The Photometric System
The photometric camera for the SDSS consists of two TDI scanning CCD arrays, one, using 30 Tektronix/SITe 2048 x 2048 CCDs in a 5 by 6 array for five-color photometry, and the other using 24 2048 x 400 chips for astrometry and focus monitoring. We here discuss these arrays, the filter system, the scheme for focusing the telescope during an imaging scan, and the CCDs and associated electronics for the camera and the spectrograph.
There are several important considerations in selecting the filter system. First, one wants the largest color baseline possible. It is desirable but not necessary to use filters for which there already exist a large amount of data. For example, the Thuan-Gunn (1976), (see also Schneider et al. 1983) uvgriz system, which has been used for much of the work at Palomar on distant clusters, avoids the 5577Å night sky line, and the i band cuts off in the near-infrared just shortward of the major OH bands. The g and r filters more-or-less match the J and F bands of the new Palomar Sky Survey. The uvgriz filters provide coverage of the total wavelength interval and are almost uniformly spaced in log lambda . They are, however, fairly narrow, and all except u could be usefully widened for a system like ours in which efficiency is of paramount importance. We have designed a filter system similar to ugriz which should transform to and from it with little difficulty, in which the g , r and i filters are as wide as practicable consistent with keeping the overlap small. If we wish the u band to be a “good” one which is almost entirely contained between the Balmer jump and the atmospheric cutoff (we can clearly not do anything about that end), it cannot be significantly wider than Thuan-Gunn u . We call the new system u’ g’ r’ i’ z’ . The response curves and sensitivity data with the coatings and CCDs we will use are shown in Figure
The SDSS system response curves. The responses are shown without atmospheric extinction (upper curves) and as modified by the extinction at 1.2 airmasses (lower curves). The curves represent expected total quantum efficiencies of the camera plus telescope on the sky.
The characteristics, are, in brief: u’ , a band with 3540Å effective wavelength, 570Å full width at half-maximum, and 60% peak transmission (such a filter can be made with available glasses; it has a small redleak problem which can be solved with suitable coatings); g’ , a very wide blue-green band centered on 4770Å (full width at half-maximum 1370Å); r’ , a red band of the same width centered at 6230Å; i’ , a far-red band centered at 7630Å with a width of 1530Å, and z’ , a near-infrared band centered at 9130Å with a width of 950Å. These bands share the advantages of the uvgriz system of avoiding the 5577Å [OI] line, and the i’ band cutting off before the catastrophic OH brightening and water absorption set in just longward of 9000Å. All the filters except u’ are easy to make and have very high central transmission; building the u’ filters proved to be difficult, but the results are quite satisfactory.
The filters are all made with one or more Schott colored glass elements in combination with a single-surface interference film which serves different purposes in different filters. The u’ filter is made of 1 mm of UG11 and 1 mm of BG38 with an interference layer which both acts as an antireflection coating in the filter passband and an efficient red blocker in a region a few hundred Å wide around 7000 Å; the glass filters have sufficient red rejection outside this region, but the uncoated filter has a peak transmission of about two percent there. The coated filter has a maximum transmission of about 10-4 outside the primary band. The g’ filter is made of 2 mm of GG400 and 3 mm of BG38, with a film which acts as a short-pass element cutting on at 5500 Å; the BG38 blocks the red efficiently out to the silicon limit. The three longer filters consist of a single colored long-pass element and for the two shorter ones a short-pass interference film, the r’ 4 mm RG550 and short-pass cutoff at 7000 Å, the i’ 4 mm RG695 and short-pass cutoff at 8500 Å, and the z’ 4 mm of RG830 open to the red. Each filter is brought to 5 mm physical apex thickness by means of an (quartz for u’ , BK7 for the rest) individual field flattener cemented to it; this element is necessitated by the fact that the chips are not flat, but are convex to the incoming light by about 230 microns measured center to corner. The final filters were all constructed and coated by Asahi Spectro-Optical in Tokyo; the characteristics and uniformity are superb.
The filter assemblies in turn are cemented to the flat rear face of the final corrector element. The front face of the corrector has been anti-reflection coated in horizontal stripes tailored for each filter passband, resulting in about 0.2% reflectivity over each band. The coating was done by QSP in Los Angeles. Since the filters are cemented assemblies and are themselves cemented to the corrector, there are only two optical surfaces near the CCD, and one of those has very low reflectivity. The other is the filter coating itself, which has low reflectivity in the passband but unfortunately has intrinsically high reflectivity near the long-wave cutoff over a very small wavelength region. It is these “crossover” regions which contribute most of the ghost energy. We have calculated ghost images for the camera, and they will not present a serious difficulty for the photometric reductions.
With our arrangement of CCDs operating in TDI mode at sidereal rate, we obtain near-simultaneous color data in the five bands; the time lapse from the r’ data, the first received, to the g’ , the last, is about five minutes.
It is worth noting that this color set is not any standard one, but this survey will produce two orders of magnitude more photometry than currently exists, and to adopt a system not optimized for the survey seemed folly. The filter set is in any case a good one for CCDs and has some quite nice astrophysical properties; we hope that it will quickly become a standard system. Toward that end we are in the process of producing as complete an astrophysical characterization of it (metallicity, gravity, temperature, galaxy type and redshift dependences) as we can; we clearly badly need these data ourselves. We will attempt to ensure that workers wanting sets of filters for this system will have reasonable avenues open to obtaining them.
With these filters, the CCDs with which they will be used, and with the SDSS telescope scanning at sidereal rate, the limiting magnitudes at the detection limit, S/N 5:1, will be about u’=22.3 , g’=23.3 , r’=23.1 , i’= 22.5 and z’ = 20.8 . These limiting magnitudes are for stars, note, and those for galaxies are typically between half a magnitude and a magnitude brighter at the same signal to noise ratio. A signal to noise ratio of 50:1 (and hence photometry at the 2% level) is reached for stars at 19.3, 20.6, 20.4, 19.8 and 18.3 in the five bands. Note that these are magnitudes on the AB system, referred to a spectrum with constant fnu (not an A0V star), and represent the fluxes a flat spectrum source with a given V magnitude would have near the effective wavelengths of the filters. It is these zero points we intend to adopt. These sensitivity calculations assume 1 arcsecond FWHM seeing (with a probably somewhat pessimistic non-gaussian PSF) and a sky with a V surface brightness of 21.7 mag/arcsecond² (typical of the Apache Point site) with average extinction at the zenith. A darker site helps, with roughly half a magnitude gain in limit per magnitude improvement in the sky (at these faint levels shot noise in the sky is the dominant noise source).
Front view of the SDSS camera assembly. This diagram shows the camera as it would be seen with the front cover and shutters removed, showing the 30 photometric and 24 astrometric/focus CCDs and their associated dewars and kinematic supports.
Although the u’ band is obviously useful, the other bandpasses provide a maximum wavelength ratio of 2.4, and adding u’ only increases the baseline to 3.0. Is there much advantage to using u’ versus a slightly longer wavelength band such as Johnson B or Thuan-Gunn v ? The answer is resoundingly yes, on three accounts. First, the u’ -band is much more sensitive to population and metallicity effects than any of the longer wavelength bands. For example, for normal spiral galaxies, U-B changes 1.5 times as fast as B-V . In elliptical galaxies, color gradients (which most likely reflect metallicity variations) are 3 times as big in U-B as in B-V ; one must reach from B to the K -band to achieve comparable sensitivity. The u’ -band lies shortward of the H and K break in late-type galaxy spectra and provides the best indicator of the presence of recent star-formation activity. Second, at high redshift, galaxies are usually observed in the more accessible and sensitive longer wavelength bands, which register light that, in the galaxy rest frame, is emitted in the ultraviolet. The u’ -band is shifted to g’ at z = 0.34 , r’ at z = 0.83 , and i’ at z = 1.3 . Third, the ultraviolet is essential in any color-based QSO survey to separate low-redshift ( z<2 ) QSOs from stars, particularly low-metallicity halo stars, and, incidentally, to separate low-metallicity halo stars from high-metallicity stars.
We had originally planned to devote two of the five chips in a given column to the u’ band for the sake of efficiency, but new UV anti-reflection coatings and an accompanying surface treatment developed by Tektronix/SITe for the u’ chips, together with a higher-than-expected efficiency of the u’ filter, have led us to go with the current set of five independent bands, including the new z’ filter. The new band is especially useful in the detection of very high-redshift quasars and extremely red stars, and, perhaps most importantly, distinguishing them from each other.
The Photometric Array-Layout and Performance
The photometric camera is designed to use thirty 2048² sensors in a six-wide by five-high array, so arranged that two transit swaths (strips) cover completely a stripe 2.53° wide in the five independent filters, with some overlap (about 1 arcminute) between sensors. About 8.3% of the area of a given stripe is imaged twice. The focal plane of the camera array is illustrated in Figure . Also shown there are the 22 smaller (2048 x 400) chips used to provide astrometric calibration, and the two 2048 x 400 focus monitors. The scanning is top-to-bottom in Figure , along the five closely-spaced CCD columns, and it is handy in discussing the camera to think about the geometry in this orientation. (Thus “top” of the camera will always refer to the leading edge.)
SDSS Filter Set and Photometric Sensitivity
|u’||3543||567||1.24x 10-1||1.85x 10-2|
|g’||4770||1387||4.51x 10-1||1.18x 10-1|
|r’||6231||1373||5.63x 10-1||1.17x 10-1|
|i’||7625||1526||5.02x 10-1||8.74x 10-2|
|z’||9134||950||1.29x 10-1||2.23x 10-2|
|Star saturates at AB||12.1||14.1||14.1||13.8||12.3|
|Eff Sky, mag/sec²||22.1||21.8||21.2||20.3||18.6|
|Sky + bkg count/pxl||45||401||690||1190||1120|
The expected photometric performance of the camera is outlined in Table, which presents the assumed filter parameters, derived from the curves in Figure . The quantity qt is the peak system quantum efficiency in the system, and qtdl/l is the integral of the system efficiency times d( ln lambda ) ; it is this quantity which relates the monochromatic flux averaged over the filter passband to the resulting signal:
For more detail: The Photometric Camera and the CCDs