The pixels on IR arrays are produced by generating voltages across the
substrate applied to individual electrodes. Captured photons are converted
to an effective charge, which changes the voltage in a given pixel, and it is
this voltage which is read out. Thus the output counts consist of a reset
level produced by the initial bias voltage,
dark current produced within the
substrate, and (normally) a signal generated by the source and sky photons.
Thus from an observed count at any pixel ( 
), to recover the
source signal from the object ( 
),
the reset level ( 
), dark current ( 
), and sky ( 
)
must all be measured, as well as a flatfield which measures the variations in
pixel-to-pixel response and vignetting from optical components (F(i,j)):
A DARK exposure will give 
(but make sure it has the
same exposure time as your object exposure, as the dark current isn't
linear with integration time). In IR observing, the flatfield often varies
on timescales of hours, due to changes in thermal emission from various
components, hence the best flatfield is generated from sky frames ( 
)
obtained with the object exposures, thus:
Because the sky is so bright in the IR, and because the flatfield can vary, it is important to subtract the sky from object before flatfielding. This also has the advantage of not needing to explicitly subtract a DARK from the OBJECT, as the dark current is already in the sky exposure (but a DARK is still needed to generate the flatfield). Thus the object signal is then obtained from:
WHIRCAM has two main observing modes: STARE and ND_STARE. In STARE,
one readout at the end of the exposure is made, giving 
, while
in ND_STARE, an initial readout is made immediately after the initial reset
is applied, and this reset level is subtracted from another readout at the
end of the exposure, giving 
. Thus it is important not to
mix DARKs taken in STARE mode (ie 
) with DARKs taken in ND_STARE
mode (ie just 
).
In the WHIRCAM array, Well depths, and hence saturation, vary across the chip, and in particular the odd column pixels have a lower well depth than the even column pixels. To reduce the problems of different saturation levels on the odd and even columns (which would be difficult to measure in real time), different reset levels have been applied to the odd and even columns, such that at 80% of saturation the odd and even column counts are approximately the same (see Figure 5). A side effect of this is that the flatfields are dominated by an odd-even column fixed pattern noise. Consequently, most faint images will not be seen on the raw image display. To confirm the presence of a faint image while observing, you'll need to subtract off sky and divide by a flatfield (eg. using IRCAM_CLRED - see Appendices A and B).
The sky counts in narrow band images may not give enough signal to noise to generate an accurate flatfield. In this case, use either the bright twilight sky, or dome flates obtained during the afternoon from the halogen flatfield lamp (ask your TO or SA where this is stored).