SpeX: A Medium‐Resolution 0.8–5.5 Micron Spectrograph and Imager for the NASA Infrared Telescope Facility

Abstract

We present the design, construction, and performance of SpeX, a medium-resolution 0.8-5.5 mum cryogenic spectrograph and imager, now in operation at the 3.0 m NASA Infrared Telescope Facility (IRTF) on Mauna Kea. The design uses prism cross-dispersers and gratings to provide resolving powers up to R~2000 simultaneously across 0.8-2.4, 1.9-4.2, or 2.4-5.5 mum, with a 15" long slit. A high-throughput low-resolution R~200 prism mode is also provided for faint-object and occultation spectroscopy. Single-order 60" long-slit modes with resolving powers up to R~2000 are available for extended objects. The spectrograph employs an Aladdin 3 1024×1024 InSb array and uses narrow slits and a spatial scale of 0.15" pixel-1 for optimum sensitivity on point sources. An autonomous infrared slit viewer is used for object acquisition, infrared guiding, and scientific imaging in the wavelength range 0.8-5.5 mum. The imager employs an Aladdin 2 512×512 InSb array that covers a 60''×60'' field of view at 0.12" pixel-1. SpeX was successfully commissioned on IRTF during 2000 May, June, and July. Astronomical observations are presented to illustrate performance.

Full text

362
Publications of the Astronomical Society of the Pacific,115:362–382, 2003 March
䉷2003. The Astronomical Society of the Pacific. All rights reserved. Printed in U.S.A.
SpeX: A Medium-Resolution 0.8–5.5 Micron Spectrograph and Imager
for the NASA Infrared Telescope Facility
J. T. Rayner,
1
D. W. Toomey, P. M. Onaka, A. J. Denault, W. E. Stahlberger, W. D. Vacca,
1,2
and M. C. Cushing
1
Institute for Astronomy, University of Hawaii, 2680 Woodlawn Drive, Honolulu, HI 96822; rayner@.ifa.hawaii.edu
and
S. Wang
3
University of Chicago Engineering Center, 5640 South Ellis Avenue, Chicago, IL 60637
Received 2002 August 31; accepted 2002 November 2
ABSTRACT. We present the design, construction, and performance of SpeX, a medium-resolution 0.8–5.5 mm
cryogenic spectrograph and imager, now in operation at the 3.0 m NASA Infrared Telescope Facility (IRTF) on
Mauna Kea. The design uses prism cross-dispersers and gratings to provide resolving powers up to R∼2000
simultaneously across 0.8–2.4, 1.9–4.2, or 2.4–5.5 mm, with a 15⬙long slit. A high-throughput low-resolution
prism mode is also provided for faint-object and occultation spectroscopy. Single-order 60⬙long-slitR∼200
modes with resolving powers up to are available for extended objects. The spectrograph employs anR∼2000
Aladdin 3 InSb array and uses narrow slits and a spatial scale of 0⬙.15 pixel for optimum sensitivity
⫺1
1024 #1024
on point sources. An autonomous infrared slit viewer is used for object acquisition, infrared guiding, and scientific
imaging in the wavelength range 0.8–5.5 mm. The imager employs an Aladdin 2 InSb array that512 #512
covers a field of view at 0⬙.12 pixel . SpeX was successfully commissioned on IRTF during 2000
  ⫺1
60 #60
May, June, and July. Astronomical observations are presented to illustrate performance.
1. INTRODUCTION
SpeX is a medium-resolution 0.8–5.5 mm cryogenic spec-
trograph designed and built at the Institute for Astronomy,
University of Hawaii, for the NASA Infrared Telescope Facility
(IRTF) on Mauna Kea. The IRTF is a 3.0 m telescope owned
by NASA and operated by the Institute for Astronomy, Uni-
versity of Hawaii, under contract to NASA. It is fully accessible
to the national and international scientific community. Fifty
percent of the observing time is reserved for solar-system ob-
jects, and the other 50% is available for other astronomical
studies.
The ∼1–5 mm region of the infrared spectrum is particularly
well suited to studies of astrophysically cool objects (T!
K), sources heavily obscured by dust, and objects at very2000
high redshift. This wavelength region includes key diagnostic
spectral features, notably the vibration-rotation bands of astro-
physically important molecules (such as CO, OH, SiO, CH,
CN, NH, H
2
O, HCN, CH
4
, etc.), the pure rotation transitions
of H
2
, many atomic and ionic transitions, and many solid-state
bands that are crucial diagnostics of the composition of dust,
1
Visiting Astronomer at the Infrared Telescope Facility, which is operated
by the University of Hawaii under contract from NASA.
2
Currently at the Max-Planck-Institut fu¨r extraterrestrische Physik, Gies-
senbachstrasse, Garching, Germany.
3
Currently at the CMC Flight Visions, 43W752 Route 30, P.O. Box 250,
Sugar Grove, IL 60554.
ices, and the surfaces of solid bodies. The current complement
of IRTF facility instruments includes NSFCAM (Shure et al.
1994), a 1–5.5 mm camera, and CSHELL (Tokunaga, Toomey,
& Carr 1990; Greene et al. 1993), a 1–5.5 mm high-resolution
( –40,000) spectrograph. SpeX fills the gap in spectralRp5000
resolving power by providing low- and medium-resolution
spectroscopy ( –2500) across the range 0.8–5.5 mm.Rp50
2. CHOICE OF SPECTRAL FORMATS
The guiding design principle of SpeX was to optimize ob-
serving efficiency by maximizing simultaneous wavelength cov-
erage using cross-dispersion. The spectral format was designed
to fit the 0.8–5.5 mm Aladdin array. At spectral1024 #1024
resolving powers of a few thousand, the background in the
0.8–2.5 mm regime is dominated by OH sky-line emission. In
contrast, thermal background emission from the telescope and
sky rises by a factor of 10
5
times from 2.5 to 5.5 mm. Therefore,
it is best to cross-disperse and optimize the instrument for these
two regions separately. The spectrograph modes are detailed in
Table 1 and illustrated in Figures 1, 2, and 3.
2.1. Advantages of Cross-Dispersion over Single-Order
Spectroscopy
Cross-dispersion provides three distinct advantages over
single-order spectroscopy:

SpeX IR SPECTROGRAPH FOR IRTF 363
2003 PASP, 115:362–382
TABLE 1
Spectrograph Modes
Mode
l
Range
(mm) R
a
Slit Length
(arcsec) Order Sorter
SXD .......................... 0.8–2.4 2000 15.0 FS/ZnSe prism
LXD1.9 ...................... 1.9–4.2 2500 15.0 LiF/Ge prism
LXD2.1 ...................... 2.1–5.0 2500 15.0 LiF/Ge prism
LXD2.3 ...................... 2.3–5.4 2500 15.0 LiF/Ge prism
Low-resolution (prism) ...... 0.8–2.5 250 15.0 or 60.0 Not required
Short single order ........... 0.8–2.5 2000 15.0 or 60.0 Filters
b
Long single order ............ 1.9–5.5 2500 15.0 or 60.0 Filters
c
a
Average spectral resolving power matched to a 0⬙.3 slit; Ris proportionately
lower with wider slits. Available slit widths: 0⬙.3, 0⬙.5, 0⬙.8, 1⬙.6, and 3⬙.0.
b
Available order-sorting filters: 0.91–1.00, 1.03–1.17, 1.17–1.37, 1.47–1.80,
and 1.92–2.52 mm.
c
Available order-sorting filters: 3.13–3.53, 3.59–4.41, and 4.40–6.00 mm.
Fig. 2.—LXD2.3 mode. Long-wavelength (2.3–5.4 mm) cross-dispersed for-
mat. Shown is an object minus sky image where a star has been nodded 7⬙.5
in the 15⬙.0 slit. At thermal wavelengths, telluric features emit. Depending on
sky background stability, this can result in positive (white) or negative (black)
telluric features in the spectrum.
Fig. 1.—SXD mode. Short-wavelength (0.8–2.4 mm) cross-dispersed format.
Shown is an object minus sky image where a star has been nodded 7⬙.5 in the
15⬙.0 slit. The white lines indicate the top and bottom of the slit. Note the low-
transmission telluric features.
Fig. 3.—LXD1.9 mode. Long-wavelength (1.9–4.2 mm) cross-dispersed for-
mat. Same as Fig. 2, except that the K,L, and bands are covered simul-

L
taneously by rotating the grating turret by 0⬚.95.
1. Efficient use of telescope time.—SpeX can acquire the
entire 0.8–5.5 mm region in just two instrument settings. Cov-
ering this range in single-order mode would require 10 different
settings at and would be compromised by lack ofR∼2000
overlap between orders. The two cross-dispersed modes overlap
at 2.4 mm to allow relative calibration of the two spectra. It is
the reduced number of instrument settings that significantly
improves efficiency.
2. Simultaneous wide spectral coverage.—Simultaneous ob-
servations guarantee that spectra are taken at the same time
and come from the same position. This feature removes the
uncertainty introduced by using sequential observations since
variations in seeing and guiding can introduce normalization
problems when piecing together spectra not taken simulta-
neously. Cross-dispersion is therefore optimum for gathering
widely separated lines and calculating precise line ratios. It is
also indispensable for high time resolution on variable objects
such as comets, asteroids, planets, and variable stars. Additional
efficiencies accrue from providing relatively calibrated spectra
under marginal weather conditions.
3. High throughput.—Order-sorting interference filters are
not required since order sorting is done by the prisms, and

364 RAYNER ET AL.
2003 PASP, 115:362–382
prisms have higher transmissions than interference filters.
Cross-dispersion is also much more efficient at the wavelength
of overlapping orders where order-sorting filters must have high
attenuation to avoid interference from the other orders. Also,
since cross-dispersion sorts orders spatially, wavelengths at the
edge of each order occur at two displaced positions, and signals
from these two positions can be combined to compensate for
the loss of efficiency that results from working away from the
blaze angle at the center of each order.
2.2. 0.8–2.4 mm Cross-dispersed (SXD) Mode
Without adaptive optics, the very best image quality obtain-
able on the IRTF is about 0⬙.3 FWHM at 2.2 mm. Hence, the
pixel size in the spectrograph is set to 0⬙.15 for Nyquist sam-
pling. The cross-dispersed slit length is set to 15⬙(100 pixels).
This is sufficient to allow telescope nodding of point sources
within the slit, eliminating the need for separate sky positions.
With a suitable choice of cross-dispersing prism materials
(ZnSe and fused silica), it has been possible to keepthe spacing
between orders almost constant, even though SpeX works in
low orders to ensure maximum wavelength coverage. This
results in a very efficient use of the array format.
With a slit length of 100 pixels (15⬙), about six orders com-
fortably fit across 1024 pixels, allowing for tilt and curvature
of individual orders and including some interorder spacing to
enable vertical repositioning to avoid possible bad-pixel clus-
ters on the array. To match the center of each order with the
most transparent parts of the atmosphere, the standard approach
of placing 2.2 mm in third order and 1.65 mm in fourth order,
etc., is taken. The blaze (i.e., central) wavelength in orderl
m
mis given by
l
1
lp,(1)
m
m
where the blaze wavelength in first order mminthislp6.6
1
case. The long- ( ) and short-wavelength ( ) limits (theoret-ll
ls
ically 40% of the blaze intensity, although in practice the in-
tensity is less) of each order are given by (Schroeder 2000):
ml
m
lp,(2)
l
m⫺1/2
ml
m
lp.(3)
s
m⫹1/2
With the 2 pixel (0⬙.3 slit) resolving power set to Rp2000
(120 km s ), third order spans the K-band and orders 3–8
⫺1
cover 2.4–0.8 mm as desired (see Fig. 1), but with a small gap
in wavelength coverage at 1.81–1.86 mm in the deep telluric
feature. No-gap coverage across this range would have meant
reducing the resolving power to . A resolving powerRp1500
of adequately separates OH emission lines (roughlyRp2000
30% of the pixels from 0.8 to 2.5 mm are covered by OH lines
of varying intensity).
At a resolving power of on a 3.0 m telescopeRp2000
such as IRTF, the limiting sensitivity depends strongly on array
performance. The sky background between the OH lines at the
detector is about 0.015 es pixel (for an instrument
⫺⫺1⫺1
throughput of 15% and 590 photons s m arcsec mm
⫺1⫺2⫺2⫺1
from the sky at 1.65 mm; Maihara et al. 1993), which is well
below the measured dark current of 0.2 es . The measured
⫺⫺1
read noise is 12 erms with Fowler multiple sampling (Fowler
⫺
& Gatley 1990), and so integration times of longer than
20 minutes are required to become dark current limited. (In
practice, integration time is limited by residual image effects.
See § 8.1 for further discussion.)
2.3. 1.9–5.4 mm Cross-dispersed (LXD) Mode
As was the case with the 0.8–2.4 mm cross-dispersed mode,
a suitable choice of prism materials (Ge and LiF in this case)
keeps the spacing between orders about constant. The slit length
is also 15⬙, and up to six orders can be covered simultaneously.
Unlike the 0.8–2.4 mm mode, however, there is no natural way
of matching the center of each order with the most transparent
parts of the atmosphere. Since the background is too large to
obtain useful spectroscopy on many objects in the Mband
(4.5–5.4 mm), the spectral format has been arranged so that the
Mband spans order 4 and so that the Land Kbands occupy
orders 5–10. This allows the Mband to be moved off the array
if desired. Depending on the chosen fine position of the grating
turret, orders 4–8 (5.4–2.3 mm; see Fig. 2) or orders 5–10
(4.1–1.9 mm; see Fig. 3) fit onto the array at a spectral resolving
power of matched to a 0⬙.3 slit, while again leavingRp2500
space to avoid possible bad areas of the array. Sky brightness
limits on-chip exposure times to about 30 s at 4 mm and to
about 2 s at 5 mm. Consequently, if integration times are set
by the sky background, the LXD mode is read noise limited
at wavelengths shorter than about 3 mm.
2.4. 0.8–2.5 mm Low-Resolution Prism Mode
This mode is designed for occultations and faint-objectspec-
troscopy. The 0.8–2.5 mm region is dispersed by a single prism
onto one quadrant of the array for fast readout.1024 #1024
Since no gratings are required and the spectral resolving power
is low (average matched to a 0⬙.3 slit), spectra canRp250
be acquired quickly and efficiently. In the prism mode, angular
dispersion is a function of refractive index, and resolving power
varies by about a factor of 2 from 0.8 mm( /0⬙.3 slit)Rp150
to 2.5 mm( /0⬙.3 slit). The prism mode also overcomesRp350
some shortcomings of the 0.8–2.4 mm cross-dispersed (SXD)
mode. The slit length is not restricted to 15⬙, and a 60⬙slit can
be used (although there is some curvature for slits this long).
The throughput of the prism mode is high and relatively con-
stant (see § 4.6) across 0.8–2.5 mm. By comparison, the SXD
mode has lower average throughput because gratings are less

SpeX IR SPECTROGRAPH FOR IRTF 365
2003 PASP, 115:362–382
Fig. 4.—Photograph of SpeX on IRTF. The cryostat (blue box) is mounted to the telescope by the black interface box, which contains the calibration unit. The
closed-cycle cooler is visible at the top of the cryostat. Visible on the front cover of the cryostat are motors for controlling the seven cold mechanisms. Mounted
to the side of the cryostat at the lower left is the cryostat-mounted electronics box containing clock, bias preamp, analog-to-digital converter, and fiber-optic boards.
For scale, SpeX is 1.4 m tall and weighs 478 kg.
efficient. SXD efficiency is also reduced at the edges of each
order because of the grating blaze function. The average
throughput of the prism mode is about twice that of the SXD
mode.
2.5. Single-Order Long-Slit Modes
The long-slit modes add a further degree of instrument flex-
ibility on extended objects such as planets, comets, nebulae,
and galaxies, even though the small pixel size (0⬙.15 pixel )
⫺1
is not optimal for low surface brightness objects. To keep op-
tical components within reasonable limits, the longest slit is
60⬙(400 pixels). The two single-order long-slit modes are iden-
tical to the SXD and LXD modes, except the gratings are not
paired with a cross-dispersing prism, and order sorting is done
with filters.
3. INSTRUMENT OVERVIEW
The cryostat is mounted at the Cassegrain focus of the tele-
scope by a rigid interface box, into which is built a spectral
calibration unit consisting of transfer optics, integrating sphere,
flat-field lamps, and an arc lamp. Three sections make up the
cold optics: the foreoptics, the infrared slit viewer/guider, and
the spectrograph. A photograph of SpeX mounted to the tele-
scope is shown in Figure 4. The optical layout and internal
view of the cryostat are shown in Figure 5.
In the foreoptics, the f/38.2 beam entering the cryostat win-
dow first encounters a dichroic turret before coming to a focus.
Visible-reflecting/infrared-transmitting dichroics feed the vis-
ible beam to an optical wave-front sensor package mounted on
the side of the cryostat. The optical path following the cold
telescope focus comprises a collimating mirror, fold mirrors,

366 RAYNER ET AL.
2003 PASP, 115:362–382
Fig. 5.—Left: Optical layout of the cryostat. Right: Photograph of the vacuum jacket and cold structure. The cryostat is shown with its front and back covers
removed. Immediately inside the vacuum jacket is the closed-cycle cooled radiation shield, also shown without its front and back covers. The radiation shield is
mounted to the vacuum jacket by four small fiberglass tabs. The cold structure is mounted to the vacuum jacket by four fiberglass V-trusses, three of which are
visible in this view. The interior of the vacuum jacket, radiation shield, and cold structure are all highly polished to reduce emissivity.
order-sorter filter wheel, 12.5 mm cold stop, K-mirror image
rotator, and camera lens. This system reimages a
 
60 #60
field onto a slit wheel and feeds an f/12.7 beam into the
spectrograph.
The field surrounding the slit is reflected into the slit viewer
by a slit mirror. The slit viewer consists of a collimating lens,
12.5 mm cold stop, filter wheel, camera lens, and array. A
field is reimaged onto a Aladdin 2 InSb
 
60 #60 512 #512
array, giving an image scale of 0⬙.12 pixel . In addition to
⫺1
object acquisition and scientific imaging, the slit viewer is used
for infrared guiding either on spillover light from the science
object in the slit or on an object in the field.
The f/12.7 beam entering the spectrograph is first folded
before collimation by an off-axis parabola. The collimator
forms a 24 mm diameter image of the telescope pupil on a
grating/prism mounted in a grating turret wheel. Following
dispersion, a camera lens images the spectrum onto a
Aladdin 3 InSb array, which is mounted on a1024 #1024
focusing stage. The spatial scale is 0⬙.15 pixel . Because of
⫺1
the wide wavelength range, cross-dispersion is done with
prisms, which are used in double-pass mode. The grating turret
has five positions: two grating/prism cross-dispersed modes,
two grating long-slit modes, and one low-resolution prism
mode. All the lenses are BaF2-LiF-ZnS/Cleartran achromatic
triplets, antireflection coated for 0.8–5.5 mm.
An IRTF-designed array controller runs both the spectro-
graph and imaging InSb arrays. Cryostat-mounted electronics,
consisting of preamps, analog-to-digital (A/D) boards, clock,
and bias boards, are fiber-optically coupled to the array con-
troller, which resides in the telescope control room. The
VME64-based array controller uses digital signal processor
(DSP) boards (each containing four DSPs) and a single-board
SPARC computer. One controller (“Bigdog”) runs the spectro-
graph array, and a second (“Guidedog”) runs the slit-viewer
array. Table 2 summarizes the basic parameters of the instru-
ment and arrays. A PC-based instrument computer (“Little-
dog”) mounted on the mirror cell is used for motor and tem-
perature control and for monitoring tasks. A graphical user

SpeX IR SPECTROGRAPH FOR IRTF 367
2003 PASP, 115:362–382
TABLE 2
Summary of Basic Instrument Parameters
Parameter Spectrograph Slit Viewer/Guider
Detector ....................... Aladdin 3 InSb1024 #1024 Aladdin 2 InSb512 #512
Image scale ................... 0⬙.15 pixel (along slit)
⫺1
0⬙.12 pixel
⫺1
Field of view ................. 15⬙.0 and 60⬙.0 slit lengths 60⬙.0 #60⬙.0
Gain ........................... 13.0 e
⫺
ADU
⫺1
14.7 e
⫺
ADU
⫺1
Saturation level
a
.............. 111,000 e
⫺
88,000 e
⫺
Recommended limit .......... ∼52,000 e(5% linear)
⫺
∼37,000 e(10% linear)
⫺
Read noise .................... 50 erms at 6 Hz
⫺
65 erms at 3 Hz
⫺
12 erms with 32 samples
⫺
20 erms with 32 samples
⫺
Dark current .................. 0.2 e
⫺
s
⫺1
1.5 e
⫺
s
⫺1
Readout rate (standard)
b
...... 0.51 s per sample 0.24 s per sample
Readout rate (fast)
b
........... 0.1 s per sample 0.12 s per sample
a
For a detector bias of 0.4 V, and assuming integration time kreadout rate.
b
For full array. Proportionately less for subarrays.
interface (GUI) runs on a two-screen Unix workstation con-
nected to the array controller and instrument computer by
Ethernet.
4. OPTICS
4.1. Concept
The classic collimator-camera approach is adopted in the
design, and the instrument comprises three major optical as-
semblies: foreoptics, spectrograph, and slit viewer. Each of
these three assemblies contains its own collimator and camera.
The main factor complicating the design is the requirement of
efficient and achromatic performance across 0.8–5.5 mm. All-
reflective designs are very expensive and beyond our budget.
All-refractive designs are generally easier to fabricate and align,
but they suffer from reflective losses due to the difficulty of
optimizing antireflection coatings for such wide wavelength
coverage. We have chosen a hybrid system of collimating mir-
rors and fixed camera lenses. The optics were designed using
the ZEMAX optical design program from Focus Software,Inc.
4.2. Foreoptics
All the required focal reduction (3.1 : 1) is done in the fore-
optics. A cold pupil image is formed in front of the narrow
spectrograph slits rather than behind them. This avoids the
effects of diffractive blurring of the pupil image at the cold
stop by the slits and guarantees good baffling.
4.2.1. Windows and Dichroic Beam Splitters
The up-looking CaF
2
cryostat entrance window is 76 mm
#76 mm to comfortably accept an field. This is
 
80 #80
the maximum field size that the wave-front sensor package can
use. CaF
2
is used because of its good transparency, low scatter,
and low dispersion. A four-position turret includes two CaF
2
dichroics and an open position. One dichroic cuts on at 0.8 mm
and the other at 0.95 mm; both are optimized for transmission
out to 5.5 mm. The dichroics are placed 200 mm inside the
entrance window. This minimizes the solid angle of warm flux
viewed through the window that could be scattered at the di-
chroics and thus increase emissivity. Following the dichroic,
the transmitted infrared beam comes to a focus at a

60 #
(33.4 mm #33.4 mm) field stop located 440 mm inside

60
the entrance window. This has the advantage of defocusing
dust spots on the (warm) window that can be a significant
source of noise at wavelengths longer than about 3 mm. When
the instrument is not in use, a pick-off mirror in the calibration
box is positioned over the entrance window to keep it clean.
The reflected optical beam comes to a focus about 80 mm
outside a side-looking CaF
2
window and on the field lens of
the wave-front sensor package. A cold PK-50 blocker,mounted
in the beam immediately inside this window, prevents thermal
background from entering through the exit window. Both win-
dows are tilted by 5⬚to minimize optical ghosts.
4.2.2. Collimator
The beam diverging from the infrared focus is collimated
by a spherical Zerodur mirror with a focal length of 473 mm
and a diameter of 60 mm. In order to clear the incoming beam,
the collimator is tilted 4⬚.5. Following two fold mirrors, the
collimator images the telescope pupil onto a cold stop.
4.2.3. Order-Sorter Filter Wheel and Cold Stop
A 15-position filter wheel is located in the collimated beam
immediately in front of the 12.5 mm diameter cold stop. This
wheel contains order sorters for the spectrograph, additional
filters for the slit-viewer imager, and an annular stop for spec-
troscopy on objects brighter than about first magnitude. Placing
filters prior to the slit keeps out-of-band photons out of the
spectrograph. Placing filters close to the cold stop and in the
collimated beam keeps them small, maximizing the number of
filters in the wheel, and eliminates the need to refocus when
changing filter. The prism modes require no order sorting (or
blocking), and so any filter in the slit-viewer filter wheel may

368 RAYNER ET AL.
2003 PASP, 115:362–382
be used for IR guiding. The cold stop is located at the front
focus of the foreoptics camera lens to keep the system
telecentric.
4.2.4. Field Rotator
A K-mirror field rotator fits into the space between the cold
stop and foreoptics camera lens. The K-mirror consists of three
inclined fused-silica plane mirrors mounted together in a drum.
The beam enters and exits the rotator along its rotation axis
through the center of two 30 mm diameter rotation bearings.
An important goal of the rotator fabrication was to keep the
optical axis and mechanical rotation axis co-aligned to within
0⬚.1, which is equivalent to an image wander radius of 5⬙,so
that an object close to the slit does not move appreciably across
the field when the position angle is changed. (The
 
60 #60
IRTF is an equatorial telescope, so there is no image rotation
during tracking.) A 3⬙image wander radius was achieved by
building to tolerance and without any adjustment.
4.2.5. Camera Lens
Immediately following the rotator is the foreoptics camera
lens, which reimages the telescope focal plane on the slit. The
camera lens consists of a BaF
2
-LiF-ZnS/Cleartran achromatic
triplet lens. Imaging is achromatic across the range 0.8–5.5 mm.
The calculated average Strehl ratio across the field is 0.85,
0.95, and 0.99 at 0.9, 1.65, and 4.8 mm, respectively. The lenses
were optimized with all spherical surfaces. To keep the mount-
ing (and therefore the alignment) simple, optimization was not
allowed to migrate the lenses more than a few millimeters apart.
All three elements are 35 mm in diameter. The resulting lens
is mounted as a vacuum-spaced triplet in a single lens mount.
Scaled versions of this lens are used for the spectrograph and
slit viewer. Optics for Research fabricated all the cryogenic
lenses.
4.3. Spectrograph
The spectrograph uses a collimator and camera of equal focal
lengths to reimage the slit at 1 : 1 onto the detector. The off-
axis parabolic collimator places the 24 mm diameter pupil on
the dispersing elements, and the dispersed beam is then re-
imaged by the triplet camera lens onto the array.
4.3.1. Slit Wheel and Slit Mirrors
Slit mirrors are inclined at 45⬚to the incoming f/12.7 beam
to reflect a field into the slit viewer for IR guiding
 
60 #60
and imaging. IR guiding is done on objects in the field or on
the spillover light from the science object in the slit. Conse-
quently, the slit mirror quality must be good at the edge of the
slit. Aligning the slits at 45⬚is very advantageous for the overall
instrument layout; however, it does increase the potential for
scattering off the edges of the slit. To minimize scatterinto the
spectrograph, we chose a metal-coated substrate-type slit. A
gold coating with slit aperture is lithographically applied to the
front side of an antireflection-coated CaF
2
substrate. Since the
gold coating is less than 1 mm thick, the knife edge is extremely
sharp. A backside metal coating is used to absorb the 3% re-
flection off the back side of the slit. At 1 mm thick and
25 mm in diameter, the substrates are flat enough for high-
quality focal and pupil-plane images in the slit viewer and thin
enough to add little astigmatism and spherical aberration into
the f/12.7 beam entering the spectrograph. Ten different slits
with widths ranging from 0⬙.3 to 3⬙.0 and an imaging mirror
are contained in the 12-position slit wheel. The slits were fab-
ricated by Max Levy Autograph, Inc.
4.3.2. Off-Axis Collimator Mirror
The f/12.7 beam entering the spectrograph box is collimated
by a diamond-turned nickel-plated aluminum mirror made by
SSG, Inc. The mirror is an off-axis parabola (OAP) with a focal
length of 300 mm and a diameter of 44 mm. The angle between
the incident and reflected beams is 10⬚. Since the spectrograph
employs slits that are matched to the telescope’s diffraction
limit, truncation of the beam at the slit results in diffractive
expansion of the beam. For a 0⬙.3 slit and a geometrically sized
collimator, the light loss is about 12%. The size of the colli-
mator is a compromise between light loss and aberrations in
the camera lens next in the light path due to the increased
diameter of the optical beam. The collimator is therefore over-
sized by 5%, which reduces the light loss to 6% without adding
significantly to the aberrations.
4.3.3. Prisms
In general, the choice of grating (or grism) versus prism for
cross-dispersion depends on the required order separation and
the spectral range to be covered. Gratings usually have higher
dispersion but suffer from changing efficiency when used to
cover a wide spectral range. The spectral range of a grating
cross-disperser is widest in first order ( ; see eqs. [2] andmp1
[3]). For a grating cross-disperser used to cover the range
0.8–2.5 mm, the efficiency falls from a peak of about 0.7 at
1.25 mmto0.4at0.8mm and 2.5 mmfora5⬚angle of incidence
(Schroeder 2000). Another disadvantage of a grating is the less-
than-optimum use of detector area. When working in low or-
ders, the order spacing significantly increases with wavelength
(Schroeder 2000). For example, to cover the 0.8–2.5 mm range
in six orders starting with third order at 2.2 mm and using the
slit length and image scale optimum for SpeX (see § 2.2) would
require an array height of about 2000 pixels, twice that avail-
able, even though each order is only 100 pixels high.
Prisms have high and constant efficiency over their trans-
parency range, but lower dispersion than gratings for the same
collimated beam diameter. To compensate for the lower dis-
persion, prisms are used in double-pass configuration, a feature
that also makes the design more compact. Order separation
with wavelength is determined by the dispersion of the prism

SpeX IR SPECTROGRAPH FOR IRTF 369
2003 PASP, 115:362–382
TABLE 3
Gratings
Parameter
0.8–2.5 mmModes 1.9–5.5 mmModes
XD Single Order XD Single Order
Orientation ........................... Littrow Ⳳ5⬚Littrow Ⳳ5⬚Littrow Ⳳ5⬚Littrow Ⳳ5⬚
Blaze l(mm) ........................ 6.6 6.6 20.0 20.0
Working orders ...................... 3–8 3–8 4–10 4–10
Order sorter .......................... FS/ZnSe prism Filter LiF/Ge prism Filter
Blaze angle (deg) ................... 10.25 10.25 12.73 12.73
Line frequency (mm
⫺1
) ............. 53.90 53.90 22.04 22.04
Working R(slit width) .............. 2000 (0⬙.3) 2000 (0⬙.3) 2500 (0⬙.3) 2500 (0⬙.3)
Beam footprint (ellipse) (mm) ...... 25 #30 25 #25 25 #42 25 #25
Grating size (mm) ................... 45.0 #45.0 45.0 #45.0 45.0 #55.0 45.0 #55.0
Grating substrate .................... Fused silica Fused silica Fused silica Fused silica
Coating .............................. Gold Gold Aluminum Aluminum
material rather than by geometry. In the infrared, it is very
difficult to find single materials with the dispersion needed to
keep the order spacing constant. Our new solution uses two
prisms in series, each made from different materials. For
a dispersing prism used at or near minimum deviation, the
angular dispersion is given by (Schroeder 2000)
dbtdn
p,(4)
dladl
where ais the collimated beam diameter and is thedn/dl
dispersion. The order spacing is made constant by matching a
material in which changes in such a way that orderdn/dl
spacing increases with wavelength, with a material in which
changes in such a way that order spacing decreases withdn/dl
wavelength to exactly compensate. This results in much better
use of detector area compared to grating or single prism cross
dispersers.
By ray-tracing potential prism designs, we have found the
best combination of materials to be fused silica (FS) and ZnSe
in the 0.8–2.5 mm range and Ge and LiF in the 1.9–5.5 mm
range. The average throughput of these broadband antireflec-
tion-coated dual prisms across their range of use is 0.87 and
0.80, respectively, compared to an average of about 0.5 for a
typical cross-dispersing grating.
Each dual-prism consists of two right-angled prisms mounted
back to back. Because of differences in thermal contraction,
the back-to-back prisms cannot be optically contacted, and a
small vacuum gap of about 1.0 mm exists between the two
prisms. To avoid interference fringes forming in the gap, the
prisms are wedged by 0⬚.3. The prisms are circular in cross
section (55 mm in diameter) and are mounted in thermally
compensating mounts in the same way as the lenses. All the
prisms were fabricated by Optics for Research.
4.3.4. Gratings
The gratings are illuminated in near-Littrow orientation (in-
plane) with a tilt of 5⬚to allow the diffracted beam to clear
the incident beam. This orientation is usually not as optically
efficient as illuminating gratings in quasi-Littrow since the grat-
ing rulings are shadowed. However, the shadowing effect is
not significant in SpeX because the blaze angles are small. The
big advantage of near-Littrow over quasi-Littrow orientation is
that the reimaged slit is not tilted with respect to detector col-
umns, thus simplifying spectral reduction. In practice, a slight
amount of tilt results from predispersion by the prisms, and
some curvature is a natural consequence of the finite-length
slits, but these effects are small, about 0.5 pixels over the 15⬙
slit and about 2 pixels over the 60⬙slit. Custom gratings were
fabricated by Zeiss and Hyperfine, Inc. The properties of the
gratings are summarized in Table 3.
The system slit-limited resolving power is given by (Schroe-
der 2000)
dsin v⫹sin v
1ir
Rp,(5)
rfDcos v
r
where is the collimated beam diameter, ris the anamorphicd
1
magnification, f(radians) is the slit width subtended on the
sky, Dis the telescope diameter, is the grating angle ofsin v
i
incidence, and is the grating angle of reflection. For thesin v
r
short-wavelength gratings, Ris set to 2000 at the center of each
order for a 0⬙.3 wide slit and changes from about 1750 to 2250
across the array as a result of the changing angle of reflection.
For the long-wavelength gratings, Ris set to 2500 at the center
of each order for a 0⬙.3 wide slit and changes from about 2200
to 2800 across the array as a result of the changing angle of
reflection.
4.3.5. Grating Turret
Each of the five spectroscopy modes occupies one position
in the grating turret. Spectroscopy modes are selected by ro-
tating the continuous-drive grating turret to place a selected
grating/prism combination into the collimated beam. Rotation
of the turret moves spectra in the cross-dispersion direction,
and fine rotation control is designed to position spectral orders

370 RAYNER ET AL.
2003 PASP, 115:362–382
Fig. 6.—Optical layout of the SXD mode components in the grating/prism
turret.
to a precision of 1 pixel (0⬙.15), about as precise as seeing
allows. In the dispersion direction, the grating turret is fixed,
and positioning of spectra on the array is determined by the
precision of the detented slit mechanism. The prisms are over-
sized so that turret rotations of Ⳳ1⬚do not vignette the beam.
This flexibility permits orders in the range 0.8–5.5 mmtobe
placed anywhere on the array to avoid bad pixels. Figure 6
shows the optical configuration of the SXD mode, consisting
of back-to-back fused silica and ZnSe prisms and a grating.
The LXD mode, consisting of back-to-back LiF and Ge prisms
and a grating, is similar.
4.3.6. Camera Lens
The spectrograph camera is a scaled-up and reoptimized
version of the foreoptics camera lens. Each element of the BaF
2
-
LiF-ZnS/Cleartran all-spherical triplet lens has a diameter of
80 mm. Lens aberrations are small compared to diffraction.
4.3.7. Array Focus Mechanism
Even though the beam at the grating/prisms is collimated,
small focus shifts occur when switching between differentspec-
troscopy modes as a result of differences in flatness of the
grating and prism faces. Using a flatness tolerance of l/4 at
0.6 mm recommended by the prism vendors, a ZEMAX Monte
Carlo tolerance analysis produced focus shifts of up to Ⳳ1
mm when changing grating/prism modes. The reason for the
sensitivity is the number of surfaces that change when changing
modes (up to nine because of the double-pass design). To com-
pensate for these small shifts, the spectrograph array is mounted
on a focus stage with a range of Ⳳ2.5 mm. In use, a focus
range of about 1 mm is required, consistent with the tolerance
analysis.
4.4. Slit Viewer
The slit viewer reimages the slit field at 0.8 : 1 demagnifi-
cation onto the InSb detector, using similar BaF
2
-512 #512
LiF-ZnS/Cleartran triplet lenses. The image scale is 0⬙.12
pixel
⫺1
. All the lenses are 35 mm in diameter. The collimator
triplet forms a 12.5 mm diameter pupil image immediately in
front of a 15-position filter wheel. Following the filter wheel,
the camera triplet focuses the field onto the detector. Imaging
is achromatic across the range 0.8–5.5 mm. The calculated av-
erage Strehl ratio across the field is 0.85, 0.95, and 0.99 at 0.9,
1.65, and 4.8 mm, respectively.
4.5. Calibration Box
4.5.1. Optics
The calibration optics are designed to illuminate the telescope
focal plane inside SpeX with a 50 mm diameter flat field. This
is the size required to cover the longest slits (60⬙). A 10.0 mm
diameter exit port (field stop) of an integrating sphere serves as
the uniform (Lambertian) flat field. This field is magnified
to 50.0 mm diameter by a 151.5 mm focal length LiF/BaF
2
achromatic doublet lens placed just under one lens focal length
(121.2 mm) from the exit port. A concave 648 mm focal length
spherical mirror used slightly off-axis reimages this field at
1 : 1 onto the f/38.2 telescope focal plane inside the cryostat.
At the same time, the mirror images the 15.8 mm aperture stop
of the lens onto the telescope secondary mirror to create an
exit pupil identical to that of the telescope (f/38.2) so that the
calibration light sources illuminate the instrument identically
to the telescope. There are three folding flats, one of which is
mounted on a linear translation stage that moves into the tele-
scope beam and picks off the beam from the calibration lamps
when required.
4.5.2. Integrating Sphere and Input Sources
The 50 mm diameter custom-made integrating sphere from
Labsphere has three entrance ports, one for a 10 W quartz-
tungsten-halogen lamp (3200 K) and 0.1 W bulb (placed side
by side) for the 0.8–2.5 mm region, one for an infrared source
(1100 K) for the 1.9–5.5 mm region, and one for a low-pressure
argon discharge lamp. With the exception of the pick-off mirror,
the calibration box contains no moving parts. Argon lines from
the discharge lamp are visible out to about 4 mm. Beyond
4mm, the thermal background from the warm calibration box
overwhelms the signal and wavelength calibration uses atmo-
spheric features. Integration times of no longer than about 3 s
are sufficient to obtain a signal-to-noise ratio (S/N) greater than
200 in the flat fields in all modes.
4.6. Instrument Throughput
Broadband antireflection (BBAR) multiple-layer coats for
the lenses and prisms were designed and applied by Thin Film
Labs. The lenses need to work efficiently across the entire
0.8–5.5 mm range of the instrument. The prisms are optimized
for smaller wavelength ranges, so their BBAR performance is
better. All the mirrors, with the exception of the OAP, use a

SpeX IR SPECTROGRAPH FOR IRTF 371
2003 PASP, 115:362–382
TABLE 4
Estimated Throughput of 0.8–2.4 mmCross-dispersed (SXD) Mode at 1.7 mm
Element Efficiency Explanation
Primary mirror ........... 0.99 Aluminum on Cervit
Secondary mirror ........ 0.99 Aluminum on SiC
Window .................. 0.97
2
0.4–5.5 mm single-layer protection coat CaF
2
Dichroic .................. (0.9) CaF
2
substrate
Mirror collimator ......... 0.99 Protected silver on Zerodur
Fold mirrors (2) .......... 0.98
2
Protected silver on silica at 45⬚
Cold stop ................. 0.95 Mask edge of 3.0 m primary mirror
K-mirror (3) rotator ...... 0.98
3
Protected silver on silica at 45⬚
Triplet lens ............... 0.975
2
0.8–5.5 mm BBAR on BaF
2
(witness sample)
0.975
2
0.8–5.5 mm BBAR on LiF
0.95
2
0.8–5.5 mm BBAR on ZnS/Cleartran (witness sample)
Fold mirror (1) ........... 0.98 Protected silver on silica at 45⬚
Slit ........................ 0.97
2
0.8–5.5 mm BBAR on CaF
2
substrate at 45⬚
Fold mirror (1) ........... 0.98 Protected silver on silica at 45⬚
Mirror collimator ......... 0.99 Diamond-turned Al, gold coat
Double-pass prism ....... 0.975
2
0.8–2.5 mm BBAR on ZnSe at 40⬚
0.98
2
0.8–2.5 mm BBAR on ZnSe at 0⬚(witness sample)
0.99
2
0.8–2.5 mm BBAR on fused silica at 0 ⬚(witness sample)
0.985
2
0.8–2.5 mm BBAR on fused silica at 40⬚
Grating ................... 0.75 Estimate at peak, near-Littrow 5⬚tilt
Triplet lens ............... 0.975
2
0.8–5.5 mm BBAR on BaF
2
(witness sample)
0.975
2
0.8–5.5 mm BBAR on LiF
0.95
2
0.8–5.5 mm BBAR on ZnS/Cleartran (witness sample)
Detector .................. 0.95 Typical of best Aladdin arrays at 1.7 mm
Total throughput ......... 0.29 0.8–2.5 mm cross-dispersed mode at 1.7 mm
Fig. 7.—Throughput of the spectrograph, measured in the prism (dash-dotted
line) and SXD (black line) modes.
protected-silver reflection coating (FSS-99) by Denton Vac-
uum. The OAP from SSG, Inc., uses the company’s standard
vacuum-deposited bare gold coating. The slit mirrors also use
gold coatings.
The throughput in the prism, SXD and LXD modes, was
measured on an A0 V star through the 3⬙.0 wide slit to minimize
seeing losses. Figure 7 is a plot of the throughput in the prism
and SXD modes. In the telluric bands, the effects of atmo-
spheric transmission have been removed by using a model of
the atmosphere. The estimated throughput of SpeX at 1.7 mm
in the 0.8–2.4 mm XD mode is given in Table 4. The peaks of
the SXD blaze function were designed to be at 6.6 mmm
⫺1
(see eq. [1]); however, the two custom gratings made for this
mode have measured first-order blaze peaks at about 6.15 mm.
Consequently, the blaze peak in higher orders is shifted to
slightly shorter wavelengths than planned. Allowing for this
shift, the measured throughput (0.30 at 1.52 mm) and estimated
throughput (0.29 at 1.7 mm) are in good agreement. The de-
crease in throughput at shorter wavelengths is probably due to
the combination of decreasing quantum efficiency (QE) of the
array (including its own single-layer antireflection coating;see
§ 8.1) and decreasing efficiency of the BBAR coatings on the
transmissive optics (although witness samples indicate good
performance of the lens and prism BBAR coatings down to
0.8 mm). The average throughput of the SXD mode in the range
0.8–2.4 mm is about 0.18. By contrast, the average throughput
of the prism mode is about double the SXD mode with a peak

372 RAYNER ET AL.
2003 PASP, 115:362–382
of 0.4 (at 1.7 mm, where the array antireflection coating is
optimized), since it does not require a grating for dispersion.
Because of the opacity of the atmosphere in the 2.4–5.5 mm
region, it is not possible to produce a continuous throughput
plot for the LXD mode equivalent to Figure 7. By measuring
the throughput in clear parts of the atmosphere, we find an
average throughput in the LXD mode of 0.15 with order peaks
at 0.2. This is slightly lower than the SXD mode and is consistent
with decreasing QE toward longer wavelengths (see § 8.1).
Assuming nominal performance of the optics and the array,
the average throughput of the slit viewer should be about 0.25.
This is lower than for an optimized camera because of the
additional losses in the foreoptics. The foreoptics alone have
an estimated throughput of 0.66. We measure an average
throughput of 0.20 at J,H, and K, slightly lower than expected.
This is probably due to a slightly lower than normal QE of the
engineering-grade array used (see § 8.1).
4.7. Stray Light and Baffling
Stray light and thermal background from the sky and tele-
scope are controlled by using a cold Lyot stop in the foreoptics
in front of the spectrograph and slit viewer. The spectrograph
and slit viewer each contain their own cold stops.
A number of further measures are taken to control stray light
and thermal background from inside the cryostat. (An unfiltered
5.5 mm cutoff detector detects ∼10
4
times as many thermal
photons from an object at room temperature compared to an
unfiltered 2.5 mm cutoff detector.) Except at the entrance win-
dow and wave-front sensor exit window, the light paths are
entirely enclosed in aluminum enclosures consisting mostly of
baffled tubes. These enclosures are designed to prevent stray
light and thermal background from cryostat components
warmer than about 80 K (such as the warm vacuum jacket,
drive shafts, and wires) from entering the light path, where it
can be scattered toward the detectors. A goal of the design was
to keep the instrument background below the detector dark
current (∼0.2 es ) because in several observing modes, the
⫺⫺1
noise is limited by dark current. Since the light path needs to
be easily assembled and disassembled, and efficiently evacuated,
the structures were built using re-entrant tubes containing lab-
yrinth-like evacuation ports. Drive shafts enter through rotary
light shields, and all access covers have tongue-and-groove seals.
Because of the wide wavelength coverage of SpeX, large
changes in either object or sky brightness can occur across
spectra. For example, the change in sky brightness across the
LXD modes is ∼10
4
. Consequently, any ghost reflection off a
bright part of the spectrum onto a fainter region can seriously
degrade data quality. The two main areas of concern are nar-
cissus ghosts from the camera lens and prisms. Ray-tracing of
the former shows that these ghosts are grossly out of focus and
do not pose a problem. The latter are avoided by tilting the
spectrograph array by 1⬚.0, which is sufficient to redirect any
ghost reflections off the prisms but small enough to incur no
detectable defocus across the array. The measured ghosting in
SpeX is at the level of a few percent and is normally removed
when the sky beam is subtracted.
5. CRYOSTAT
5.1. Mechanical Design
The vacuum jacket consists of a four-sided center section
weighing about 90 kg onto which two large O-ring sealed
covers, each weighing about 35 kg, are bolted. The four-sided
center section is electron-beam welded from four plates of
2219-aluminum (see Fig. 5). The external dimensions of the
vacuum jacket are 112 cm #79 cm #61 cm. Since no hard
connections are made with the vacuum jacket except at the
rigid 25 mm thick top plate, stability concerns are limited to
keeping the flexure well below the yield strength of the alu-
minum and to keeping the flexure of shafts connecting to warm
motors mounted on the vacuum jacket within the specifications
of the flexible couplings. This is easily achieved using a vacuum
jacket wall thickness of 12.5 mm. A spreader bar has been
added between the two large covers, without which the wall
thickness would have to be considerably larger. The maximum
observed deflection upon evacuation is 2 mm.
SpeX is attached to the telescope multiple instrument mount
by a rigid interface box mounted to the top plate of the vacuum
jacket (see Fig. 4). A spectral calibration unit is built into this
box. The walls of the calibration box are 25 mm thick, and
the box is 28 cm deep. The top plate of the vacuum jacket
contains the entrance window, liquid nitrogen fill port, and
mount for the closed-cycle cooler. The cryostat-mounted elec-
tronics box and wave-front sensor box are mounted on opposite
sides of the cryostat center section.
The 6061-T6-aluminum cold structure is supported on four
V-shaped fiberglass support trusses that connect to the cold
structure near the center of gravity (see Fig. 5). The base of
the support trusses is bolted directly to the inside surface of
the rigid top plate of the center section. In the interests of
stiffness, the cold structure consists of a series of boxes bolted
to a backbone that provides cooling and support. The total
weight of the cold structure is 140 kg. The cold structure is
enclosed by a 3 mm thick aluminum radiation shield. The
30 kg radiation shield is supported separately by fourfiberglass
tabs. The total weight of the instrument at Cassegrain focus is
478 kg (not including the wave-front sensor).
5.2. Mechanisms
There are seven cryogenic mechanisms in the cryostat. We
use external warm motors with vacuum feed-throughs in pref-
erence to internal cold motors because it is cheaper and, in our
view, involves less risk. The advantage of being able to position
mechanisms manually for testing, and in case of failure, is
highly valued. The motors are mounted on one of the two
covers, and the optics and detectors are accessed by removing

SpeX IR SPECTROGRAPH FOR IRTF 373
2003 PASP, 115:362–382
the opposite cover. Motors need not be disassembled for most
troubleshooting inside the cryostat.
The mechanisms fall into three different categories: discrete
position wheels with detents, continuously variable wheels, and
a linear flex stage. Since failure rates for mechanical limit
switches in previous IRTF instruments have been unacceptably
high, we decided to use noncontact Hall effect devices for
position sensing. Each of the wheels is equipped with one
stationary Hall effect sensor. Wheel positions are encoded by
placing magnets on the edge of the wheels. Opposite polarities
mark home positions and discrete (element) positions. Theout-
puts of the Hall effect sensors are passed through comparator
circuits and function much like limit switches.
The dichroic turret, order-sorter filter wheel, slit wheel, and
slit-viewer filter wheel are all wheels with detented positions
that are driven by compliant worm-gear mechanisms that use
400 step revolution
⫺1
stepper motors. In order for the detent
to properly locate the wheel, some angular compliance is pro-
vided by allowing the worm to slide along the supporting shaft,
thereby providing the angular “slop” for detent engagement.
The detent rolls on the periphery of the wheel and is forced
into the detent with a spring. The array focus stage employs a
continuous antibacklash linear position mechanism with an alu-
minum alloy flex stage providing Ⳳ2.5 mm of focus travel.
The flex stage itself is fabricated using electric discharge ma-
chining and provides compliance in one direction and very
high stiffness in all others through the use of thin sections. The
stage is driven with a screw that in turn is coupled with a worm
and worm gear to a motor shaft connected to a 400 step rev-
olution
⫺1
stepper motor. The grating turret and K-mirror image
rotator mechanisms require continuous rotary motion and em-
ploy worm and antibacklash worm-gear mechanisms driven by
Animatics DC servos with 4000 steps revolution
⫺1
to achieve
high-resolution positioning. In both mechanisms, a position is
determined by counting steps from a home position indicated
by a Hall effect sensor.
5.3. Cryogenic Design and Performance
The cooling scheme adopted in SpeX is a simple and con-
servative design that meets all the temperature and cooling
requirements. Cooling the cold structure to liquid nitrogen
(LN
2
) temperatures is adequate to keep the instrument back-
ground below the best expected InSb detector dark current of
0.1 es , since all but a half-viewing angle of 12⬚can
⫺⫺1
be shielded with a baffle at the temperature of the detector
(30 K). A closed-cycle cooler is used to cool a radiation shield
surrounding the cold structure, and the cold structure is attached
to an LN
2
can that provides a very stable temperature. The use
of an LN
2
can also speeds cooling. The surface areas of the
vacuum jacket, actively cooled radiation shield, and cold struc-
ture are 4.0, 3.5, and 2.0 m
2
, respectively. The actively cooled
shield need only be cooled to a temperature that reduces the
radiation load on the cold structure down to the level of the
other parasitic loads, such as the G10 fiberglass support trusses,
which can be accommodated by the LN
2
can.
We use a two-stage 1050-CTI cooler to cool the radiation
shield and detectors. The first stage has a rated cooling of about
70 W at 80 K, and the second stage has a rated cooling of
about 8 W at 20 K. The first stage cools a highly polished
3 mm thick 6061-aluminum cold shield to less than 100 K.
The resulting radiation load from the radiation shield onto the
cold structure is less than 0.5 W. The estimated heat load on
the cold structure consists of 3 W from the G10 support trusses,
1 W from the motor shafts, 1 W from the wiring, 2 W from
the entrance window, and 2 W from penetrations in the radi-
ation shield. Since the heat load from the radiation shield is
so small compared to the other parasitic heat loads, anynormal
variation in the closed-cycle cooling to the radiation shield has
little effect on the temperature of the enclosed cold structure.
Larger temperature excursions of typically 0.5 K result from
changes in the level of LN
2
as a result of boil-off, but this is
too small to have any measurable impact on instrument per-
formance (e.g., focus or instrument background). The calcu-
lated hold time of the 10 liter capacity LN
2
can compares well
with the measured hold time of about 48 hr. The bulk of the
90 kg cold structure cools to 80 K in about 24 hr, at which
time all the mechanisms function normally. Since there is little
conduction cooling through the bearings of the filter wheels
and turrets, the surfaces of these wheels are painted black to
speed cooling by radiation. The instrument background mea-
sured at the spectrograph detector falls to 300 esat24hr,
⫺⫺1
5es at 48 hr, and bottoms out at ∼0.1 es after72hr.
⫺⫺1⫺⫺1
About 150 liters of LN
2
is required to reach the operational
temperature at 48 hr.
The spectrograph and slit-viewer arrays are located close
together, at the end of the cryostat opposite to the closed-cycle
cooler. The array mounts are cooled via a 0.8 m long copper
strap with a cross-sectional area of 25 mm
2
. Each of the array
mounts, including cold baffle snouts, weighs about 0.9 kg and
takes 11 hr to cool to the operating temperature of 30 K. Using
heater resistors, each mount is controlled to K30.00 Ⳳ0.01
by a Lakeshore model 330 temperature controller. Warmupalso
takes 48 hr. After turning off the closed-cycle cooler and
blowing out the LN
2
, warm-up of the cryostat is accelerated
by circulating warmed dry air through the LN
2
can.
6. ARRAY CONTROLLER AND DATA SYSTEM
The array controller for SpeX can be separated into six major
hardware components: a VME64 instrument controller, a cry-
ostat-mounted electronics box (containing preamps, A/D con-
verters, clock, and bias boards), an array electronics power
supply, an embedded computer/motor controller, a temperature
controller, and a workstation displaying the SpeX GUI. The
VME64 instrument controller consists of two independent com-
puters. One computer, known as Bigdog, runs the spectrograph
InSb array, and a second, known as Guidedog,1024 #1024

374 RAYNER ET AL.
2003 PASP, 115:362–382
runs the IR slit viewer/guider InSb array. For motor512 #512
and temperature control, and for monitoring tasks, Bigdog and
Guidedog communicate with the embedded computer known
as Littledog. The location and interconnections of these com-
ponents are illustrated in Figure 8.
6.1. VME64 Instrument Control Computer
Bigdog and Guidedog are housed in a VME64 rack mount
located in the telescope control room. Other than the shared
chassis and power supply, the two controllers are not coupled.
The chassis has two independent VME backplanes, and each
controller operates independently. Each controller houses
a workstation-class single-board computer running the Unix/
Solaris 2.6 operating system, DSP-based single-board com-
puters, a fiber-optic link to the cryostat-mounted electronics,
and a board for time-stamping data. The IXZ4444 DSPs per-
form the time-critical tasks of pixel clocking, pixel data ac-
quisition, and low-level pixel processing. The Unix computer
performs upper level control functions, including loading code
into the DSPs and commanding them, controlling the instru-
ment via the GUI, and higher level data processing and display.
Guidedog is identical to Bigdog except that its smaller array
requires only one IXZ4444 DSP board and one fiber board.
Each IXZ4444 DSP board contains two independent groups
of two DSPs. The first group of “clock” DSPs outputs clocking
and bias patterns in the form of 32 bit words to the fiber channel
interface board, which then transmits these patterns across the
optical fiber to the clock and bias boards located at the cryostat.
The second group of “buffer” DSPs receives 16 bit digitized
pixel values from the A/D converters at the cryostat via the
optical fiber and fiber channel interface board. The buffer DSPs
perform the pixel processing steps of co-adding and time
stamping.
In Bigdog, two DSP boards are used to increase the pixel
processing speed. One board clocks and buffers the top half
(quadrants 1 and 2) of the array while the second1024 #1024
clocks and buffers the bottom half (quadrants 3 and 4). This
architecture allows the full array to be read out in a minimum
integration time of 0.1 s and proportionately shorter for
subarrays.
The architecture also allows data to be streamed onto the
local disk without interruption and with minimal gaps between
images (or spectra). This is called “movie mode.” In this mode,
images are accumulated in one of the DSP board’s SRAM
modules. When this location is full, storage is switched to the
second SRAM and the first SRAM is read out. By alternating
between the two SRAMs, images can be streamed onto the
VME bus for storage onto the local hard disk. Each SRAM
readout is saved as a three-dimensional FITS image. The max-
imum storage rate for a image is 2 Hz. Movie1024 #1024
mode is normally used in the prism configuration where a
0.8–2.5 mm spectrum is matched to quadrant two of the spec-
trograph array so that small subarrays can still acquire a large
spectral range. In this mode a subarray can be512 #100
streamed to the local hard disk at about 10 Hz (and propor-
tionately faster for smaller subarrays) for occultation spectros-
copy. The only limit to the occultation time is then the storage
capacity of the disk.
Guidedog works in a similar fashion except that the one
quadrant array requires only one DSP board. Images512 #512
from Guidedog can be streamed to disk independently of Big-
dog. For occultations, this allows spillover at the slit to be
monitored.
The array data are read out over the VME bus, formatted
into FITS, displayed on the X-Windows graphics display, and
stored on disk. For the normal mode of observing, where in-
tegration times are longer than a few seconds and the overhead
in storing data is relatively short, data are sent across the Eth-
ernet LAN for storage on the larger capacity network disks (80
Gbytes), rather than stored in the local disks on Bigdog and
Guidedog (each 8 Gbytes).
6.2. Cryostat-mounted Electronics
The cryostat-mounted electronics provides for the complete
control of array readout and digital conversion of pixel analog
signal values. The spectrograph and guider electronics are
housed in a single 19 inch rack-mount chassis that is mounted
to the side of the cryostat to keep the high-frequency analog
lines to the arrays as short as possible. Array clocking patterns
and programmable biases (most of the biases are hard-wired)
are transmitted over the fiber-optic link from the clock DSPs
and passed down a backplane into a clock-level-shifting circuit
that translates the TTL signals into 12 bit digital-to-analog
voltages to drive the arrays. Each of the 32 spectrograph array
outputs and each of the eight guider array outputs are condi-
tioned by a preamplifier circuit and output to its own 2 MHz,
16 bit A/D converter. The converted digital values are then
transferred through another backplane onto the fiber-optic inter-
face board for transmission to the pixel-buffering DSPs.
6.3. Embedded Computer and Motor Controller
Littledog is an embedded computer containing peripheral
electronic boards for motor control, temperature monitoring,
spectrograph calibration box control, AC power control, and
monitoring. It operates under command from Bigdog and
Guidedog using the telescope’s Ethernet LAN. To keep the
analog lines short, Littledog is located on the mirror cell next
to the cryostat. Littledog is an i486/100 MHz PIC/ISA com-
puter with 24 Mbytes RAM, a 10 Mbyte s Ethernet, and an
⫺1
IDE 500 Mbyte hard disk. The operating system is Linux. The
chassis also houses power supplies for stepper and servo mo-
tors, as well as the stepper motor driver modules.
6.4. Observing Workstation
Observers at the telescope log in to Bigdog and Guidedog
from the facility workstation located in the observer’s area.

SpeX IR SPECTROGRAPH FOR IRTF 375
2003 PASP, 115:362–382
Fig. 8.—Locations and interconnections of the cryostat, computers, and array controller components.
The SpeX GUIs run on Bigdog and Guidedog, but the display
windows are exported to the observer’s workstation. This work-
station has two screens, one for Bigdog and one for Guidedog.
By logging into Bigdog and Guidedog, remote observing can
be done from any appropriate workstation connected to the
Ethernet interface.
7. SOFTWARE
7.1. Instrument Control
The instrument control software consists of four major com-
ponents: the instrument control (IC) application, which runs on
Bigdog and Guidedog; the motor control and monitoring
application, which run on Littledog; the X-Windows user in-
terface (XUI); and the quick-look Data Viewer.
7.1.1. Instrument Control (IC)
Independent versions of the instrument-control application
run on the Bigdog and Guidedog computers. Its purpose is to
run the command parser, to run the data acquisition task, and
to monitor Littledog functions. The command parser accepts
user commands from the GUI (XUI) and executes them. Some
commands are routed to the data acquisition task and some to

376 RAYNER ET AL.
2003 PASP, 115:362–382
Littledog. The acquisition task is responsible for controlling
the DSP clocking and data buffering. When the data are ready,
the acquisition tasks assembles them into FITS images, saves
them on disk, and/or sends the them to Data Viewer for display.
A monitoring task makes the Littledog status (motors and tem-
peratures) available at the IC for display by the XUI. The IC
application is written in C.
7.1.2. Motor Control and Monitoring
A suite of applications runs on Littledog to initialize, po-
sition, and monitor the eight motors and position sensors and
to monitor the temperature controllers and remote power con-
trol units. All the code is written in C.
7.1.3. X-Windows User Interface
The X-Windows user interface (XUI) provides an easy-to-
use GUI to control the IC using various icons, buttons, menus,
and software widgets. It also provides a simple way to set up
macro files built from the IC’s command set. For the benefit
of IRTF observers, the look and feel of the SpeX GUI was
designed to be very similar to those of the other IRTF facility
instruments. Like the IC, there are two independent XUIs, one
for Bigdog and one for Guidedog. The ICs allow multiple
versions of the XUI to execute, which allows for remote support
of observers by IRTF staff or remote participation by observers.
SpeX has been operated successfully from the US mainland
and Europe. The XUI is written in C and uses the GTK⫹
widget library. Sample screens from Bigdogxui and Guide-
dogxui can be viewed at the SpeX Web page.
4
7.1.4. Data Viewer
Data Viewer (DV) is a general-purpose FITS viewer for
viewing images and spectra. In addition, it incorporates func-
tions allowing interaction with the IC/XUI for guiding, tele-
scope pointing, and setting subarrays. In normal operation, DV
is displayed side by side with Bigdogxui and Guidedogxui. DV
can also be used as a stand-alone FITS viewer. DV is written
in C and uses the GTK⫹widget library.
7.2. Observing Modes
SpeX has three fundamental observing modes, each of which
can be selected with the XUI: basic, slow guide, and movie
mode.
7.2.1. Basic
Basic is the fundamental array data acquisition mode. A
“GO” command in Bigdogxui takes a spectrum, and a GO in
Guidedogxui takes an image, each with the following adjustable
parameters: on-chip integration time, co-additions (co-adds),
nondestructive reads, array clock speed, telescope nod pattern
(beam A, B, or beam A followed by beam B), and number of
4
http://irtfweb.ifa.hawaii.edu/Facility/spex.
cycles (repeats of a GO command). There is no intrinsic limit
to the maximum on-chip integration time, but it is set in soft-
ware to 1800 s. The minimum integration time is 0.1 s, which
is proportionately less for subarrays.
7.2.2. Slow Guide
Slow guide is the guiding mode of the infrared slit viewer
in Guidedogxui. It takes images continuously and sends them
to the DV and to a stand-alone video display for fast display.
The slow guide mode includes several adjustable parameters:
integration time, guide algorithm (peak or centroid), gain (frac-
tion of calculated telescope offset to be moved), guiding on or
off, sky subtraction on or off, and guiding in both beams or
just A or B—all of these can be changed on the fly to fine
tune the guiding. Slow guiding sends telescope offsets across
the Ethernet to the telescope control system at a maximum
frequency of about 1 Hz. This speed is adequate for telescope
tracking corrections. Anything faster requires correction
through an optical fiber connection to the tip-tilt hexapod sec-
ondary mirror. This “fast guide” mode is not yet implemented.
7.2.3. Movie
Movie mode is available in both Bigdogxui and Guidedog-
xui. It includes all the parameters of basic mode with the ad-
dition that the GO command takes images continuously and
streams the data to disk until stopped by the user (as described
in § 6.1). Since Bigdog and Guidedog are independent, spectra
and images can be acquired simultaneously. Movie mode is
designed for occultations, but it can also be used to construct
high spatial resolution images on bright objects by selecting
and then combining images with the best seeing from movie
sequences.
7.3. Spectral Reduction
Data are viewed at the telescope as array images on the
observing workstation. The IDL GUI-based spectral reduction
package Spextool can be run on another workstation. Using
observer-selected files, Spextool produces one-dimensional
spectra that have been flat-fielded, wavelength calibrated, and
corrected for telluric absorption. For the XD modes, Spextool
also merges the cross-dispersed orders. The design of Spextool
and the data reduction procedure are described in a paper by
M. C. Cushing et al. (2003, in preparation). A new method for
correcting near-infrared spectra for telluric absorption using
A0 V stars, which is incorporated into Spextool, is discussed
in Vacca, Cushing, & Rayner (2003). Spextool can be down-
loaded from the SpeX Web page at the IRTF Web site.
8. LAB TESTING
8.1. Array Testing and Selection
SpeX received its arrays from the NASA Planetary Astron-
omy Infrared Detector Array Infrastructure Project (PAIDAI),
which was led by Dr. Mike Mumma of the Goddard Space

SpeX IR SPECTROGRAPH FOR IRTF 377
2003 PASP, 115:362–382
Fig. 9.—0.8–2.4 mm stellar to substellar temperature sequence obtained with the SXD mode. Main-sequence stars are plotted in blue, brown dwarfs in green,
and the planet in red.
Flight Center. PAIDAI funded a foundry run of Aladdin
and arrays at Raytheon Infrared Op-1024 #1024 512 #512
erations (RIO). As a result of tests at RIO and PAIDAI partner
NOAO, we selected the Aladdin 3 array SCA412202. Aladdin
3 devices employ an improved (SBRC-206) multiplexer and
have generally better read noise, speed, and reduced odd/even
column structure compared to Aladdin 2 devices (SBRC-152
multiplexer). All Aladdin arrays are coated with a single-layer
SiO antireflection (AR) coat on InSb, optimized at 1.7 mm.
Reflection losses at the array are at minimum a few percent
at 1.7 mm, increasing to about 0.30 at 5.0 mm and 0.36 at
0.85 mm (where the single-layer AR coat has a maximum re-
flectance equal to that of bare InSb). Measured QEs of the best
Aladdin arrays are typically about 0.95 at 1.7 mm. Testing of
SCA412202 in our lab test cryostat, and later in the instrument
on the telescope, confirmed the expected performance at 30 K
and 0.4 V bias: dark current 0.2 es , a read noise of
⫺⫺1
50 erms (reset-read-read) measured at a readout rate of 6 Hz,
⫺
which is reduced to about 12 eat slower speeds and with
⫺
Fowler multiple sampling (implemented with up to 32 non-
destructive reads at the beginning of an integration followed
by another 32 nondestructive reads at the end of the integra-

378 RAYNER ET AL.
2003 PASP, 115:362–382
Fig. 10.—0.8–2.4 mm spectrum of the Wolf-Rayet star WR 158 obtained with the SXD mode at . The inset shows detail in the 0.82–1.11 mm regionRp2000
including a P Cygni profile in the He ilineat1.080mm.
tion). At 0.4 V bias, the full-well depth is 111,000 e, which
⫺
is linear to 5% up to about 52,000 e. An algorithm to correct
⫺
for nonlinearity is incorporated into Spextool. Cosmetically,
the array has two long 1 pixel wide cracks (which have neg-
ligible effect on data) and very good uniformity. The operating
temperature of 30 K was found to be the best compromise
between dark current, which decreases with temperature, and
read noise, which increases with temperature.
A generic problem with Aladdin arrays is residual image,
which manifests itself as enhanced dark current when a bright
object or high-background exposure is followed by a low-back-
ground exposure. Although the blanked-off dark current of the
spectrograph array is 0.2 es , the effective dark current
⫺⫺1
during regular observing is up to 1 es (or a few percent
⫺⫺1
of the previous exposure, whichever is greater) since object
integrations are usually bracketed by brighter telluric standard-
star integrations and calibrations (flat fields and arcs). Wehave
found that executing global array resets at 1 Hz when the array
is not integrating reduces residual image effects significantly
(although still only to the level discussed above). However,
since global resets dissipate power locally on the chip, the array
cools during long integrations when global resets are stopped,
causing the bias level and spatial structure to change during
the first integration of a series. Consequently, the first (Abeam
minus B beam) bias subtraction is not perfect and adds some
noise, somewhat offsetting the theoretical gain in S/N obtained
by integrating longer. To overcome the bias change, werecom-
mend that observers do a 60 s integration before starting in-
tegration sequences with on-chip integration times longer than
a few minutes. This “preimage” minimizes the bias change
during the actual integration sequence. Also, the preimage is
short enough that the global array reset procedure still reduces
residual image effects.
The mask developed for PAIDAI contains one 1024 #
die in the center of the InSb wafer, two dies1024 1024 #512
at the top and bottom, two dies on one side, and five512 #512
dies on the opposite side. Because of their distance256 #256
from the center of the wafer, the dies are usually of512 #512
lower quality than the die. We selected the Aladdin1024 #1024
2 array SCA46509 for use in the slit viewer/guider.512 #512
At 30 K and 0.4 V bias, SCA46509 has a dark current of 1.5
es , a read noise of 65 e
⫺
rms (reset-read-read) measured at
⫺⫺1
a readout rate of 3 Hz, a well depth of 88,000 e, which is linear
⫺
to 10% up to about 37,000 e, and good uniformity.
⫺
8.2. Cryostat Assembly and Testing
Eight cold tests of the cryostat in the lab were required over
the period 1998 September to 2000 April before the instrument

SpeX IR SPECTROGRAPH FOR IRTF 379
2003 PASP, 115:362–382
Fig. 11.—0.9–4.1 mm spectrum of the methane dwarf SDSS 1254⫺0122 (T2 V) obtained with the SXD mode at and with the LXD mode atRp1200 Rp
. A 0.8–2.55 mm spectrum obtained with the prism mode at is overplotted (red). The inset shows the 1.15–1.30 mm of the SXD spectrum.940 R∼250
was shipped to the telescope for commissioning. In order, these
tests verified cryogenic performance, optical alignment and
stray light, flexure, and finally all-up instrument performance.
The only significant optical adjustments required were posi-
tioning of the arrays and alignment of the prisms. Our design
philosophy was to build to tolerance and pin all optical as-
semblies, with some shimming if necessary, rather than build
more complex adjustable mounts. The fully assembled cold
structure was inspected for light leaks. This was done by
placing small xenon flashlights at key locations inside the
sealed light path and inspecting the cold structure inside a
darkroom using image-intensified goggles. Light leaks were
spotted at several locations and fixed. Flexure tests of the
instrument were done in all axes on a lift/tilt table. Initially,
flexures of up to 10 pixels were measured on the spectrograph
array, but these were largely removed by stiffening the springs
in the grating turret and slit wheel mechanisms and by bal-
ancing the rotation of the grating turret. The final values are
2 pixels in the east, 0.5 pixels in the west, 0.5 pixels in the
north, and 0.75 pixels in the south, when tilting to 60⬚from
vertical. Movements of the reimaged slit were all less than
0.5 pixels.
9. OPERATION AT THE TELESCOPE
SpeX was commissioned on IRTF during 2000 May, June,
and July and became generally available for Time Allocation
Committee awarded telescope time in 2000 August. Since then,
SpeX has operated on IRTF for about 50% of the available
time and with very few problems.
9.1. Observing Procedure
The general observing procedure is as follows. First, the
observer selects the required slit size. The object is then imaged
with the slit viewer/guider and placed in the slit. There are
several options for guiding. Autoguiding can be done with the
infrared slit viewer either on spillover from the object placed
in the slit or on an object elsewhere in the field of
 
60 #60
view and at any position angle of the rotator. In the cross-
dispersed and prism modes, guiding can be done through any

380 RAYNER ET AL.
2003 PASP, 115:362–382
Fig. 12.—2.1–4.8 mm spectra of YSOs obtained with the LXD mode at . L1551 IRS 5 is an embedded protostar, and BP Tau is a classical T TauriRp1500
star. A template star of similar spectral type (K5 V) is shown for comparison.
filter in the slit-viewer filter wheel. The magnitude limit for
autoguiding on spillover depends on slit width and seeing, but
the limit is about in an integration time of 10 s. TheJp15
limit for manual guiding on spillover is about in 10 s.Jp17
Alternatively, guiding can be done on-axis in the visible using
the tip-tilt wave-front sensor package mounted to the side of
SpeX. The beam is reflected into the wave-front sensor by the
cold dichroic inside SpeX. The tip-tilt sensor package will be
replaced with a wave-front sensor running a 36-element AO
system in 2003. Another option is to use the facility off-axis
visible guider.
Once guiding is started and the spectroscopy mode is se-
lected, spectrograph integrations can begin. In the SXD and
prism modes, where sky background is low, on-chip integration
times up to several hundred seconds can be used for faint
sources. (See § 8.1 for how to optimize the array for long on-
chip integration times.) In the LXD modes, sky brightness lim-
its on-chip exposures to 30 s at 4 mm and 2 s at 5 mm. For
optimum telluric correction, standards should be taken within
about 0.05 air masses of the object. Flat-field and argon arc
calibrations are usually obtained at every standard star position
even though flexure is very small (typically about 0.2 pixels
for 0.1 air mass differences). This is done by executing a cali-
bration macro, which takes about 3 minutes to complete.
Using the slit viewer/guider to obtain infrared images for
science is straightforward, and the observing procedure is the
same as for other infrared cameras.
9.2. Example Results
SpeX is being used for a wide variety of projects. Here we
illustrate performance using data taken from an infrared spectral
survey of the H-R diagram being conducted by three of us
(Rayner, Vacca, & Cushing) and by observations of young
stellar objects (YSOs) by J. Muzerolle et al. (2003, in
preparation).
Figure 9 shows a 0.8–2.4 mm stellar through substellar tem-
perature sequence. The observations were done in the SXD mode
at and at for the T dwarfs. On dry MaunaRp2000 Rp1200
Kea nights, correction through the strong telluric bands at 1.1,
1.4, and 1.9 mm is excellent (e.g., K5 V, M4.5 V, and T5 V).
The gap in wavelength coverage at 1.81–1.86 mm is necessary
to maintain a maximum resolving power of (seeRp2000
§ 2.2). The other gaps on some objects are due to regions of

SpeX IR SPECTROGRAPH FOR IRTF 381
2003 PASP, 115:362–382
poor telluric correction that have been removed. Integration times
ranged from a few tens of seconds on the brighter dwarfs
( ) to about 2 hr on the T2 V ( ; Leggett at al.J∼5Jp14.88
2000). With the exception of the telluric bands, all the high-
frequency structure seen in these spectra is real. See, for example,
the many FeH absorption features identified in the spectra of
late-M and L dwarfs by Cushing et al. (2003).
At the top of the temperature sequence, the O9 V (35,900 K)
star shows the Paschen line series of atomic hydrogen plus
several helium lines. Paschen and Brackett hydrogen lines
strengthen through B5 V (15,400 K) to A2 V (8800 K) before
weakening with decreasing temperature. Refractor y metal lines
strengthen from F3 V (6700 K) to G2 V (5800 K) through
K5 V (4300 K) at which temperature carbon monoxide ab-
sorption at 2.3–2.4 mm increases. At M4.5 V (3100 K), re-
fractory metal absorptions weaken as a result of condensation
and rain-out, while alkali metal and molecular absorptions (metal
hydride, water, and carbon monoxide) begin to strengthen.These
trends continue through M8 V (∼2300 K) into the brown dwarf
regime, which starts at about L4 V for field dwarfs and which
is represented here by an L5 V (∼1800 K). Note the strong and
broad water features at 1.4 and 1.9 mm. The L-T transition region
(Leggett et al. 2000) is characterized by the onset of methane
absorption at about 1.3, 1.6, and 2.2 mm. These features are first
seen in the sequence shown here at T2 V (∼1300 K) and are
very strong in the T5 V (∼1100 K). Further dramatic changes
are not expected to occur until water vapor begins to condense
at about 400 K. The reflected solar spectrum of Uranus (60 K)
is dominated by strong methane and water absorption.
A 0.8–2.4 mm spectrum of the Wolf-Rayet star WR 158 ob-
served in the SXD mode at is shown in Figure 10.Rp2000
The emission features are due to mass loss in a strong stellar
wind. The inset shows detail in the 0.82–1.11 mm region in-
cluding a P Cygni profile in the He iline at 1.080 mm.
A 0.9–4.1 mm spectrum of the methane dwarf SDSS
1254⫺0122 (T2 V) is shown in Figure 11. The spectrum is the
combination of a 1.9–4.2 mm LXD observation taken through
the 0⬙.8 wide slit ( ) and the SXD observation fromRp940
Figure 9. SDSS 1254⫺0122 is very faint at 3–4 mm(

Lp
from Stephens et al. 2001) and required an integration12.12
time of about 2 hr. However, the fundamental methane ab-
sorption at 3.3 mm is clearly detected. Overplotted in the figure
is a spectrum obtained with the low-resolution prism mode
through the 0⬙.3 wide slit ( ; see § 2.4). The prism spec-R∼250
trum covers the range 0.8–2.55 mm with no gaps. The prism
mode is very efficient and overcomes some of the limitations
of the SXD mode, where read noise and dark current (enhanced
by persistence) of the Aladdin 3 array prevent background-
limited performance at . Note the excellent corre-Rp2000
spondence between the prism and SXD spectrum, confirming
that the cross-dispersed orders in the SXD mode are being
combined correctly. The inset shows the 1.15–1.30 mm region
of the SXD spectrum.
Figure 12 shows 2.1–4.8 mm spectra of the embedded proto-
star L1551 and the classical T Tauri star BP Tau compared to
a standard star of similar spectral type (K7 V), observed in the
LXD mode by J. Muzerolle et al. (2003, in preparation). Note
the excellent telluric cancellation. There is clear continuum
infrared excess emission arising from accretion disks in both
YSOs. In BP Tau, weak photospheric are seen in the Kand L
bands plus hydrogen emission lines from the accretion disk. In
L1551, carbon monoxide absorption at 2.3–2.4 mm is from the
disk, and broad water absorption at 3 mm and carbon monoxide
ice at 4.7 mm are probably due to ice-covered grains in the
circumstellar envelope.
More details of SpeX can be found at the SpeX Web page.
The information includes an S/N calculator, a user’s manual,
and engineering photographs and drawings.
Building SpeX for IRTF has been a team effort. It isa pleasure
to thank successive IRTF Division Chiefs Bob Joseph, Tom
Greene, and Alan Tokunaga, and instrument technicians Greg
Ching and Darryl Watanabe, all kept under control by IRTF
project assistant Ms. Karan Hughes. Major parts of SpeX were
made in the Institute for Astronomy (IfA), University of Hawaii
at Ma¯noa by machinists Ray Gruber, Sean Kawamoto, Kelly
Collins, Randy Chung, and Danny Cook, under the direction of
Lou Robertson. At the telescope, SpeX was handled by the IRTF
day crew, Paul Jensen, George Koenig, Lars Bergknut, Danley
Lee, Imai Namahoe, and Sammie Pung, and observing was as-
sisted by telescope operators Paul Fukumura-Sawada, Bill Gol-
isch, Dave Griep, Charlie Kaminski, and Paul Sears. Money
matters were somehow managed by Gale Yamada and Chris
Kaukali. Special thanks also to Don Hall for suggesting a me-
dium-resolution cross-dispersed IR spectrograph for IRTF in the
first place. Funding for SpeX was provided by the National
Science Foundation, NASA, and the University of Hawaii.

382 RAYNER ET AL.
2003 PASP, 115:362–382
REFERENCES
Cushing, M. C., Rayner, J. T., Davis, S. P., & Vacca, W. D. 2003,
ApJ, 582, 1066
Fowler, A. M., & Gatley, I. 1990, ApJ, 353, L33
Greene, T. P, Tokunaga, A. T., Toomey, D. W., & Carr, J. S. 1993,
Proc. SPIE, 1946, 313
Leggett, S. K., et al. 2000, ApJ, 536, L35
Maihara, T., Iwamuro, F., Yamashita, T., Hall, D. N. B., Cowie,
L. L., Tokunaga, A. T., & Pickles, A. 1993, PASP, 105, 940
Schroeder, D. J. 2000, Astronomical Optics (San Diego: Academic
Press)
Shure, M. A., Toomey, D. W., Rayner, J. T., Onaka, P. M., & Denault,
A. J. 1994, Proc. SPIE, 2198, 214
Stephens, D. C., Marley, M. S, Noll, K. S., & Chanover, N. 2001,
ApJ, 556, L97
Tokunaga, A. T., Toomey, D. W., & Carr, J. S. 1990, Proc. SPIE,
1235, 131
Vacca, W. D., Cushing, M. C., & Rayner, J. T. 2003, PASP, in press