Re: Tech Note 66: Degree of Crowding on Mark IV images

[Tom Droege <droege@wwa.com>]

First I would like to observe with Michael that our CCDs do not melt, and 
the night sky is not lit up like noon.

What do we do when we "find" stars?  We find areas in the sky where the 
light flux peaks.  This give us an RA, Dec. location in the sky coordinate 
system.  We adjust the exposure so that we do not get a saturated value 
everywhere.  When we do this, we find such peaks at a fraction of the pixels.

These peaks contain light from all the stars in the area that we include in 
measuring the peak.  We do not measure one star we measure many.   We 
cannot avoid this.   This would not be a problem if we could measure the 
same star in exactly the same way every time.

We cannot do this.  For example, one evening we might measure a star 
sitting in the center of a pixel, then next evening it might be sitting at 
the junction of 4 pixels.  The result is an entirely different looking 
PSF.   This is not so much a problem for measuring a particular star, since 
photons are mostly conserved by our detector. ( I am not even certain about 
this.)  But a star close to the measured star may contribute more or less 
photons to the measurement depending on how the two stars are located on 
the pixel grid for successive measurements.

The is an elementary statement of the problem.   Michael and Andrew point 
out that this problem affects a significant number of star measurements.

Note that we may not be able to "see" the star that is near by our measured 
star and which is contributing a variable number of photons to the 
measurement because it is in the noise.  This (I think) does not prevent a 
variable bias to the measurement.

One way to detect this, is to keep raw data around each measured star.  We 
might keep 7 x 7 pixels.  Suppose we measure 250 stars per square degree, 
or 10M stars.  Two filters and 100 measurements per star gives 200 
GBytes.   Not an impossible number.   This is "only" 400 CD ROMs.

Given this data, we could stack the 100 images for each measured 
star.  This would bring problem near by stars out of the noise by a factor 
of 10(?) where they could be identified.  This is something that could be 
automated.

Note that we only have to examine the 1% of so of stars that appear to be 
varying.

I would do this as a batch process after candidate variable stars have been 
identified by 100 or so measurements.  I would then load the raw data disks 
and a list of stars to be tested.  Just grind through all the raw data 
outputting the arrays to be stacked for each star.   One hopes to have a 
200 CD juke box to do this.   Or one could output the 7 x 7 pixel array as 
stars are found on the first star finding pass.  This would reduce the 
stored data over raw by a factor of 10.  I can think of arguments to do it 
each way.  For example, if we reprocess all the disks, one can then use the 
latest software and do a consistent analysis while searching out the arrays.

As one can see, I am still a fan of keeping the raw data.

Tom Droege

At 02:27 PM 6/1/00 -0400, you wrote:

>   Inspired by Tom's questions and Andrew's calculations, I wrote
>a program to place stars randomly on a CCD and calculate the
>distance to nearest neighbors.  The results are in Tech Note 66:
>
>      http://a188-L009.rit.edu/tass/technotes/tn0066.html
>
>   The Note shows results for cases from 1,000 to 9,000 stars
>scatted over an image.
>
>                                           Michael Richmond