![]() |
Persistence On The IR Detectors |
|
Both the SPEC InSb array and the SCAM HgCdTe array are subject to some charge "persistence," although this is more often a problem on the SPEC detector because of the need to observe bright calibration stars and faint science objects in sequence.
The image below shows a particularly bad example of the persistence effect. In this case, the original lamp flat was taken using the 42 x 0.76 arcsecond low-resolution slit and the Nirspec-7 filter. The integration time was the minimum 0.25 seconds, and 10 coadds were taken. CDS sampling mode was used. Peak counts on the original spectrum were about 25,000 per coadd, which is beyond the nominal 1% nonlinearity limit of 18,000 per coadd.
Most importantly, a signficant attempt was made to flush the persistent charge. After the original spectrum was taken, the flush command was used with the default 50 coadds, three times; flush was also run a fourth time with 200 coadds (see below for more information about the flush command). After all that flushing, the image here was taken as a dark exposure, with itime=200 seconds and 1 coadd. The region originally exposed to the flat lamp has about 100 DN/pixel more "dark current" than the unexposed regions.

This image illustrates the importance of not overexposing the detector.
Try to avoid saturation and over-exposure. In particular, on the SPEC detector, try to avoid exposing to levels above 6000 DN per coadd in CDS or MCDS sampling mode. Your data will still be within the 1% linearity regime as long as the counts are below 18,000 DN per coadd, but the more you fill the wells the longer the persistence will persist.
The persisting charge can be reduced by both rapidly repeated readouts and by long integration times followed by readouts. The rapid readout treatment is of course faster, but seems to be less effective when the original image is more severely over-exposed. It certainly is worth trying it first, even in bad cases, as it takes comparatively little time.
Experience shows that repeated readouts will reduce the persistence considerably. There is a simple command that will automatically take a test image with a large number of coadds at the minimum integration time, and then restore the itime and coadds parameters to their previous values. To run the command, go to any terminal window that is logged into the NIRSPEC host computer waimea in the account "nirspec". If you are in front of a computer in Remote Ops 2 that is logged in as "nirspec", you can open a window on waimea by putting the mouse pointer on the blue background, holding down the right button, going into the Login Windows menu, and letting up on xgterm 'waimea' (NIRSPEC). At the command line in your waimea window, enter the following command:
flush    for the SPEC detector, or
flush2    for the SCAM detector.
to run a test image of 50 coadds at the minimum integration time. The flush and flush2 commands accept an optional parameter of the number of readouts to coadd. For severe persistence you may need to set this value to 100 or even 1000. To command, for example, 200 coadds on the SPEC detector, use the command:
flush 200
For the SPEC, BE SURE that you do NOT have a bright star on the slit or a calibration lamp turned on when you run flush! If there is a bright light source illuminating the slit, you will not be "flushing" out any charge because you are dumping more charge in between readouts.
For the SCAM, BE SURE that you do NOT have a bright star or other light source anywhere on the field of view when you run flush2! If there is a bright light source illuminating any part of the field of view, you will not be "flushing" out any charge on those pixels because you are dumping more charge in between readouts.
There is also a component of the persistence that behaves like increased dark current. This component seems to be reduced only by integration time on the detector. If bad persistence occurs, and the "flush" command does not completely remove it, put in the BLANK in the filter wheel and take several long (300 s) darks. Examine each dark carefully before starting the next, to see if the persistence has been acceptably removed.