The spectrograph uses a 1024x1024 Aladdin 3 Insb array. The
   saturation well-depth is about 8000 Data Numbers (DN) for
   the default bias of 0.4V, the gain 13.0 electrons/DN, the 
   dark current 0.2 electrons/sec over long integrations and the
   readnoise 50 e RMS per read. With 32 non-destructive reads (NDRs)
   (minimum itime about 17 sec with default SlowCnts of 20 - see below)
   the readnoise is reduced to about 12 e RMS. The number of NDRs
   is automatically selected by the software. Shorter itimes
   reduce the number of NDRs. If efficiency is a concern then the number
   of NDRs can be set manually (see C). The SlowCnts parameter allows the array
   to be readout at different speeds. With 20 SlowCnts the minimum
   ITIME is 0.51sec, with 3 SlowCnts the minimum itime is 0.18sec.
   The penalty for faster read out is higher readnoise although
   in practice the reason for increasing the readout rate is because
   of higher background where readnoise is not an issue. At 0.4V bias
   the array is linear to about 5 percent up to 4000DN. FOR GOOD 
   LINEARITY CORRECTION (in Spextool) AND TO SAFELY AVOID SATURATION,
   IT IS STRONGLY RECOMMENDED THAT COUNTS BE KEPT BELOW ABOUT 4000 DN.

   The SpeX guider uses a 512x512 Aladdin 2 Insb array. The saturation
   well-depth is about 5000 Data Numbers (DN) for the default bias of
   0.4V, the gain 13 electrons/DN, the dark current 1.5 electrons/sec
   over long integrations and the readnoise 65 e RMS per read. Up to 
   2500DN the device is linear to about a 10 percent. The full guider 
   array (512x512 pixels) minimum ITIME is 0.24sec using the 
   default value of 12 SlowCnts. With 3 SlowCnts the minimum ITIME
   is 0.12sec. However, at this speed fixed pattern noise becomes
   noticeable. Subarrays can be used to reduce the ITIME further
   For example, a 256x256 subarray has a minimum ITIME of 0.04sec
   using 3 SlowCnts. For this particular device, increasing the
   bias does not usefully increase the well-depth. TO SAFELY AVOID
   SAFELY AVOID SATURATION, IT IS STRONGLY RECOMMENDED THAT COUNTS
   COUNTS BE KEPT BELOW ABOUT 2000 DN (although IR guiding works well 
   on saturated images).
	
   Observers should note that the guider array is currently not corrected for
   linearity. To get photometry to 1% level the object and standard need to be
   exposed to approximately the same well depth, otherwise photometry can be off
   by up to 10%. Work is underway to include linearity correction (ask). Flat fielding
   can be done on the sky or dome. For wavelengths longer than 2.5 microns flat fielding
   does not work but dithering images can achieve a few percent photometry.

   Spectrograph array 1024x1024

   For the default number of SlowCnts (20) the spectrograph array
   takes 0.51sec to read out per NDR. If the ITIME is long enough,
   up to 32 NDRs will be taken. The maximum number of NDRs is set
   in the SETUP menu. For example, the minimum ITIME        
   for 32 NDRs is 0.51x32=16.3sec, so if ITIME=17sec the program
   will automatically set NDR=32 and the real time for the 17sec
   on-chip integration is 17+16.3=32.3sec - an efficiency of about
   50%. The trade-off is between readnoise and efficiency. In theory,
   the readnoise is reduced by the square root of the number of
   reads. In practice, we see little reduction beyond 32 NDRs
   (50 e RMS for 1 NDR compared to 12 e RMS for 32 NDRs for the
   spectrograph array - see plot). The observer has the option of
   manually setting the number of NDRs in the Setup menu to improve
   efficiency which might be an issue for time resolved observations.

   Example, SlowCnt 20 (default)
            ITIME=4.0sec, default NDRs=4.0/0.51=7, 
            real time=4.0+(7x0.51)=7.6sec per coadd
            efficiency = 53%

            SlowCnt 20 (default)
            ITIME=4.0sec, manually set NDRs to 2
            real time=4.0+(2x0.51)=5.0sec per coadd
            efficiency = 80%

   Example, SlowCnt 20 (default)
            ITIME=120sec, default NDRs=32,  
            real time=120+(32x0.51)=136.3sec per coadd
             efficiency = 88%

   With SlowCnts manually set to 3, the array read out takes 0.18sec
   so the efficiency is higher (but so is the readnoise due to the 
   faster read out of the array).

   Example, SlowCnt 3
            ITIME=4.0sec,  NDRs=1, 
            real time=4.0+(1x0.18)=4.18sec per coadd
            efficiency = 96%

   The setup for the fastest read out possible is:
            SlowCnt2 
            ITIME=0.1sec,  NDRs=1,
            real time=0.1+(1x0.1)=0.2sec per coadd
            efficiency = 50%
   This works well on very bright stars.


   IF YOU CHANGE THE DEFAULT SETTINGS OF NDRs AND SlowCnts
   REMEMBER TO CHANGE THEM BACK FOR LONGER ITIMES OTHERWISE
   NOISE WILL BE ADDED.


   Guider/Imager array 512x512

   For the default number of SlowCnts (12) the guide array
   takes 0.24sec to read out per NDR. If the ITIME is long enough,
   up to 16 NDRs will be taken. The maximum number of NDRs is set
   in the SETUP menu. For example, the minimum ITIME for 16 NDRs
   is 0.24x16=3.84sec, so if ITIME=3.9sec the program
   will automatically set NDR=16 (if the default has not been 
   changed) and the real time for the 3.9sec on-chip integration
   3.9+3.84=7.74sec - an efficiency of about 50%.

   The best setup for thermal imaging in the L' filter is
   ITIME=0.2 SlowCnt=9 (NDRs automatically set to 1).
   real time=0.2+(1x0.2)=0.4 per coadd
   efficiency = 50%

   The efficiency in this mode can be increased by decreasing
   SlowCnts to the acceptable minimum of 3 for the guider which 
   has a minimum ITIME of 0.12sec, but at the cost of increased
   column structure (noise):
   ITIME=0.2 sec, NDRs=1
   real time=0.2+(1x0.12)=0.32 per coadd
   efficiency= 63%
   

   Depending upon the number of counts 1024x1024 data files are
   2MB or 4MB in size. They are saved in FITS format as
   filename0001.a, filename0002.b etc., where .a and .b refer to the
   the telescope beam position and 'filename' is string of unix
   characters up to 20 in length. All the instrument parameters
   are saved to the fits header (integration time parameters,
   telescope position, mechanism positions etc.). Depending
   on the ITIME, up to several hundred 1024x1024 images can be
   acquired per night, so observers should be prepared to
   backup ~2GB of data per full night of observing.

   1024x1024 images take about 0.25sec to save to /scrs1 across the
   network and about 0.10sec to save to the /data local disk. The
   time taken to read the TCS information varies from 0.1-0.5sec
   so for the best efficieny TCS communication can be turned off
   in the Setup menu. DV display can also be turned off but the
   effect on efficiency is small. 



   Data is usually saved to the 70 GB scratch disk on stefan eg.
   /scrs1/bigdog/your_name/date, where bigdog indicates spectrograph 
   data. The path is setup by clicking AUTOSAVE ON in the 
   bigdog xuiwindow and then typing in the path. Similarly
   for the IR guider/imager e.g. /scrs1/guidedog/your_name/date,
   where guidedog indicates guider data. These data are owned by
   bigdog and guidedog, irrespective of the path name, and so you
   must open a bigdog or guidedog window to erase data. For better
   efficiency movie mode data are usually saved to the local disks
   on bigdog or guidedog eg. /data/your_name/bigdog/date and
   /data/your_name/guidedog/date. The local /data disk have much
   smaller capacities and so the space must be managed carefully.
   



   Note that the header parameter DIVISOR is the product of NDRs
   and COADDS. Fits files must be divided by ITIME x DIVISOR to
   get DN/sec. The SpeX Data Viewer (DV) is set by default to divide by
   the DIVISOR (under SETUP) but check to make sure. Data reduction
   packages such as IRAF do not automatically divide by ITIME X DIVISOR
   to get DN/sec, this must be done explicitly. Observers frequently
   make this mistake and find that their counts (DN/sec) are too large
   by a factor of the DIVISOR. Spextool automatically accounts for the
   DIVISOR. 

   It is important to make sure that signal levels on the both
   arrays remain in the linear operating range. FOR THE SPECTROGRAPH
   COUNTS MUST BE LESS THAN ABOUT 4000DN. FOR THE GUIDER ARRAY
   COUNTS MUST BE LESS THAN ABOUT 2500DN. (An exception to this is
   when guiding. Guiding works even on saturated objects.)
   The spectrograph calibration macros keep signal levels below about
   4000DN. 

   Check the signal by running the cursor across the spectrum or
   image and by drawing an x_linecut through the image or spectrum.
   Make sure that the DV window is set to 5000 when checking for
   saturation and then zoom on window to measure individual pixlels.
   Always check to make sure that dividing by the DIVISOR is set on
   in DV 'SETUP' window(see below).

   Always do this on an unsubtracted object frame since
   subtracting off a sky frame can lead to an erroneously low
   value (particularly when observing at thermal wavelengths).
   If the signal level is too high reduce the ITIME. In the
   guider the signal can also be reduced by using a narrower filter
   in the guider filter wheel. A complication is that in heavy
   saturation the measured signal can actually fall to zero. 
   (Signal = first read - last read = saturation - saturation = 0)
   However, heavy saturation is usually obvious, appearing as a 
   central dip or valley in the PSF.

   Signal levels do not change with NDRs or COADDS since DV is set 
   by default to divide by the DIVISOR (= NDRs x COADDS). However,    
   dividing by the DIVISOR can be turned off in the DV 'SETUP'
   window, in which case saturation level will vary with NDRs and
   COADDS. Always check to make sure that dividing by the DIVISOR
   is on.