2.1

AIROPA concept

AIROPA models the field-dependent PSF using two separable components that contribute to phase aberrations:

Once these two components are characterized, starting from an empirical on-axis PSF, the corrected PSF_final can be written as:^{8, 9}

where PSF₀(r = 0) is the empirical central PSF; PSF_instr(r) is estimated through a grid of phase maps in relation to the position on the detector (r); PSF_atm(r) is estimated with AIROPA atmospheric code.^5–7 This factorization allows to use the on-axis PSF₀ (with no anisoplanatism and no instrumental aberrations) as a reference for PSFs throughout the field of view (off-axis).

AIROPA has been applied to NIRC2 and is being tested with simulations and on-sky data (Witzel et al., in prep.). This approach is very well suited to solve the OSIRIS PSF estimation problem: adapting AIROPA to OSIRIS allows to correctly make use of the imager that functions in parallel to the IFU.

2.2

Instrumental aberrations

We derived the phase aberrations as a function of field location using phase diversity data. These are obtained by taking a sequence of out-of focus images on the detector with a fiber light source. This measurement is repeated at several positions covering roughly the whole detector field resulting in a grid which samples the static instrumental aberrations (phase maps grid,⁶ Witzel et al. in prep.).

For OSIRIS we characterized the instrumental aberrations of both the imager and IFU in the Spring of 2017. For the imager, we took a grid of nine by nine locations across the detector (see Fig. 2).

Figure 2.

Grid of fiber images taken on OSIRIS imager showing the locations where the instrumental aberrations were mapped on the detector. The image show a superposition of all the in-focus fiber images.

At each detector position we took images at 2.5, -2.5 and -5 mm out-of-focus. The dataset is taken at 2.173 μm and 20 mas plate scale. We made this measurements on the previous OSIRIS imager detector before it was replaced at the end of 2017. The new detector is already in place and almost ready for use and we will be acquiring the same dataset in the near future to be able to characterize it.

For the IFU we only considered the central position since the field of view is small and we do not expect much variations. We took images at 2.2, -2.8 and -5.3 mm out-of-focus as shown in Fig. 3. We took the data at 20 mas platescale (the one usually used at Keck to do image sharpening procedure) and in Kn3 band (the preferred band for Galactic Center observations).

Figure 3.

Phase diversity data on the OSRIIS IFU. Fiber images at 2.2 mm (left), -2.8 mm (center) and -5.3 mm (right) out-of-focus positions.

The phase diversity data is fed to a software that uses the Gerchberg-Saxton method,¹⁰ an iterative algorithm for retrieving the phase. This is the same method used at Keck for image sharpening. The only differences are that we used higher signal-to-noise data, a larger number of iterations of the code, and more carefully used the random number generator.

An example of two phase maps retrieved on the imager is shown in Fig. 4. The imager detector shows only small variations in phase aberrations across the detector.

Figure 4.

Examples of phase maps at different positions onto the imager detector of OSIRIS.

The OSIRIS imager and IFU phase maps are compared to a phase map previously obtained on NIRC2 detector on Fig. 5. The wavefront errors of the three phase maps are compared in the following Table:

Figure 5.

Phase maps at the OSIRIS imager’s upper-right corner (left) and at the IFU’s center (center). On the right NICR2 central phase map is shown for comparison. All phase maps are show at the same scale and color-bar.

The OSIRIS IFU shows a much higher wavefront error that the other detectors, and an higher value than expected for the instrument. This might be due to the dataset and needs more investigation.

3.1

Procedure to predict the IFU’s PSF from imager

The procedure to obtain the IFU PSF starting from the imager involves several steps and can be briefly describes as follows:

Figure 6.

Simplified illustration of the procedure to apply AIROPA-IFU using 2017 Galactic Center data as an example.

The same process can be repeated for any wavelength channel of the IFU. Fig 6 (bottom left and center) shows the predicted IFU PSF at two different wavelengths, as applied to Galactic Center data obtained in 2017.

3.2

Wavelength sampling

Predicting the IFU PSF requires the ability to predict the PSF at multiple wavelengths. The Kn3 band has 433 spectral channels, while Kbb band has 1665 (these are the filters most frequently used for Galactic Center observations). Running the algorithm over 1000 times is not very efficient. We investigated how fast the PSF changes for different spectral channels (taking into account diffraction and atmospheric effects) in Fig. 7. We examined the difference between a PSF predicted at the first spectral channel of Kbb band and a PSF predicted 10, 50, 100 or 1000 channels away. Across 10 channels the PSF shows very little variation, across 100 channel the residuals start to show significant variations. We find that sampling every 50 channels seems to be sufficient to account for the PSF variation with the wavelength in this context.

Figure 7.

Residuals of IFU’s PSF sampled at 10, 50, 100, or 1000 channels. All images are shown at the same scale and color-bar.

3.3

Test AIROPA-IFU on Galactic Center datacube

We tested AIROPA-IFU on Galactic center data to evaluate its performance. In particular, we used data obtained with OSIRIS on May 18th, 2017. The IFU dataset has been acquired with 15 minutes exposures in 35 mas platescale and using the Kbb filter, which covers the wavelength rage 1.965–2.381 μm. The corresponding field of view is 0.665 × 2.24 arcseconds. The raw data were reduced with OSIRIS pipeline.¹¹ The parallel imager took 10 exposures of 2 seconds and 10 co-adds each, with the Kn3 filter. The imager pixel-scale is 20 mas. The 10 exposures were combined and the final field of view is 20.4 × 20.4 arcseconds.

The dataset is centered on a faint star S2-36,¹² located very close to a brighter star GCIRS16 CC. Due their proximity, S2-36 spectrum is dominated by GCIRS16 CC halo and it is very hard to extrapolate the fainter star spectrum without a good knowledge of the PSF halo. The standard procedure used to extract spectra in OSIRIS data is to simply use a circular aperture since we do not have any accurate knowledge of the PSF.

With AIROPA-IFU we address this problem and the parallel imager is used to predict the desired PSF. We apply the method as described in the above sections an use the predicted PSF to fit the IFU data (Fig 8, left) to model the observations (Fig 8, center). For comparison, we also fit the same data with a multiple 2D-Gaussians model without any knowledge of the PSF shape (Fig 8, right). The model obtained with the AIROPA-IFU well reproduces the observations: both stars are recovered, the PSF shapes are very consistent and have a well-reproduced cores. The 2D-Gaussian model is incapable of recovering the fainter star and is inadequate to describe AO data.

Figure 8.

Application of predicted PSF to spectro-imaging data of two close stars at the Galactic center: spectroscopic observations (left) can be modeled through the AIROPA-predicted PSF (center) and compared to multiple 2D-Gaussians profile fit (right).

However, the S2-36 spectrum extracted with the newly predicted PSF is still dominated by the brighter star spectrum. We interpret this as an indication of the fact that the PSF model is still not perfect.

On the other hand, the spectral extraction is showing significant improvement for very bright stars. Fig. 9 (left) compares the spectrum of the brightest star in the field (GCIRS16 C) extracted with a circular aperture to the one extracted with the newly predicted AIROPA-IFU PSF. The signal-to-noise ratio improves of 50% when using AIROPA-IFU. On Fig. 9 (right) the same comparison is shown for the fainter star GCIRS16 CC. In this case the spectrum shows some unusual artifacts. This, together with the difficulty in recovering S2-36 spectrum, shows that the method still needs testing and refinement, especially for faint stars.

Figure 9.

Comparison of star spectrum extracted with an aperture (black) and with the AIROPA-IFU predicted PSF (red) for a bright star (left) and a fainter one (right).

Some of the possible problems might reside in the IFU phase map that, as previously discussed, still need investigation. We will be acquiring a new dataset in the near future to assess whether the phase maps are accurate. There is also some uncertainty on the precise relative orientation of imager and spectrograph that we are currently addressing. The on-axis imager PSF extraction also need some refinement. Finally the code itself still needs some more testing and these are preliminary results. However the significant improvement in signal-to-noise ratio for the very bright star already shows the potential of this approach.