MuSCAT: a multicolor simultaneous camera for studying atmospheres of transiting exoplanets

We report a development of a multi-color simultaneous camera for the 188cm telescope at Okayama Astrophysical Observatory in Japan. The instrument, named MuSCAT, has a capability of 3-color simultaneous imaging in optical wavelength where CCDs are sensitive. MuSCAT is equipped with three 1024x1024 pixel CCDs, which can be controlled independently. The three CCDs detect lights in $g'_2$ (400--550 nm), $r'_2$ (550--700 nm), and $z_{s,2}$ (820--920 nm) bands using Astrodon Photometrics Generation 2 Sloan filters. The field of view of MuSCAT is 6.1x6.1 arcmin$^2$ with the pixel scale of 0.358 arcsec per pixel. The principal purpose of MuSCAT is to perform high precision multi-color transit photometry. For the purpose, MuSCAT has a capability of self autoguiding which enables to fix positions of stellar images within ~1 pix. We demonstrate relative photometric precisions of 0.101%, 0.074%, and 0.076% in $g'_2$, $r'_2$, and $z_{s,2}$ bands, respectively, for GJ436 (magnitudes in $g'$=11.81, $r'$=10.08, and $z'$=8.66) with 30 s exposures. The achieved precisions meet our objective, and the instrument is ready for operation.


Introduction
Transiting planets, which transit in front of their host stars, are especially important research objects among exoplanets, as when combined with radial velocity measurements, they can provide us various information about the nature of exoplanets such as the mass, radius, density, orbital obliquity, and atmosphere. Most of transiting planets have been discovered by transit surveys which monitor the brightness of hundreds of thousands of stars. Several groups have worked or been working on ground-based transit surveys, [1][2][3] and CoRoT 4 and Kepler 5 have performed space-based transit surveys. Moreover, the second epoch mission of Kepler, namely K2, 6 is now ongoing, and next generation space missions TESS 7 and PLATO 8  High precision multi-color transit photometry is known to be useful for such follow-up observations to discriminate whether transit-like dimming is caused by a true planet or by an eclipsing binary. 9 This is because a true planet is almost dark in all wavelength, while an intervening body of an eclipsing binary is bright itself and its brightness changes significantly with wavelength. Thus false positives caused by eclipsing binaries can be spotted by observing significant wavelength dependence in transit depths. On the other hand, transit depths of a true planet also have wavelength dependence. Most of the wavelength dependence comes from the stellar limb-darkening, but the apparent planetary radius also has weak wavelength dependence which reflects the nature of its planetary atmosphere. High precision multi-color transit photometry is known to be useful to measure the weak wavelength dependence in transit depths to study atmospheres of transiting planets. This kind of study is known as transmission spectroscopy, and numbers of multi-color transit observations for this purpose have been reported so far. [10][11][12][13][14][15][16][17] Multi-color simultaneous cameras are very fruitful for the studies described above for two reasons. First, multi-color simultaneous cameras provide not only higher efficiency but also more feasibility to accomplish aimed studies than single-color cameras, since observable transits from a specific ground site are very limited. Second, simultaneity of multi-color transit photometry is important to avoid systematic differences of transit depths due to luminosity change in host stars possibly caused by existence of starspots, plages, stellar activity, and so on. For the reasons, multicolor simultaneous cameras such as GROND, 10,11 BUSCA, 12 ULTRACAM, 13 SIRIUS, 14,15 and MITSuME 16,17 have been actively used for transit observations.
Considering the fact that more interesting transiting planets will be discovered in the near future by advanced ground-based surveys, and also space-based surveys, like K2, TESS, and PLATO, developments of new multi-color simultaneous cameras are highly desired. We here report a development of such an astronomical instrument named MuSCAT (Multi-color Simultaneous Camera for studying Atmospheres of Transiting planets), which is now installed on the 188cm telescope at Okayama Astrophysical Observatory (OAO) in Japan.
The rest of this paper is organized as follows. We first describe designs of the optical system of MuSCAT and its components (Sec. 2), and introduce the control system of MuSCAT (Sec. 3).
We then report characteristics and performances of MuSCAT shown in engineering observations (Sec. 4). We discuss on some capabilities for future upgrade of MuSCAT (Sec. 5), and finally summarize this paper (Sec. 6).

Scientific Requirements and Design Policies
We have designed MuSCAT considering the following conditions. As we plan to use MuSCAT for validations of transiting planets discovered by transit surveys, at least 2 colors are necessary to discriminate eclipsing binaries from transiting planets. Considering the cost and available research grants, we adopt a design for a 3-color simultaneous camera with the 3 colors in optical wavelength where CCDs are sensitive. For transit observations, it is important to obtain good comparison stars in the field of view (FOV) to achieve high precision transit photometry. For the reason, we have designed the MuSCAT FOV as wide as possible for the 188cm telescope. We have also took care of the throughput (TP) of the instrument to achieve high photometric precision. To achieve higher sensitivity to the utmost extent, we have carefully selected and designed the MuSCAT optical system including astronomical bandpass filters, dichroic mirrors, and CCDs.

Optical Design
Layouts of the optical system of MuSCAT are shown in Fig. 1. MuSCAT adopts a 45 • plane mirror and an Offner relay, which consist of SiO 2 -protected aluminum mirrors, and inserts 2 dichroic mirrors in the light path to take images of 3 bands in optical wavelength simultaneously. MuSCAT maintains space for another dichroic mirror around the Cassegrain focus so as to accommodate near-infrared (NIR) channels as a future upgrade capability. For high precision transit photometry, it is desired to have the widest possible FOV to get good comparison stars. The Offner relay enables us to distribute F conversion lenses around the relay. The lenses contribute to achieve wider FOV by converting F-number from F18 at the telescope to F5.5 before the relay and further converting to F4.0 just before CCD cameras after the relay. Aberration correction is performed by 4 lenses before and 1 lens after the Offner relay, and finally accomplished by a plane-convex lens just before CCD cameras. Anti-reflection coating for optical wavelength is applied to the lenses. The lenses are designed to correct aberration of the whole system including both the 188cm telescope and MuSCAT. Thereby the optical system of MuSCAT provides a good imaging quality throughout the FOV. Wavelength divisions are performed by 2 dichroic mirrors after the Offner relay. The dichroic mirrors are wedge-shaped to reduce astigmatism. Astronomical bandpass filters are inserted just before the last lenses. Lights of astronomical objects are divided into 3 colors and detected by 3 CCD cameras. A picture of the actual MuSCAT installed on the 188cm telescope is presented in

Dichroic Mirrors and Bandpass Filters
The 2 dichroic mirrors are manufactured by Asahi Spectra Co.,Ltd. The size of the first dichroic mirror (DM1) is 113 mm by 108 mm, and the depth of 11.8 mm with the wedge angle of 7 min 52 sec. The size of the second dichroic mirror (DM2) is 90 mm by 88 mm, and the depth of 9.9 mm with the wedge angle of 12 min 33 sec. Anti-reflection coating is processed on the back sides of the dichroic mirrors so that the DMs transmit remaining lights (namely, not reflected ones) almost completely.

Cameras and Detectors
MuSCAT equips 3 CCD cameras manufactured by Princeton Instruments. The first one is a PIXIS: 1024B model, used as the ch 2 CCD camera in r is applied to the vacuum windows of CCD cameras and its transmittance is shown in Fig. 6. The data of the QEs and the transmittance of the BBAR coating are provided by the manufacturer. We summarize nominal specifications of the CCD cameras in Table 1. We also present actual measured values of gains, full well, and read noise as well as an upper limit of dark current based on data taken during engineering observations (Sec. 4) in Table 1.

Total Throughput
Efficiencies of the F conversion optics and the Offner relay are roughly estimated as 60% in g ′ 2 band, 61% in r ′ 2 band, and 50% in z s,2 band. Based on the transmittance and reflectance of DMs, filters, BBAR coating, and QE of CCDs in the previous subsections, we calculate expected total throughput of MuSCAT. We plot the wavelength dependence of the expected total throughput of MuSCAT in Fig. 7 and present machine-readable values in Table 2. A comparison of the expected total throughput with a measured one is presented in section 4.3.

Spot Diagram
Simulated imaging performances of the MuSCAT optical system are shown in spot diagrams in Figs. 9 and 10. Fig. 9 plots spot diagrams for on-focus cases which indicate spot radius of all wavelength are well less than 1 arcsec throughout the current FOV and the potential FOV.
This imaging performance is thus sufficient for the Okayama Astrophysical Observatory where the typical seeing is about 1.5 arcsec. Fig. 10 shows spot diagrams for defocused cases where the secondary mirror is shifted by 1.5 mm, which makes spot radius expand to about 4 arcsec. It is well known that defocusing is very useful for high precision transit photometry for isolated sources, 18 and thus those cases are more realistic for transit observations. The panels imply that images are almost circular throughout the FOV and suitable for aperture photometry.   measurements. For the other information than the time, PC1 refers to a telescope-control PC, which gathers all the up-to-the-second information.
PC2 provides a user interface. Observers can send an observing command to each camera from this PC specifying the exposure time, number of images, gain and readout speed settings, and so on. We note that one can select optimal settings for each camera. Namely, one can set a images are instantaneously sent to PC2 to be displayed in SAOImage DS9. PC2 also has a function of self autoguiding (see Section 3.2), and has a data storage.

Self Autoguiding
To achieve a 0.1% level high precision relative photometry, it is essential to receive stellar fluxes by the same pixels during a set of observations to mitigate the incompleteness of the pixel-to-pixel sensitivity correction. Because the tracking of the 188cm telescope is not perfect, an autoguiding system is critical to keep the stars on the same positions on the detectors. However, MuSCAT has no guide camera to capture a guide star located in the surrounding area of the FOV of the science cameras. We therefore have developed a self autoguiding system which uses scientific images for guiding. [14][15][16][17] Specifically, the stellar centroids of several bright stars are measured on one of the three band images soon after the image is obtained. Then the mean stellar displacement on the latest image relative to a reference image is calculated to feed back to the telescope. All the above processes are done within a few seconds after the end of exposure. We note that the CCD channel to be used for the stellar centroid calculation is selectable. We usually set the exposure time for the guiding channel to 30-60 seconds such that the feedback result will be well reflected in the next image and that the guiding frequency is high enough. As shown in Section 4.4, the autoguiding system can stabilize the stellar centroid positions within ∼1 pixel for bright stars (magnitude less than about 12). This is important to reduce systematic errors caused by incompleteness of flatfielding. We find no large difference in the autoguiding performance between in-focus and out-offocus observations, meaning that the guiding performance is not limited by the degree of defocus but limited by the change of stellar PSF shapes due to seeing variation and the mechanical accuracy of telescope driving (tracking and fine moving).

Results of Engineering Observations
We

Detector Characteristics including Bias, Flat, and Linearity
We have took hundreds of bias, flat, and linearity test frames during engineering observations in order to learn detector characteristics of MuSCAT. We have observed lights of a filament lamp projected onto a matte whiteboard on a wall of the 188cm telescope dome for flat frames and For reference, we present median flat images in g ′ 2 (median of 433 flat frames) and r ′ 2 (median of 490 flat frames) bands in Fig. 12. We note that the unique feature has a very good repeatability with the fractional fluctuation of much less than 0.1%, which has little impact on the photometric precision required for our purposes. We have also exposed a very bright star of V = 6 to check for the existence of image persistence on the CCDs and ghost patterns due to reflection by the lenses.
We found no apparent features that can affect the photometry. We have derived full well values, gains, readout noises for the data. The values are presented in Table 1.
We have also tested the linearity of MuSCAT CCDs for each readout speed and each gain. Our method is based on a previous study for the CCDs of High Dispersion Spectrograph (HDS) of the Subaru telescope. 19 First, we have created linearity test frames which have gradational counts on the CCDs, by opening only a half of the tertiary mirror cover and inserting a black plate into the  light path in front of MuSCAT. Fig. 13 shows an example of a linearity test frame. Second, we monitor counts of the filament lamp until the filament lamp is stabilized. We note that it takes about 2 hours until counts become nearly-unchanged. We then start linearity test exposures as follows.
We first determine an exposure time for each CCD which gives counts from bias level to saturation level gradationally on the CCDs. We define frames with the above exposure time as "A" frames and ones with a half of the exposure time as "B" frames. We then take A and B frames alternately until obtaining 20 frames each. We have repeated such exposures for each gain and each readout speed, namely for the gain modes of 1, 2, 4 e − /ADU, and for the readout speed of 100kHz and 2MHz. Subsequently we subtract a median bias frame for each gain and each readout speed. We then make a new frame which computes photon counts of each pixel in an A frame minus twice of photon counts for the same pixel in a B frame using adjacent A and B frames (39 pairs in total for each gain and each readout speed). We define those frames as "C" frames (namely, C = A − 2 × B for each pixel). To visualize the linearity of the CCDs, we plot electron counts (namely, photon counts × gain) of pixels in A frames as X-axis and electron counts of the same pixels in C frames as Y-axis. An example of such a figure is shown in Fig. 14. We finally fit the plotted data with a linear function (Y = aX) using the data up to X=64000, and the best fit linearity slopes are summarized in Table 3. Based on the above test, we have confirmed that MuSCAT CCDs have a good linearity within ∼0.21% at a maximum up to the saturation level. The result means that the effect of non-linearity is well negligible even for high precision transit photometry if counts of stars do not change drastically during observations. In the case we need to correct non-linearity, we will use those data for non-linearity corrections.

PSF and Distortion
We obtained images of the open cluster M67 with MuSCAT during the first light observation.
On this night, the sky condition was photometric and there was no moon. representative PSF for each grid using stars inside the grid. We also note that the seeing in g ′ 2 , r ′ 2 , and z s,2 was 2.1 ′′ , 1.8 ′′ , and 1.7 ′′ , respectively, which was slightly worse than the typical seeing at the site of ∼1.5 ′′ in optical bands. Thus imaging quality was not limited by MuSCAT itself but by the seeing. We have confirmed that the PSF is nearly circular throughout the FOV and that MuSCAT does not have unexpected large aberration or imaging problems.
We have also derived distortion maps of images as differences in the pixel scale on the CCDs using one of output options of SCAMP. 23 The derived pixel scale distortion maps are presented in Fig. 18. The figure indicates that the pixel scale distortion is limited within about 0.3%, which is negligible for standard aperture photometry.

Sensitivity and Efficiency
We also estimate limiting magnitudes for the g ′ 2 , r ′ 2 , and z s,2 bands using the images of M67. The measured sky brightness in g ′ 2 , r ′ 2 , and z s,2 were 19.9 mag arcsec −2 , 19.5 mag arcsec −2 , and 18.7 mag arcsec −2 , respectively. For each band, 10 × 60 s images were obtained with dithering. We note that we used high speed readout mode (2 MHz). We apply bias-flat correction and stellar position alignment to the data and stacked them into a single image for each band. We conduct photometry for ∼100 stars on each stacked image by using the DoPHOT package, 24 which performs an analytical PSF fitting. The measured instrumental magnitudes of these stars were then compared to the SDSS 9 catalog 25 for photometric calibration. We note that we here ignore color terms and simply approximate that the g ′ 2 , r ′ 2 , and z s,2 bands are identical to the g ′ , r ′ , and z ′ bands, respectively. Limiting magnitudes with 10-min exposure are estimated as the signal-to-noise (S/N) ratio reaches 10, yielding g ′ lim = 21.7, r ′ lim = 21.7, and z lim = 19.8. We note that z lim is affected by higher readout noises in z s,2 band (see Table 1) as we used the high speed readout mode. For the S/N calculation, we simply adopt the photometric errors returned by DoPHOT. We show a plot for SDSS magnitudes v.s. photometric errors in Fig. 19.
In addition, total throughput including the airmass, 188cm telescope, and MuSCAT is measured by the same data. We estimate total throughput as follows. First, we measure the zero-point magnitudes on the stacked M67 images as ZP(g ′ 2 )=28.63, ZP(r ′ 2 )=28.71, and ZP(z s,2 )=27.08, which correspond to 10 electrons for all bands. Next, we estimate the expected incident flux coming from an astronomical object with the above magnitudes into the effective area of the primary mirror of the 188cm telescope. Finally, comparing the expected flux with the detected one (10 electrons), we estimate the total throughput in g ′ 2 , r ′ 2 , and z s,2 bands as 20%, 28%, and 13%, respectively. telescope, and MuSCAT, as shown in Table 4.

High Precision Photometry
In order to check for the photometric performance of MuSCAT, we observed two stars; GJ 436 26,27 and WASP-12, 28 both hosting a transiting planet. Observations were carried out on 2015 March 2 UT, during out-of-transit phases for both targets. There was no cloud, but the sky level was relatively high due to a bright waxing moon with the age (lunar phase) of 11.5. We used the low  Table 5. For each observation, the FOV was adjusted so that several similar-brightness stars were simultaneously imaged. The stellar image was defocused such that the FWHM of PSF became ∼33-38 pixels and ∼31-36 pixels for GJ436 and WASP-12, respectively. The self autoguiding system (see Sec. 3.2) was activated by using g ′ 2 -band images for both targets. The stellar centroid changes in r ′ 2 band during the observations of GJ 436 are displayed in Fig. 20, showing that the stellar positions were quite stable with the dispersion not exceeding ∼1 pixel.
The observed data were reduced by using a customized aperture-photometry pipeline. 31 The applied aperture radius, the number of comparison stars used for the relative photometry, and the unnormalized flux ratio of the target star and the ensemble of the comparison stars are summarized in Table 5. Note that the applied aperture radius was determined such that the dispersion of the resultant light curve was minimum. We show the resultant normalized light curves of GJ436 and WASP-12 in Figs. 21 and 22, respectively. The black dashed line indicates the best-fit linear function. Photometric precisions, which we define as the root-mean-square (rms) of the residual light curve from the liner fit, achieve 0.101%, 0.074%, and 0.076% in the g ′ 2 , r ′ 2 , and z s,2 bands for GJ 436, while those for WASP-12 are 0.16%, 0.16%, and 0.15%, respectively. These rms values are listed in Table 5.
To see how the photometric performance of MuSCAT has been achieved, we calculate the error budgets for these observations as shown in Table 6. In the table, σ target , σ comp , and σ sky indicate photon noises arising from the target-star flux, comparison-star flux, and sky-background flux, respectively, calculated assuming the Poisson (photon) noise. σ read is the noise contributed from the read-out noise listed in Table 1. σ scin is the scintillation noise, for which we apply the following equation, where D is the diameter of the primary mirror of the telescope in cm, Z is the zenith distance, h = 372 m is the height above sea level of the observatory, h 0 = 8000 m, and T is the exposure time in seconds. 32,33 All the remaining (unknown, or difficult to assess) components of the photometric error are treated as σ unknown , which is calculated as where rms is the same as that listed in Table 5. Among these noise sources, all but σ unknown are basically unavoidable. Possible causes of σ unknown can be the difference of atmospheric transparency between toward the target star and toward the comparison stars, the modulation of scintillation noise, the incompleteness of flat-field correction, and so on. We indeed find that σ unknown is the major noise source in some of the light curves, but is still limited in degree of about 30-40% in rms 2 , meaning that σ unknown is not a very limiting factor for the photometric precision. In other words, the most part of the photometric precision ( 60% in rms 2 for all three bands) can be explained by the theoretical noise models. We therefore consider that the expected photometric performance of MuSCAT has been well achieved.
We also note about the time-correlated noise (so-called the "red" noise) in the observed data. For high-precision photometry such as transit observations, a treatment of the red noise would be very important. 34 We calculate a red-noise factor, which is the ratio of measured rms in binned data to the one expected from the rms in unbinned data, for our observations. We find 1.3 in average, which is a typical value for ground-based transit observations. Although one hour observations are not sufficient to evaluate the red noise in detail, we consider that the level of the red noise of MuSCAT is similar to other ground-based instruments. We will thus take into account the red noise for future science observations with MuSCAT.

Upgrading and Transferring Capability
Although the current MuSCAT is ready for operations, it still has upgrading capabilities. First, MuSCAT can be upgraded in terms of FOV by replacing the three 1k×1k CCD cameras, which give 6.1×6.1 arcmin 2 FOV, with 2k×2k CCD cameras, which will provide 12.7×12.7 arcmin 2 FOV. Such a wider FOV would be desirable to find good comparison stars especially for very bright targets which will be discovered by the TESS mission. Second, MuSCAT has space for another dichroic mirror to add NIR channels (see Sec. 2.2). Additional NIR channels enable us to take images from optical to NIR simultaneously like GROND. Such a capability will enhance scientific merits for transmission spectroscopy in the light of efficiency and simultaneity.
MuSCAT also has a transferring capability. The current instrument is optimized for the 188cm telescope at Okayama Astrophysical Observatory whose F-number is F18, but MuSCAT can be transferred to or can make a copy of itself for other telescopes by replacing F conversion lenses.
The field of view of MuSCAT is 6.1×6.1 arcmin 2 with the pixel scale of 0.358 arcsec per pixel.
One of the prime aims of MuSCAT is to confirm whether candidates of transiting planets discovered by transit surveys, including such as K2, TESS, and PLATO, are truly planets or false positives due to eclipsing binaries. Another prime aim of MuSCAT is to measure the wavelength dependence of transit depths in visible bands, providing rough information about exoplanetary atmospheres: such as the feature of the Rayleigh scattering by hydrogen dominated atmospheres, the feature of the Mie scattering by hazy atmospheres, or the flat feature of cloudy atmospheres. 16,35 The capability of multi-color simultaneous transit photometry is well suitable for those aims.
Since MuSCAT can achieve 0.1% photometric precision with 30 s exposure for stars brighter than ∼10 mag as shown in Sec. 4.4, MuSCAT will work effectively for the purposes especially for bright TESS transiting planets. In addition, MuSCAT would be also useful for follow-up observations of supernovae and gamma-ray bursts, and monitoring variable stars, and so on. The instrument is ready for operation at Okayama Astrophysical Observatory. Single pixel ∼70000 e − Selectable readout speed (readout time) 100 kHz (10.0 s), 2 MHz (0.58 s) Measured read noise: @100 kHz 3.9 (ch 1), 3.8 (ch 2), 4.2 (ch 3) e − @2 MHz 11 (ch 1), 12 (ch 2), 24-27 a (ch 3) e − a : An additional pattern (not random) noise is seen only in the 2MHz mode of the ch 3 CCD. This is scheduled to be repaired (the repair has been completed at the time of publication).