The 4kx4k CCD version of the Y4K Camera went into operation at
the Yale 1-m in July 2005, and we have been privledged to be working with
some of the first images obtained with it. Data for our program is obtained
in queue mode by the 1-m observers, and provided on the Yale SMARTS 1-m
ftp site the next day. In the course of our reductions, we have learned
a lot about this instrument.
We have also
developed IRAF reduction scripts
for the basic CCD processing,
and automatic photometry of objects. We are making these all available
in a tarball.
Things like directory names have been hard-wired for the photometry;
if you know a little bit about
scripts they should be readily adaptable by others. Enjoy.
- Gain 1.5e/ADU. The effective read-noise on our data is 14e.
- Scale 0.289"/pixels (from the Y4K characteristics page.
- FOV: 20'x20'; N is down; E is right.
- Linearity good to 38,000 ADU.
- Shutter correction: a comparison of 10 sec and 30 sec dome flats
definitely show that the edges and center are exposed differently,
as illustrated in this
image from the Ohio State web site.
However, at 10 sec the effect is everywhere less than 0.3%. If you are
using shorter exposure times than that on your objects, be prepared
to make the correction. For the Galactic cluster data I have
correction images for 0.3 sec and
3 sec exposures.
- PSF variations. The point-spread functions on our images definitely
vary with position on the chip, and in particular degrade strongly in
the lower-right (NE) corner. If the fwhm is about 3.5 pixels over
most of the chip, it may be as bad as 4.5 pixels in the lower right.
- Bias variations. The bias level, and bias structure, are both quite
unstable. Examples are given here. As discussed below, this results in not being able to
do photometry along the middle rows/colums. See below.
- Dome flats vs sky flats. The dome flats do NOT illuminate the chip
the same as the sky (See:
Over a 2000x2000 region centered at (1500,1500)
the agreement is about 1%, but quickly degrades outside of this,
with a 6-7% difference at the edge. Photometry shows that the sky
flats are right. Thus, if you rely upon dome flats for your photometry,
you will need to restrict yourself to about a quarter of chip.
The illumination function does appear to be stable, and I have generated
a correction function based upon about 50 sky frames.
- Best seeing. There are clear variations of the PSF across the detector, and my impression is that
there is a slight tilt of the camera within the focal plane. Out of 1744 images obtained between
August 2005 and Jan 2006, the median delivered image quality was 1.65". The 10 pecentile BEST images
were 1.3" and better. There is a hard wall at 1"; no images had average FWHM across the frame that
were better than that. Here's the histogram:
Basic Reduction Steps
The CCD is read out through 4 amplifiers, each with its own bias level,
and each with a slightly different gain. So, in principle the following
steps should work fine:
At that point, one expect to see a seemless image with sky level, and on which one can do photometry even on stars on the boundaries. This was certainly the case for the old Tektronix chips read out by 4-amps deployed
elsewhere on Tololo.
- Break the image apart into 4 sections, each containing its own overscan region.
- Fit the overscan and subtract; trim the image.
- Put the image back together.
- Use the zero-second (bias)
exposures to remove any residual 2-d structure in the
- Divide by the normalized flat-field.
However, the bias structure of the Y4K Camera is quite unstable
during the night. No matter how I fit the overscans, the combined
bias frame did not match the
actual bias structure in program frames to better than 3-4 ADUs. This
always left small "jumps" at quadrant boundaries.
The implications of this are greater than one might guess: if
using a 5-pixel radius aperture on a seam, the DIFFERENCE in contribution
of the sky would be roughly 500e-, or 5% for a star with 10,000 e-!
So, our reduction procedure simply masks out a pixel-region along
the seams, preventing photometry in this region.
The problem is well illustrated by comparing the average
bias levels in each of the four quadrants for ten consecutive zero second
See the example here.
For each quadrant the variations are 10-20 ADUs. (For a typical
CCD one might expect variations of the order of 1 ADU.) Furthermore,
the structure of these biases vary considerably, as can be seen in
plots of the overscans. By fitting a high-order spline to the overscans,
one almost takes this out, but not quite. `
A further complication is that the image headers are lacking any
CCD section/trim/overscan information. Thus, ``standard" IRAF routines,
Steve Heathcote's powerful "quadproc", which are strongly header-driven,
will not work with these data without a lot of fiddling. So, we wrote
Our procedures are as follow. The scripts must be definited
in your loginuser.cl file, located where ever you keep your
IRAF home directory (i.e, with your login.cl). Note that the following
are hard-wired to work only on UNBINNED (1x1) images.
Note that there are several associated files:
- Download all the program images, flats, and zeros for a particular
night into a separate subdirectory.
- Load imred, bias, digiphot, daophot.
- Run y4kdoit.cl, which then calls the following:
The "fixed" images are now suitable for
- y4ktrim.cl gets rid of any binned images, and then
does the breaking apart, overscan correction, and reconstruction:
- y4kbreak.cl breaks the CCD images apart.
- y4ktproc.cl fits each of the 4
overscan regions with an order 15 cubic
spline! My preference is to fit these with a constant, but the overscan
here doesn't allow it.
- y4knew.cl reconstructs the 4 quadrants
into a single image.
The new image begins with "T" for "trimmed".
- Next, the y4kzero.cl script is used to average together the trimmed bias frames,
and subtracts them from all the flats and object frames. The resultant
images begin with "Z" (for zero-subtracted).
- y4kflat.cl combines the flats for a single filter, normalizes it, and then divides the normalized flat into all the object
exposures with that filter number. If you have multiple filters, call
this multiple times. The resultant images begin with "F".
- y4ksky.cl takes each flattened object and corrects for
the illumination function. My correction image can be found
- y4kshut.cl makes the shutter correction (currently only works for 0.3sec and 3.0sec.); the correction images are:
- y4kfix.cl takes each flattened
object image and prepares it for
photometry by setting values in bad regions to 50,000 ADU. The bad regions
are defined by a pixel mask image that I constructed using "ccdmask"
on some divisions of long and short dome flats and then "or"ed with
the result of running ccdmask on the sum of many, many sky flats that
had been divided by their respecive dome flats. I also threw in
an extra 2-pixel
stripe down the middle for good measure. My version of the bad pixel
mask can be found here:yalerev.pl.
- allFWHMS: lists the FWHMs of all of the images
- ERRORLOG: lists any errors caught
- allskies/*.fits contain the skyflats / domeflats