Data Set 20: progress report (revised)

[Andrew Bennett <andrew.bennett@ns.sympatico.ca>]

This was supposed to be an end-of-year report 
from the Bennett independent processing effort
bu things were going so badly that I put it off.

1) Dark/Flat processing is back working again after
some very depressing bug chasing. At last, with the
dome flats and lots of averaging, processed data has
lower noise than raw. A major improvement, Tom: let's
keep on doing it this way!

2) I finally got tired of retuning my matching routines
every time Tom kicked the mount - it took 3 sessions of
tuning to get through the 3 days of data the first time.
I took a fortnight off and translated Michael Richmond's
matching routines into Pascal. Better. They went through
the whole of Data Set 20 starting from the RA and Dec in
the FITS header and with no further tuning, without missing 
one image. At 0.3 seconds/image they are both faster and 
more reliable than what I was using.

3) I am now managing to keep some image data, like "median"
and "sigma" accessible to the photometry program. Yes - I
know IRAF does all this sort of thing automatically but this
is an *independent* effort. Read *pig-headed* if you like.
I am using simple pruning criteria like median or sigma
less than 150% of darkest, quietest image and thus manage to
throw away all the data for Days 2239 and 2246 leaving
only 2199. Looking at the raw photometry suggests that this
is a good idea but it does rather cut down on the ability to
detect variables!

4) Even after throwing away all this data, things don't look
too good. The V-band noise floor is perhaps a little better
than before - at least I have got it below 0.01 mags; some of 
the results before I got the bugs out of Dark/Flat processing
were running more like 0.1 mags ...  But the I-band is not as
good as I got with Data Set Tom. It should be a lot better
because stars are measured in more less the same place on
the CCD each time instead of scattered randomly.

5) I have the first version of Ensemble Photometry running
but the results are awful so far. Using only *good* sets
for which the same ensemble of calibration stars is available
in every case drastically reduces the amount of data and looks
worse than the raw data! I will keep looking for the bugs ...
but I fear that the problem is the small number of ensemble
sources that are clean on every image: the increased statistical
scatter from using fewer sources outweighs the improvement in
systematic error from using the same sources each time.
Or, as I said, I still have bugs.

Andrew Bennett, Avondale Vineyard