NAME
dafd - do true amplitude frequency equalization
SYNOPSIS
dafd [ -Nntap ] [ -Ootap ] [ -nsns ] [ -nene ] [ -rsirs ] [
-reire ] [ -flfl ] [ -fhfh ] [[ [ -G ] [ -nfnf ] [ -wlwl ] [
-ffftap ] ] [ [ -exppwr ] [ -fxppwr1 ] [ -fmfm ] [ -E ] [ -P
] ] [ -W [ -wffile ] [ -frfreqs(1), -wtweight(1) ] [ -
frfreqs(2), -wtweight(2) ] [ ... , ... ] ] ] [ -F ] ] [ [
-R ] [ -smsm ] [ -sfsf ] [ -epep ] ]] [ -I ] [ -V ] [ -? ]
DESCRIPTION
dafd performs true amplitude spectral balancing by computing
a frequency boost function using one of four possible modes:
(0) sum of agc'd narrow gaussian bandpass filtered versions
of input data, (1) either exponential with frequency or fre-
quency powered function, (2) linear segments in frequency
defining decibel increase in amplitude, (3) simple flatten-
ing of the amplitude spectrum, applying the same function
(in the frequency domain) to every trace, (4) applying the
rho function (used in slant stacking to preserve frequency
content after stacking). The final time output is the
post-filtered to roll the ends of the boost function.
The response of this type of frequency equalization is
excellent compared with standard equalizers (see e.g. dctvf)
based on gain correcting narrow passbands. Side lobes are
usually much lower, and of course the correction is true
amplitude.
dafd gets processing controls from the command line. Reason-
able defaults are set up.
Command line arguments
-N ntap
Enter the full path of the file containing the data
set. Default is a pipe in.
-O otap
Enter the output file name for the filtered traces.
Default is a pipe out.
-ns ns
(integer) First trace in record to process (default =
1). Traces prior to this in each record will be passed
unprocessed.
-ne ne
(integer) Last trace in record to process (default =
all). Traces after this in each record will be passed
unprocessed.
-rs irs
(integer) First record to process (default = 1).
Records prior to this will be passed unprocessed.
-re ire
(integer) Last record to process (default = all).
Records after this will be passed unprocessed.
-fl, -fhfl,fh
(real) Low cut and high cut frequencies in Hz for
post-boost bandpass filter. These are the 6db points
of the Butterworth filter. The defaults are set dif-
ferently for the 2 modes described below: Mode 1 - the
low cut default = 2Hz, the high cut default = .7
Nyquist; Mode 2 - the low cut default = .( times the
lowest frequency control point, the high cut default =
1.2 times the highest frequency control point.
Mode 0: agc and sum of narrow gaussian bandpass filters
-G If present use the summation of agcd gaussian bandpass
filtered versions of the input data. This method can
not be undone by a subsequent inverse ( -I ) dafd
operation.
-nf nf
Number of frequency bands between -fl[] and -fh[]. The
gaussian filters will overlap at the 6db (half ampli-
tude) points so that summations of all the filters will
yield unit response.
-wl wl
Length of agc window in ms. Each bandpassed version of
the input data will be agcd prior to summation.
default = 500
-ff ff
Optional output file name for gaussian filters. The
format of this file is 2-col: frequency - amplitude
suitable for XY plotting, e.g. usp xgraph [filename]
Mode 1: simple amplitude boost functions
-exp pwr
Exponent: for frequencies greater than fm the exponen-
tial boost function: G(f) = exp [pwr*(f-fm)]; for power
function boost: G(f) = 1 + (f-fm) ** pwr. For frequen-
cies less than fm the gain functions are G(f) = exp
[pwr1*(fm-f)] and G(f) = 1 + (fm-f) ** pwr1. Default =
0
-fxp pwr1
Exponent: for frequencies less than fm the exponen-
tial boost function: G(f) = exp [pwr1*(fm-f)]; for
power function boost: G(f) = 1 + (fm-f) ** pwr1. For
frequencies greater than fm the gain functions are G(f)
= exp [pwr*(f-fm)] and G(f) = 1 + (f-fm) ** pwr.
Default = 0
-fm fm
Exponent: the exponential or power function will be
hung on this frequency, i.e. the boost about this fre-
quency will be determined by the exponents -fxp[]
(f<fm) and exp[] (f>fm). This is useful if the input
spectra have a single predominant peak. Default is 0."
-P If present use use power frequency function [f**pwr]
-E If present use exponential gain frequency function
[exp(pwr*f)]
Mode 2: linear decibel boost defined by freq/weight(db) pairs
-W Must be present on command line to trigger mode 2.
There are two ways to get freq - weight pairs into the
program: (1) in a file, or (2) on th command line.
-wf file
Optional file containing frequency - weight pairs, one
pair per line (whitespace separating the numbers).
-fr freq(1), -fr freq(2), ..., -fr freq(nfs)
Command line method of inputting frequency points
defining line segments over the amplitude spectrum
corresponding to weights (in decibels) to be applied to
the amplitudes (see weights below). There must be one
frequency control point for each weight below. Weights
between frequency control points will be linearly
interpolated.
-wt weight(1), -wt weight(2), ..., -fr weight(nfs)
Weights in decibels corresponding to the frequency
points above. These db values can be estimated by
looking at a specal display of some sample traces.
Mode 2 is particularly suited to correcting amplitude
spectra that decay with increasing frequency away from
some central maximum (i.e. the higher frequency
regime).
Mode 3: simple flattening of amplitude spectrum
-F If present flatten the amplitude spectrum before
filtering between the user specified low cut and high
cut frequencies. This mode can not be undone by the
inverse option -I
Mode 4: rho filter frequency enhancement
-R If present use the rho filter method of frequency
enhancement. This filter was developed for use in slant
stacking (tau-p) as a way of compensating for the
degradation in the higher frequencies cause by the
stacking process.
-sm sm
Envelope smoothing window length. Default = 200. Very
insensitive to output results."
-sf sf
First frequency to use. Default = 0.0. For the inverse
rho filter operation this should be set to 1.0.
-ep ep
Frequency power function. Default = 1.8. The classic
rho filter operation compensates for stacking which is
basically an integrator having a typical 1/freq
response. The compensation for a 1/freq function then
should be a linear function of freq but there may be
other reasons why you might want to increase this value
(get higher resolution).
-I If present use the inverse gain function, i.e. 1/G(f).
Note: the inverse will be unstable for exponents
greater than about 2.
-V If present give verbose printout
-? Query mode. With this flag, dafd will give a descrip-
tion of the command line arguments and stop the pro-
gram.
DISCUSSION
When deciding on the mode to use and designing the particu-
lar function (except for the flattening option) it is very
useful to pick a portion of data - a few sample traces - and
use specal to display the 1-d amplitude spectra (use log
scale). The straight line segments may be chosen by using
the cursor on the plot to pick frequency - weight (db)
pairs. In this option the amplitude spectra will be boosted
by this many db at this frequency (interpolated between con-
trol points). This method tends to break down in the lower
frequencies, e.g. boosting between 0 - 10 Hz with respective
db weights 30 - 0. Chances are the amplitude spectrum is
highly nonlinear in this region so straight line segments
can be inadequate.
BUGS
Beware of applying an overly ambitious boost function; the
result is usually a very noisy output data set.
See Also
dctvf (standard agc-pass band frequency equalizer using
ormsby filters). For more sophisticated amplitude condi-
tioning see fxbl, spbl and stel
AUTHORS
P. R. Gutowski, Amoco Production Research Center, 1992
COPYRIGHT
copyright 2001, Amoco Production Company
All Rights Reserved
an affiliate of BP America Inc.
Man(1) output converted with
man2html