Re: Quick not on Disk 18a

[Stupendous Man <richmond@a188-l009.rit.edu>]

  I mentioned that raw images from the I-band camera on Disk 18a
contained pixel values with peaks every 8 and 64 counts.

  Tom replied that this is entirely within spec for the ADC chips,
and showed that it shouldn't lead to big errors in photometry if 
one considers carefully its effect on the pixels on which light from
a single star falls.  True.

  I was trying to point out a different facet of this effect: 
if one builds a histogram of the pixel values, and then analyzes
the histogram to determine a sky value, one might find a spurious
result.  Let me illustrate with a little ASCII art:


           true pixel values                   values with sticky ADC

      |        ...                         |          . .
      |       .   .                        |
 Num  |                                    |            
      |      .     .                       |        .     .
  of  |                                    |           .
      |     .       .                      |         .   .
 pix  |                                    |      .        . .
      |    .         .                     |      
      |   .           .                    |    .  .        . .
      | ..             ..                  |  .. .             ..
      ________________________             ___________________________

         pixel value                            pixel value


  Fitting a gaussian to the histogram on the left is easy.  Fitting
a gaussian to the histogram on the right is NOT easy.  

  By binning the pixel values, one evens out the shape of the 
distribution.  Tom is right that it's a good idea to choose a binning 
factor which is not a power of two.  I perform this binning only
when trying to fit a sky value to the data in each image.  Since
the typical distribution of pixel values in a good night-sky image
is several tens of ADU wide, binning by, say, 10 ADU, doesn't
really hurt my ability to find a good sky value.  After all, 
once I've fit a gaussian, I can determine its peak to much
less than one bin width.

  I don't perform binning in any other steps.  Come to think of it,
though, maybe I should when I measure a local sky value for each 
star during the photometric measurement process.  I'll have to check
on that ....

                                         Michael Richmond