1.1 Introduction

The Keck II Deep Imaging Multi-Object Spectrograph (DEIMOS) is a general purpose, faint object, multi-slit, double-beam spectrograph whose key features are wide spectral coverage (up to 5000 Å per exposure), high spectral resolution (~1 Å), high throughput, and long slit length on the sky (~ 32'.6, sum of both barrels). DEIMOS operates in three modes: direct imaging, single-object spectroscopy, and multi-object, long-slit spectroscopy. Intended as the principal optical spectrograph on Keck II, DEIMOS is a powerful multi-purpose instrument tuned to many of the scientific questions that drove the construction of the Keck Telescopes.

DEIMOS is designed for high performance across the blue and visible regions for l > 4200 Å. The blue limit is conditioned by the planned use of silver-coated optics on Keck II. Use of silver optics on both the spectrograph and telescope boosts throughput redward of 4500 Å by 30% over a comparable spectrograph on Keck I. Further gains come from the use of very large cameras and detectors.

Throughout this report we often refer to LRIS and HIRES. LRIS is the faint-object spectrograph on KECK I and was built by Caltech. HIRES is the high-resolution spectrograph on Keck I and was built by Lick. DEIMOS has much inheritance from both: LRIS with respect to optics, and HIRES with respect to instrument control systems, data electronics, and software.

1.2 Basic Design Philosophy and Goals

DEIMOS is fundamentally optimized for faint-object spectroscopy of 1) individual point-sources, 2) low-surface-brightness extended objects, and 3) samples of faint objects distributed on the sky. It is also designed for efficient direct imaging. To these ends, the following performance goals have been established:

1.2.1 High Spectral Resolution

The major enemy of faint-object spectroscopy in the red beyond 7000 Å is contamination by bright night-sky OH lines. Faint galaxies at the photon limit of Keck are 10% of the brightness of the sky continuum flux but only 1% of the brightness of the OH lines. Minimizing OH contamination and subtracting residual OH to high accuracy are major requirements.

OH lines are intrinsically narrow delta-functions distributed irregularly throughout the red region of the spectrum with a typical spacing of ~15 Å. Figure 1.1 shows a typical tracing of the OH lines made with a FWHM of 4.8 Å. At this resolution the OH lines begin to overlap. An important planned operating mode for DEIMOS therefore uses the 1200 mm-1 red grating. Such a grating yields a net spectral FWHM of only 1.0 Å near 8000 Å (see Table 1.4), which is narrow enough that much of the spectrum will be uncontaminated by OH. The high resolution needed for good OH subtraction also yields high scientific benefits, as described in Appendix A on sample scientific programs.

1.2.2 High Mechanical Stability

Residual OH contamination must be subtracted to high accuracy. The major obstacle in current Lick and Keck spectrographs is caused by fringing in the CCD detectors, coupled with mechanical flexure in the spectrograph. Fringing is due to internal interference between the front and back surfaces of the detector. Fringing is present in every thinned CCD yet manufactured and sets in strongly at about 7000 Å, increasing to the red. Figure 1.2 shows typical fringing in the present Kast red detector at Lick. From an extensive battery of tests on various detectors in the laboratory and at the telescope, we have determined that fringe amplitudes in spectroscopic mode typically range from 10-30%, and the spacing (wavelength) is often as short as 10 px.

If fringing is present, mechanical flexure in the spectrograph/detector will couple to it in an unfortunate way. Flexure moves the image in wavelength on the detector so that the wavelength of light illuminating a pixel changes. This changes the phase of the fringes, so that the observation and its flat-field no longer coincide. Flat-field can amount to a few percent.

The above scenario is the prevailing explanation for flat-fielding errors in astronomical CCD spectrographs. So far we have accepted this explanation and adopted an extremely tight stability goal of 0.1 px for wavelength shifts. Recent thinking has caused us to question this explanation and, even if true, to loosen our spec from 0.1 px to 0.25 px. This issue is discussed further in Chapter 7. For the body of this report we continue to adopt a stability goal of 0.1 px.

The basic need for a tight stability goal is driven by the assumption that fringing will in fact be typical in the DEIMOS CCD detectors. This seems reasonable since no fringe-free thinned CCD has as yet been demonstrated. However, developments in progress may succeed in creating a thinned yet fringeless CCD. If that could be accomplished, we might be able to loosen the stability requirement and simplify the proposed active Flexure Compensation (FC) System in Chapter 7.

However, it is also clear that having an intrinsically stable, low-flexure instrument will offer important broader advantages. The proposed FC system can only steer the image in x and y but cannot compensate for rotation of the image. That must be controlled by good structural design. Furthermore, even if fringing were eliminated, it still would be desirable to be able to place the image on the detector in a reliable and repeatable way. This is so because, unlike other astronomical spectrographs, DEIMOS will not have continuously variable grating tilt mechanisms. Each grating will have only three (or at most four) separate tilt positions, a simplification made possible by the extremely wide spectral coverage per exposure. Hence DEIMOS lends itself naturally to "discrete observing modes." If stability were high enough to permit long-lived calibrations for each of these modes (the Space Telescope model), then calibration overhead on a nightly basis would be much reduced and remote observing would also become much easier.

Thus, with or without FC, we place a high premium on achieving an extremely rigid spectrograph with the goal of getting as close as we can mechanically to the ultimate stability requirement of 0.1 px. To that end, we have chosen to simplify the design problem by placing the instrument on the Nasmyth platform, where its axis of rotation remains fixed (horizontal) with respect to gravity.

1.2.3 Large Detectors, Wide-Angle Cameras, and Long Slits

High resolution by itself gains little if limited to short stretches of spectrum. Wide spectral coverage at high resolution needs many pixels (large detector) and a camera with a wide acceptance angle. These considerations are based purely on single-object spectroscopy without reference to long slits. However, once in place, these goals also enable very long slit lengths. By optimizing conditions for single objects, we have automatically created the conditions for a powerful multi-slit spectrograph. We have further exploited this situation by adopting a double-beam layout with two identical sides, and hence twice the total slit length.

1.2.4 High Overall Observing Efficiency

As stated, we seek to minimize instrument calibration time and overhead. Further efficiencies include rapid object acquisition, good guiding, and fast centering of multi-slit masks. An automated acquisition plan for finding objects and centering them on the slit masks has been developed. A novel TV-reticle design will eliminate flexure between the TV and the focal plane to ensure good offset guiding. Other proposed conveniences include continuous monitoring of telescope focus and aberrations, rapid changeover between spectrograph configurations, procedures to monitor optical alignment and throughput, and algorithms to quickly set up and focus the spectrograph.

High efficiency also requires high photon throughput and sensitive detectors. The baseline detectors consist of thinned chips for high QE; they will also be anti-reflection coated to increase QE further. All-silvered optics increase the throughput by 30% over aluminum. We have searched extensively for the most transparent camera glasses that meet our optical needs. A final requirement is good optical coatings, which are important since we have two extra surfaces in the design (a window), plus a multi-element camera. Strategies to address these issues are described in the following chapters.

1.2.5 Flexibility, Cleanliness, and Ease of Maintenance

The final design goals consider long-term performance. DEIMOS is an expensive facility instrument that will likely be in service for many years. One may think of it as an optical bench that may someday be put to other uses. As the mechanical design is being developed, we are trying to keep open the possibility of switching cameras one day, to install blue cameras, should that be desirable. Other possible uses of DEIMOS include narrow-band imaging and drift-scanning. In all cases we are trying to look ahead to anticipate future uses of the DEIMOS platform to make sure that upgrades are not precluded by the present design. Consistent with this longterm view, we avoid reliance on known short-lived components and systems, whether hardware or software. Required maintenance should be infrequent, easy to carry out, and minimize risk to personnel and to the instrument. Components that require replacement should be inexpensive and easy to install. Since we are using silvered optics, the interior of the spectrograph should be exceptionally clean, which is made more difficult by the fact that the focal plane is wide open in direct imaging mode. We have therefore decided to install a transparent window immediately behind the focal plane to allow the spectrograph to be continually purged with dry nitrogen.

To summarize, the principle performance goals of DEIMOS are:

A final point concerns detectors. We are aware that the called-for 8K by 8K detectors are beyond the current state of the art. Despite our best plans, we may not succeed in getting full detectors in time for first light, and we may have to go to the telescope with an interim solution. Nevertheless, we feel strongly that it is the right choice to push the optical and mechanical design of DEIMOS to the outer limits permitted by the Keck telescope. This strategy is advisable because the current generous funding package for the Keck II instruments is not likely to recur in future. If we designed to current detector sizes, the resultant instrument would still be very costly yet would soon have to be replaced as detector capabilities increased (this has already happened with the Keck I first-light instruments). Hence, in planning DEIMOS we have taken advantage of the current funding window to create a flexible, upgradable platform for faint-object spectroscopy that should remain state-of-the-art into the foreseeable future.

1.3 Basic Layout

The current design has the following basic parameters:

The basic instrument layout is shown schematically in Figure 1.3. Light enters from the Keck II tertiary mirror at left, passes the curved slit mask in the focal plane (open for direct imaging), strikes a collimator mirror, and is divided into two beams by a V-shaped pair of "tent" mirrors on the optical axis. After striking a grating (a flat mirror in direct-imaging mode), light enters one of the two refractive cameras, passes a filter slide and shutter, and is imaged on one of two mosaic detectors. Two offset TV guiders (described in Chapter 2) pick off starlight at two opposite points in the focal plane. One of them is also convertible into a direct slit-viewing TV. The entire instrument sits in its own cradle instrument rotator on the Nasmyth platform (Figure 1.4). The cradle bolts securely to the platform on a kinematic mount for observing and unbolts and slides to a stow position when not in use (Figure 1.5). In Figure 1.4, the tall structure beside the spectrograph is the Electronics Vault, a thermally controlled enclosure for the instrument control electronics.

End-on views of DEIMOS are shown in the left part of Figure 1.3. At bottom is a face-on view of the active area of the focal plane. The two shaded areas show the entrance apertures for the two beams. Their outer extent is roughly set by the 10' radius of the Keck f/15 focal plane. The middle of the field, which is obstructed by the tent mirrors, is occupied by the offset TV guider optics and the slit-viewing optics (not shown).

The short rectangles show the grating trays. The choice of grating/mirror sizes is discussed in Chapter 2. A large circular disk around the circumference at the front of the instrument provides both support and the driving surface for the instrument rotator. The rear bearing is smaller and is integrated with the cable entry.

The assumed detector arrays are mosaics of 2K x 4K CCDs with 15m pixels. Each mosaic has 8 CCDs in a 2 x 4 array, for a total of 8K x 8K pixels. The image of the slit-mask holder on the detector array is shown in Figure 1.6. It is offset so that direct images will fall on only half the array, saving data storage space. Spectra will fill the whole array. Each individual CCD is three-side buttable with 1 mm gaps on the long sides and a goal of £ 100m gaps (7 pixels) on the short side. The spectral direction is vertical and spans the small gap.

The total slit length covered by the detector is 17'.4, and the total active slit length (not counting detector gaps) is 16'.6. The nominal telescope focal plane overhangs the detector slightly at each end and is not used. The camera FOV (radius 11.4) also overhangs the detector so that the images will still be good well into the corners. The extra imaging area at either end of the slit but off the detector will be used to locate the auxiliary CCDs for the Flexure Compensation system (Chapter 7). These are the small rectangles at left and right of the main array.

The last major mechanical component on the spectrograph is the Slitmask Cassette Holder/Changer mechanism. Two of these (one for each side) sit in front of the focal plane (Figure 1.3). A total of 10 slots will be available on each side, 2-4 of them containing observatory-supplied slits (long slits and focussing grids) and 6-8 of them loadable by the user. When installed in the focal plane, each slitmask can flip up and down to permit quick changeover between imaging and spectroscopic modes.

The last major system component is the Slitmask Cutting Machine. The present choice for this machine is a laser cutter with three-axis drives, which permit cutting slits of arbitrary size and shape on a curved focal plane surface. Each of these components, plus electronics and software, are described in the following chapters.

1.4 Summary of Expected Capabilities

Tables 1.1-1.5 summarize the performance goals and specifications of the design. Table 1.1 describes the basic parameters, overall layout, and certain aspects of image quality. Table 1.2 presents detailed parameters for imaging, and Table 1.3 does the same for spectroscopy. Table 1.4 is a more detailed look at achievable spectral dispersions and resolutions. Table 1.5 summarizes the currently proposed discrete grating tilts and spectral ranges for each one.

Noteworthy aspects of the design include:

Future upgrades under study:

1.5 Assessment of Major Challenges

For the Committee's benefit, we list here our own rank-ordering of the major challenges in constructing the instrument. Each of these is discussed further in the following chapters.

  1. ) Detector Mosaic: The baseline detectors are beyond the current state of the art. An independent two-pronged development attack is being mounted from separate Keck funds at essentially no cost to DEIMOS. If these paths fail, we will install small interim double mosaics of thinned CCDs and of size 2K x 8K to permit efficient single-object spectroscopy and minimal long-slit capability (4'). This would require 4 detectors total and would not exhaust the detector budget. Money saved would be banked for future developments. This strategy would satisfy single-object users but would be a major blow to multi-slit spectroscopy programs.
  2. ) Optics Fabrication: The main perceived risk is breakage of the CaF2, which costs approximately $25,000 per blank. We will probably have Optovac edge-finish and rough-grind each piece to minimize our total exposure. The aspheric camera elements are comparable in difficulty to previous aspherics we have made, but there is a cumulative risk to schedule and budget in doing a total of six of them (two beams).
  3. ) Mechanical Control of Spectrograph Flexure: This is another unfamiliar item at the level of stability we are trying to achieve. However, the proposed FC system provides some insurance, as does the possibility of a future fringeless CCD. The failure mode is more graceful than preceding high-risk items.
  4. ) Slitmask/Changer Mechanism: This is a complicated mechanism, must operate in all orientations, must be extremely reliable, and is the least familiar of the high-risk items. We have never made a mechanism like this before. On the positive side, we can construct an early prototype and life-test it in the lab.
  5. ) CCD Data Handling System: The data rate is 32 times greater than the current LRIS and HIRES systems, which are our previous benchmarks. To a great extent we are relying on future computer hardware improvements that are not on the market yet. These include bigger and faster disks, faster CPUs, more memory, faster networks, and more capacious archive media. We are also relying on the inclusion of particular upgrades in the second-generation UCSD CCD controllers. On the positive side, failure is again more graceful than other high-risk items.
  6. ) Total Software Burden: The total instrument system is highly software intensive, more so than either HIRES or LRIS. Operating efficiency depends to a great extent on the availability of convenient software packages (e.g., object acquisition and centering, rapid spectrograph setup and focussing, telescope focus monitor). Many of these software tasks are new and involve higher data rates, quick-look display and analysis of huge images, on-line analysis of TV images, etc. Each task is in itself little problem, but the total requirements are large. Careful priorities need to be set. To compensate, other software elements are familiar given previous experience with HIRES, including motor control, the basic architecture of the UCSD controllers, and the architecture of the data-handling system.
  7. ) Slitmask Cutter: This is a high-technology piece of expensive equipment that is vital to the multi-slit mode. We may be relying on an unfamiliar vendor who is new in the business. On the positive side, we can purchase this item early and exercise it in the lab.
  8. ) Total Budget Overrun: Again, the concern is not so much for any one task but rather that the total instrument contains many difficult challenges. There is little fat in the budget to deal with a major disaster. We propose to address this by putting the most difficult tasks up front where possible, where they can get early attention. However, some items such as detectors are paced by considerations beyond our control.

1.6 The Second Beam

The budget presented in this report refers to a single-beam instrument, which is all that we have funding for at present ($5.3 M). A component-by-component analysis indicates that a second beam can be constructed for an additional $1.5 M. A major task for the coming year is to raise this money privately. However, the current schedule does assume two beams. To avoid rework, we will have to know by approximately next summer whether we are going ahead with the second beam. After that, the transition will begin to cost money. We are buying the extra glass needed for the second beam up front, as that seems prudent. All other purchases for the second beam will be delayed until next summer.

1.7 A Final Comment on the Double-Beam Design

This section provides some optional background on the reasoning behind the double-beam design. The obvious disadvantage of the double-beam layout is loss of the center of the telescope FOV, where images are best. We have considered this point but have concluded that this outcome is inevitable for any efficient spectrograph/imager at the Keck/ Nasmyth focus. Here is why.

1.7.1 Reflecting Collimator Designs: Two Beams vs. One

Assume that any spectrograph or imager will contain a section where the light is made parallel, and so will have a collimator. The argument is general, however, and is not predicated on the light's being exactly parallel.

The key design decision is whether this collimating element should be refractive or reflective. A reflective collimator, as in LRIS and DEIMOS, offers three practical advantages. It is simple and low-risk to build, it is efficient (when silvered, it passes 95-98% of the light), and it is cheap (the DEIMOS collimator is estimated at only ~$100 K). Epps has demonstrated a general principle of wide-field reflective collimators that, for maximum FOV, the focal plane curvature of the collimator should roughly match that of the telescope. Such a collimator should be placed with its axis on the telescope axis, and its focal length should be roughly equal to or slightly shorter than the telescope focal-plane radius of curvature. Both LRIS and DEIMOS are laid out on this principle.

Such a collimator images the entire focal plane, but its pupil image is naturally located back near the center of the telescope focal plane. That is where the grating must sit. Such collimators therefore tend to produce a FOV that is donut-shaped, with a central obscuration caused by the grating at the pupil. Realistically, the grating must also be supported by a slide or turret, so the total obscuration is larger. To deal with this, there are only two choices:

1) Leave the pupil where it is and image the remaining unobscured FOV with a single-beam instrument. This leads to an offset FOV, as in LRIS. However, the big 11.4 camera in DEIMOS would considerably increase the FOV over LRIS. If the grating turret were circular, the bigger field would curl around the focal plane in two "horns" encircling the turret. The total area for direct imaging would be slightly more than one-half the Keck 20' FOV, but there would be only one slit, of length ~20'.

2) Split the donut-shaped FOV in half, as in DEIMOS. Send the two halves of the donut in opposite directions by diverting them with a central tent mirror. The donut halves get inverted by the extra reflection at the tent mirror and now face each other back-to-back at the focal plane. Obscuration by the tent mirror exceeds that of a grating and its mount slightly so that more of the central field is lost. The total imaging area is only roughly the same as in the first approach despite the two beams, but now one has two slits for twice as much spectroscopy. Thus imaging is no worse and spectroscopy is twice as good, but at the cost of building two beams instead of one.

To summarize, the two-beam approach does not in any way compromise imaging for the sake of spectroscopy -- the imaging FOV is essentially identical to a maximized single-beam, LRIS-style design. The two beams do cost more but in return offer double the spectroscopy, more flexibility in terms of the total complement of filters and gratings that can be carried, plus the opportunity for serendipitous parallel programs with the second beam.

This, briefly, is the philosophy behind the two-beam design. Since faint-object multi-slit spectroscopy is important at Keck yet very time-consuming, it seems reasonable to invest a modest incremental cost ($1.5 M) to double the throughput for this very important activity.

The two-beam design further offers good use of detector real estate for multi-slit applications. Cameras naturally possess a circular field of view. If one fills the whole FOV with direct imaging, the usual approach, there will be no pixels left over for spectra. Conventional designs with square FOVs in the telescope focal plane therefore waste well more than half of their area when used for spectroscopy. A two-beam design whose active focal plane areas are long and narrow does not suffer from this drawback. In DEIMOS, roughly 1/4 of the horizontal detector real estate is used for imaging and 3/4 for spectra. Thus, all multi-slit spectra are guaranteed to share at least three-quarters of their spectral elements in common, an important advantage in many applications.

1.7.2 Refractive Collimator

A refractive collimator would be attractive because it would form a straight-through image of the center of the Keck FOV. This would exploit the best Keck images and also form an accessible pupil on the Nasmyth platform that could easily feed other instruments. However, there are four drawbacks. First, preliminary explorations by Epps show that the optical train and the distance to the pupil would both be rather long, roughly twice the focal length of the current collimator. This does not leave much room for other instruments at the rear of the platform, or for a future Adaptive Optics system in front. Second, the unit's length would make it hard to design and support. Third, even if coated it would lose more light than a highly efficient, silvered reflective collimator. Finally and perhaps decisive, it would definitely cost more, about $1 M.

These considerations mandate against the choice of a refractive collimator.