adapted and updated version of "10 µm interferometry on the VLTI
with the MIDI instrument", by Ch.
Leinert, U. Graser, L.B.F.M. Waters, G. Perrin, B. Lopez, V. Coudč du Foresto,
A. Glazenborg-Kluttig, J. C. de Haas, T. M. Herbst, W. Jaffe, P. J. Lčna, R.
Lenzen, R. S. Le Poole, S. Ligori, R. Mundt,
J. W. Pel, I. L. Porro, and O. von der Lühe, published on Proc. SPIE,
Vol. 4006, "Interferometry in Optical Astronomy",
p. 43, 2000.
HTML
version edited by S. Ligori (Feb 25, 2002)
Interferometry with the ESO Very Large Telescope
Interferometer (VLTI) on Paranal will include the four 8 m telescopes of the
VLT (called 'unit telescopes' or UTs) as well as three relocatable outrigger
telescopes of 1.8 m diameter (called 'auxiliary telescopes' or ATs).
Observations with the VLTI started in 2001 (see ESO press release).
MIDI, the midinfrared interferometric instrument covering the astronomical N
band from 8 µm to 13 µm, will be the first scientific instrument available at
the VLTI. Half a year later, the nearinfrared closurephase imaging
instrument, called AMBER2, will arrive on
Paranal. A dualfeed system called PRIMA1, 3, which will add faint source and phasereferenced
imaging as well as an astrometric mode to the VLTI, will follow in 2003.
The high thermal background and the reduced spatial
resolution at this comparatively large wavelength are obvious disadvantages
connected with 10 µm interferometry, and the need to cool the detector and its
environment to very low temperatures adds technical complexity. Probably for
these reasons, only two 10 µm interferometers have been built before. The
wideband interferometer SOIRDETE5 could not
calibrate its observed fringes because of strong fluctuations in the thermal
background. ISI6 succeded with a narrowband
heterodyne technique7, 8,
but is limited
in calibrating fringes to N = 1.8 (200 Jy). Therefore
direct interferometry in the 10 µm band, as envisaged by MIDI, still will be a
new undertaking, promising new results or results on new type of objects. This
attempt will be particularly promising at the VLTI which offers
·
big apertures of 8 m diameter (important because in
the backgroundlimited case the S/N ratio scales ~D2 )
·
long baselines (up to 130 m, corresponding to a
spatial resolution of lambda/D = 16 mas)
·
four telescopes (i.e. six readily available baselines)
In this combination the VLTI is unique among
presentday observatories, including the Keck interferometer, which has
slightly larger telescope dishes but only one 75 m baseline. These prospects
led us to propose the midinfrared interferometric instrument for the VLTI.
Most of the parameters defining the operation of the
instrument are given by the atmosphere, some by the design of the delay line
optics. However, the choice to observe only one baseline at a time was made to
keep sensitivity.
Telescope diameter |
UTs |
8 m |
|
ATs |
1.8 m (used for bright sources) |
Baselines |
UTs |
47...130 m |
|
ATs |
8...200 m |
Resolution |
Lambda/D for 100 m baseline |
20 mas |
Wavelength coverage |
|
N
band (8 µm - 13 µm) |
|
expandable to |
Q band (17 µm - 26 µm) |
Field of view |
|
± 1" |
Airy disk (FWHM) at 10 µm |
for 8 m telescopes |
0.26" |
|
for 1.8 m telescopes |
1.14" |
Sampling time for fringe measurements |
determined by fringe motion |
100 ms |
Atmospheric stability |
for chopping |
200 ms |
Differential dispersion |
in 100 m of air |
0.9 µm (10 µm to 20 µm) 46 µm (1.6 µm to 10 µm) |
Input
beam diameter |
from UTs |
80 mm (compressed to 18 mm) |
|
from ATs |
18 mm |
Detector (320x240 pixels) |
pixel
size |
50
µm |
|
full well capacity |
≈ 2 107 electrons |
|
read noise |
≈ 1000 electrons |
Background at UTs (in Airy disk) |
from sky |
1.6 1010 photons/s |
|
from VLTI mirrors |
1.2 1011 photons/s |
Limiting N magnitude without fringe
tracking |
with UTs |
4 mag (1 Jy) |
|
with ATs |
0.8 mag (20 Jy) |
Limiting N magnitude with external
fringe tracking |
with UTs |
9 mag (10 mJy) |
|
with ATs |
5.8 mag (200 mJy) |
Table 1. Basic parameters of the MIDI instrument
Figure 1. Optical concept
The scientific promises of a midinfrared
interferometer strongly depend on its limiting sensitivity. Our goal is to
achieve the sensitivity limited by the photon noise of thermal emission from
the VLTI train (1.5 x 1011 photons/s per Airy disk of radius
lambda/D = 0.26" with a VLTI transmission of 40%), assuming that the fringe
measurement happens within one coherence time of 100 ms, that the
instrumental transmission of MIDI is 60% and that 75% of the light fall into
the Airy disk and are useable for the interferometric measurements. This gives
the following performance on the UTs with broad band observations (limiting
brightness scales like sqrt(n) when splitting the band into n spectral
channels):
Point
source sensitivity:
·
N = 4 mag (1.0 Jy) in selffringetracking mode, with
8 m telescopes and tip/tilt correction
·
N = 9 mag (10 mJy) with external fringe tracking
Accuracy of visibility:
·
5 % faint sources
·
1 % bright sources, beam cleaning by spatial filtering
Phase measurement: by external referencing
The very large gain in sensitivity possible with a
fringe tracker (fed by an adaptive optics system) cannot be overemphasised.
To reach these goals, we want to limit the
instrumental loss in sensitivity to correlated flux (visibility loss) to
less than 25 %. This requires alignment of the beams to 1% of their diameter, ±
5" in their axes, and an optical quality of the surfaces of lambda/8
peaktovalley, measured at optical wavelengths, e.g. 632 nm10.
Figure 2. Design of the beam combining optics inside
the cryostat. M1 and M2 are confocal paraboloids, M3, M4
and M5 flat mirrors. The beamcombining 50% : 50%
coated surface is indicated by a pointer. Not shown: a cold
shutter located about halfway between the cold stops
and the entrance windows.
A schematic representation of the layout of the
instrument is given in Figure 1. Since radiation at 10 µm is dominated by
thermal emission from the environment, most of the instruments optics has to be
in a dewar and to be cooled to cryogenic temperatures (4 K 8 K for the detector,
40±5 K for the cold bench, 77 K for the outer radiation shield).
Good suppression of the thermal radiation of
background and surroundings requires to design for both a cold pupil and a cold
field stop inside the instrument.
We choose to have beam combination in (or close to)
the pupil and to detect the signal in the image plane. This appeared to us the
most promising way to support the baffling against the very high thermal
background and to
reduce the effects of variable pixel responsivity,
which is common among midinfrared detectors. This concept is shown
schematically in Figure 1:
From the left, the afocal beams from two telescopes of
the VLTI are approaching the instrument. Their nominal diameter is 80 mm, and
they are reduced to 18 mm by a beam compressor provided as part of the VLTI
infrastructure, represented here for simplicity by two lenses.
After the four folding mirrors of a small
experimentinternal delay line, the compressed beams enter the cryostat (cold
box) through the entrance window (Dewar window). The telescope pupil
is imaged by the VLTI delay line
optics onto a cold pupil stop to provide the needed
suppression of thermal emission from outside the beams. Next, an intermediate
focus is formed, where different masks or spatial filters can be introduced for
additional suppression of unwanted radiation. (If no spatial filters are used,
the detector pixels, which are much smaller than the Airy disk, still provide
an alternative way to limit the spatial region admitted for the measurement).
Then the beams are recollimated and move on to combine on the surface of a
50%:50% beam splitter which coating is indicated in the
Figure on the lower back side of the ZnSe plate. This
is the heart of the instrument.
From the beam combiner onwards, the two interfering
beams have a common optical axis. Actually, there are two of such overlaid
beams, one outgoing to each side of the beam combiner. These two outputs are
modulated in
flux depending on the optical path difference of the
interfering beams, but with opposite sense in because of energy conservation.
We measure the degree of coherence between the
interfering beams (i.e. the object visibility at the actual baseline setting)
by artificially stepping the optical path difference between the two beams
rapidly over at least one wavelength within the coherence time of ≈ 0.1
s. This is done with help of the piezodriven roof mirrors forming part of the
small delay lines just outside the cryostat. Then, for both channels on the
detector the fringe amplitude is determined from the modulating signal, usually
in one of the methods known from stellar polarimetry9.
The large and not precisely
known thermal background forces us to determine the
total flux separately by a chopped measurement, chopping between the object and
an empty region of the sky. Dividing fringe amplitude by total flux the
visibility is obtained.
For good accuracy of the visibility determination, the
relative fluxes of the combining beams have to be known at the time of the
fringe measurement. To this end, part of the incoming light can be extracted
out of both of the
incoming beams before beam combination, and these
socalled 'photometric beams' are also imaged onto the same detector as the
interferometric signal.
Spectral resolution is obtained by filters or with a
prism or grism in the parallel beam after beam combination. This provides
moderate spectral resolution and at the same time keeps the detector from
saturating.
In principle, for the measurements with MIDI the full
interferometric field of ±1" is available simultaneously.
This is a small area on the sky, but the point to
emphasise is that there is at all an interferometric field beyond the Airy disk
of a single telescope.
For detector we use the 320x240 Si:As Impurity Band
Conduction (IBC) array manufactured by Raytheon. Pixel size is 50 µm x 50 µm,
peak quantum efficiency over the 228 µm response range is > 40%, the active
area > 95%, and less than 1% bad pixels are expected. Power consumption is
≈ 60 mW, and maximum frame rate 700 Hz. Detector control is quite
flexible but complicated by the large number of controllable parameters. Our
laboratory
tests will have to verify the critical quantities of
full well capacity of 1.1 x 107 e- and the read noise.
This will require fine tuning of signal size for best S/N ratio15.
The design of the cold optics16
is shown in Figure 2. For focal plane mask, filter, dispersing element, and
camera different elements can be selected by moving a slider or a wheel.
cameras |
field camera |
lambda/D
= 3 pixels |
|
spectroscopic camera |
lambda/D = 1x2 pixels (two in spatial direction) |
|
pupil viewing camera |
pupil
diameter = 40 pixels |
dispersing elements |
grism |
resolution
per pixel = 520 |
|
prism |
resolution per pixel = 60 |
|
none |
for observations with filter |
filters |
10 filters |
narrowband to full N band |
focal plane |
open |
full ±1" field |
|
slits |
14 lambda/D |
|
diaphragms |
14 lambda/D |
|
10 µm monomode fiber |
best beam cleaning17 |
In addition, special setups are available for testing
and alignment. The positions for future 20 µm optical components already are
provided, allowing an easy upgrade to this second wavelength range.
Figure 3 shows how the mechanisms associated with the
different optical components are distributed within the 40 K envelope, and how
they are mounted on the socalled cold bench base plate. The main design
criterion has been to have a stable, stressfree construction out of one single
material, in this case the aluminum alloy Al 6061T651, where the annotation
T651 indicates a special annealing heat treatment. Homologous contraction
during cooldown (0.42% from room temperature to 35 K) will preserve the
alignment of the optical paths within the 40 K envelope, deferring the
alignment and adjustment tasks to the warm parts of the instrument. The 40 K
radiation shield connected to the cold bench and the outer radiation shield connected
to a liquid nitrogen tank at 77 K are not shown in Figure 3 for clarity, nor is
the boxshaped cryostat. The outer dimensions of this cryostat are partially
outlined by the the front end of the eight motor/encoder units and by the
larger one of the two flanges on the closed cycle cooler. With dimensions of
850 mm x 598 mm on the table, 804 mm height, and a weight of 300 kg 350 kg
this is an instrument
of moderate size.
Figure 3. Design of the cold part of MIDI. The size of
the cold bench to which the optical elements and the moving devices are mounted
is 596 mm x 428 mm. The weight of the complete assembly, excluding motors and
closed cycle cooler, is 28.9 kg.
Figure 4. MIDI setup in the interferometric
laboratory. The optical table carrying the cryostat, external optics and
auxiliary equipment has dimensions of 1500 mm x 2100 mm.
Figure 4 shows how the cryostat18,
with turbomolecular pump mounted, is arranged on the optical table of MIDI in
the interferometric laboratory. The closedcycle cooler, connected to the
cryostat by a bellow (appearing black) is rigidly mounted to a separate massive
block of cast material just next to the MIDI table. With respect to Figure 4,
the delay lines of the VLTI would run left/right near the bottom the page.
Dedicated beam compressors more to the right of the MIDI table in the
interferometric laboratory keep the image of the telescope pupils on the cold
pupil
stops inside the MIDI dewar with alongitudinal
precision of ≈1 cm. The beams from the telescopes (any telescope pair can
be chosen for the two input beams) approach the cryostat from the lower part of
the figure, passing over the socalled warm optics bench, a separate
breadbord with two corners cut out, and mounted directly on the large optical
table. The small fourmirror internal delay lines close to the dewar are laid
out asymmetrically in order to compensate for asymmetries in the way the beams
are directed towards the instrument. The two alignment plates on the warm
optics bench are removed during the observations. In addition to the
calibration provided by the VLTI system LEONARDO, which is situated a couple of
meters upstream of the instrument, MIDI has two calibrating
units of its own: a blackbody screen which can be
moved in and out of the fieldofview of the instrument, and a laser in the
neighbouring room, the radiation of which can be made replacing the
interferometric beams by moving in a small plate carrying the necessary optics.
The path of the laser beam through the laboratory wall to the warm optics bench
can be seen in the upper right of the figure.
Figure 5. The standard fringe sampling mode (``ABCD
method'').
The optical path difference (OPD) between the
interfering beams is modulated by the equivalent of one wavelength, either by
having four steps, separated by lambda/4 (middle panel, measurement taken at
dotted position), or by a continuous linear sweep (ramp) one wavelength (lower
panel, integration over the shaded interval); and this modulation pattern
is repeated as needed. In both cases, the phase of the
modulating signal signal is equal to tan-1
[(A-C)/(B-D)]+p/4,
and the visibility proportional to sqrt[(A-C)2 + (B-D)2],
with proportionality factors of 1.0 and 2*sqrt(2)/p,
respectively.
Because 10 µm arrays have the bad reputation of
unstable flat fields, the main observing modes for MIDI are using onthepixel
comparison for background subtraction. In the socalled ``ABCD'' method (see
Figure 5) the optical path difference (OPD) between the incoming beams is
stepped or scanned through one wavelength. The detector pixels in the Airy disk
of a sufficiently unresolved object then will experience a sinusoidal flux
variation, with four measurements being taken per period. Since the method
works from differences between these flux values measured
on the same pixel(s), it is comparatively insensitive
to flatfield effects.
If the interferometric measurements are performed with
dispersion, different ways of deriving the modulation can be applied. For
nonzero OPD, constructive and destructive interference will follow each other
when moving along the spectrum. This gives rise to a ``channeled'' (i.e.
transversed by dark lines) appearance of the spectrum. The
distance between adjacent minima or maxima in the
channeled spectrum is constant in optical frequency space and inversely
proportional to the momentaneous OPD value. With some knowledge or assumption
on the nature of the OPD fluctuations and on the wavelength dependence of
visibility for the object then both, the OPD history during the measurement and
the object visibility can be obtained by a fitting procedure11. The expectation is that by the associated increase
in effective integration time, when compared to selffringe tracking
measurements, one magnitude
or more can be gained in limiting sensitivity. These
socalled coherent integration procedures, which will as a side effect
allow slow (a few s) OPD tracking, will also be available for observations with
MIDI.
Table 2 gives an indication how the limiting magnitude
for selffringe tracking depends on the circumstances of the observations. Read
noise and transmission of optics lead to the result that the achievable S/N
ratio and hence limiting magnitude depends on spectral resolution, which should
not occur in the truly background limited case.
Table 2. Estimated point source sensitivities for 100
ms integration time
Therefore, best sensitivity, N = 4.3 mag (slightly
less than 1 Jy) may be expected for broad band observations. On the other
extreme, one could decide to observe with fibers (60% coupling efficiency),
using the grism with photometric channels (taking 35% of the light) and reading
out the detector at its maximum rate of 4 ms per measurement. The resulting
limiting magnitude for selffringe tracking then would be N = 1.1 mag (110 Jy)
for an individual resolution element and N = 0.1 mag (43 Jy) for the combined
signal of 10 resolution elements.
In all cases, external fringe tracking over a time
interval T will improve the sensitivity considerably, theoretically by at least
a factor of sqrt(T/0.1s) or already by 2.5 mag for even a modest 10 s interval
of fringe stabilisation.
The scientific Potential of MIDI observations will be
discussed in more detail elsewhere in these conference proceedings13. Apart from the spatial resolution given by the
geometry of the VLTI array, it strongly depends on the sensitivity of the
instrument and on the accuracy of visibility calibration. Before fringe
tracking will be available on the VLTI in 2003, the limiting magnitude will be
N = 4 mag (1 Jy) at best (see Table 2). When fringe stabilisation becomes
available, fringe motion due to atmospheric turbulence will be compensated, and
interferograms can blindly be added together without smearing the fringe
signal. A total integration time of 15 min should improve the sensitivity by a
factor of 100 to N = 9 mag (10 mJy). With this kind of performances,
extragalactic studies come within reach of the instrument. However, it has to
be stressed that these are point source sensitivities, related to correlated
flux, and a somewhat extended source may have to be much brighter to produce
measurable fringe
signals. Moderate accuracies of ≈ 5% 10% in
visibility are expected in general, although on bright sources, using
photometric monitoring of the input beams by means of the photometric channels
of the instrument, accuracies of 1% appear feasible.
Objects |
Topic |
Number with(without) fringe tracker |
required telescopes |
Active Galactic Nuclei |
dust tori |
5* (≈ 12) |
UTs only |
Young Stellar Objects |
geometry and disk structure |
6 (≈ 15) |
UTs mostly |
Extrasolar planets |
detection by shift of light center |
(2) |
UTs only |
AGB stars |
spatial distribution of dust components |
≈ 10 |
|
Smaller programs |
Galactic center, hot stars very low mass stars |
(≤ 5) |
UTs mostly |
*NGC 253,
NGC 1068, NGC 3758, NGC 5128, Circinus (J1413-6520)
Table 3.
Outline of first scientific observations planned with
MIDI
This will require a similar accuracy in total flux
determination, which has to be achieved by chopping against an empty part of
the sky, a complicated procedure within an interferometer. From these
considerations, we identified as main goals for the first two to three years of
observations with MIDI (see Table 3) the study of nearby active galactic nuclei
and the study of circumstellar material around young stars.
Defining the spatial distribution of different dust
components around asymptotic giant branch stars will be another important
topic. Following the orbits of close young binaries for mass determination and
studying the history of dust formation in red giants are also promising
studies, but because of the long time base needed they are less well suited as
one of the first science programms. Measuring the spatial separation of an
extrasolar planet from its central star is an exciting project, but very, very
challenging for an instrument like MIDI, because the signature will be well
below 1% even for the biggest extrasolar planets known14.
Finally, small projects on individual targets of astrophysical interest will
help to demonstrate the scientific capabilities of the instrument.
MIDI is being built by a european consortium led by
the MaxPlanckInstitut für Astronomie in Heidelberg (PI team),including
astronomical institutes in the Netherlands coordinated by the Netherlands
School for Astronomy
(CoPI), Observatoire de Paris (CoPI for the French
contribution), the KiepenheuerInstitut für Sonnenphysik in
Freiburg, the Thüringische Landessternwarte in
Tautenburg, and Observatoire de la Côte d'Azur (chair of the science group).
The final design review for the optics has been passed
in July 1999, the remaining parts of the final design review happened at the
end of February 2000. Integration of the instrument in Heidelberg has been
completed in July, 2001, and the first fringes were obtained in the lab on
October 30th, 2001. Shipment to Paranal is now foreseen in October, 2002, and
commissioning with the siderostats of the VLTI is planned starting before the
end of 2002. Commisioning with the 8 m telescopes will then follow in the first
half of 2003, and with the 1.8 m auxiliary telescopes later the same year. MIDI
therefore should begin science observations by the end of 2003.
The possible future development of the instrument
includes
·
Addition of a wavelength band at 20 µm. Space for the
necessary optical components has already be reserved in the optical and
mechanical design
·
Phase measurements by external referencing. This has
to await the implementation of the VLTI subsystem PRIMA1,3 , which will become operational in 2003
·
High spectral resolution (lambda/Delta lambda = 10
000), to allow gas phase and kinematical studies.
·
Simultaneous measurements on several baselines for
closure phase imaging
·
A nulling option. The very accurate beam cleaning to
be expected from the use of singlemode fibers should be an advantage when
developing this option12
While the first two upgrades can be obtained with
modest modifications on the present instrument, the three
latter ones will require substantial changes or a
complete redesign.
1. A. Glindemann and 17 coauthors, The
VLTI - The Observatory of the 21st Century, ESO Messenger No. 98, pp. 1-7,
December 1999
2. R. Petrov, F. Malbet, A. Richichi, and
K.H. Hofmann, AMBER, the NearInfrared/Red Focal Instrument, ESO
Messenger No. 92, pp.11-14, June 1998
3. A. Quirrenbach, V. Coudč du Foresto, G.
Daigne, K.H. Hofmann, R. Lattanzi, M. Osterbart, R. le Poole, D. Queloz, and
F. Vakili, PRIMA - Study for a DaulBeam Instrument for the VLTI
interferometer, in Astronomical Interferometry, R. D. Reasenberg, ed.,
Proc. SPIE 3350, pp. 807-817, 1993
4. Ch. Leinert, and U. Graser, MIDI -- the
Midinfrared interferometric instrument for the VLTI, in Astronomical
Interferometry, R. D. Reasenberg, ed., Proc. SPIE 3350, pp. 389-393, 1993
5. Y. Rabbia, D. Mekarnia, and J. Gay, Infrared
interferometry at Observatoire de la Côte d'Azur, France, in Infrared
Technology XVI Proc. SPIE 1341, pp. 172182, 1993
6. W.C. Danchi, M. Bester, C.G. Degiacomi,
P.R. Mc Cullogh, and C.H. Townes, Visibility curves at 10 microns wavelength
for stars with dust shells, Amplitude and intensity spatial interferometry,
Proc. SPIE 1237, pp. 327334, 1990
7. W.C. Danchi, M. Bester, C.G. Degiacomi,
P.R. Mc Cullogh, and C.H. Townes, Location and phase of dust formation in
IRC +10216 indicated by 11 micron spatial interferometry, ApJ Letters 359,
L59L63, 1990
8. M. Bester, W.C. Danchi, C.G. Degiacomi,
C.H. Townes, and T.R. Geballe, Distribution of dust about o Ceti and ff
Orionis based on 11 micron spatial interferometry. ApJ Letters 367,
L27L31, 1991
9. M. Shao et al., The Mark III stellar
interferometer, Astron. Astrophys. 193, 357371, 1988
10. I. Porro, U. Graser, Ch. Leinert,
Estimated performance for 10micron interferometry at the VLTI with the MIDI
instrument, in Working on the fringe: an international conference on
Optical and IR interferometry from ground and space eds. S. Unwin and R.
Stachnik, Astronomical Society of the Pacific Conference series, San Francisco
1999, p.34
11. J.A. Meisner, Coherent estimation
of complex fringe visibility: a generalized approach, in Interferometry
in Optical Astronomy, Proc. SPIE, Vol. 4006, p. 1068, 2000
12. G. Perrin, Ch/ Leinert, U. Graser,
L.B.F.M. Waters, and B. Lopez, MIDI, the 10 µm interferometer of the VLT: a
step towards midinfrared interferometry in space, Proceedings of the
DARWIN conference, Stockholm, November 1999
13. B. Lopez et al., Astrophysical
potentials of the MIDI VLTI instrument, in Interferometry in Optical
Astronomy, Proc. SPIE,
Vol. 4006, p. 54, 2000
14. B. Lopez and R. Petrov, Direct
detection of hot extrasolar planets with the VLTI using differential
interferometry, in Interferometry in Optical Astronomy, Proc. SPIE, Vol. 4006, p. 407, 2000
15. S. Ligori, U. Graser, B. Grimm, R.
Klein, Design and tests of the MIDI detector subsystem, in Interferometry
in Optical Astronomy, Proc. SPIE, Vol. 4006, p. 136, 2000
16. J.W. Pel, A. GlazenborgKluttig, J.C.
de Haas, H. Hanenburg, and R. Lenzen, Cold optics of MIDI: the midinfrared
interferometric instrument for the VLTI, in Interferometry in Optical
Astronomy, Proc. SPIE,
Vol. 4006, p. 164, 2000
17. G. Perrin, V. Coud'e du Foresto, and
M. Olivier, Spatial filtering for 10 µm interferometry, in Interferometry
in Optical Astronomy, Proc. SPIE, Vol. 4006, p. 1007, 2000
18. R.R. Rohloff, U. Graser, N. Ortlieb, M. Lebong, and W. Laun, Cryo design for the VLTI MIDI instrument, in Interferometry in Optical Astronomy, Proc. SPIE, Vol. 4006, p. 277, 2000