The focal plane of Gaia contains 106 CCDs, each with 4494 lines and 1966 light-sensitive
columns, leading to it being called the ‘billion pixel camera’. The pre-processing requires
calibrations for the majority of these CCDs, including SM, AF, and BP/RP, in order to
model each window during image parameter determination. Where effects cannot be
adequately modelled, the affected CCD samples can be masked and the observations
flagged accordingly. The CCDs are affected by the kind of issues familiar from
other instruments such as dark current, pixel non-uniformity, non-linearity, and
saturation (see Janesick 2001). However, due to the operating principles used by
Gaia such as TDI, gating, and source windowing, the standard
calibration techniques need sometimes to be adapted. The use of gating generally demands multiple
calibrations of an effect for each CCD. In essence, each of the gate configurations
must be calibrated as a separate instrument.
An extensive characterisation of the CCDs was performed on ground, and these
calibrations have been used in the initial processing. The effects must
be monitored and the calibrations redetermined on an on-going basis to identify changes,
for instance the appearance of new defects such as hot columns. To minimise disruption
of normal spacecraft operations, most of the calibrations must be determined from
routine science observations. Only a few calibrations demand a special mode of operation, such
as ‘offset non-uniformities’ and serial-CTI measurement (e.g., Section 1.3.3, Section 1.3.3, Section 2.3.5, and Fabricius et al.2016).
There are two main data streams used in this calibration: 2D science windows and Virtual Objects (VOs; Section 1.1.3).
The 2D science windows typically contain bright stars, although a small fraction of faint stars
which would otherwise be assigned a 1D window are acquired as 2D (known as Calibration Faint Stars, CFSs).
VOs are ‘empty’ windows which are interleaved with the detected objects, when
on-board resources permit. By design, the VOs are placed according to a fixed
repeating pattern which covers all light-sensitive columns every two hours, ensuring
a steady stream of information on the CCD health. The VOs allow monitoring of the
faint end of the CCD response while the 2D science windows allow us to probe the
Dark signal (or dark current) is the charge produced by each column of a CCD when it
is in complete darkness. While such condition was achieved during the on-ground
testing, it is not possible to replicate in flight as there are no shutters on Gaia.
The observed VO and science windows must therefore be used to determine the dark
signal for each gate setting, although these also contain background, source and
contamination signal, bias non-uniformity, and CTI effects. A sliding window of
50 revolutions is used to select eligible input observations, for instance those not containing
multiple gates or charge injections. The electronic bias (including non-uniformity) is
subtracted from each window and a source mask is created via an N-sigma clipping of the
de-biased samples. The leading samples in the window are also masked to mitigate
CTI effects. A least-squares method is then used to estimate a local background
for the window (assumed to be uniform), and this in turn can be subtracted to
provide a measure of the dark signal in each CCD column covered by the window. In this
manner, measures can be accumulated for each column over the 50 revolution interval, and
then a median taken to provide a robust dark signal value.
In an ideal device, there would be a linear response between the accumulated charge
and the output of the Analogue-to-Digital Converters (ADCs) at all signal levels. In reality, the response typically
becomes non-linear at high input signals for a variety of reasons (see Janesick 2001).
Although the linearity has been measured before launch, a calibration has not yet
been implemented in the daily pipeline due to the uncertainty in determination of the
input signals, which require detailed knowledge of a range of coupled CCD effects.
In the meantime, a conservative linearity threshold has been used to allow masking
of samples which may be within the non-linear regime. A related topic is the pixel
non-uniformity which represents the variation in sensitivity across the CCD. In Gaia,
we observe only the integrated sensitivity of the pixels within a particular gate so
this is known as the Column Response Non-Uniformity (CRNU). Similarly, no in-flight
calibration has yet been performed apart from the extreme case to identify ‘dead’ columns.
These are columns which appear to have zero sensitivity to illumination and can be
found using bright-star windows. The accumulated samples for a dead column have
a distribution which is consistent with the expected dark signal plus readout noise. The
CCDs used on Gaia have been selected for their excellent cosmetic quality (see also Section 1.3.3).
At the highest signal levels, various saturation effects occur on the device and within the ADC.
There can be large differences in the effective saturation level across a single device,
or even between neighbouring columns, for example due to variation in the Full Well Capacity (FWC).
For reasons beyond the scope of this document, the saturation level can oscillate or
jump depending on the readout sequencing. An algorithm has been developed to
measure the lowest observed saturation level for each gate and column to allow
conservative masking of samples. A Mexican hat filter is applied to the accumulation
of samples from bright-star windows to identify overdensities of data at particular
signal levels, using analytical significance thresholds. The lowest significant
peak is then taken as the saturation level. If no peak is found, then the maximum
observed sample for that column is used.
The calibrations discussed above are computed daily in the framework of the First-Look system
(see Section 2.5.2) and, if they are judged to be satisfactory,
the corresponding software libraries are subsequently used in the pipeline.