---

Gemini Near-Infrared Spectrometer

Operational Concepts

Version 1.3 Oct. 4, 1996

Preface

This document describes the Gemini Near-Infrared Spectrometer (GNIRS) being built for the Gemini 8-m telescope on Mauna Kea by NOAO, and its scientific capabilities.

The first part of the document reviews the primary science requirements and describes the instrument itself. The second part of the document describes how the instrument would be used for typical observing scenarios.

In this version, certain assumptions are made regarding the system capabilities that are subject to change at or soon after Preliminary Design Review. In some cases, details of Gemini interfaces are also still subject to further definition.

1. System Overview

1.1 Summary of Science Drivers

The Gemini 8-m telescopes will be unique in two respects as far as infrared performance is concerned. First of all, they are expected to have unprecedented image quality, as well as an adaptive optics (AO) system, and will thus to be capable of taking advantage of the very best seeing conditions for observations at high spatial resolution. Second, the low emissivity, good image quality, and large collecting area should give them excellent sensitivity, especially for measurements of point sources. In addition, the idea that the telescopes will be operated using flexible queue scheduling requires that the instruments be capable of adapting rapidly to changing conditions and programs. It is important to realize in this regard that the baseline telescope performance in terms of image quality and emissivity is expected to be good enough that, for most of the thermal infrared (longward of 1.9 m m), the best sensitivity for measurement of point sources will be achieved without the AO module.

The desire to exploit the optimum conditions as well as the highest possible angular resolution - i.e. near diffraction-limited observations - led the Gemini Science Committee to specify a pixel scale of 0.05 arcseconds, matched to a slit size near 0.1 arcseconds.

The initial specifications for the GNIRS clearly gave highest priority to taking full advantage of the 8-m telescopes' high spatial resolution. It was, however, recognized that a configuration capable of taking advantage of the best seeing conditions is not optimal for typical conditions, and that one needs to be able to work under such conditions while retaining the ability to use the high resolution configuration when conditions merit.

A versatile instrument would also offer the ability to change spectral resolution, since some programs require only modest resolution but want to maximize spectral coverage, while others want higher resolution, either to resolve weak features or to take advantage of lower background in between strong atmospheric emission lines. It would also be desirable to be able to trade off spatial and spectral coverage - in one direction, by using cross-dispersion and a short slit to increase spectral coverage, in the other by use of integral field or multi-slit techniques to provide additional spatial coverage for a reduced spectral region
 
 

1.2 Instrument Description

The main instrument capabilities are summarized in the table below.

Image Scale:             0.05 arcsec/pixel (long camera
                                0.15 arcsec/pixel (short camera)

Slit length                 50 arcsec (long camera)
                              100 arcsec (short camera)

Slit widths                 0.10, 0.15, 0.20, 0.30, 0.45, 0.6, 1, 3 arcsec + "open"

Wavelength range     0.9-5.8 m m

Resolutions:             2000,6000,18000 (long camera)
                                667, 2000, 6000 (short camera)

Other options present: cross-dispersion (0.9-2.4 m m)
                            Wollaston prism for polarization analysis

Detector                 1K x 1K InSb 27 m m pixels (ALADDIN)

In addition, a rear slit viewing configuration will be used to facilitate acquisition of objects, and the internal (IR) wavefront sensor will provide tip-tilt correction and fine focus correction. Overall efficiency of the instrument depends not only on instrumental throughput, but on ease and speed of set-up, and on configuration stability and repeatability as well. It is important to minimize the amount of observing time that must be used for real-time calibration sequences.

1.2.1 Instrument Design Drivers

The sensitivity of the Gemini telescopes will be available for scientific observations only if the instruments maintain the same properties of low emissivity, excellent image quality, and in addition, high throughput.

The design emphasizes the reduction of stray light of all sorts - both stray background and scattered light from the object(s) being measured. It is critical to minimize excess background on the 8-m telescopes, since there are wavelength regions where total emissivity of sky plus telescope will approach the telescope design goal of 2%. Under these conditions, scattering of 1% of the light at the dewar window (for example) would be disastrous. Within the instrument, internal backgrounds must be kept negligible. With the small pixels specified for the instrument and the possibility of working at resolutions of 6000 and higher, there will be interesting regions of the infrared spectrum where intrinsic background is exceedingly small (<<0.1e^-/sec). In all probability, the limiting factor will be detector dark current, but it is necessary to ensure that instrumental background is kept well below this value since improvements in detector technology will almost certainly occur during the lifetime of the instrument.

Instrumental background can be of two main forms. One is leaked light, primarily at longer wavelengths. This is light that gets in through feed-throughs or around filters (for example). This would also include emission from badly-shielded, warm components inside the instrument, such as internal motors or heater resistors. The other is light that is scattered within the instrument itself. For example, below 2 m m the contrast between bright airglow lines and the adjacent continuum is known to be many orders of magnitude. If even a few percent of the bright line emission is scattered into the continuum regions, the overall sensitivity of the instrument suffers greatly.

The instrument design is intended to minimize both sources of excess background. The use of an InSb array for the single detector in this instrument creates a particular design challenge since it detects long-wavelength light out to 5.7 m m, and will also be used to observe at moderate-to-high resolution at wavelengths less then 2.2 m m where the natural background is very low. For example, at R=2000 in the J band the average natural background will be less than 0.1 e^-/sec while one square mm of dewar at 0^o C with an emissivity of 1% will emit 10¹¹ photons/sec/sr into the InSb band pass, which would blind the array to the small J band signal, even if attenuated to a part in a billion! Stray light is eliminated fundamentally by provision of separate fore-optics and careful baffling. This was also a consideration in the decisions on aspects such as the location of the order-sorting filter wheel.

Scattered light has been less widely recognized as a potential problem in instrument design, largely because it only becomes a problem with low dark current and enough pixels to have large dynamic ranges in the background (or object) spectrum in a single exposure. Considerable emphasis was given to this area. Preference was given to the use of off-axis reflective optics where possible because these provide less scattered light than refractive surfaces, and where refractive materials are used, the design was restricted to low-index materials, which can be efficiently anti-reflection coated. The design also minimized the number of refractive elements, and avoided placing them in locations where scattered light is particular deleterious (e.g., no field flatteners close to the detector).

In addition to minimizing background, it is also important to maximize instrument throughput. Part of the emphasis on using reflective surfaces stems from this consideration, since in the infrared (unlike visible wavelengths) reflective surfaces are inherently more efficient, since they are achromatic and have lower losses per surface. The relatively slow focal ratios involved allow an unvignetted design, thus avoiding the principal drawback of reflective designs.

The third aspect considered is operational efficiency. The very narrow slits to be used require that objects be accurately and quickly positioned in order for the observer to achieve maximum efficiency. Furthermore, many interesting objects may not have visible counterparts. These considerations led to provision in the design for viewing of the slit with a mirror in place of the grating. This provision will allow the observer to rapidly and accurately center (or re-center) a faint object on the slit. Viewing can be done in both broadband filters, for acquisition of continuum objects, and in a limited selection of narrow-band filters, corresponding to bright, astrophysically-interesting emission lines. These will allow centering on emission features in regions such as supernova remnants, H II regions, or H-H objects. This design assumes that the array controller is capable of operating the ALADDIN arrays so as to minimize thermal transients due to integration time changes (for example), as well as compensating for any such transients that cannot be avoided.
 
 

1.2.2 Spectrometer Optics

1.2.2.1 Fore-Optics

The window to the dewar will be a very weak lens to adjust the focus of the telescope exit pupil (the telescope secondary) on the internal Lyot stop of the instrument. Given its low optical power, it is tolerant of flexure between the warm mechanical support and the cold optics. Since this window is also the interface between the instrument, which is cold and under vacuum, and the exterior, which is warm and at atmospheric pressure (or halfway there on Mauna Kea), it is necessary to deal with thermal and pressure effects in designing the window and its mount.

An all-reflective Offner system forms an image of the telescope pupil on the secondary of the Offner optics while it re-images the telescope focal plane on the slit plane of the instrument with 1:1 magnification. More precisely, because the telescope pupil is only a few meters away, the power of the entrance window/field-lens is required to make the image of the telescope pupil exactly coincide with the Offner secondary. The system preserves the f/16.2 focal ratio of the telescope and more importantly has an exit pupil at infinity.

A cold Lyot stop is placed at the Offner secondary to minimize thermal background. Since there are some aberrations in the fore-optics, the Lyot stop size is a compromise between minimizing background and maximizing throughput.

1.2.2.2 Filters/Slits

A wheel for order-sorting filters is installed in the output beam from the Offner optics about 200 mm before the slit plane. The filter complement is I1, I2, J, H, K, L, M, Xdisp, dark (9 filters). The designations I1 and I2 correspond to the pass-bands defined by the 7th and 6th orders of the 6.6-m m-blaze gratings. The JHKLM pass-bands correspond very roughly to the equivalent photometric pass-bands, but in all cases are matched to the corresponding grating order or the full width of the atmospheric window, whichever is less. The KLM pass-bands in particular will be wider than standard photometric pass-bands. Xdisp is a long-wave blocker for use with the cross-disperser. In the document below these filters are referred to as blocking filters or order-sorting filters, as distinct from the photometric or acquisition filters.

A second wheel is provided for filters used for acquisition. These comprise a set of filters corresponding to standard broad-band photometric bands (JHK - only K_S is provided initially, because the J and H order sorters are a near match to the Gemini J and H photometric filters). Longer wavelengths are not contemplated because these would require faster readout rates than are otherwise needed in the spectrometer; an L filter could be added if the controller supplied by Gemini would support its use. Also present are a small selection of narrow-band filters that may be useful for acquisition of emission-line objects or features. These are (provisionally) designated as 1% filters corresponding to the H I lines at 2.16 and 1.28 m m, the H₂ line at 2.12 m m, He I 1.08 m m, and the [Fe II] line at 1.64 m m, plus a 2% filter for the PAH feature at 3.28 m m. This particular set of filters is based on current appreciation of science to be done with the instrument, and may be expected to evolve with time.

All filters will have the same optical thickness, since refocus of the instrument as filters are changed is not provided (nor desirable). While there is provision for additional filters, it is expected that filter changes will be infrequent, since they require opening the instrument.

The filters are spaced back from the focal plane and tilted at a 2^o angle. They are in front of rather than behind the slit so that out-of-band radiation is removed from the beam before passing through the slit and so that in-band light reflected from the filter will not tend to form a ghost image (as would be the case if it were reflected back toward the slit plane).

A lens will be placed in one of the filter wheels, which will form an image of the Lyot stop on the detector, to check beam alignment with the telescope.

The slit mechanism consists of a detented slit selector wheel, with one open position (~30 arcsec diameter) for field viewing and an independent Dekker wheel for shortening the slits close to the slit wheel.

1.2.2.3 Collimator

The collimator is an off-axis segment of a paraboloid of focal length 1500 mm. The angle between the axis and the principal ray of the beam is currently 4 degrees. This angle affects the optical performance of the collimator, and the final value of this angle will be chosen during detailed design. The tradeoff is between unvignetted field in the cross-slit direction and the radius of the field which meets the Gemini imaging specification.

The collimator in particular represents a compromise. Due to the off-axis parabolic design of the collimator the image quality at 25 arcseconds along the slit at the mid-line of the array is barely compatible with the Gemini specification that 85% of the light be contained within a 27 m m pixel. In the corners of the array it exceeds this geometrical optical criterion by about 20% (although the camera designs can and do partially correct these aberrations). However, the unobstructed, simple all-reflecting nature of the collimator, its ease of alignment, and its excellent control of the telescope pupil offers significant advantages over other designs (i.e. refractive or cassegrain style) in the area of stray and scattered light. The scientific basis for the geometrical optical constraint is that the PSF of the AO corrected telescope not be degraded. Beyond not wishing to compromise the on-axis performance for the wide field performance we note most models of the performance of AO systems predict that the performance will degrade rapidly over a field of 30 arcseconds due to decorrelation of the atmosphere. The Strehl ratio at 2.2 m m at the end of the long camera slit will be a very respectable 0.93. While a perfectly diffraction limited image by the telescope would be 1, actual Strehl ratios with the AO system on Gemini at 25" will be much lower.

1.2.2.4 Grating/Cross-Disperser/Wollaston Prism

The grating mechanism will select between five grating positions using a detented or clamped mechanism. Subsequently the angle of the grating selected can be scanned about an axis close to its face by an amount adequate to cover the useful range of the grating. The need to minimize flexure and maximize reproducibility led to a decision to limit the grating tilt adjustment to a series of precise, detented settings rather a less-precise "continuous" motion. The detent spacings correspond to intervals of ~0.4 array width with the long cameras, and thus to ~0.13 array width with the short cameras. The zero-point will be set so that the atmospheric windows can be individually covered with one setting at R=2000 (except for the 3 m m window, which is too broad to fit on the array in one setting).

We are proposing using the gratings in multiple orders. Our experience is that the performance near the blaze of the gratings is not strongly dependent on order. On the other hand, the decreasing spectral range means that the efficiency is dropping more rapidly off the blaze than would occur with first order gratings. The problem is that there are simply too many desired modes to consider 1st and 2nd order gratings for all bands and resolutions. The proposed solution gives good throughput at blaze and, in the cross-dispersed mode, the loss of light to adjacent orders can be at least partly recovered by combining the overlapping parts of the orders from orders 4 up, where the array covers more than a free spectral range. In any case, our choice of blaze tends to make the order cross-over occur in atmospheric absorption bands.

Approximate descriptions of three gratings to satisfy the specifications listed earlier are as follows

Grating changes will be possible but will not be done routinely, as they will require opening the instrument and re-alignment of any newly-installed grating. Such changes would be in the nature of upgrades rather than configuration changes.

The cross-disperser turret selects amongst a set of fixed prisms or a mirror, to provide cross-dispersion in wavelength or in state of polarization.

The list of devices comprises:

1. 1. prism to cross disperse the light between 0.95 and 2.5 m m across the chip with the long camera
2. 2. same as 1.) but for the short camera
3. 3. a mirror, for use with the full slit-length
4. 4. a Wollaston prism (MgF2) to split the two polarization states. The separation will allow use with both cameras.

1.2.2.5 Cameras

The cameras are:

1. 1. Blue Long Camera; 1305 mm focal length (.05 arcsec pixels). The camera is a simple doublet lens optimized and coated for 0.95-2.5 m m, followed by a third element closer to the detector.
1. 2. Red Long Camera; 1305 mm focal length (.05 arcsec pixels). The camera is a simple doublet lens optimized and coated for 3.0-5.0 m m.
1. 3. Blue Short Camera; 435 mm focal length (.15 arcsec pixels). The camera is a four-element refractive design (two doublets).
1. 4. Red Short Camera; 435 mm focal length (.15 arcsec pixels). The camera is a four-element refractive design (two doublets).

1.2.2.6    System Efficiency

The table that follows shows calculated system efficiencies in long-slit mode (no cross-dispersion) for some wavelengths and gratings. (This table will be updated to reflect actual component performance as instrument construction proceeds.)

The overall system efficiency is generally greater than 40%. The exceptions are the shortest wavelengths, where reflectivity of gold is starting to decrease slightly, and where the gratings are used in 7^th order. The other low point is the extreme red end of the K band, where the transmission of the glass (SF6) used in the cameras substanially reduces transmission in the short camera. (The long blue camera transmission is also down, but much less.)

1.2.3 Auxiliary Optics

1.2.3.1 WFS Feed

The internal infrared wavefront sensor (on-instrument wavefront sensor, OIWFS) is fed light from a pierced mirror located immediately after the dewar window and before the spectrometer fore-optics. The mirror transmits light though a central hole to the spectrometer and reflects light from the surrounding field to the WFS. The hole will pass light to the full length of the short camera slit, and will be wide enough to minimize diffraction effects; a central portion will be wide enough for use in acquisition; the field of view is to be determined, but will likely be ~30" diameter.

Following the mirror, a series of lenses are used to re-image the focal plane onto the WFS itself. This design will accommodate a Shack-Hartmann sensor. The optics also form an image of the telescope secondary on a gimbaled mirror, which is tilted so as to place any portion of the unvignetted field of view of the telescope onto the wavefront sensor. Thus the sensor itself can have a minimal field of view and does not need to be physically moved inside the instrument.

A small filter wheel is also provided for band-pass and neutral density filters, and may also contain aperture stops.

1.2.3.2 Slit Viewer/Acquisition

Slit viewing is provided by means of a flat mirror mounted in one of the positions of the grating turret. This design requires that changes between gratings be made both reasonably rapid (under 30 seconds) and highly repeatable.

The slit viewing system also constitutes a rather basic imaging system. It could provide an alternative for observers who need photometry of the objects they take spectra of, or who need to do quick imaging of a field in order to select objects for spectroscopy. It is not intended as a substitute for the facility IR Imager for programs with a significant imaging component.

1.2.4 Upgrade Options

1.2.4.1 User Filters/Slits

The spectrometer has been designed to allow for a small number of additional filters and slits. This permits additions of filters not included in the initial complement, but which later prove to be desirable. It would also be possible in principle for a user to supply a filter for a specific program - for example a narrow-band filter matched to a particular planetary spectral feature. Because installation of any filter requires opening the instrument, it would have to appropriately justified and scheduled; a user would not be able to make last-minute requests of this nature. Also, the filter would have to be parfocal with the regular filters.

An additional slit capability might be in the form of a limited coronagraphic capability, which may be of interest even though the slit wheel is located after some of the optics.

1.2.4.2 Additional Gratings

Provision for at least one additional grating is being made in the grating turret. There are several reasons for not attempting to fill this position initially, but the main one is that we expect experience with the instrument will be needed to decide which grating (or gratings) is optimal. Some possibilities are listed below:

- Higher spectral resolution grating. The design of the spectrometer limits the maximum blaze angle that can be used efficiently, but some increase may be possible. Also, users might be willing to get higher spectral resolution even at the price of some vignetting. Immersed gratings have the potential to circumvent these problems, particularly since high-index infrared optical materials are available. High-efficiency immersion gratings ruled directly on such materials are not currently available - at least not in the size needed - but may well be in a few years’ time.

- Echelle format grating. At the higher resolutions (6000 and 18,000) it is not possible to observe the full spectrum at once with the gratings currently planned. A grating blazed for longer wavelengths could be used like an echelle grating, in combination with the prism cross-disperser, to place the entire 1-2.5 micron spectrum on the array. Obviously, the usable slit length decreases by a corresponding amount, so this option is really of interest only at the time a 1Kx1K array is available. For example, to fit the entire spectrum onto the array at R=6000, 15 or more orders would be needed (e.g. from order 11 to 27 for a 26-m m-blaze grating), with a corresponding slit length on a 1K array of perhaps 50 pixels.

1.2.4.3 Integral Field Unit

A rough conceptual design has been developed for an integral field unit, and space for this implementation has been left within the instrument. Details of specifications, implementation and fabrication need further development, but the basic concept is similar to that described by Content (1996, in press).In this particular implementation, the spectrometer slit plane is re-imaged (and magnified) onto an image slicer, which would produce approximately 20 slices. Light from the slicer is directed to a parabolic mirror, which produces a series of pupil images on a linear micro-lens array. The micro-lenses form images of the individual image slices. The light path is folded so that the row of "slice" images is at the collimator focus.

The IFU would intercept the light before it reaches the spectrometer slit wheel, "process" it, and redirect it to the slit plane. A mechanism is required to insert and remove the IFU, so that the other configurations of the spectrometer (including acquisition) are available.

1.2.4.4 Camera and Detector Upgrade

The design of the camera turret permits substitution of other cameras of similar design and focal length. This would most likely be done in conjunction with an array change involving either a change in pixel size or more pixels (for example, a pair of 1Kx1K arrays butted together at one edge). This would represent a major upgrade, and should definitely not be thought of as a simple "camera change" like those commonly done on optical spectrographs. Note that the ability to upgrade the cameras comes at a certain price in compactness for the turret
 
 

1.2.5 Mechanical Design

1.2.5.1 Design Drivers

The mechanical design requirements for the instrument are extremely demanding. On the one hand, it is necessary to be able to reconfigure the instrument rapidly, in order to accommodate changes in program due to changing conditions (whether these changes are for a single observer or different programs selected from a queue). On the other, the mechanisms must be extremely stable and repeatable - and, of course, highly reliable. Ultimately, some compromises may have to be made between these two aspects.

1.2.5.2 Mechanisms

1.2.5.2.1    Dark Slide

A dark slide will be provided in front of the window. Its purpose is primarily to protect the window when the instrument is not in active use. Since it will be at ambient temperature, it will not be used for "dark" exposures and will not be light-tight or hermetic. A manual override will be provided for this mechanism; it is the only mechanism on the instrument for which such an override will be provided.

1.2.5.2.2 WFS Scanner mirror

This will be a gimbaled mirror located at the image of secondary formed in the WFS optical train. It will be scanned in two axes so as to permit imaging any point in a 3.5 arcminute diameter field onto the sensor. Relatively small but precise motions will be required, since the WFS will be used to effectively control small telescope offsets when centering objects or offsetting them while observing.

1.2.5.2.3 Filter Wheels

Two filter wheels will be provided for blocking and acquisition filters. These are relatively large filters, since they must pass the full length of the long-camera slit. A third wheel will probably need to be added if an integral field unit is added at a later date; since the IFU field will be small the filter wheel would be able to handle more filters. (It must, however, have an "open" position capable of passing the full field of the spectrometer for use with other configurations.)

The WFS contains a separate filter wheel for its own filters and aperture stops.

1.2.5.2.4 Slit Wheels

A pair of wheels will be used to define slit width and length; they can be though of as a slit wheel and a Dekker wheel. Discrete slits will be used rather than a continuously variable slit in order to ensure repeatability. The slit wheel will be located at the spectrometer focal plane; the Dekker wheel will therefore be slightly out of focus.

The slit wheel must be highly repeatable, so as to keep wavelength shifts under 0.1 pixel and also keep flat field variations under 1%.

1.2.5.2.5 Grating Turret

The grating turret is unquestionable the most demanding mechanism in the instrument. It performs two functions: it selects different gratings, and for a given grating it provides tilt adjustment to scan the useful wavelength range.

The grating motions must be stable and repeatable to 0.1 pixel on the array (with the long camera), which implies that flexure and other uncertainties must be held to sub-micron levels. In order to achieve this, both the grating selection and grating tilt positions are set by detents rather than attempting to achieve the required repeatability with stepper motor and gear train. The grating tilt interval is a compromise between flexibility and the need to provide a robust yet compact mechanism. The tilt interval corresponds to 0.4 array widths with the long-focus cameras (and thus to 0.13 array widths with the short cameras). This allows one to piece together spectra with reasonable overlap at high resolution, provide some flexibility in setting on separated pairs of lines at intermediate resolution, and also allows for coverage of a single atmospheric "window" with one grating setting at low resolution (except L, which does not fit on the array).

The grating turret has provision for five gratings. Three gratings are specified; the fourth position will be used for the acquisition mirror, and the fifth will be left available for an additional grating, as described above (1.2.4.2).

1.2.5.2.6 Cross-Disperser Turret

A cross-disperser turret is also provided. This contains four elements: a mirror, for long-slit spectroscopy; a long-camera cross-disperser; a short-camera cross-disperser; and a MgF₂ Wollaston prism for polarization analysis.

The cross-disperser prisms will put a complete 0.95-2.4 m m spectrum at R=2000 on the array when used with the relevant cameras. No provision is made for cross-dispersion of longer wavelengths. (In fact, the prisms will probably absorb long-wavelength light.) The Wollaston prism will use the full width of the array with the long camera, and will thus underfill the array with the short camera.

The tolerances on the cross-disperser turret are similar to those for the grating turret, but it is only required to move in a single axis, and to set to a small number of discrete positions. (Nominally one per element.)

1.2.5.2.7 Camera Turret

A single, four-position turret will be used to select between the four cameras in the spectrometer. This is probably the largest mechanism in the instrument, but its tolerances, though still demanding, appear to be less stringent than those of the grating and cross-disperser turrets.

1.2.6 Instrument Control

Instrument control is primarily in two areas: interaction with the array controller, and control of mechanisms (and thermal control).

The interaction with the controller is essentially in the form of setting up and initiating integrations, and responding to the end of integrations. There is no apparent need for close synchronization of mechanical functions with array functions.

The mechanical functions can be thought of as responses to re-configuration requests initiated by the user (or as part of an observing sequence). For the most part, it is not of importance whether the different changes involved in reconfiguration are carried out in parallel or sequentially. If sequentially, their order is usually not important. Any exceptions are likely to involve minimizing the exposure of the array to high backgrounds, and configuration routines may need to take this into account.

What is important is to minimize the telescope time used by reconfiguration. The goal is to have any reconfiguration take no more than 3 minutes, and to have common reconfigurations (for example, from acquisition to observing) take 60 seconds or less.
 
 

1.2.7 Data Analysis System

Data reduction and quality assurance will make use of the facilities provided by the Gemini Data Handling System and the array controller. Descriptions of how these would be used are provided in section 2.3.

2 Observing Procedures

All observing modes are generically similar. An observing configuration requires specification of the position of all the mechanisms listed above (exceptions being focus and dark slide). In addition, the positions of the instrument rotator and PWFS must also be specified for any observation.

The array control parameters must also be specified. This would include, at least, the microcode to be used, the integration time, the number of integrations and how (if at all) they are to be co-added. For some programs or configurations one would not read out the entire array - or at least would not save the full image.

An observing sequence is the basic unit of observation, and normally consists of a series of observations taken with a fixed configuration, with the only difference being that the telescope is moved between observations. (This would require compensating motions of the OIWFS and PWFS but otherwise no "reconfiguration".) The observer would not normally need to intervene to carry out in the individual steps in an observing sequence, but there would be provision for termination or modification of observing sequences in progress. Observations forming an observing sequence would normally be reduced together, rather than as independent images.

2.1 Sample Observing Scenario

In this section we describe how a specific observing program would be carried out. The example is a program originally described in the NOAO proposal, but it should be noted that most programs will be carried out in a very similar way. Differences are noted in the following section (2.2).

One key point to note is that from the point of view of instrument operation, there is no difference between queue or service observing and so-called classical observing. In all cases, an observer would (or should) start the night with a list of objects to be observed, each of which would have associated with it a complete instrument configuration including all required guide stars. Differences between the observing modes lie primarily in the way in which observations are evaluated or program objects are selected, and these decisions are made at a higher level than the instrument control.
 
 

2.1.1 Program Science

In this program, we would obtain moderate resolution, high signal-to-noise 2-m m spectra of early M giants within the bulge of M31 and M32. Using the CO band head, the Na I blend (2.207 m m) and the Ca I blend (2.263 m m) these spectra will reveal for the first time the metallicity distribution within luminous spheroidal systems other than the bulge of the Milky Way. Because metallicity and age are degenerate when determining ages from main sequence turn-off colors a determination of the stars' metallicities is needed to determine their age. Since the progenitors of giants are normal main-sequence stars, the metallicities of giants will allow us to test whether the bulge of M31 and M32 have formation histories similar to that of the bulge of the Milky Way, or whether luminous spheroidal systems have diverse formation histories. Understanding the star formation histories of these three nearest luminous spheroids will give us great insight into the process by which the spheroidal regions of galaxies form.

The spectral energy distributions of M giants peak in the near-IR. Coupled with the superb seeing in the near-IR, this reduces confusion in these crowded fields because the giants appear several magnitudes brighter relative to the background of main sequence stars than they do optically. Near-infrared spectroscopy therefore appears to be a very attractive means of obtaining abundances. At moderate resolution (R~2000) a number of spectral features appear to be useful for determining relative abundances in giants. We plan to concentrate on three K-band features which have now been well studied in giants in the Galactic bulge, disk and globular cluster system. These are the Na I blend at (2.207 m m), the Ca I blend at (2.263 m m) and the CO band head (2.3 m m). Despite the fact that the Na and Ca features are complex blends (Gaffney and Lester 1992, ApJ, 394, 139), observations of cluster, disk and bulge stars in the Galaxy indicate they can be used as reliable indicators of relative abundance when coupled with effective temperature derived from J-K colors (Frogel and DePoy, private communication; Whitford, Terndrup and Frogel 1990, "Bulges of Galaxies", p. 19; Terndrup, Frogel and Whitford 1991, ApJ, 378, 742). But it does appear to be necessary to observe only giants fainter than M_bol = -3 (K=8 in Baade's Window in the Galactic bulge). For sources brighter than this, it appears that the presence of large numbers of AGB stars and LPVs makes abundance determination very messy.

The distance moduli of M31, M32 and the bulge of the Galaxy are 24.2, 24.2, and 14.2 respectively, with an uncertainty of about 0.2 magnitudes. Thus, we wish to observe M giants with magnitudes fainter than about K=18. To obtain metallicity determinations accurate to about 0.2 dex requires a S/N of greater then 75 with a resolution of about 2000. This will be a very challenging observation even with the very powerful combination of the excellent image quality and low emissivity of the Gemini telescope and the high throughput and very low background of our proposed infrared spectrometer. Our program will be to observe stars with 18<K<18.5 with a S/N>75. We would like to observe stars at varying distances from the centers of M31 and M32. This will allow us to test if there are radial metallicity gradients as well as measure the abundances in these two luminous spheroidal systems. In total we would like spectra of about 30 stars to make a first attempt to understand the distribution of metallicities in these fields. The crowding in these fields is fairly serious with the average separation between K=18 stars being about 2 to 10 arcseconds depending on the distance to the center of the galaxy (DePoy et al. 1993, AJ, 105, 2121). With adaptive optics we should have a field of about 30" diameter around the natural guide star which is well corrected. Within this field with good seeing we should have D(50)<0.1" and D(85)<0.4", resulting in a mean separation between sources of about 10 to 50 seeing disks, which makes the field crowding manageable. We have identified several stars with V<14 which should be usable by the adaptive optics system.
 
 

2.1.2 Observing Program Strategies

Our sources will have K magnitudes of between 18 and 18.5. We will need the Gemini telescope with a fully functional adaptive optics (AO) system to achieve the excellent image quality these observations will require. We will select objects that lie close to stars that are suitable guide stars, and will observe using the reflective slit option for maximum throughput. We will employ the R=2000 grating with a K-band order sorting filter along with a two pixel wide (0.1") long slit. Given that the background in the K band rises exponentially, toward longer wavelengths, due to thermal emission, a single number for signal to noise is not meaningful. Our features of primary interest are at 2.207 m m and 2.263 m m, which are in a region between strong OH emission and the onset of the bright thermal background. We predict a background of about 1 electrons/sec/pixel. Using one of the guide stars, around which we have selected our fields, we should be able to achieve a D(50)<0.1" in good conditions. Thus, our slit losses will be about 55% of the light. With multiple non-destructive reads the array noise should be less than 10 electrons. Thus, to achieve background limited performance we will need single exposure times of over 100 seconds. From past experience we have found that sky variability will not create serious background subtraction problems (i.e. the residual background is small enough to be subtracted by fitting a function along the slit) at this resolution for exposure times of 600 seconds. With such an exposure time the shot noise from the sky will be 25 electrons and the addition of the 10 electrons read noise brings the total noise to 26 electrons. It is therefore legitimate to assume background-limited performance, as long as dark current and stray and scattered light are much less than 1 electron/sec/pixel. We assume that the dark current is less than 0.1 electron/sec/pixel and that the instrumental internal background is less than this. Thus, for a K=18.5 star the total noise after a 600 second exposure will be 27 electrons/pixel. For a stellar profile the slit loss will be 55% and we will obtain a total stellar signal of 490 electrons per pixel (summed perpendicular to the dispersion). The total signal to noise of an optimally extracted spectrum will be 17 per resolution element (2 pixels). For a signal to noise of 75 per resolution element we will need 20 exposures, or about 3.3 hours. With the overheads outlined below and the need to observe a hot star nearby on the sky for telluric absorption correction we estimate that a total of 4.5 hours will be needed for these observations. Given the source density it should be possible to align the slit so that 2 or more sources will lie along the 30" region with images are well corrected by the AO system. (The range in wavelength to be observed is small, so differential refraction across the region is not significant. It is obviously necessary to compensate for differential refraction between the guide wavelength and the wavelength being observed.) Thus, to reach our goal of measuring 30 stars will require about 68 hours of telescope time.

There are overheads associated with these observations since various types of calibration data must be taken during the night. In particular, absorption by telluric lines is significant in this spectral region and it will be necessary to observe a bright, hot star every few hours to remove them effectively from the data. For calculation of our overheads, we assume that all mechanisms within the instrument and the telescope can be reconfigured simultaneously. Further increases in efficiency would be realized if the reconfigurations can be done while moving from one object to another. A typical observing sequence is outlined below.
 
 

2.1.3 Observation Details

Note: all times listed (except for integration times) should be considered tentative at this time. Also note that configuration sequences and observing sequences will usually be pre-programmed (e.g. as configuration files or observing macros) so as to require only a single command from the observer. In what follows, unless it is indicated otherwise, it should be assumed that any reconfiguration of the instrument will involve a single command, and any observing sequence will also involve a single command. It should also be possible to terminate or otherwise modify observing sequences while in progress (see the section on data quality assurance). For completeness, we also include in the description below some diagnostic observations that would not be required on a nightly basis.

It is also important to realize that many operations involving the telescope that are described below will also be performed as a single automated sequence. Examples are guide star acquisition and focus, or the process of moving guide probes when the telescope makes small offsets during an observing sequence.

2.1.3.1    During the Day.

These represent observations intended either to provide daily calibration data or verify normal operation of the instrument. For queue observing, calibrations would be obtained for the program(s) most likely to be carried out that night. In this case, we consider only those observations required for the specified program.

a. Configure instrument to desired observing configuration

    1. Slit (0.1" long-slit)

    2. Filter (K blocker)

    3. Select grating (12 l/mm for R=2000)

    4. Set tilt for K band coverage

b. Verify Correct Operation (2-3 hours)

    a. Measure linearity using continuum source. Verify gain and noise intercept by fitting gain curve.
    b. Measure noise performance and system dark current

    a. Insert 0.1" long slit
    b. Move grating to mirror position
    c. Image slit. Verify focus and note location on slit viewer detector for use later, during object centering

    a. Configure instrument to observing mode.(<1 minute)

1. Slit 0.1"
2. Filter to K
3. R=2000 grating

    b. Insert facility calibration unit, select appropriate continuum source
    c. Obtain 10 images with 100,000 electrons signal each (<5 minutes)

  a. Filter to "dark" position
  b. Obtain 10 images at each exposure time to be used (in this case, 600 and ~10 seconds).

    a. Configure instrument for slit viewing while slewing to focus source (<1 minute)

1. Slit to open (30" diameter)
2. Filter to K_S acquisition
3. Behind the slit viewing mirror in position
4. Move guide probes to guide star locations

    b. Focus telescope on peripheral wavefront sensor (PWFS)
    c. Focus AO wavefront sensor (AOWFS).
    d. Adjust focus (< 1 mm) using OIWFS.
    e. Take test image with science array.

    1. Telluric absorption calibration star (<10 minutes)

        a. Configure instrument for slit viewing while slewing to source (<1 minute)

            1. Slit to open

            2. Filter to K acquisition

            3. Rear slit viewing mirror in position

            4. Move all WFS probes to guide star locations

            5. Set rotator to desired position angle

        b. Setup (<3 minutes)

        c. Configure instrument to observe source (<1 minute)

            1. Grating in position (R=2000 for long blue camera)

        d. Observing telluric absorption calibration star (<5 minutes)

            1. Obtain spectra (<20 seconds)

            2. Offset star along slit (<10 seconds); move all guide probes to follow guide stars

    3. Calibrate OIWFS sky if needed before moving

2. Source observation (110 minutes)

a. Configure instrument for slit viewing while slewing to source (<1 minute); same as 1.a above
b. Setup (<3 minutes); same as 1.b above
c. Configure instrument to observe source (<1 minute); same as 1.c above
d. Observing source (105 minutes)

1. Obtain spectra (600 seconds)
2. Offset star along slit (<10 seconds); move all guide probes to follow guide stars
3. Calibrate OIWFS sky if needed before moving
4. Repeat offset and observe 5 times
5. Verify source centering after 50 minutes (<3 minutes)

6. Obtain spectra (600 seconds)
7. Offset star along slit (<10 seconds)


8. Repeat offset and observe 5 times.

3.Alternate telluric absorption calibration star observation (<10 minutes) and program star observations (105 minutes) until desired signal to noise reached. End with telluric absorption star observation.

1. Verify flat field (<6 minutes): same procedure as in afternoon
2. Verify darks (~2 hours): same procedure as in afternoon

Any infrared spectrometer will need a similar set of calibration observations. Our design reduces the overheads in a number of ways. First, the rear slit viewing system makes source centering both very accurate and rapid. Secondly, our design will be mechanically stable: the telescope pupil and spectrometer beam will flex by less than one percent. When coupled with a design optimized to reduce scattered light, this should make the flat field very stable. and thus it will only need to be measured at the start and finish of the night. We will require that the slit mechanism reproduce at the 3 m m level (0.1 pixels in the detector plane). This will assure that objects can be accurately centered and that the flat field will not change due to movement of small imperfections on the slits. The mechanical stability should also make it necessary to verify source centering only once every one or two hours in order to maintain the source centered to better than 0.1 pixels on the slit.
 
 

2.1.5 Observing Command Sequences

As emphasized above, maximum efficiency is achieved if routine command sequences are fully automated, so that the observer needs to initiate an extended sequence rather than each command within the sequence. Low-level command sequences and standard high-level sequences would be available as "canned" routines or menu items, but it should be possible to generate non-standard high-level sequences as well; the observer (or the astronomer in the case of queue-scheduled observations) would need the ability to down-load and test such sequences prior to use.

These routines may be processed either at the level of the Observatory Control System or the Instrument Control System. In either case, the processing should appear transparent to the observer.

2.1.5.1    Examples of Low-Level Sequences

Guide star acquisition. Once a star is placed within the field of the OIWFS, the sequence of events whereby the probe is centered and guiding and fast focus are initiated will take place automatically. This is a situation where non-standard sequences would not be contemplated.

Offsets while guiding. Typical observing scenarios call of the object to be repeatedly offset small distances along the slit. When this is done, the guide probes need to be accurately offset in the contrary direction, in order to retain precise positioning. This sequence, while basic, would obviously treat the distance and direction of offset as an input variable.

Reconfiguration. There are many instances where one will be switching between two or more configurations frequently while observing. Examples would be changing between object acquisition and spectrum acquisition, or changing between wavelength settings. In this case, the observer would click on a pre-programmed configuration from a menu. The observer would previously have defined which configurations were present in the menu.

2.1.5.2    Example of High-Level Sequences

Data Acquisition. The most frequent high-level sequence would be a sequence alternating an observation with an offset. An example might be a set of 5 300-second exposures with 5 arcsec offsets along the slit between them. Some level of "nesting" is desirable for cases where one wants to repeat the basic sequence several times before checking centering; this could also involve changing wavelength (configuration) from one sequence to the next.

2.2 Additional Scenarios

The instrument is optimized for the sorts of scientific observations envisaged by the Gemini Science Committee - observations at high sensitivity and high spatial resolution. Examples of these abound. One such case is the sample program described above, which envisages sensitive observations of individual stars against a complex stellar background. Here one needs both the spatial resolution to separate individual stars from background contamination, and the sensitivity to get high signal-to-noise at good spectral resolution.

Another case would be studies of the inner regions of active galactic nuclei, such as the regions around Seyfert galaxy nuclei or quasars. Here one needs the best possible angular resolution in order to get the best possible spatial resolution on these distant objects (0.1 arcsec at the distance of NGC 1068 corresponds to roughly 10 parsecs). Equally important is the fact that the design will minimize scattering within the instrument of light from the bright nucleus, when one of the effects that one wishes to observe is scattering of nuclear light from material within the galaxy close to the nucleus.

For the common case where observations are required of relatively isolated point sources, the instrument will provide significant gains over an instrument/telescope combination optimized for lower image quality, especially if the combination has higher emissivity. In this regard, it should be noted that for point-source spectroscopy, the added emissivity expected to be introduced by the adaptive optics module will usually more than offset the increased fraction of light passing through the slit at wavelengths beyond 2 m m. At shorter wavelengths, in contrast, the background is not dominated by thermal emission and the AO module should normally provide increased sensitivity.

For most observations with lower resolution and the long camera, differential atmospheric refraction is a consideration. On Mauna Kea, differential refraction across the K band (2.0 to 2.5 m m) is 0.02 arcsec (20% of the slit width) for a zenith distance of 45^o; the effects are more pronounced for the J and H bands. As a result, observations will need to be made near the zenith or with the slit oriented at the parallactic angle, or some means of correcting for light losses (e.g. wide slit observations) is needed.

There are many programs where other configurations of the instrument would be used, described below. The basic observing procedures for all of these would be very similar. Only those options that will be present in the first-light instrument are discussed.
 
 

2.2.1 Short Camera

The coarser pixels provided by the "short" camera provide optimum efficiency for measurement of point sources under most conditions, while retaining spectral resolution. Typical gains in signal to noise will be of order half a magnitude or greater compared with using the "long" camera and 0.1 arcsec slit. An adaptive optics system producing moderate to high Strehl ratios could get most of the light from a point source down a narrow slit at shorter wavelengths, where increased thermal background is not a factor - but at these wavelengths dark current is as large as sky background, so reduced sky provides little gain. Opening the long camera slit to the same width would provide the same slit losses, but at the price of a loss of a factor of three in either spectral resolution or spectral coverage. Not only is this equivalent to reducing the 1024 array to only 340 pixels, but there may be a significant noise penalty due to spreading the light onto many more pixels due to read noise and dark current.

The short camera provides even greater gains - up to a factor of three - for extended objects, whose angular sizes are greater than the telescope image size. Most extragalactic objects, even those at high redshift, are well-resolved on scales of a few tenths of an arcsec. (It is in fact their "fuzzy" nature that allows them to be distinguished from foreground stars in deep infrared surveys, most of which are carried out with effective pixel sizes in the range 0.5 to 1 arcsec.) The coarser pixel scale also allows use of a longer slit, although the optical performance does not permit a full factor of three increase.

The wider slit width of the short camera makes it less sensitive to differential refraction. For observations in a single band differential refraction of 20% of the slit width does not occur in the J band until zenith distance exceeds 40^o, and in the K band one can go to 70^o.

Since it is possible to switch quickly between the two scales, inclusion of the short camera does not preclude use of the long camera when better than average seeing conditions present themselves. This is a pronounced advantage over designs in which the cameras are only interchangeable by opening the instrument.
 
 

2.2.2 Cross-Dispersion

Use of cross-dispersion is by now a well-established concept. The instrument will use an efficient prismatic cross-disperser which will give coverage from 0.9 m m (for overlap with CCD spectrometers) to beyond 2.4 m m where the background begins to rise exponentially. Cross-dispersion offers several significant advantages:

- Multiplex advantage. Instead of having to measure three or more separate spectra, one can cover the whole wavelength region from below 1 m m to 2.4 m m with a single grating setting. The maximum multiplex advantage is a factor of five, though under normal circumstances the lower backgrounds at shorter wavelengths mean that the real gains are more like a factor of 2-3.

- Simultaneity. For scientific programs that need to compare intensities of features from one wavelength region to another, cross-dispersion is indispensable. This is because seeing and transparency variations will inevitably cause the normalization of separate observations to be different. One can of course correct for this by observing each wavelength region with a very wide slit - under photometric conditions - and then using these data to correct the data taken with a narrow slit. The time required to do this is such that a 4-m telescope equipped with an efficient cross-dispersed spectrometer will be eminently competitive with the Gemini telescopes without cross-dispersion. Example of programs that require such observations abound - for example, at low redshift, comparison of Brg to Pab for reddening determinations in obscured objects; comparison of the He I lines at 2.058 and 1.083 m m, where the ratio is an indication of the optical depth of the Lyman continuum. At high redshifts, measurements of the Balmer decrement in quasars or proto-galaxies require measurements in the H and K windows.

- Higher efficiency. Since order-sorting is not required, the filter used needs only to block out long-wavelength radiation, and can thus have substantially higher transmission than a broad-band order-sorting filter. The presence of multiple orders on the array simultaneously means that one can use light from more than one order at a given wavelength. This is of particular importance near the edges of the atmospheric windows, where the grating is not run at peak efficiency and there will be significant power in an adjacent order. Furthermore, the regular order-sorting filters will cut off the edges of the bands, at wavelengths where the atmosphere may not yet be completely opaque. This is especially true for a dry site like Mauna Kea, and for the shorter wavelengths. Astrophysically interesting lines do not always occur cleanly in the midst of atmospheric windows, so the ability to observe "into the grunge" is at times desirable (consider cases such as the H₂ v=1-0 Q-branch, near 2.4 m m, [Si VI] 1.962 m m, or, worse yet, Paa at 1.87 m m.)

The cross-dispersers will work below 2.5 m m only (and there will be transmission losses beyond 2.4 m m). At longer wavelengths, the changes in background are too great to try to include with the shorter wavelengths - and suitable materials do not exist in any case. The maximum slit length available with the cross-disperser is, of course, much shorter than in the simple long-slit mode; since the cross-dispersion is not uniform the maximum slit length is ~100 pixels if a 1Kx1K array is used. This is good enough for measurement of point sources with the short camera, and with the long camera under good conditions (which is when one would be using it anyhow).

Because one is observing over a large wavelength range, differential atmospheric dispersion will be greater than the slit width with the long camera at even moderate zenith distances. Under these circumstances, it will be necessary to observe with the slit oriented at or close to the parallactic angle (in elevation) so that light is centered on the slit at different wavelengths. A suitable choice of observing parameters will allow one to observe a range of position angles on any object, but for most objects this range will be less than 180 degrees (though typically greater than 90 degrees).

It is important to emphasize the size of the effect - if one wants to cover the full range 0.9-2.5 m m, the differential refraction on Mauna Kea is 0.02" only 5^o from zenith, 0.06" at 15^o, and 0.3" (the full width of the short camera slit) at ~52^o.

2.2.3 Polarization Analyzer

The mechanism proposed to handle the cross-dispersers (one for each camera) will also include the mirror for long-slit work. In addition, a MgF₂ Wollaston prism will be mounted on the same mechanism, to allow polarimetry with a modest slit length (>10 arcseconds). An adequate slit length is important for polarization work, since the most interesting polarization studies for the 8-m telescope are likely to be those involving polarization structure in extended objects - disks around young stellar objects, or polarization around Seyfert nuclei.

For work with the analyzer, a rotating half-wave plate will be needed inside the instrument support structure, as part of the telescope facility. For these observations, the object would be observed primarily at different wave-plate rotation angles, with slit position a secondary variable. (Although the converse is equally feasible if desired.)

2.2.4 Higher Spectral Resolution.

The first-light instrument configuration provides a maximum resolution of 18,000 (fully sampled), with the long camera. At this resolution, with the narrow slit, the primary sources of noise below 3 m m will be dark current and read noise, so that very long exposures would be preferred, and the stability of the dark current and array "bias" levels is likely to be critical.

The intermediate spectral resolution - R=6000 - is provided by the short camera as well, so there

will be a substantial drop in limiting magnitude to go to R=18,000 unless the object can be used with the AO system, which may well be the case.

2.2.5 Different Kinds of Objects

The scenario outlined above contemplated observations of point sources that could be readily seen and centered in the slit. Other kinds of observations are, of course, possible:

-Invisible objects. While the sensitivity of the acquisition mode should be very good, one can imagine observing objects for which imaging would be slow, and yet there would be some hope of obtaining a spectrum (for example, a faint emission-line source). In these cases, one would need to have an accurate offset from the object position to an object that is detectable by the slit viewing system, and would offset from the reference position to the object position using the WFS.

-Extended objects. There are many possibilities. If the object fills the slit, it is necessary to offset to a sky position when observing, since it is no longer possible to observe multiple positions on the slit. Programs may require multiple slit positions, either stepped or rotated about a common reference point. Use of the OIWFS and the slit viewing system should make such set-ups exteremly accurate and not too time-consuming.

2.3 Data Reduction and Data Quality Assurance

Data reduction and quality assurance procedures will follow methods developed at NOAO during the last seven years for the reduction of data obtained with long-slit spectrometers. In this regard, it is important to realize that there is not a rigid division between the procedures needed for data quality assurance and those used for data reduction. In particular, for long observations of faint objects, which will generally be lengthy and with low data rates, a skilled observer should be able to perform much of the data reduction in near-real-time, and should be able to accurately monitor signal to noise or other data parameters in order to use telescope time most efficiently. At the other extreme, for calibration observations, including most standard-star observations, the data reduction overheads will be longer than the rather short observations, so that the observer will end up using conservatively estimated values for integration times or numbers of frames.

2.3.1 Data Quality Assurance

Data quality assurance can be thought of as two different functions: verifying that the instrument is functioning properly, and verifying (or ensuring) that the required signal to noise (or other measurable quantity) is being obtained on a given object. It is certainly possible to estimate the latter if the object’s magnitude and morphology are known and the seeing profile, transparency, and sky background are accurately known, and such estimates should be used in setting up an observation in a night’s program (whether queue or classical) - but the combined uncertainties plus uncertainties in the spectrum to be measured make absolute reliance on such formulae quite risky.

The first stage of data quality assurance is primarily directed at confirming proper operation of the instrument through the quick-look task. Normally one would display a frame after subtraction of a reference image and division by a flat field; statistics would be displayed for a selected area in the image but visual inspection would also be useful in looking for low-level fixed-pattern noise. It is important to realize that the "reference" image will almost never be a calibration frame such as a dark or bias; instead, a preceding frame in the sequence of observations would be used so as to provide first-order sky subtraction. (Note that programming of the frame to be used should be part of the set-up of the observing sequence, and ought not to require observer intervention.) Since the telescope is being moved between observations, the displayed image should consist of a positive and a negative spectrum offset by some amount (unless the object is quite extended, so that the reference position is blank sky, in which case only one spectrum will appear). Sky subtraction will doubtless be imperfect, so weak residual sky lines will probably be present as well. In order to examine array performance, the statistics area should be chosen in a well-behaved portion of the array, preferably one with no spectrum or at least no strong features.

For bright objects, such as standard stars, this level of quality assurance should be enough, if one uses observing parameters conservatively estimated from ambient conditions. For faint objects, where larger amounts of telescope time are involved - tens of minutes to hours - it is clearly desirable to continue processing data to the point of performing proper sky subtraction and then extracting and co-adding spectra in order to accurately estimate signal to noise (usually the relevant parameter). An estimate based on data taken part-way through an observing sequence would be used to modify number of observations in the sequence in progress. It is not necessary (nor even desirable, because of the statistical bias introduced - the "stop when you reach ten sigma" effect) to use all the data to be obtained in making these estimates. Since this processing cannot be performed in real-time (because of the need for subsequent frames), it will be performed in near-real-time, and may lag the data acquisition process by minutes or even tens of minutes. Details of this processing are discussed in the data reduction section, below.
 
 

2.3.2 Data Reduction - Sample Program

Data reduction will follow standard methods. All frames are linearized and bad pixels are masked. Then, a flat-field image will be constructed by subtracting "lamp-off" images (images viewing the calibration source with the lamp off) from the flat-field images obtained of the calibration continuum source. Next, the spectrometer response and lamp wavelength response will be removed by first summing the flat-field signal in the cross dispersion direction. The summed signal is then fitted with a low-order function and the fitted function is used to divide the flat-field images. Finally, the flat-field images are averaged after rejecting deviant pixels values to form the final flat field. Flats from the start and end of the night should first be reduced separately, and compared to verify instrument performance, and should then be combined. (Flats from the beginning of the night only can be used for quality assurance reductions during the night. If overall stability is good enough - to be verified in practice - it may not be necessary to obtain end-of-night flats.)

The reduction of the telluric calibration star spectra proceeds as follows: First, a median image is formed of all the star observations after rejecting deviant pixel values. Since the star is observed at a different slit position each time, this median image contains the median sky signal as well as the dark. This median sky image is then subtracted from each of the star observations. Each median-subtracted stellar image is then divided by the flat-field image. Finally, the stellar spectra are extracted using an optimal extraction algorithm which rejects bad pixels and also fits the sky on either side of the source to remove any residual sky which may be present due to sky variability. The wavelength calibration of each spectrum is obtained by extracting the same trace from the median night sky image. The pixel positions and wavelengths of the night sky lines are then fitted using a low order fit to derive a dispersion solution. The dispersion solution is then used to generate dispersion-corrected spectra. The extracted, dispersion-corrected stellar spectra are averaged following rejection of deviant pixels. Finally, any stellar absorption lines are removed from the spectra by dividing the spectra by an appropriate model atmosphere. Since only model atmospheres for fairly hot stars are well understood at this level of accuracy in the near-IR, the spectral type of the telluric star must be chosen accordingly.

The object observations are reduced similarly to the telluric calibration star spectra. As a final step, they are divided by the spectra of the telluric star observations which bracket the observation. For this particular program, reduction of every 50 minutes of integrations (5 slit positions) would be done while the next 50 minutes are being obtained, in order to confirm estimated signal to noise for the particular star.

Flux calibration standards would be observed and reduced like program objects, and the resulting spectra would be used to determine a flux calibration. Since division by the telluric standards has removed atmospheric features, the flux calibration curve should be a smooth, low-order function. For this particular program, the relevant quantities are equivalent widths of absorption lines, so a flux calibration is not strictly necessary.
 
 

2.3.3 Data Reduction - General Issues

Philosophically, the key point is that it is not worth worrying about single frames at all. One takes a series of images (an observing sequence) to produce a single observation of a given total integration time. The concern is being able to reduce the observing sequence and not about reducing a single piece of it. Taking a quick-look difference is usually all that is needed for quality control on individual images during an observing sequence. There are several slightly different recipes for data reduction, and the optimal recipe depends on the nature of the program, observing conditions, and the actual performance of the telescope and spectrometer.

2.3.3.1 Preliminary reductions. These are not dependent on the data type.

1 - Linearize

2 - Mask bad pixels based on low response, high dark current, poor stability

3 - Dark subtract with an equal integration time and co-add images. (This step only really required for those cases where both frames are likely to be scaled before addition or subtraction to adjust sky levels; where this is not the case, the final "reference" frame subtraction will also remove dark.)

4 - Division by flat-field

5 - Distortion correct and wavelength calibrate - Keep track of masked pixels while doing this.

These steps require: flat field images, dark images at appropriate exposure times, and wavelength calibration images (which will normally be night sky emission-line spectra).

2.3.3.2 Sky Removal and Spectrum Extraction

The data basically separate into two classes:

    -data taken very quickly so the night sky does not significantly vary

    -data taken with long exposures where the night sky does vary.

The reason for the two classes is that if the sky varies significantly during a series of exposures taking a median does not make sense. Unlike imaging you cannot scale spectra when combining them into a median since different night-sky features will vary by different amounts. Thus, medians only make sense when the sky has not had time to vary significantly.

For quick observations (e. g. a telluric absorption or flux standard) one can take exposures of the star at multiple positions along the slit and then take a median to form a sky frame. The reduction procedure would be as follows:

    a - Steps 1 - 4 of the preliminary reductions (2.3.3.1)

    b - Median-combine the images and subtract the median from each frame.

    c - step 5, using the sky frame created in (b)

    d - Trace and extract the spectrum from each image, fitting a low order polynomial in the cross dispersion direction to
           remove residual night sky variability.

    e - combine the extracted spectra to form the final spectrum

    f - compute the variance from the variance in the spectra as they are combined.

This should probably be considered the default recipe for near-line data reduction as used for quality assurance, although it should be possible to specify other preferences.

For faint objects, where one needs to take background limited exposures, one takes a series of spectra with the object shifting along the slit each exposure. Now, though, one should average the adjacent spectra to produce the off. The exact number to average is not critical since the noise in the final difference will be dominated by the noise in the original image and the systematics due to sky variability. For such an object the procedure is as follows:

    a - Steps 1-4 of the preliminary reduction (2.3.3.1)

    b - Average the (four) temporally closest images and subtract them from the image

    c - Step 5 on the sky frame produced in (b)

    d - Remove the residual sky emission by fitting a polynomial which excludes the object positions in the cross dispersion
    direction.

If the field is crowded enough to have additional objects on the slit, a straight average may lead to confusion, and a median - or scaled median - will prove to be the lesser evil. Standard reduction recipes will need to be provided for all of these cases.

If the object is bright enough to extract well in a single observation it can be extracted and combined on a frame by frame basis:

    e - trace and extract each spectrum

    f - combine the extracted spectra excluding bad pixels to form the final spectrum

    g - compute the variance of the final spectrum from the variance of the combined spectra

If the object is not bright enough to extract from a single image the images must be combined before the extraction:

    h - shift in the cross-dispersion direction and combine all images excluding bad pixels

    i - generate a variance image by computing the variance of each pixel in the combined image.

    j - Trace and extract the final spectrum and its variance.

If this recipe is applied to the observations in a sequence the variance can also be estimated by combined the spectra produced by several sequences.

For extended objects the procedure is the same except you would move off to sky on every other exposure.

For cross-dispersed spectra the procedures are basically the same, except that one must first "strip out" and straighten the individual orders to make them look just like normal 2-D spectra. Then it is just a matter of book-keeping to keep track of which spectra come from which order.

2.3.3.3 Absorption and Flux Correction

The final step of the reduction is then to use the telluric standards and any flux information you have to flux calibrate and correct for telluric absorption. While not always part of quick-look reduction telluric cancellation can determine the overall quality of data and should be carried out if time permits.

    a - Divide the object spectrum by the telluric correction star

    b - Divide the telluric absorption corrected spectra by a model atmosphere for the given stellar type of the telluric correction
    star

    c - Apply a flux correction based on the observations of stars with known flux.

This procedure works best when the object and the standard have been observed over the same range of slit positions, since this compensates for any curvature effects in the cross-dispersion direction. It will work less well when signal levels are variable, as would be the case with variable seeing or transparency. In these cases, or where strong atmospheric absorption present and there is reasonably high signal to noise on the objects, it may be preferable to match object spectra to standard spectra taken at similar slit positions, perform the division, and then combine the results from the different positions.

NOAO Intranet Services

NOAO Copyright  Statement