-----------------------
General CCD description
-----------------------

These CCDs have a number of unique features.

1. They are capable of "orthogonal transfer", i.e. the charge can be
   noiselessly clocked horizontally as well as vertically, which permits
   tracking of image motion and fast readout of guide stars.

2. They are each split into four regions with different clock lines so
   that each region can be clocked independently.  These regions have
   no interruption of the metrology of the pixels however.

3. They have serial registers at both ends, and OPTIC is normally run
   using one amplifier at each end, hence four amplifiers altogether.
   In many ways it looks like four 2k CCDs butted together.  This is
   the nomenclature normally used instead of the "chip1 and chip2" above.

               2048             2048
        <===============> <===============>
    --- +---------------+ +---------------+
     |  |  CCD1 lower   | |  CCD3 lower   |  516  (sic)
     |  |---------------| |---------------|
     |  |               | |               |
   2052 |               | |               |
     |  |  CCD1 upper   | |  CCD3 upper   | 1536
     |  |               | |               |
     |  |               | |               |
    --- |---------------| |---------------|
     |  |               | |               |
     |  |               | |               |
     |  |  CCD0 upper   | |  CCD2 upper   | 1536
   2052 |               | |               |
     |  |               | |               |
     |  |---------------| |---------------|
     |  |  CCD0 lower   | |  CCD2 lower   |  516
    --- +---------------+ +---------------+
        <===============> <===============>

Various useful parameters include:

    pixel size                                   15   um
    readout time (unbinned)                      25   sec
    readout time (binned 2x2)                     8   sec
    CCD thickness (enhanced red response)        40   um
    read noise                                    4   e-
    gain                                          1.4 e-/ADU
    full well                                   >80K  e-
    non-linearity   presently             <1% to 30K ADU
    gap between CCDs                              1   mm ~ 72 pixels
    dewar hold time                             >24   hour

The results of a readout through four amplifiers are always rotated
and flipped in the output file into a physically consistent picture.
The bias strips for all four regions are found on the right hand side
in two stripes, with the left stripe from the left chip and the right
stripe from the right chip.  The top of the output image corresponds
to the side of the dewar opposite the connectors.  With a normal view
of the sky (e.g. a lens or a Cassegrain focus) the parity of the
output image is conventional, i.e. east is found counterclockwise of
north.  With another reflections off a flat mirror the parity is flipped.
The gap between the CCDs is not represented in the output image in any way.

Note that readout parameters refer to each quadrant and each
amplifier, so that reading out 2048 x 2052 gets the entire mosaic.
Also offsets are amplifier-centric, i.e. a read with an offset skips
pixels upstream of the amplifiers in the serial and parallel
directions, which may cause a surprising piece of the CCDs to be read.
The amplifiers of the four quadrants are normally found in the lower
left and upper right corners, but if the "geom" specification in the
"status" command or the "CCD?GEOM" FITS header is 2 instead of 0 the
amplifier is found on the opposite side of serial register.

Although the four regions in each chip are fundamentally the same, the
"lower" regions enjoy the ability to reach the amplifiers without
disturbing the upper regions.  Therefore the "lower" regions are
potential locations for guide stars, and the control software will
permit you to specify zero or one stars in each lower region to be
used as guide stars.  

When exposing, the software quickly clocks a small patch which you
specify around each selected guide star first horizontally and then
vertically over to amplifier.  This does not disturb the image in the
upper regions or any unselected lower regions.  The software then does
a readout of the small patch, analyzes the guide stars, possibly
performs shifts on the non-guide regions, and then pauses for another
guide star image to collect.  The overhead involved in reading out
these small guide patches varies from 10 msec to 30 msec, depending on
whether they are all close to their amplifiers or on the other side of
the chip.