can be justified in cases where the data set
contains both some good (high signal-to-noise) traces and
other traces that also contain line-spectral noise caused by
the near surface, by coherent ambient noise or by hardware
or processing side effects Empirical observations, modeling
and recent theoretical work by Joe Pinter confirm the line-
spectral nature of the near surface coherent noise and the
resulting high data variance (instability). Note that this
problem also occurs in shallow marine data. Other noise
events such as migration smiles can also be attacked.
The spectrum of the high quality traces may therefor be used
asa "target" towards which the other traces are modified
without changing the phase of the data (which carries the
information we want to preserve). The process can also make
considerable improvements to data that has trace-to-trace
and shot-to-shot variation due to near surface variation, a
task very close to that for which Don Wagner's original stel
program was written.
The process makes significant improvements in data for two
main reasons: (i) Coherent noise is by far the largest noise
problem we face since the energy in true random noise is
relatively insignificant. Much coherent ambient noise has
near-line-spectral character as does most shot generated
near-surface noise. (ii) The only remaining processing prob-
lem of first order significance for data quality is signal
instability (the variance problem). As well, most high
fidelity techniques require stabilization preprocessing in
order to produce precision results. There is little point in
attempting a deconvolution process that improves the data
five percent if the signal (and noise) is seventy five per-
cent unstable causing the assumptions behind deconvolution
to be wrong, or if the nature of the coherent noise (very
non random) is such that operator design is affected by it
(and its variance) far more than by the signal.
STEL gets its parameters from command line arguments and,
optionally, a target trace from an input file (-T argument).
Usually the target trace is one of the input traces of each
record or is a stack of a user selected range of traces.
Program traf may be used to produce such files from USP
data. The arguments specify the input, output, target
option, and optionally, the percentage reduction of ampli-
tude difference and the start and end trace numbers when
target is a stack. A "restore amplitude" scaling option is
also selectable.
Concerning True Amplitude Processing and Scaling
When line spectral noise is present, one can not produce a
"true amplitude" section due to large and varying amounts of
the source energy which feeds into the coherent noise. For
this reason, the results of stel may approximate "true
amplitude" even though the amplitude of the signal part of
the trace is changed. There may be cases (when the program
is used for other purposes) when the original average ampli-
tude of the spectra should be restored (even though noise
energy has been removed) and in such a case the user can
utilize the -R option. Using -R we multiply the complex
spectrum by the multiplicative inverse of the average (over
frequency) scalar multiplier before the inverse FFT is done.
Results will not be "true amplitude" either since the origi-
nal trace amplitude included noise contributions and lower
(relative to the target trace) signal levels. The best stack
will always result when the signal is most stabilized with
respect to phase, frequency and amplitude and this will usu-
ally occur when the restore option is not used (pre-stack
use of stel). When stel is used post-stack, the use of the
restore option may replace noisy traces with weak (but
higher S/N) traces. When the goal is pure stablization of
amplitude spectra, (data has very low levels of coherent
noise) it may be better to use the retore option. It is
recommended the user test both, as post-stack processing is
fast cheap and easy. A maximum entropy spectral analysis
will reveal much concerning the spectra of the signal and
noise and any strong variations from trace to trace or
record to record. This information will permit a more
rational selection of parameters. Since Fourier methods are
relatively poor for line spectral techniques it is hoped in
future to have some sort of maximum entropy version of stel
which will permit us to deal with the variance problems that
these spectral methods so clearly reveal. Programs linespec
and seelines are available to do this type of analysis.
Simplified Illustration of Process With and Without Restore
(-R)
: BEFORE : +
: Spectrum of : + Spectrum of Input
: Target (good) : + Trace With Coherent
: Trace : + Noise Energy At 3 Hz
: (the "truth") : ++ and 10 Hz
AMP : AMP : ++ + ("true"
: : ++ + amplitude?)
: * : +++ +
: ***** ** : +++ +
:******* ***** : +++ ++
:*************** ** * : +++ ++
:************************ :++++++ ++ ++++
***************************. +++++++++++++++++++++++++
FREQ 15 Hz FREQ 15 Hz
: AFTER :
: Spectrum of (No -R) : Spectrum of Output
: Target (good) : Trace After stel
: Trace : (beta = 90 percent)
: (as above) : Letter b shows cut
AMP : AMP : and fill limits.
: : b ("true" amplitude?)
: * : b++
: ***** ** : b++b b
:******* ***** : b+++bbb bbb++
:*************** ** * :bb+++bbbbbbbbb++ b b
:************************ :++++++bb++bbb++++bbbbbb
***************************. +++++++++++++++++++++++++
FREQ 15 Hz FREQ 15 Hz
: AFTER :
: Spectrum of (With -R) : Spectrum of Trace
: Output With : After stel process
: Restore Option : with beta = 90
: : (as above)
AMP : AMP :
: Restore mult = 0.5 : s Ave mult = 2.0
: ("true" amplitude?) : sss
: : ssss s
: r : sssssss sssss
: rrr rr :ssssssssssssssss s s
:rrrrrrrrr rrrrrrr r :sssssssssssssssssssssss
rrrrrrrrrrrrrrrrrrrrrrrrrrrr sssssssssssssssssssssssss
FREQ 15 Hz FREQ 15 Hz
Concerning Iterated Use (if you must).
Iterated use may be reasonable when (-op1) the target
trace is a stack of several traces or when different
targets are used for each iteration. In the case of the
stacked target, The spectrum of the output will tend to
converge to a result which emphasizes the most trace-
to-trace stable Fourier components (which are usually
the lowest frequency components) of the signal. Follow-
ing this process with an appropriate band pass filter
may show structural relationships previously hidden.
Iterated use requires that the signal in the traces
stacked into the target trace have little or no NMO or
trace-to-trace arrival time differences (or all but the
lowest frequencies of the signal will be reduced). When
iteration is used, it is extra important to be aware of
the effects of structure and statics on stacked target
trace spectra. The user may iterate application by
running stel several time using the last output as the
next input. This permits changes in arguments between
applications. Use of the -K (kaskade) option permits
the repeated application within a single run. In this
case, parameters remain constant for each application.
Use of -K3 for example will cause three iterated appli-
cations. Iterated use with different parameters usually
make more sense than using the -K option.
Concerning Use In Other Domains.
Tests in the Radon domain are currently under way.
Results will be included in updates. Users are asked to
report any interesting new forms of application (or
problems). Our advise and/or opinions are as always
available free. (Contact Dennis Bjerstedt or Don
Wagner.)
Command line arguments
-N ntap
Enter the input data set name or file immediately
after typing -N unless the input is from a pipe in
which case the -N entry must be omitted. This
input file should include the complete path name
if the file resides in a different directory.
-O otap
Enter the output data set name or file immediately
after typing -O. This output file is not required
when piping the output to another process. The
output data set also requires the full path name
(see above).
-T ntarg
Enter the optional (-op2) file name for the tar-
get trace file. By default, this file must be a
text file formatted for eight columns of ten digit
integers. The total number of samples in the file
must be greater than or equal to the number of
samples in the traces of the input data set. This
file may be a usp-formatted dataset consisting of
a single data trace, if the -usp command-line
option is also specified.
-usp This flags the optional (-op2) -T[ntarg] dataset
as being a usp-formatted dataset instead of a text
file. The target trace file will consist of a
standard line header and a single data trace. If
more than one trace is present, the program
ignores them, reading only the first trace.
-B ibeta
This is the factor by which the difference between
the amplitude spectrum (at each frequency) of the
trace and target is reduced (in percent). Thus
-B80 causes a reduction of the difference of
amplitude by 80 percent. Integer values from 1 to
100 (-B1 to -B100) are allowed.
-op iopt ]
Enter the option number which specifies the method
of getting the target trace as follows:
1 Use stack of traces form trace nfirst
to nlast.
2 Use file specified with -ntarg for
trace input.
100+n Use trace n as the target trace. Thus,
-op233 will cause trace 133 (=233-100)
to be used.
-nf nf ]
Enter first trace number to stack to form target
trace.
-nl nl ]
Enter last trace number to stack to form target
trace.
-tt jtapt ]
Enter the number of samples desired in the restore
mute process. Default is 10 samples.
-R Requests that amplitude spectra be scaled
("restored") via multiplication by inverse of
average multiplier found for complex FFT com-
ponents. This compensates overall trace amplitude
for the fact that signal has increased relative to
noise (assuming the data has near surface coherent
noise). Using -R restores (nearly) the trace
amplitude variance of he input. This can be good
or bad according to your goals. See note on "true
amplitude processing" above. With noisy data, true
amplitude processing is either not possible or is
not advisable as it will incorrectly represent the
subsurface geology. In such cases, the only viable
path is to select some form of scaling that is
optimal for the set of compromises you prefer.
This is one of several fundamental differences
between coherent and random noise problems with
important implications for improving both acquisi-
tion and processing.
-K ktm
Enter the number of times to repeat the stel
filter application. The stel parameters remain
the same for all applications. This option will
only be permitted with -op1 (the default stack
option).
-P Enter the command line argument "-P" if you wish
all records in the input passed to the output.
Only records irs to ire inclusive will have the
stel filter process applied. For the rest, output
is the same as input. The filtered and unfiltered
records may now have noticably different ampli-
tudes so consider scaling issues.
-V Enter the command line argument "-V" to get addi-
tional printout.
-? Enter the command line argument "-?" to get on
line help. The program terminates after the help
screen is printed.
BUGS
??????????????????
AUTHORS
Dennis Bjerstedt
COPYRIGHT
copyright 2001, Amoco Production Company
All Rights Reserved
an affiliate of BP America Inc.
Man(1) output converted with
man2html