[vsnet-chat 2096] Re: faster read-out is really better ?

[aah@nofs.navy.mil]

Ulisse Munari has raised interesting points, as he often does!
  If you plot readnoise vs. readout speed, it is true that
the readnoise typically increases as you increase the readout
speed.  However, it is a pretty shallow function, and the
difference between a 50Kpix/sec readout rate and a 100Kpix/sec
rate is pretty small for most science-grade CCDs.  Most cameras
just set their rate by the speed of the ADC.  You must weigh
the increased telescope efficiency of a fast readout (which
means higher signal/noise in the same amount of clock time)
versus the slightly higher readnoise.  As long as you are
in a regime where the background noise approaches or exceeds
the readnoise, the efficiency factor wins.  For spectroscopy,
where there effectively is no background, you want the readnoise,
your dominant noise source, to be as low as possible and so
will generally use longer read rates.
  However, I'm surprised that you take >3min for reading a 1024x1024.
That would imply a 5.1Kpix/sec read rate.  I would be more
concerned about the increased liklihood of cosmic ray hits
and the differential dark current during readout, in addition
to the loss of efficiency.  Have you checked your readnoise
curve for your detector to see where the optimal read rate resides?

Ulisse also wrote:
>There is one additional aspect to consider, among many others:
>if you leave the telescope unattended, I suspect the field of view 
>drifts across the CCD, with the variable and the comparison stars
>migrating over regions of higher-and-lower response of the chip
>(them being caused by intrinsic pixel-to-pixel variation in CCD 
>response, by vignetting effects and by randomly distributed
>shadows by the dust grains sticked to the CCD's window or on the
>filters). Unless you have FULL CONTROLL over the flat-fielding
>of your CCD, the expected errors may be QUITE LARGE.
  This is also true.  Proper flatfielding is very important for
proper photometry.  If you use autoguiding with no flexure, the
stars will stay on nearly the same pixels.  Otherwise, the
pixel-pixel variations (nonlinear response and dark current, etc.)
will influence the photometric accuracy.  For astrometry, it gets
even worse.  There are pixel-pixel variations in pixel size on a CCD,
and often a step function at the replication boundaries.  We reposition
a field on subsequent nights to within one pixel in order to achieve
the sub-mas accuracy required for our parallax work.
  Paul Warhurst suggests high-quality flats rather than accurate
registration of frames.  You need both.  For millimag work, you
rarely can get flats good enough, especially on a night-to-night
basis.  Accurate registration will help in getting good differential
results, even if a small zero point would be required to get your
data on someone else's system.
  Note that the above discussions are rarely important for most
amateur work, especially if 0.01-0.02mag precision is the goal.
Just look at the differential results between two well-exposed
comparison stars...that will tell you how well you are currently
doing photometry.
  (interesting. between Paul, Ulisse and I, we have almost a full
day spread in local time.  The wonders of Internet!)
Arne

VSNET Home Page

Return to Daisaku Nogami

---

vsnet-adm@kusastro.kyoto-u.ac.jp