CCD readout modes and characteristics
The chip used for ESPaDOnS is a 2kx4.5k 0.0135mm square pixel ccd
manufactured by eev (42-90 series).
The one tested up to now is an engineering grade with a rather
high number of cosmetic defects. The science grade chip that cfht allocated to
ESPaDOnS (referred to as eev1 in cfht dialect) is supposedly much better for cosmetics.
Cfht unofficially agreed that this detector would be dedicated to ESPaDOnS as much as possible,
and that it will remain mounted all the time on the instrument to optimise the instrument
stability and preserve the thermal and mechanical equilibrium within the instrument as
much as possible.
In order to cover a wide enough range of astrophysical applications, we decided to
implement several readout speeds.
This flexibility is usually not offered on other
cfht instruments, but we thought that ESPaDOnS users could greatly benefit from it.
For the brightest objects for which photon noise will dominate, achieving the smallest
possible readout noise is not crucial; short readout times are much more important,
either to improve the overall duty cycle of the observing session (eg when short
exposures are required to avoid saturating the chip or to ensure a high frequency
temporal monitoring). For the faintest objects that are usually exposed for longer
time chunks, having short readout times is much less critical; decreasing readout noise
as much as possible is in this case very important as it impacts very heavily on the final
quality of the collected data.
Four readout modes were selected to cover all potential needs of
The fastest reads out the full chip in 25s with a readout noise of 7.5e,
while the slowest achieves the lowest possible noise of 2.5e, reading out the whole
chip in 90s. For each of these readout modes, we determined the noise by measuring
the rms deviation in various 100x100pxl portions of the chip in a bias frame.
The gain was measured by ratioing slightly out-of-focus flat field images taken in
identical conditions and by computing the slope of the inverse variance (ie the squared
signal to noise ratio) as a function of adu counts. The graph on the right shows one
of such fits in the particular case of the 'slow' readout speed (points representing
measurements troughout the image while the full line depicts the linear fit to the
points). In all cases, good linearity was observed up to the saturation level.
The following table summarises the measured characteristics of each readout speed.
|speed || gain (e/adu) || noise (e) || time (s) || saturation (adu) |
|fast || 1.85 || 7.4 || 25 || 58,000 |
|normal || 1.40 || 4.2 || 40 || >65,535 |
|slow || 1.27 || 2.9 || 65 || >65,535 |
|xslow || 0.84 || 2.5 || 90 || >65,535 |
As all thinned ccds, eev chips are known to exhibit severe fringing patterns when illuminated with infrared light.
This is quite obvious from the image on the right showing some of the reddest flat field orders obtained with ESPaDOnS. On this image,
the orders run vertically, and each of them show the expected cross order structure for the polarimetric mode (two spectra per order and
three slices per spectrum). The fringing signature is that the flux along the orders is found to exhibit very strong variations (with an
amplitude of as much as 50%) on very small scales (a few tens of pixels). These patterns are however observed to flat field out properly,
leaving no apparent residuals in the intensity or polarisation spectra even when images are corrected using flat field frames with different
count levels. The only signature of this effect is that the error bar in the reduced spectrum is found to vary, as expected, by relative
amounts of as much as 25% on the same spatial scales.
© Jean-François Donati, last update Dec 12 2005