The AO systems realize the out-running idea of Babcock (1953) to compensate in real-time the atmospheric turbulence effects by suitable fast deformations of the optics. The exhaustive treatment of the AO subject in its various facets and an extensive list of relevant works can be found in the monograph by Tyson (1991) and in the review by Beckers (1993).
To improve the information content of a set of AO images it is necessary to deconvolve them from the Point Spread Function (PSF) as usual, and this is even more needed since each image, as well its PSF, is characterized by a variable amount of compensation. Thus the main problem faced when analyzing AO images, is the determination of the instantaneous PSF of the optical system during the acquisition. The PSF is affected by the diffraction pattern of the telescope, by residual static aberrations and by the AO servo-loop correction. Static aberrations can have a slow (tens of minutes) variability in time, due to optomechanic flexures and thermal effects. The servo-loop residuals instead have rapid variability induced by short (seconds) and long (minutes or tens of minutes) seeing changes, see for instance Rigaut et al. (1991), Tessier et al. (1994), Christou & Bonaccini (1996a).
It has been found experimentally that in general "short'' exposure IR images (some hundredth of seconds for sufficiently bright sources) have a larger variability in the PSF, while "long'' exposure images (from some seconds to some minutes) benefit from a more stable PSF. During AO observations, to apply deconvolution, the PSF of the system has to be retrieved either from a point source within the field, or from an offset point source observed separately (usually called the PSF calibrator star). Methods to derive the PSF from the AO wavefront sensor (WFS) signal have been proposed and proven successful under certain conditions of SNR (e.g., Véran et al. 1997).
In all of these cases however, the PSF is determined with limited
accuracy. Either because it has been measured at a different time, at a
different
position within the field and with different fluxes on the AO wavefront sensor,
or because (in the case of WFS retrieval) it does not detect small scale
artifacts of the image. In general the PSF is formed by a central core
with Airy rings, superimposed on an extended and sometimes irregular halo.
In the case of the ESO ADONIS adaptive optics system, in use at the 3.6 m
telescope at La Silla, several bumps on the PSF halo are present with
static and dynamic components.
In the images we examine in this paper, the three brightest bumps of variable
intensity (
) and position, due to
uncompensated triangular coma aberration, roughly individuate the deformed
first Airy ring.
Hence to recover the image, methods allowing the use of a relaxed or "best fitted'' PSF in the deconvolution are necessary as the PSF is not accurately measured. This is the case of the Iterative Blind Deconvolution (IBD) technique, which yields at the same time the best fitted PSF and image. But deconvolution belongs to the class of unstable inverse problems, i.e., the solution by any classical method of the linear system discretizing the convolution integral equation is not continuously dependent from the data even if the PSF is exactly known. Thus every efficient numerical method of deconvolution in order to reach an approximate and stable solution must use statistical or deterministic constraints on the solution itself (such as non negativity and smoothness, maximum likelihood or maximum entropy, minimum norm and so on, see, e.g., Lucy 1994) and these constraints necessarily involve a controllable amount of bias in the solution; see for instance Titterington (1985) or Bendinelli et al. (1986). For IBD to work, since the PSF is less defined, stronger constraints are required. These are imposed not only on the solution (as in the usual deconvolution with well known PSF) but also on the PSF itself. They are based on a priori knowledge or well founded hypotheses, as pointed out by Jefferies & Christou (1993) or Thiebaut & Conan (1995).
The IBD technique has been successfully applied by one of us to Adonis AO data, using the IDAC program developed by Matt Chesalka and Keith Hegge at Steward Observatory. This algorithm starts from a good estimate of the PSF, and not from a blind guess of it, and yields at the same time the best fitted PSF for each frame, and the image. This method has been tested on point sources and on extended objects, with simulations and real case data (e.g., Christou & Bonaccini 1996b, 1997). It is however slow in converging, and not necessarily the more efficient software for all astrophysical observations with AO.
The aim of this work is to show that when dealing with a set of AO images of a close binary star, the magnitudes and the positions of their components may be derived in a faster and more direct way than with IBD.
The method we present is based on a two stage procedure which at first gives photometry and separation of the components by a non linear fit, then gives the detailed PSF by deconvolution of the observed images from the sum of the two delta functions which represents the true intensity distribution.
It has been
applied to 3 sets of 48 images each (in the J, H and K bands
respectively) of the binary star 
Canis Majoris ( CMa), which
were taken with ADONIS at the ESO 3.6 m telescope during a technical run.
Each set is composed by the
first 48 of a series of 200 short exposure images
(
= 0.05 s, time
step
0.74 s) already analyzed by
Christou & Bonaccini (1996b, 1997) by means
of IBD using IDAC.
This case is a particularly difficult one, as the companion is located on the first Airy ring of the H-band image, and difficult to disentangle from triangular coma features.
Copyright The European Southern Observatory (ESO)