[Aavso-photometry] In defense of the ST-9 & CCD choiceConsiderations

[arne]

Tomas L. Gomez wrote:
> This interesting discussion about pixel size came at a very appropriate
> time for me: I was about to buy an ST9 camera for my new
> LX200R 8-inch, but thanks to this mails I am now pondering about
> going for the ST8!. This will be my "travelling telescope", meaning
> that I will take it to dark places in the countryside, so I guess that
> the seeing will be good (before I buy, I will check this).
> 
> I have a couple of questions:
> 
> 1) What do you think about defocusing a bit in order to increase
> the size of the FWHM?
> 
This works, and is a common technique for both increasing the dynamic
range in an image (millimag work on exoplanets) as well as working with
stars that would normally be too bright.  However, as a standard practice,
defocussing goes against one of the reasons you paid for that quality
telescope - it is like having poor optics.  You will not be able to
work effectively in crowded fields, and for SCTs, the star profiles
become non-Gaussian (donuts if you really defocus) and harder to centroid
and measure with typical software.  It is better to go with smaller
pixels from the start if you feel you will need to defocus often.

> 2) As far as I understand, both ST8-XME and ST9-XE cameras have a
> CCD in which each pixel has two detectors (a "normal" gate and another
> blue sensitive gate). If the usual rule is to have an FWHM of 2-3 pixels,
> but, to avoid that this pixel structure spoils the photometry,
> I guess one should increase this rule to, maybe, 3-4 pixels, isn't it?
>
The "rule" is to have a minimum of 2-3 pixels to not undersample.
More pixels per fwhm is ok, but you always have to compromise. More
pixels per fwhm means a smaller field that you can cover with a given
CCD, and finding comparison stars will be more difficult.  In addition,
for faint objects, more pixels means more total readnoise, and you will
reach a point where readnoise is the dominant noise source.  3-4pixels
is ok, and probably a good choice for your typical seeing, as you will
then have about 2 pixels per fwhm on your best seeing nights.

At the USNO-Flagstaff, we typically run about 3 pixels per fwhm.  On
the 1.55m telescope, this means 0.3arcsec pixels for the arcsec seeing.
We get poorer astrometry when the seeing gets good!  We have a poor
seeing cutoff at 2.5arcsec (which is 8 pixels/fwhm), not because the
astrometric centroiding gets worse, but because the signal gets spread
over too many pixels and we can't take long enough exposures to compensate.
When you undersample, the analytical centroiding doesn't have enough
pixels to get a good image profile, and subpixel variations start
becoming important.  For front-illuminated chips, the subpixel variations
are more of a problem than for our back-illuminated chips.  Again,
any camera and pixel scale will produce decent photometry, but the
closer you optimize your system for a given project, the higher
the potential data quality.

> 3) Is it OK to take a flat with bin 1x1, do a "software bin" to 2x2, and
> use this flat for pictures taken at bin 2x2?. What about darks?
> 
I would not recommend this.  If you are binning in software, then you
can certainly bin flats and darks in software as well.  If you bin
on-chip, then there will be subtle differences in all calibration
products and it is easier to take flats/darks in the binned mode than
to have to worry about how much effect these differences make.

> As far as I can see, the only disadvantage of this method is that a
> "software bin"
> will have more noise that a bin done on the camera, but taking more flats
> this will overcome this problem.
> 
Readnoise is seldom a problem with flats since they have high flux
levels - the Poisson noise of the illumination source is the dominant
error source, even during software binning.  Where software binning
is different than onchip binning is when working with faint objects,
where the readnoise is comparable to the star signal or sky signal.
Then decreasing the effective number of pixels by 4x or 9x reduces
the effective readnoise.  You have to look at the crossover point,
perhaps doing this empirically rather than with math.
Arne