SCORE: A Mid‐Infrared Echelle Format Spectrograph with No Moving Parts

NASA's Space Infrared Telescope Facility (SIRTF) is a multiuse infrared observatory that will be equipped with two cameras and a spectrograph. It is scheduled for launch on 2001 December 1. The Infrared Spectrograph (IRS) consists of four modular grating spectrographs. By design, the spectrograph modules have no moving parts. This is a major simplification in terms of fabrication, testing, and operation. Inevitably, this approach places firm constraints on the resolution, slit length, and wavelength coverage that can be achieved. Two of the modules operate at low resolution (R = λ/Δλ = 50) and have long slits. The other modules trade slit length for increased resolution, R = 600. The characteristics of the individual modules are presented in Table 1. The predicted on‐orbit SIRTF performance of the high‐resolution module is shown in Figure 1, as are the predicted and achieved ground‐based performance of SCORE.

Zoom In Zoom Out Reset image size

As a further simplification, the modules are designed to function upon assembly with no adjustments being necessary. In short, they are designed to "bolt and go." A prototype of one of the modules, the high‐resolution, short‐wavelength module, has been built to demonstrate that the required optical performance can be achieved with the bolt‐and‐go philosophy and to serve as a test bed for the development of the data reduction software. In order to support the software development, we have changed the ruling spacing on the predisperser grating to cover just the ground‐based 10 μm window on the array. This demonstration module is called the SIRTF Cornell Echelle (SCORE). It has been used very successfully on the Palomar 200 inch (5 m) telescope¹ and is the subject of the present paper.

SCORE is based on Si:As BIBIB detector arrays developed at Rockwell International (now Boeing). The configuration and current performance characteristics of the 128 × 128 arrays are listed in Table 2. Two orthogonal views of the SCORE optical path are shown in Figure 2. The optical path splits at the slit plane. The light passing through the slit proceeds onto the spectrograph optics. The light reflected by the front surface of the slit plate is reflected to the imaging train. These two paths are discussed separately.

Zoom In Zoom Out Reset image size

2.1. Spectrograph Path

The entrance slit is matched to the diffraction limit, 2λ/D, at 10 μm or 1 ^'' for the 200 inch telescope at Palomar. A broadband interference filter passes wavelengths in the ground‐based 10 μm window. The first grating operates in first order and serves as the cross‐disperser for the second grating, which operates in 15th through 29th order. In this way an echelle format spectrum is formed on the detector array. Neither grating is strictly an echelle since both have blaze angles well below 45°. All of the optical elements with the exception of the slit, the order‐sorting filter, and detector are diamond‐machined aluminum. The diffraction gratings were manufactured by Diffraction Products, while the mirrors were made by Optical Filter Corp. (OFC). Both the collimator and camera mirrors have conic surfaces with surface accuracies of less than 100 Å rms. More detailed information on the spectrograph components can be found in Table 3. The spectrograph housing was fabricated from aluminum so there is a minimum change in the focus position between room temperature and the operating temperature, 4 K. To increase the stability of the optical system further, the housing was machined as three separate parts: the main spectrograph housing, a separate chamber for the collimator, and a light‐tight cover. The mirrors and gratings each have three mounting "feet" (Fig. 3). The feet of each element are screwed directly to bosses on the outside of the housing, while the optical surface protrudes through a hole in the housing wall.

Zoom In Zoom Out Reset image size

This scheme proved so accurate and stable that it was not necessary to change the focus of the system after it was properly shimmed at room temperature to achieve proper focus. The spectrogaph focus was tested by reimaging a 200 μm diameter room temperature pinhole onto the slit. By scanning the pinhole across the slit we produced a slit beam profile. The FWHM of this profile agreed with expectations of an in‐focus slit. We also measured the FWHM of a set of unresolved NH₃ lines taken in the lab, again finding the result consistent with an in‐focus slit. Judgment of the stability of the slit focus and the position of the spectrum on the detector rests upon monitoring these parameters after numerous thermal cycles, Dewar shipments, and other events that can cause misalignments. We have observed no significant changes in focus or spectrum position.

A series of baffles is arranged inside the housing to trap scattered light and light from unwanted diffraction orders. The housing interior and the baffles were painted with Chemglaze black according to the manufacturer's instructions.

2.2. Slit Viewer Path

SCORE is mounted inside a cylindrical cryogenic Dewar manufactured by Precision Cryogenics, Inc. and is cooled to liquid helium (LHe) temperature for proper operation of the detector arrays. LN2 capacity is 3.7 liters and LHe capacity is 5 liters, which yields consistent LHe hold times of 24 hr or more on the telescope. Detector temperatures during operation are stable to better than 1 K. The f/12 input focal ratio of SCORE was matched to the f/72 Cassegrain focal ratio of the chopping secondary on the 200 inch telescope by a pair of ZnSe lenses. As a convenience, only one lens was cooled to low temperatures. The other lens was at the ambient dome temperature. We estimate this increased the system noise by approximately 10%. The pair also creates an image of the telescope's exit pupil inside the Dewar. This cold Lyot stop helps reduce stray background radiation. In addition, there are a total of five ambient temperature mirrors including the telescope primary and secondary. Clearly, the background radiation can be significantly reduced by cooling some of these surfaces. In SIRTF the IRS will not reimage the telescope pupil since it will be operating much closer to the ground‐based optical situation in which the telescope emissions can be ignored and the sky brightness is relatively low. In addition, the SIRTF telescope is well baffled so the entrance slit cannot view the sky directly.

The entrance slit in the Palomar configuration subtends 09 × 17 on the sky. A monochromatic image of the slit on the array subtends 2 × 4 pixels. The field of view was chopped north‐south by 4 ^'' at a frequency of about 10 Hz and was nodded approximately every 10 s for most of the observations.

The operation of the Rockwell/Boeing arrays is discussed elsewhere (Seib, Salcido, & Reynolds 1994). Because of the very large background flux seen by SCORE from the telescope and the atmosphere as compared to the very low SIRTF background, the SCORE detector was equipped with a multiplexer with much deeper wells, 6 × 10⁶ versus 2 × 10⁵ electrons, than will be used in the SIRTF modules. The SCORE array was operated by a set of control electronics that are similar to those used by SpectroCam, a 10 μm spectrograph/camera built by Cornell for the 200 inch telescope (Hayward et al. 1993). A block diagram of the electronic system to operate SCORE is shown in Figure 4. In short, the instrument user interface is on a Sun Sparc station in the Observatory's data room, while the array control is accomplished by a dedicated Pentium PC mounted in the telescope's Cass Cage. The two computers are connected by a local Ethernet link.

Zoom In Zoom Out Reset image size

In the SIRTF application the array will be sampled slowly with typical integration times of 10–500 s. Multiple (8–64) nondestructive reads will be made during each integration to mitigate the effects of detector read noise. However, the high ground‐based backgrounds require that even with a deep‐well multiplexer the SCORE array has to be read every 60 ms. This is achieved by a hardware co‐adder configured to accumulate up to 128 reads before their sum is sent on to the PC and from there to the Sun workstation via the Ethernet link for display and archiving.

The echelle array and the slit viewing array are clocked simultaneously using the same wires to supply the constant voltages and the clock signals. Only the detector bias can be independently varied. This too is a partial test of the technique to be used in the IRS on SIRTF where all four IRS arrays will be clocked by a single set of redundant wires.

Techniques for extracting optical echelle spectra rely on the long slit lengths and low nonsource noise that characterize those data (Marsh 1989; Horne 1986; Mukai 1990). Many of the so‐called optimal extraction techniques endeavor to increase the final signal‐to‐noise ratio obtained by computing a slit profile of the source that varies smoothly with wavelength and then weighting individual pixels along the slit accordingly. This profile is a measure of the distribution of the source intensity along the slit and is affected by the source's angular size and structure, the slit throughput, and various instrument and alignment and differential optical effects. Applying these techniques to SCORE required modification to account for the short slit, significant order cross‐talk, and background‐dominated noise characteristics.

SCOREX is a suite of IDL‐based routines that implements an optimal extraction of SCORE cross‐dispersed data. The most significant differences between these data and their optical counterparts are the sky signal, the noise sources, and the degree of cross‐talk between adjacent orders. Sky data are often recorded at the extremities of the slit in optical data in a single measurement, whereas sky background dominates SCORE data and is removed by rapid chopping on and off source. The sky signal also dominates the noise in SCORE data. Sky noise is far larger in magnitude than either read noise or direct source noise—the dominant noise source in optical spectra. And finally, optical crossed spectra employ much larger focal plane detectors, which mitigates the moderate order overlap at shorter wavelengths that characterizes SCORE data.

The first step in the reduction process is calibration, which need be performed only when new observing conditions are encountered or the instrument has been modified or disturbed. The first step of calibration involves stretching, shifting, and rotating a theoretical order overlay mask, based upon a conventional ray‐trace of the system. This three‐parameter space is searched with a conjugate gradient technique to maximize a merit function that gives equal weight to total selected efficiency and total number of pixels. This allows the orders' varying curvature and slant to be matched effectively. To prevent the resampling noise discussed by Mukai (1990) that occurs when resampled "sky" pixels cross the physical detector boundaries, each pixel along the slit is specified as a fractional pixel, i.e., a linear combination of itself and the pixel directly above or below it. Next, regions of overlapping wavelength in adjacent orders are aligned with sharp atmospheric or NH₃ lines to ensure wavelength continuity across order boundaries. The overall wavelength map is then calibrated using the bright and abundant atmospheric lines, along with a calculated atmospheric transmission spectrum convolved to SCORE's resolution. The wavelength‐dependent line tilt from vertical is measured and fitted with a polynomial, and a map of the instrument's throughput efficiency is constructed from a pair of blackbody images taken in the lab.

Individual source extraction does not require further calibration, except for standard star extinction. Reduction of individual sources involves first correcting each fractional order image with the efficiency map obtained during calibration. This is equivalent to flat‐fielding of the spectrum and prevents efficiency effects from contaminating the optimal extraction process. The order images are line‐tilt corrected to ensure that slit images line up with the order image pixels. Care is taken to assure the total line flux is preserved, which requires renormalization, since the line tilt is a relatively strong function of wavelength. Each order is then individually optimally extracted.

The optical extraction algorithm discussed by Horne (1986) requires iteration to achieve a self‐consistent solution, since the variance invoked for weighting the pixels along the slit at any wavelength depends on the final flux extracted. In the optical case, this dependence is strong. In our case, since sky noise dominates the data, this iteration is not critical to achieving the best signal‐to‐noise ratio. Iteration does prove useful, however, in excluding corrupt or bad pixels and in alleviating further complication of noise cross‐talk at short‐wavelength orders. Weighting by the inverse of the variance of the predicted flux at each pixel is equivalent to weighting by the square of the signal‐to‐noise ratio in that pixel, diminished by the square of the flux at that wavelength.

A slit profile is fitted using low‐order polynomials, iteratively removing outliers. This manner of estimating the profile allows SCOREX to be used both for extended and point sources, since no source profile is assumed a priori. Using the fit profile, cosmic‐ray hits and uncorrected blemishes can be removed (since, in general, the distribution along the slit of such artifacts deviates grossly from the computed profile). The initial flux estimate is made by averaging along the slit. For a revised estimate of the flux, the slit pixels are weighted inversely by the variance estimate derived from the profile and this flux. The new flux is then used to generate a new variance estimate, and the process is repeated until the flux converges. As mentioned, since the pixel variances depend only weakly on the source flux, this convergence is rapid. Those pixels that do not match the expected profile with sufficient confidence are conservatively collected in each subsequent iteration and are eliminated from the calculation to avoid spurious features in the final spectrum.

Individually extracted orders are then spliced together, maintaining optimal signal‐to‐noise ratio by weighting overlapping fluxes in various orders by the efficiency at that order and wavelength. The differing wavelengths in overlapping regions are interpolated onto a subset chosen to maintain the resolution. A "noise efficiency" is subtracted from the individual weights to avoid divide‐by‐zero artifacts at the order perimeters, where the noise becomes dominated by nonlinear noise sources. This provides an effective efficiency cutoff below which overlapping data will be disregareded in calculating a combined flux at that wavelength. On most orders, data to somewhat less than the 40% efficiency level can be used successfully.

Calibrator stars are reduced and extracted in a similar fashion. The final flux output using any given calibrator observation depends on its exact placement in the slit. Therefore, as a final calibration for point sources, if the chopping amplitude was sufficiently small to allow an image of the source to appear in the camera field (as is standard procedure for point sources), photometry is used to provide a flux correction and eliminate the position‐dependent slit throughput uncertainty.

Taken together, the SCOREX routines provide simple, flexible extraction of SCORE data. Global calibration need be performed infrequently, and individual spectra require no addition calibration steps, which reduces the complexity of rapid production and interpretation of SCORE spectra.

SCORE has been operated at Palomar on several occasions. The operational performance of SCORE is very close to the predicted performance shown in Figure 1. Figure 5 shows the imager view of NGC 7027 and the raw signal from our most recent run. The dark rectangle on the edge of the nebula is caused by the slit aperture. Vignetting of the upper left‐hand corner causes the nebula to appear nonsymmetrical. Figure 6 shows the spectrum of NGC 7027 taken from Palomar on SCORE's first "night" during evening twilight. Additional information and pictures of SCORE can be viewed on our SIRTF web page at http://astrosun.tn.cornell.edu/SIRTF/irshome.htm.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

SCORE has clearly demonstrated the "bolt‐and‐go" philosophy that underlies the IRS design strategy for SIRTF. We have successfully developed a software package to extract and calibrate the data from SCORE and will use this to guide the development of the SIRTF ground processing efforts.