The BAR5 aperture location on the 50CORON aperture has a demonstrated inner working angle of 0.25" and has achieved reliable high contrast results for bright circumstellar disks. As part of an effort to push the limits of contrast with STIS, the STIS team has investigated the BAR5 location for high contrast by conducting an aggressive set of observations designed to achieve a working contrast limited by the photon noise in the wings of the STIS point spread function (PSF) at the BAR5 inner working angle with a goal of obtaining contrasts of 10-6 or better. All data obtained in this special calibration program (14426) have executed successfully and are publicly available from MAST.
The program is designed to observe the star HD 38393 over the course of nine orbits. Each orbit is at a different spacecraft orientation, and the orbits are arranged in three sets of three continuous orbits separated by roughly one or two months. Each orbit consists of several exposures that are dithered on a 3x3 point grid with a spacing of 0.25 pixels to mitigate uncertainties due to pointing and slit repeatability. The first set of observations executed on October 21, 2015. This dataset is ideal for testing other post-processing methods. Interested members of the astronomical community are invited to improve upon our results and can contact the STIS team at the HST Help Desk if they are interested in making their techniques available to the wider high contrast imaging community.
Our initial results from this program demonstrate that better than 10-6 contrast is possible across the STIS bandpass at angular radii >0.6” and better than 10-5 contrast is possible at angular radii >0.25”. This is sufficient to detect most stellar and substellar companions to solar-type stars and potentially even some young exoplanets, given STIS’ sensitivity at wavelengths out to ~1 micron. It will be sufficient to observe compact circumstellar material around various types of stars and nebulosity around compact galaxies. This technique is appropriate for targets where complementary total intensity visible information is needed with ground-based extreme AO coronagraphs, or where such coronagraphs cannot operate with comparable performance, such as in the northern hemisphere or for primary targets with R>10-11.
For this calibration program, the STIS team chose to execute sub-pixel dithers in a 3x3 grid. The reasons for this were two-fold. First, dithering behind a coronagraphic mask works to reduce residual speckles when the mask is sensitive to miscentering. Secondly, it allows for an operational test of similar dithering that will be used with JWST coronagraphs. For STIS and BAR5, small mis-centerings on the order of 0.25 pixels (~12 mas) are possible, which can cause significant differences in the amount of light leaking around the edge of the BAR5 mask. For classical PSF subtraction, this can allow better matching between a target star and reference star, while for post-processing (such as with LOCI or KLIP) it allows needed diversity in the PSFs. This principle was first developed in response to the tendency of coronagraphic masks on the MIRI instrument to degrade performance with the expected target acquisition accuracy of JWST (Soummer et al., 2014). Simulations of the MIRI response to sub-pixel grid dithers showed up to a factor of 10 improvement over raw contrast thanks to dithering. Being able to test this algorithm on-sky before the launch of JWST will provide a useful test of this idea.
Sub-grid dithering, however, makes accurate centroiding of all the images important. For STIS, this is straightforward. Because the masks and Lyot stop in the instrument are not optimized, significant diffraction spikes from the telescope support structure remain in coronagraphic images. While this is not ideal for imaging purposes, it does serve as a unique external probe of the stellar centroid. The diffraction spikes are nearly at 45 degree angles from the vertical direction on the detector. Therefore, it is possible to measure the vertical location of the spike on the detector as a function of detector column and linearly fit the spikes to infer a stellar centroid. If one subtracts off the PSF halo, and uses a mask that encompasses only the diffraction spikes, accurate centroids can be obtained with this method, typically with uncertainties of 0.05 pixels, or 2.5 mas.
A comparable method has also been implemented by JHU graduate student Bin Ren, who has been working on the 14426 calibration program by implementing KLIP post processing on the data. He has implemented a Python script to do Radon transform centroid finding (as detailed in a paper led by Laurent Pueyo) He writes:
centerRadon is a python script which finds the center of stars based on the Radon Transform. The code starts searching the center from an estimated location, then for each position, it sums over different angles (i.e., making use of the diffraction spikes) and creates a cost function, where the center of a star should have the maximum intensity. The default values are set for the STIS coronagraphic images of Hubble by summing over the diagonals, i.e., 45° and 135°, but it can be generally applied to other instruments--see the docstrings for detailed explanation of the functions.
Bin's code is available on github. The STIS team has verified that the radon centering matches the diffraction spike centroiding results within the uncertainties of both methods.
The figure above shows the measured centroids using centerRadon in STIS detector pixels of the 14426 program for all nine orbits. The non-repeatability of the stellar position is on the order of 0.25 pixels. Additionally, there also appears to be a slight non-repeatability in dithering (on the order of 0.1-0.2 pixels).
How to Implement Sub-Pixel Dithering in APT
Because the BAR5 finger is angled, dithering is somewhat more complicated than the default STIS dither patterns. To fully see how to implement a 3x3 sub-pixel dither, one can download the 14426 Phase II, or retrieve the phase II within APT. In any case, the relevant area is in the creation of a new pattern. If you click on the "Patterns" item on the APT Phase II sidebar, you can select a new pattern: "Line", then fill out the pattern object as follows in this screenshot:
The key points are that the bar is angled at a position angle with a position angle on the detector in the POS-TARG frame with an angle of 78.245 degrees, and you want a vertical LINE subpattern that is oriented 90 degrees to that angle executing for each dither point in your main LINE pattern. You then want to dither +/-0.25 pixels (or 0.01269") away from the target center point. It's also important to remember to set the initial POS TARG to be X=-0.0150 and Y=-0.0098 to ensure that the star initially lands on the lower left dither point, since we did not choose to center the pattern. Here's a screenshot of the special requirements pane of one of the 14426 visits:
Prior to Cycle 24, to use the BAR5 position you needed to specify a POS-TARG relative to the BAR10 position in APT, because the BAR5 location was not a supported position in APT. For current versions of APT, one can directly select the BAR5 position. Please be aware that instructions for selecting BAR5 within APT will be available in a version later than 24.0. Direct any questions about using BAR5 in APT to the HST Help Desk .
Classical ADI performance
PSF subtraction for each dither position within each visit was done using adjacent orbits. For example, visit 4 used either the visit 5 or visit 6 image at the same dither point as a PSF reference. The better subtraction was determined by eye and a mask for each dither point was constructed. All nine separate spacecraft orientations were de-rotated and combined. This procedure is most reminiscent of classical azimuthal differential imaging (ADI), and is particularly well suited for STIS. STIS' bandpass is unfiltered and is entirely due to the quantum efficiency of the detector--thus PSF subtraction with STIS is often critically dependent both on the relative color of the host star to its reference as well as to changes in telescope state. In the case of 14426, our use of the host star as the reference guarantees that the only subtraction residuals we observe are due to changes in the telescope. If one uses a reference star, contrast performance may be degraded depending on how much the reference star departs from the target star’s SED across the STIS passband (0.2-1.0 micron).
To test the STIS band contrast performance of our observations with ADI, we inserted artificial point sources using calculated TinyTim PSFs assuming a G spectral type SED as a function of radius and azimuth ranging from 0.25" to 1.2". We scaled the point sources relative to a source with the equivalent of a total aperture corrected V band magnitude of 26.05 (~1 ct/s for GAIN=4 with STIS). To determine if a point source was recovered, we obtained aperture photometry at its location and determined whether the source was detected at a S/N >5, using an annulus at the radius of the source to determine the estimated noise. At each radius we calculated the median contrast obtained. We compared our 9-roll results with 3-roll and 6-roll only results to gauge the improvement of increasing the number of orientations. One benefit is the increase in available azimuthal coverage and final inner working angle, since diffraction spikes and the BAR5 occulter need to be masked out and degrade performance at the edge of the bar. Additionally, on average the contrast improved close to the expected photon-limited aimprovement of increasing the exposure time, which demonstrates that a contrast benefit is also obtained for multiple spacecraft orientations. This was not necessarily true for the innermost angular separation, potentially because of small number statistics.
The 810 total images (9 orientations x 9 dither positions x 10 CCD readouts) were aligned using centerRadon, and were grouped into three mutually exclusive sets of 270 based on their vertical detector position: 697.50, 698.25, and 699.0 (See Figure 1). To minimize the influence from the BAR5 occulter and residuals from the diffraction spikes, three separate masks were created on 87 x 87 sub-arrays of the data.
A reference library of PSFs were generated for each vertical grouping and spacecraft orientation from the other spacecraft orientations to determine the contrast achieved within 12 pixels (~0.6”). Following the KLIP algorithm we generated model PSFs based on the most common principle components of the STIS quasi-static PSF and subtracted these from the data. We injected artificial point sources to determine the SNR=5 contrast level for a range of azimuths, using the first 50 KLIP modes. Exterior to 0.6”, we used all images together and selected the 170 images with the most similar PSFs from other spacecraft orientations and used these as the final reference library. In this case, the 170th mode was used to subtract off the PSF and artificial scaled point sources were also injected and recovered to determine contrast.
The following table and figure show the results compared to the raw PSF wings, the ADI reduction, and the photon limit of the PSF wings. The photon limit was calculated by assuming a 3x3 photometric box and the assumption of Poissonian statistics for a total exposure time of 180s, using the observed azimuthally averaged PSF wing flux in e-/s at each radius.
|Pixel||Angle (arcsec)||ADI Contrast||KILP Contrast||Photon Limit|
Comparison to Ground-based Extreme-AO coronagraphs
We compare our results to published performance of the Gemini Planet Imager (GPI) and Subaru ScEx-AO to demonstrate our results relative to existing ground-based telescopes. It is important to note that STIS primarily is sensitive to wavelengths complementary to the NIR, where these two instruments conduct observations. Thus, while contrast performance may be comparable or better, STIS’ sensitivity to a given object depends sensitively on its optical SED. The closest instrument to STIS on the ground would be VLT’s SPHERE/ZIMPOL instrument, which can conduct high contrast imaging in polarized visible light from the southern hemisphere. However, ZIMPOL’s performance has not been explicitly published at this time. GPI has successfully detected, at a SNR=8, the companion to 51 Eri in the H band at a contrast of 1.6E-6, implying a SNR=5 contrast of 10-6 at an angular radius of 0.45”. STIS obtains a similar contrast level, to within a factor of 3 at this separation. SCExAO’s published contrast capabilities in the northern hemisphere for H band obtain 10-5 contrast from 0.16-0.35”, comparable to STIS’ contrast performance. Exterior to 0.5”, STIS+BAR5 exceeds the reported performance of SCExAO’s H-band performance.