For a while, if you were in trouble and no-one else could
help, you could turn to The
Equaliser, who would turn up and resolve your problems.
In seismic processing, our problems are often amplitude
related.
High amplitudes on a screen display or plot can make it
difficult to see fine detail elsewhere, even after we have applied amplitude
recovery methods such as linear gain and spherical divergence corrections.
At this point, we can call on our very own Equaliser: the
Automatic Gain Control or AGC.
An AGC balances out
amplitudes across a whole trace, using a sliding window. An average value for the amplitude is
calculated inside the window, and then a scale factor calculated to normalise
this to a fixed value, usually 1.0. The
window then slides down by a sample and the process is repeated.
The window (or gate) is typically 250-500ms in length; short gates have a more pronounced effect. This
is an exceptionally “brute force” approach but it is very effective – we used it a lot on hard-copy wiggle-trace plots, which were particularly
difficult to scale properly.
 |
| A marine shot record, with no AGC (left) and AGC applied (right). |
AGC is great when you haven’t got the time to scale a plot
properly – just drop it on the data and you can see everything!
AGC will “blow up” the noise in the water column (or above
the direct/refracted arrivals on land data) which can surprise the
inexperienced. It can be a good idea to
have muted this off before showing the displays to clients.
That said, a simple AGC has some downsides. Yes it is fast to calculate, but uses the mean
value as the basis for the average amplitude. As a result you get a “scaling shadow” just above and just below a
strong event. You
can get around this by using a “robust” (or median) AGC – it takes a bit longer
to run but by using the median value instead of the mean, it has fewer issues.
 |
| Normal AGC (left) compared to a robust AGC (right) – the robust AGC doesn’t have the scaling shadows around the strong amplitude refracted and direct arrival events. |
Of course sometimes the AGC scaling shadow can actually be
useful – it makes it easier to pick out refractions on noisy land data for
example.
In general, AGC is cheap and
nasty as a gain recovery method.
While it lets us see structure clearly, the subtlety of amplitude
variation is lost. With interpretation
workstations allowing you to toggle an AGC on or off, and the rise of
quantitative interpretation methods like AVO
and inversion, AGC has fallen out of favour.
It is more common now to use a combination of
spherical divergence correction and linear gains to recover amplitudes more correctly. The former corrects for the spreading of the
source wavefield with time, and the latter for losses arising from things like
scattering, inelasticity, as well as mode conversion of P-waves into S-waves or
refractions.
These “true amplitude recovery” processes are usually tested early
on in a processing sequence (often in combination) and retain much of the “amplitude
character” of a trace that AGC can swamp.
In
addition we can also select parameters for these gain recovery techniques based
on quantitative analysis – such as amplitude decay curves or even in a
statistically robust way using surface-consistent scaling techniques, which
helps with their accuracy.
 |
| Robust AGC (left) compared to amplitude recovery (right); in this case a T2 spherical divergence correction coupled with a 1 dB/second linear gain function. |
However, AGC should not be neglected, as it remains a useful
trick in the processors toolbox for managing difficult datasets with big
amplitude variations.
Consider
these shots – the RMS amplitude of each trace is plotted over the top as a
graph:
 |
| A few shot records, one of which has a nasty surprise in the middle! |
As you can see from the RMS amplitude plots, one of
the traces has extreme values; this doesn’t show all that well on the seismic
display, and you might miss it. Apply an
FK filter, however, and you get a nasty result!
 |
| The same shots with an FK filter applied; the extreme amplitudes have given us an impulse response function of the FK filter, spreading the high amplitude noise across the whole shot. |
The dipping noise has gone from the dataset, but the high amplitudes of a single trace have produced an impulse response function for the FK filter that has contaminated the whole shot. It is an extreme example, but highlights the issue.
We
can combat this using a removable AGC. This is like a regular AGC except the scalar applied to each window is
stored; we have one scalar value for each sample so this can be time
consuming. We can then “back off” the
AGC, recovering the original amplitudes by removing the scalar values one gate
window at a time.
 |
| The same shots with an FK filter applied with a ”removable AGC wrap” – that is an AGC before the FK filter which stores the gate scalars, and an “un-AGC” afterwards to recover the original amplitudes. |
This “AGC” wrap approach can save a lot of time when
managing a large number of spikes, or when data has big variations in amplitude
that cannot be compensated for easily in other ways. It is also reasonably “AVO friendly” in that
the amplitudes are well preserved.
While this example is extreme, the AGC wrap “equaliser” can
be a good way to deal with noisy land datasets when you need to apply FK or
Tau-P domain processes, and is well worth remembering as an alternative to
time-consuming despikes or geologically driven scaling.
It
doesn’t have as good theme
music as the 80’s original though.
By: Guy Maslen
