The GI2T has been described in two recent papers
(Mourard et al. 1994a; Mourard et al. 1994b, M94b hereafter). In stellar interferometry,
the multi-dimensional coherence volume of the incoming wavefront must be
correctly sampled in order to maximize the signal to noise ratio of
high angular resolution collected data. GI2T uses two 1.5 m primary
mirrors much larger than the r0 Fried's parameter in the visible
(Fried 1965) characterizing the spatial coherence area. In practice, the
light beams from GI2T's telescopes are recombined in an image plane after
output pupil remapping. The multi-speckle interferograms feed a
spectrograph whose entrance slit is one speckle wide and about 10 speckles
high. A high magnification of these interferograms on the focal detector
is done in order to correctly sample the interference fringes. In
addition, the exposure time of the detector must be short enough to
correctly sample the temporal variation of the atmospheric turbulence,
i.e. its coherence time. Besides, the multi-r0 operational mode of
the GI2T results in an auxiliary dimension of the coherence volume related
to the spectral correlation of the fringes as a function of wavelengths.
This limits in practice the total bandwidth of the interferograms to a few
nanometers (Berio et al. 1997). These overall constraints demand a fast
detector with a large number of pixels. The CP40 camera (Blazit 1987)
was built in the 1980's for the purpose of speckle and long-baseline
interferometry with large apertures as for the GI2T. It is a 2-stage
intensified photon-detector, with the output phosphorus screen coupled to a
mosaic of 4 Thomson 
CCD's which are readout at the
standard TV rate of 20 ms. This camera is installed at the focal plane of
the low-resolution spectrograph of the GI2T beam-combiner (
).
Among the problems that one may encounter with ICCD cameras, the PCH
has dramatic consequences in the estimation of fringe visibility. For the GI2T (M94b), we estimate
the visibility as the ratio of high frequency to low frequency energies in the average
spectral density of short exposures. In practice, the first step of data processing
consists in computing the average two-dimensional AC of these short exposures.
The PCH appears at the center of the AC. The shape of this hole is not stationnary in time and depends
strongly on the average number of photons per short exposure. Based upon data collected on more than
10 stars with different visual magnitudes, we have checked
that the PCH converges to a 2D inverse gaussian function extending out to
points (Fig. 1 (click here)) for large numbers of averaged AC.
It is superimposed to the top of the fringe pattern which is symmetrized in the
AC process. In a second step the power spectrum is obtained by Fourier
transforming the AC. This transformation dilutes the PCH artifact
over the whole spatial frequency domain where a large fraction of its energy extends
beyond the cut-off frequency of the interference signal.
Figure 1: The central part of the averaged AC of photon-noisy
multi-speckle data recorded on 
Cephei with the GI2T. The
central depression corresponds to the PCH artifact
Our method aims at fitting a 2D polynomial function to the Fourier transform of the
PCH in order to reconstruct the original unbiased power spectrum. Note that no assumption
is made on the exact shape of the PCH Fourier transform.
The fit uses the noise background, made of the photon noise and the PCH Fourier transform,
over the large frequency domain beyond the fringe peak support.
This is more efficient than in the direct space because the PCH has a narrow
support at the center of the interferogram AC.
As already mentionned, the power spectrum of a diluted pupil such as the GI2T one includes
3 components: a low-frequency contribution (the sum of the AC of each aperture)
and two symmetrical high-frequency components (the CC of
apertures). The adopted correction method is based on masking these
components, to fit a 2D polynomial distribution to the noise
background and to substract it from the raw power spectrum, in order to obtain
the original photon-unbiased power spectrum. As a by-product, one corrects also for the photon-bias
in the averaged power spectrum.
In practice, the masked parts represent only a few percents of the power spectrum and besides we assume
the continuity of the Fourier transform of the PCH, thus the fit
interpolation in these parts is correct.