In Gundalf C8.3k-C8.3m, we added the ability to tailor the filter menus since there are so many filters available (some 3000 or so). Gundalf does not remember context as part of its stateless design for security reasons, so to remember a set of filter choices in between invocations of Gundalf, start Gundalf up and immediately edit the filter menus using the "Edit Filter Menus" option in the main menu. Then use the "Save Session" option to save the session with some suitably memorable name, such as my-default-filters.ses. Remember that downloading files is not done by Gundalf but by your browser using html5 so if you wish to choose a directory, you may have to set an option in your browser.
Now, each time you start up Gundalf, immediately restore your my-default-filters.ses session file using "Restore session" and your chosen filter menus will be enabled.
The nature of the modelling algorithm means that there is a degree of iteration in this leading to small errors in long delays as used in popcorn arrays. They are typically around 1%.
If you request Gundalf to do this, it simply switches off the source ghost. The result is somewhat unfamiliar looking to a geophysicist but this viewpoint has value. It is free of uncertainties about the reflection coefficient and for single guns, is a good approximation to a notional source. The normal statistics of Peak to Peak and Primary to Bubble must be adjusted to Zero to Peak and Primary to Bubble (from zero). The period is unaffected. They are most commonly used in calibration to compare with near-field hydrophone measurements, (concurrent far-field measurements are normally used for source depth monitoring.)
SPL and SEL are commonly used by marine biologists to measure sound impact on sea life. SPL is either the zero to peak, or peak to peak (which Gundalf uses), expressed in dB. relative to 1 muPa at 1m. Geophysicists tend to use bar-m, so to convert, just calculate 20 log10(bar-m) + 220. Correcting this for geometric spreading by subtracting 20 log10(distance in m.) gives an approximate idea of the SPL as a function of distance.
SEL is a bit less accessible. Its the sum of pressure2 taken over the significant part of the signature, (usually taken as window containing 5-95% of the total) normalised to 1s. duration, and expressed as dB. relative to 1 muPa2-s at 1m. It is similar to the rms pressure but has no simple relationship to typical airgun signature measurements such as peak to peak and primary to bubble. Marine biologists favour it because it is very useful in measuring cumulative exposure to sound. Check out Help -> Units and Help -> Audiogram for more information.
The upper and lower bounds of the envelope are not single signatures. Instead Gundalf computes the largest positive and negative amplitude at each time sample for any of the modelled signatures and displays these extremal values. In general then, the envelope is a composite of many signatures, although it could of course be just one. In other words, none of the modelled signatures lie outside the envelope at any point.
In short no. The reason we do this is that the internally developed filters work at any sample interval which seems most appropriate on start up. If you want to associate specific filters with models, its best to create a session file from the main menu and save all the context, including sample interval, filters, file locations and all the other relevant parameters together.
We don't know, although Windows job scheduling seems crude in the extreme. You may even find that the signature window reports as not responding when such plots are being produced. Please be patient. The window does indeed come back and the plot is produced as normal. It is worth noting that the Gundalf software is identical on both systems and yet on the same hardware, our Linux machines are around 12 times faster than our Windows 7 machines, and around 4 times faster on low-resolution plots for typical arrays.
Finally, note that there is a slight asymmetry in the interpolated plots caused by the interpolation algorithm as it has to interpolate dimensions which are divisible by 2. Other interpolation methods exist but are generally more time-consuming and the effect is negligible.
Note that this is no longer relevant in the Cloud version which uses an entirely different system.
The dip (also known as "take-off" angle) is specified from the vertical. In other words a dip of 90 degrees is the horizontal. Dips are positive measured away from the vessel. The Gundalf system is a right-handed system so zero azimuth points directly away from the boat and increases in the clockwise direction when viewed from above. A blue arrow on the dip-azimuth plots indicates the boat direction.
If the impulse response of an internally generated Gundalf filter is displayed in the signature window by convolving with a spike, the first sample is not zero and it appears to be truncated. In fact the filter is perfectly OK. Minimum phase filters do not have to have a zero as the first sample, (consider the minimum phase wavelet (1,0.5) for example). Gundalf filters are designed to be maximally flat in the passband. Some filters such as the DFSV filters, start with zero but significantly dip in the passband. The philosophy in Gundalf is to produce the filters with the closest frequency domain response to the user's requirements.
Until 7.1a, guns were numbered as the user numbered them. This was fine in most circumstances but could get confusing as guns were deleted and added. In 7.1a we rationalised this but chose a gun ordering which ordered strictly in x,y. This led to a non-intuitive gun numbering for clusters in drop-out reports for example. From 7.1b onwards, guns are numbered in y,x within sub-arrays if the user specified them and if the user does not specify sub-arrays, Gundalf will try to identify them. This corresponds to the conventional way users expect guns to be numbered as it causes y direction clusters in sub-arrays to be numbered sequentially. NOTE: If a sub-array is reversed, this implies that its numbering is reversed also.
Historically, the passband of the internal filters of Gundalf has always been calculated from the 3db points reflecting their electrical engineering heritage. Externally defined filters are simply used as they are supplied and will reflect the policy of the filter designer. This may be either 3db or 6db, as in the SEG definition.
Applying a zero-phase filter to a (typically) minimum phase airgun signature only gives a minimum phase output if the spectral bandwidth of the airgun signature is smaller than the equivalent bandwidth of the filter. The reverse is usually true so the output is mixed phase. Gundalf reports always plot from time zero for causality so the onset of the mixed phase wavelet may be removed. It is best to use minimum phase wavelets with airgun signatures to preserve causality. This can also affect the amplitude spectrum slightly as Gundalf only uses the causal part of the signature to compute these spectra in the reports. If you want to see the anticipation part of the signature, use the signature window which will show the full signature, pre- and post- time zero.
The far-field signature of a marine seismic source array is zero at all frequencies at 90 degrees from the vertical in the infinite far-field. That's why Gundalf doesn't calculate it. Its a consequence of the ghost cancelling the primaries exactly at all frequencies, (the effect is known as Lloyd's mirror in the environmental world). When its straight down, the ghost is exactly the negative of the direct arrival at infinity but there is always a time-difference in arrival so the combined signature is non-trivial and it makes sense to calculate the equivalent signature at 1m, (in bar-m). At 90 degrees however, the time difference tends to zero at infinity and they ultimately cancel. For the same reason, the signature at the sea surface is always zero at any range.
It does however make sense to calculate the far-field at a fixed position, say 2km away at the same depth as the array, although it is typically very small, but the natural output is in bars and Gundalf can already do this, (just enter the appropriate observation position).
The normalised cross-correlation, being a measure involving squares of numbers, is very heavily influenced by the peak and trough of airgun signatures, because they are the largest numbers - the bubble trail does not actually contribute much to its calculation. Dropping a couple of guns out of a marine source array of perhaps 32 guns certainly affects the peak to peak but makes only a small difference to the overall shape of the signature. However, normalising the cross-correlation gets rid of any scaling effects on the peak to peak (it is scale invariant) and the remaining differences are sufficiently small that the normalised cross-correlation is usually at least 0.98 – 0.99, and for a big array of say 36 guns or more, is often 1.0 to 3 significant figures, for all single-gun and double-gun drop-outs. It would make more sense to use a statistical measure of comparison which was sensitive to relative amplitude differences, such as semblance (due originally to Taner). However it is debatable whether any such measure would be particularly useful in practice as the predictive deconvolution used in seismic data processing is a great healer and this is a much more attractive alternative than shutting down a seismic survey vessel on the grounds that the normalised cross-correlation due to drop-out has reduced a little.. The following reference gives some idea of the resilience of a marine seismic airgun array signature to various kinds of noise:-
Les Hatton (1986) "Weather and the 3 microbar limit on the NW European Continental Shelf, First Break, August, 1986
Clear the visible gun canvas of guns, then go to Behaviour -> Preferences -> Grid Display Options ... and select "Centre grid in x". You can then enter guns with negative x values. In essence, Gundalf restricts you to entering guns with positions corresponding to the visible grid.
If the source ghost and the cable ghost are at the same depth, they do indeed amplify the signal. Consider a source ghost (1,0,-1) and a receiver ghost with the same depth (1,0,-1). Convolving the two gives a signal (1,0,-2,0,1), amplifying the trough by a factor of two. If the source and receiver are at different depths, this constructive addition does not take place. For example (1,0,-1) * (1,-1) = (1,-1,-1,1).
The location of the browser in the Windows registry is incorrect. To reset, just choose Behaviour -> Set locations -> Forget browser location. Then choose the Help page again and you will get a dialog asking you to find your browser. Navigate your way to "C:\Program Files\Internet Explorer" and choose Iexplore.exe. Now look for your Help page again and it should appear OK. You should not have to do this again.
Note that this is no longer relevant in the Cloud version.
Values of optimisation parameters are not ordered if they lie within the required range. So for example, if you are trying to maximise the primary to bubble ratio and you set a range 10 <= P/B <= 20, any value between 10 and 20 is equally valid. This means that if the current best is 14 and it finds a value of 16, it will not reject the 14 because both are equally valid. If you want to maximise the primary to bubble, choose just a lower bound and then select the extremal box. Gundalf will then exceed the lower bound by as much as it can and 16 would be selected in preference to 14.
In Gundalf, every gun is modelled independently - the sub-array number is purely for convenience. For example, if you want to construct an array made up of identical sub-arrays, create the sub-array on its own and then build the array by continually adding this sub-array. The sub-array number is completely ignored when modelling the full array signature as it should be - all guns potentially interact.
For the report, the filtering is applied to the notional sources before shift and summation as part of the modelling algorithm itself. For the signature window, the filtering is applied after summation to allow the same model to be filtered in different ways without having to recreate it. This gives a generally small difference in peak to peak of typically less than 1 per cent for an array modelled at small sample intervals, (0.0005 sec). However for larger sample intervals, the difference increases and might be as large as 6 per cent for coarser sample intervals of 0.002 sec and many guns. Also significant differences may be observed for Wiener filtering which does not commute with linear filtering. Gundalf is calibrated on the sources shown in the reports and the signature window should therefore only be considered as a (generally good) approximation. Note that all html reports are entirely self-consistent between signature, directivity and energy budgets and should therefore be considered the correct version. This effect appears in Gundalf versions 3.1n (August 2004) and later.
The fact that a physical problem is symmetric
does not imply that its numerical mathematical solution is symmetric.
Numerical solutions involving non-linearity or forward marching in
time or retarded time interaction are rarely fully symmetric even when
the initial geometry is symmetric. In airgun modelling, every gun
interacts through its pressure field with every other gun in the notional
source inversion. The algorithm has to start somewhere at time zero and
the discretisation error can manifest itself as a small numerical asymmetry.
Gundalf automatically reorders the guns to minimise this effect and make it
consistent but uneconomically small time intervals would have to be used
to make it effectively disappear.
It is worth noting that the same effect can occur in 3-D migration with some algorithms. Interestingly in migration, until depth migration was understood properly, the algorithms were symmetric when the physics wasn't, representing the opposite problem.
As a final example and on a more fundamental level, the Hilbert matrix is a matrix for which the i,j th element = 1/(i+j-1), where i is the row number and j the column number. It is a symmetric positive definite matrix which even has a closed form solution. However numerically, its a nightmare and is famously ill-conditioned. Even when its handled properly, asymmetric and frequently very large numerical discretisation errors arise. http://en.wikipedia.org/wiki/Hilbert_matrix
In summary, conservation principles in physics such as symmetry and even conservation of energy may not normally carry over into their numerical solutions.
specified ?
Specifying the position of the far-field hydrophone specified the measurement position for pressure field. A receiver cable has no meaning for such measurements.
Absolutely. All calculations are done at double precision and the modelling server is regressed on both machines and agrees within 6 significant figures for all of our test models. The client interface is also identical on the two platforms except that the local standard widgets are used for things such as opening files.
Note that the Cloud version of Gundalf is full 64-bit clean and is run on standard AWS images.
Very rarely and only when we receive accredited field measurements for which a re-calibration gives a better fit over all the range of modelling scenarious we cover. Re-calibration typically affects P/P and bubble period by less than 5 percent. P/B might be affected a little more. We will tell you if there has been a re-calibration in the distribution notes for a release.
Gundalf treats every gun as potentially interacting as a proper modelling package should. Sub-array identifiers are only logical groups to help manipulate groups of guns - they are ignored during the actual modelling. As sub-arrays are moved closer they begin to interact and the far-field signature varies.
Yes, you can set the maximum time to display for the
signature and the maximum frequency to display for the amplitude
spectrum. These are used in the high resolution plots. A full range
plot of the signature and the amplitude spectrum is also produced.
You can also set the amplitude scale for the signature, the db
amplitude scale in the amplitude spectrum and the maximum dip angles
and dip increments for the directivity plots if you wish. For
additional functionality just check out the Behaviour -> Report Parameters menu on the main page.
Gundalf Designer provides a collection of features to design an array, compare arrays and generate comprehensive reports on the array properties. Gundalf Optimiser can design optimal arrays from a specified inventory of guns which satisfy geophysical objectives such as maximising the primary to bubble ratio as well as containing all the functionality of Gundalf Designer. Gundalf Optimiser can also produce the environmental noise impact report. In general, computations which are expensive or which operate on multiple array files are carried out in the Optimiser.
There are spike signatures included in the
distribution which can be input using the Signature Window
(the spyglass icon second from the right). Choose the
spike signature with the correct sample interval from the
DEMOS sub-directory, (there is one for each of the sample
intervals of the supplied filters, 0.25, 0.5, 1, 2 and 4 msec).
This will display the spike. Then choose the filter whose
impulse response you would like to study from the
Behaviour -> Filter parameters -> Band-pass menu and the press the Apply filtering button.
This will show the temporal response of the filter in the
left hand window and the amplitude and optionally the phase
spectrum in the right hand window. You can drag and select
interesting areas of the plots.
Yes. The default is 0.5 msec. Increasing this reduces the accuracy of the modelling most noticeably around the peak where both amplitude and onset time may change, (by around 10% and 2 or 3 samples respectively). Modelling at 0.5 msec gives the best compromise between speed and accuracy.
Although this policy may change, the current policy is not to interfere in any way with an externally supplied filter. External filters have their sample interval encoded into their name, for example, dfsv_0p5_0-0_64-18.flt has a 0.5 msec sample interval, a low cut and slope of 9 and a high cut and slope of 64 Hz and 18 db/octave respectively.
Internally generated filters adapt to whatever the model sample interval was.
Yes. If you check the button to suppress volume
information on the Behaviour -> Report parameters menu and then
choose to generate a short form report, the gun volume information
is absent from the generated report although the signature and
spectrum are shown as well as an array plan view.