The observing procedure for a target star consists in recording exposures
of that target with the occulting mask as well as exposures made without
the mask to measure the point spread function (PSF). The same information
for a comparison star is necessary in the reduction procedure as described
in Sect. 4.3 (click here). Usually two comparison stars are observed for
cross confirmation purposes, under the same conditions, with and without
the mask. For each objects, the sky emission is measured at a distance of,
typically, 30
away from the target object. Individual exposure
times are constrained by the dynamical range of the detector. With the 256
256 SHARP II near IR camera on bright targets (mag = 3-6), they
are typically a few seconds. Data cubes of typically 100 images are
recorded. In addition flat-field and dark exposures are recorded at the
beginning and the end of the night.
What is referred to here as "bad pixels" are pixels which behave
abnormally in comparison to the whole detector. These include pixels
with a non-linear response, dead pixels and pixels with excessive
noise or characteristics that vary from one night to another. They
are easily detected by their strong deviation from the mean local
intensities. Their value is replaced by the median value of the
neighbouring pixels. This procedure is necessary for a few hundred
pixels in the 256 256 image, and has no deteriorating effect on
the angular resolution.
With the detector used the dark signal is very stable and reproducible, but depends strongly on the exposure time (especially for short exposures). Dark exposures of equal duration as the exposure on the astronomical targets are recorded.
The sky emission is measured 30 away from the star. The implicit
assumption is that no significant variation occurs on this scale. Note
that to take into account this variation to a first order would imply
additional observations of the sky on the opposite side of the object, but
doing this would also increase the delay between the object and the
reference exposures. The sky subtraction also removes the instrumental
background.
The flat fields are derived from sky exposures, once dark signal has been
subtracted. Indeed, no spatial variation of the sky is expected over the
13 field of the camera and we checked that no background emission
is detectable on this field. This method appears to give more consistent
results than when using dome flats. This correction is especially
important for extended sources but also for precise photometry since it
involves corrections of the order of 10% of the mean signal level over the
detector.
Figure 3 (click here)a shows the typical shape of an individual image of an occulted star. We have tested some recentering and selection procedures in order to obtain the best result from a large number of such individual images. The classical shift and add recentering procedures do not improve the image quality. Indeed, along a data cube of a few minutes, no spatial drift of mechanical origin has been measured. The intensity centroid of an occulted star light is stable to better than 0.5 pixel over a 5 minutes period and very narrow (diffraction-limited) thanks to the adaptive optics system. This value is estimated by numerical procedures measuring the centroid of various ring of light in the wings of the PSF of an occulted star. Such procedures were validated in the favourable case of the observation of HR 4796 , where both the occulted star and an unocculted star appeared in the field of view. This allowed to compare simultaneously these criteria with the center of the easily fitted unocculted PSF. This lead to the consistent typical value of a 0.25 pixel rms variation. Also, the irregularities present in the wings of the PSF are variable on time-scales much shorter than our single exposures (turbulence time-scales) and cannot be removed by shift and add techniques. These irregularities can only be reduced by increasing the number of exposures.
Figure 3: Residual light from a star behind the mask,
for a single exposure a) and 200 added images b). The scale
is logarithmic: the inhomogeneities in the wings are at very low
intensities compared to the peak of the star
For image selection, criteria based on the Strehl ratio or the width
of the peak of intensity do not directly apply to the coronographic
data. The selection can be based on the level of the residual
intensity after the occulting mask, asymmetry or displacement of the
center of the PSF, or azimuthal irregularities. We have explored
these criteria, or combinations of them, to establish PSF "wing
quality indicators''. Yet none of them lead to a significant
improvement in the results. Observed data cubes appear to be so
homogeneous that we cannot clearly distinguish "bad'' images compared
to the mean ones. These conclusions do not derive from a lack of
information on the center part of the PSF. Indeed, we tested our
procedure on a field of view which contains both a bright star, placed
under the occulting mask and a fainter star (
= 3) 8\
apart. This gives the opportunity of simultaneously observing the
center part and the wings of the PSF. The data cubes do not show
important variation of turbulence conditions. Such changes can occur
in less than one minute but such an evolution can be detected in the
wings as well as at the peak of the PSF for frame selection
purposes. And, in the present case of a globally homogeneous cube,
indicators from the central part of PSF (like the FWHM, radius of 50%
energy, etc.) are of no help either. Indeed, the wing quality
indicators relate to the poor correction of the higher modes caused by
turbulence. Thus two types of indicators are not correlated in such a
case and we have checked that selection based on the central part of
the PSF does not lead to a better result for the occulted star light
profile. Finally, as long as differences in the quality are low and
uncertain, we do not want to arbitrarily cut off our sample and we
keep as many images as possible in order to maximize the signal to
noise ratio.
Then, the homogeneity of a data cube is systematically tested with a simple
criterion such as the total intensity over an image, which would indicate
any technical disorder or sudden correction change, which never
happened in our observations. Then all images are simply added.
After the mask, most of the stellar light is removed. However, the wings of the PSF are still present outside the occulting mask and are much stronger than the circumstellar emissions that we wish to image. We may be able to remove the residual light from the central object, since its shape is expected to exclusively depend on optics and correction. This step is critical and some difficulties are generally encountered.
A possible method is to use the star itself to estimate the diffracted
light profile on a region of the image where no circumstellar emission is
expected. The profile is then azimuthally extended to the whole image,
following Golimovski et al. (1993) in the observation of the
Pictoris disk. Yet the definition of the area used to evaluate the light
to be subtracted is critical. Indeed, it involves some a priori
knowledge of the observed circumstellar emission. In particular, in
case of extended emission it seriously affects the shape of detected
non central emission, especially close to the star. In order to avoid
such possible biases, which are difficult to estimate, we prefer to
use a comparison star observed under the same conditions. This
comparison star is chosen to be angularly close to the target and of
similar magnitude and stellar type in order to get the same adaptive
optics correction. For the same reason, it is observed very close in
time as the target object. We also take care to avoid any reference
star suspected to have a companion or some circumstellar material. The reduction
procedure of this comparison is exactly the same as for the main
target. The residual light from the central object is then
subtracted, after proper scaling to the brightness of the target, by a
multiplicative factor. After this procedure, only non-central emission
should remain.
The reasons why this procedure is delicate are the following: First, the gradients in the PSF feature we subtract are steep (radial dependence in r-3) so that any slight difference in the shape of the residual light from the star affects the efficiency of the procedure. A possible offset that may have occurred between the observations of the target and the comparison induces in the signal after subtraction a strong deviation from the zero in the region close to the mask with a highly positive area on one side opposite to an highly negative one. We correct this effect by recentering down to a precision of 1/8 pixel. Exploring higher precision do not clearly improve the result and the remaining deviation from zero after the subtraction of the central light comes from the following effects more than from the centering precision. Second, due to the photon noise and also to the fact that very high modes of the image correction are imperfect, some azimuthal inhomogeneities remain and vary from one single exposure to another. They are reduced by increasing the number of exposures, but the finite total exposure time still causes uncertainties to the evaluation on the shape of the residual light as can be seen in Fig. 3 (click here). This uncertainty, coupled with the photon noise, is quantified by the measure of the signal variance along a data cube, represented as a function of the distance from the star in Fig. 5 (click here) (long dashed line curve). Third, the global quality of the image correction varies on time-scales which depend on weather conditions (minutes to hours), and on the star intensity and position in the sky. Such a slight variation implies that both profiles may be not exactly similar anymore. This clearly appears when one divides the image of the object by the comparison one. The result sometimes show a dependence on the distance from the star. This occurs when the two stars are observed more than one hour apart. With shorter delays, the division is flat over most of the field of view. We can then derive a constant scaling factor. The dominant source of uncertainty is no longer due to this determination.
Finally, after all these corrections, we are able to quantify the resulting uncertainty in two ways. The first one is to measure the residual light in the reduced image (star corrected by the comparison) where no emission is seen. The second one is to apply the same method to two comparison stars observed respectively before and after the target.