There are some significant overheads associated with the operation of
the array. Thus, it is not possible to obtain 1 hour of ``on sky''
data in one hour unless you take a single exposure of one hour, which
is of course not possible. The first overhead is the time that is
needed to reset the array before each observation, the time delay
before the first read (fdly) and finally the time needed to make
the two reads for the double correlated sampling. This amounts to
about 0.5 seconds when fdly is set to zero. At this point the
data has been read into a Transputer on the instrument. To transfer
the image onto the Sun computer then takes about 4 seconds. By using
coadds you reduce the number of images being transferred to the
Sun computer and thus remove this overhead. The table below
summarizes the observing efficiencies obtained for various
combinations of coadds and integration time to obtain a total exposure
time of 100 seconds. Fetching header information from the TCS (i.e.
HA, RA, Dec etc...) takes some time ( 
2 seconds). This can
be reduced by turning tcp_off when it is not needed. Note that
when npics is set to a number larger than one the TCS
information is only read at the start of the sequence saving this
overhead for subsequent images. Finally, if you are making
observation in some type of raster pattern the time needed for
telescope motion will be an additional overhead, the length of which
will depend on how long the offsets are.
To maximize observing efficiency you want to maximize the length of individual exposures making sure to keep the brightness of the sky and objects low enough that nonlinearity effects are not to large. Secondly, you want to coadd as many images as is possible while keeping in mind that you do not want to integrate at a single position for too long a time since sky variability will make the data difficult to reduce. For a typical program obtaining deep K band images where the telescope is offset a few arcseconds every 100 seconds typical observing efficiencies will be about 75%. Finally, we should stress that coding your observing sequence into a TCL script can greatly enhance your observing efficiency (see the section on TCL scripts and the example scripts on disk).
Since there are different amounts of overhead time between images taken with and without coadds, otherwise identical images taken with and without coadds will be slightly different (a few ADU near the top of the quadrant boundaries). I believe this is due to different amounts of array heating before the first round of resets since the array clocking will not be identical in these two cases. This type of difference can be minimized by making fdly large (several seconds) since it allows the array to reach thermal equilibrium before the first round of reads, thus forgetting its previous operating history. But large fdly reduces observing efficiency and can lead to problems with nonlinearity. Thus, we recommend that sky and dark observations be taken with the identical integration times and number of coadds as your actual observations.
