Nominal detector dark noise and dark current values for individual
detectors are stored in calibration files Cal-G 21_n, where n equals
one of 1, 2, 4 or 8 for the different reset intervals. Units are 
V/s.
In the standard processing, this calibration file is used to issue
warnings when a particular detector is very different from the nominal
value. Note that 4 and 8 second resets were only used early in the
mission for calibration measurements.
Checks were made at the start of every revolution for the purpose of characterising the dark current and noise throughout the mission. It was found that they were stable for all bands except band 3. Here, the dark currents increased by up to a factor of three during the mission, while the dark noise hardly increased for most detectors. In most cases the increase in dark currents was gradual, but occasionally the dark current suddenly jumped to a higher level, e.g. detector 36 (band 3 detector 12) in revolution 150. Dark currents also decreased: after 250 revolutions of relative high dark currents, those in detector 36 dropped to one-fifth of their high value.
Some detectors had either an exceptionally high dark current and/or a high noise. These have been dubbed 'bad' detectors. Which detectors in a band are or are not good can change. For example, at the start of the mission band 3 detector 34 was poor. After PV, when its bias voltage was changed, detectors 30, 31 and 36 were the worst in that band.
The dark currents were higher at the start of a revolution for all bands, decaying with time so that after about four hours they had reached their nominal values. This was the case for bands 2, 4 and 6, with dark currents between 15% and 100% higher than nominal at the start of a revolution. This effect was less important for bands 3 and 5, while for band 1 it was only observed in a few detectors.
Normally all changes in dark current will be automatically corrected
during the pipeline processing, see Section 7.3.3. Cases
where the continuum is weak can cause problems for dark current
subtraction. When the
amount of flux falling on the detector is small, there may be cases
where the output of the detector from incident flux 
dark current
noise is less than the output from dark current noise, the
end result being negative fluxes in the AAR product.
The straightforward dark current subtraction for bands 1, 3, 4, 5 and 6 is described in Section 7.3.3. No models to correct the data for memory effects could be found for band 4 and, as a consequence, dark current subtraction for this band is the same as bands 1 and 3. For band 2 however, a method was found to model the signal falling on the detectors which incorporates a dark current correction.
An adapted version of the Fouks-Schubert model was developed by Do Kester and successfully implemented in the SWS pipeline to correct band 2 data for transient effects. The method can bring the errors (of sometimes up to 20%) down to the few percent level. This is described in Kester 2001, [22] and García-Lario et al. 2001, [11].
Section 9.2 gives a complete description of the implementation of the band 2 memory correction in the SWS pipeline.