- Obs/IPAC Interface Document Rev. J --
J. White
reported that Rev. J of the Observatory/IPAC Interface Document is
ready for signoff, which is hoped to occur by the end of this week. The
definitions contained in this document are mutually agreeable on all
sides and appear to be consistent with IPAC's requirements for
processing the survey data, and therefore the TAPELOAD code is being
written to conform with this version.
- Sparc Benchmark Suite --
G. Kopan reported that
he has prepared a benchmark suite for use in calibrating a 4-CPU Sparc
3000 against lugosi. The suite involves running the currently delivered
version of PIXCAL/DFLAT/RDFRAME/FREXAS, the prototype coadder, and
KAMPHOT. Four pairs of scans are processed, one pair per CPU. Lugosi
runs one pair in 2 hours on one CPU; it also runs all four pairs in 2
hours, using 4 CPUs. Since we apparently cannot get a 4-CPU Sparc 3000
as a loaner, and since Gene feels that it is quite important that a
4-CPU machine be used, he has prepared a tar file and placed it in the
ftp area for the Sun Benchmark division to download, run, and send back
the output files.
- Anti-persistence Study --
J. Fowler reported on
the anti-persistence study he is conducting with help from G. Kopan, R.
Cutri, and R. Beck. Selected pixel stacks have been studied via the
latest debug output from DFLAT. The effect of very bright objects is a
little different from what had been anticipated; instead of merely
requiring more trimming before averaging, the stack of values is biased
in a way that cannot be fixed by any symmetrical trimming. On the other
hand, asymmetrical trimming has been found to be a very dangerous
practice in the past, so if it is to be used, considerable caution will
be required.
J. Fowler and G. Kopan have developed a trial method for attempting
asymmetrical trimming, known as "recursive median-distance rejection".
This works as follows: The value farthest from the median is rejected;
the median is recomputed; the value farthest from the new median is
rejected, and the median is again recomputed; this is continued until
half of the values in the stack have been rejected, and then the
remaining values are averaged (the number of rejections could be a
parameter, but currently we have been trimming about half of the
values, 25% from the high end and 25% from the low end). This appears
to be stable for the pixel values we have seen, and should be a more
robust way to trim before averaging. J. Fowler will develop a testbed
and try this algorithm out; if it appears promising, it will be coded
into DFLAT and tried on scans known to produce visible anti-persistence
artifacts in coadded images.
A news item related to PIXCAL/DFLAT was pointed out by R. Cutri:
the PIXCAL/DFLAT SDS has been signed off. Roc obtained the
signature of the cognizant science team member R. Elston during a
recent collaborative observing trip.