Observation Planning Tools for the ESO VLT Interferometer 
Derek J. McKaya,b, Pascal Ballestera, Jakob Vinthera 
a European Southern Observatory, Karl-Schwarzschild-Str. 2, 85748 Garching, Germany 
b Rutherford Appleton Laboratory, CCLRC, Chilton, Didcot, OX11 0QX, United Kingdom 
ABSTRACT 
Now that the Very Large Telescope Interferometer (VLTI) is producing regular scientic observations, the eld 
of optical interferometry has moved from being a specialist niche area into mainstream astronomy. Making 
such instruments available to the general community involves dicult challenges in modelling, presentation and 
automation. The planning of each interferometric observation requires calibrator source selection, visibility pre- 
diction, signal-to-noise estimation and exposure time calculation. These planning tools require detailed physical 
models simulating the complete telescope system | including the observed source, atmosphere, array congu- 
ration, optics, detector and data processing. Only then can these software utilities provide accurate predictions 
about instrument performance, robust noise estimation and reliable metrics indicating the anticipated success of 
an observation. The information must be presented in a clear, intelligible manner, suciently abstract to hide 
the details of telescope technicalities, but still giving the user a degree of control over the system. The Data 
Flow System group1 has addressed the needs of the VLTI and, in doing so, has gained some new insights into the 
planning of observations, and the modelling and simulation of interferometer performance. This paper reports 
these new techniques, as well as the successes of the Data Flow System group in this area and a summary of 
what is now oered as standard to VLTI observers. 
Keywords: Very Large Telescope Interferometer, observation planning tools, modelling 
1. INTRODUCTION 
The European Southern Observatory operates the Very Large Telescope (VLT) Observatory on Cerro Paranal 
(2635 metres altitude) in northern Chile, comprising four 8.2-metre Unit Telescopes (UTs) and numerous smaller 
telescopes. The VLT also operates in an interferometric mode as the VLT Interferometer (VLTI), which allows 
the coherent combination of stellar light beams. The light is guided from the UTs via Coude optical trains into 
an optical delay-line tunnel. The dierence in arrival times of the light from the rst telescope with respect to 
that from the second telescope is compensated by the delay lines so that the beams have eectively zero Optical 
Path Dierence (OPD) at the detector, where the interferometric beam combination takes place. The VLTI is 
described in more detail in the work of Glindemann, et al..4 
Currently under construction are four 1.8-metre Auxiliary Telescopes (ATs), the rst of which has now 
achieved \rst light".6 When completed, these will be the main \workhorses" of the VLTI, and will be in 
constant operation (unlike the UTs, which have their own individual programmes, and will not always be used 
for VLTI observations). 
While the UTs are xed, the ATs are designed to be relocated to any of a number of \stations", thus permitting 
a larger range of interferometer-spacings (\baselines") to be obtained. The movable ATs, in conjuction with the 
xed UTs oer a range of baselines up to 200 metres.6 
In addition to the UTs and ATs, which are the intended telescopes for \normal" operation, two 0.4-metre 
siderostats have also been constructed. These siderostats may be relocated to any of the AT stations and, with 
their simplicity and stability, serve as commissioning telescopes for the VLTI and its instrumentation. 
The VLTI has been designed to be operated like a regular astronomical instrument: that is, detailed technical 
knowledge is not mandatory. This is a very important development in optical interferometry, as it allows a much 
wider community of astrophysicists to carry out their own observational experiments.8 However, it does mean 
Further author information: (Send correspondence to D.J.M.) 
D.J.M.: E-mail: dmckay@eso.org, Telephone: +49 89 3200 6548
that this community must be given the opportunity and ability to plan observations so as to accomplish their 
data acquisition without loss due to inappropriate observing parameters. Such planning is akin to the \exposure 
time calculators" which are prevalent on many telescopes. However, the provision of such a tool is not a trivial 
task. 
This paper describes the generation of a model for the calculation of observational-quality parameters, iden- 
tifying those that are important and those which have negligible eect. It describes how the model may be tuned 
with regards to dierent instruments operated on the VLTI and describes how the model compares to existing 
VLTI data. Additionally, it describes two other tools used for VLTI observation planning. 
2. VLTI INSTRUMENTS 
There are three instruments which are either operational or currently being commissioned: VINCI, MIDI and 
AMBER. 
VINCI is the VLT INterferometer Commissioning Instrument that operates at 2.2 microns. It was the rst 
VLTI instrument to be installed, and is not oered to the general community, but rather is used for commissioning 
and testing, due to its relative simplicity and stability. Since rst fringes with the 16-metre baselines using the 
siderostats (17 March, 2001) and 103-metre baselines with the UTs (30 October, 2001), most of the calibration 
data used in the work described in this paper was acquired using this instrument. The observations were 
made to assess the compliance of the instruments with the technical specications as well as to characterise 
the performance of the facility in its rst phase of development. In addition to these more technical tasks, a 
number of observations are certainly also useful for scientic purposes. In order to fully involve the community 
in analysing and understanding the data and its scientic and technical implications, ESO has decided to make 
these data available to the community through the ESO archive. 
MIDI is the MID-infrared Instrument, covering the wavelength range 8 to 13 microns and measuring spectrally- 
dispersed fringes. It allows resolutions (=B) of 20 milliarcsec at 10 microns to be achieved. The main scientic 
objectives include the study of protostars and very young stars, circumstellar discs, brown dwarfs, tori around 
galactic nuclei, the centre of our own galaxy and the search for exo-planets. 
AMBER, the third VLTI instrument, operates in the near-infrared range, between 1 and 2.5 microns, using 
two or three Unit Telescopes. The instrument has been designed for up to three telescope beams in order 
to be able to perform imaging using phase closure techniques. It is anticipated that the additional operating 
wavelength of 0.6 microns will be added at the time the ATs become operational. The main scientic objectives 
are the investigation, at very high angular resolution, of disks and jets around young stellar objects, and AGN 
dust tori with a spectral resolution up to 10000. 
3. OBSERVATION PLANNING 
In order to plan an interferometric observation and to assess its feasibility, one needs adequate tools to model the 
visibilities for a specied array conguration, taking into account constraints like shadowing eects, the range of 
the delay lines, etc.. In addition, appropriate calibration stars must be selected. Two specic tools are provided 
for this purpose: the VLTI Visibility Calculator (VisCalc) and the calibrator selection tool (CalVin). 
3.1. VisCalc 
VisCalc is a web-based tool that provides calculations of simulated dispersed visibilities based on software models 
of the VLTI instruments. The declination and spectral energy distribution, as well as the source geometry, are 
parameters used to specify the observation target. 
Visibilities are calculated analytically for uniform discs, Gaussian discs and binaries. Visibilities may also 
be calculated numerically for a user-provided brightness distribution, which is uploaded as a FITS le. The 
user-specied observation conditions include the starting hour angle and the duration of the observation, as well 
as the instrument and array conguration. Dierent results can be displayed including the uv-tracks, the input 
image and its Fourier transform, plots of visibility versus time, visibility squared versus time, loss of correlated 
magnitude and the illumination distribution. 
Screenshots from a typical VisCalc session are shown in Figure 1.
Figure 1. Screenshots from a typical VisCalc session. Part of the input page is shown to the left and part of the output 
page is shown to the right. The output not only includes tabulated values, but also plots indicating the expected visibilities 
and the path of the baseline tracks over the uv-plane (the Fourier transform of the source model is shown). 
3.2. CalVin 
The calibrator selection tool (CalVin) is used to nd suitable stars that may be used to calibrate the VLTI at 
the time of the observation. CalVin provides a similar interface to that of VisCalc, but it involves a two-stage 
selection process. On the rst input page, the target coordinates, the array and instrument congurations can be 
selected. The default search criteria are displayed on an intermediate page which allows the search parameters 
to be rened. 
On the results page, the table of matching calibrators is listed. For all matching calibrators, the visibility and 
observability information are calculated and displayed. It is then possible to use VisCalc for a more comprehensive 
calculation of the visibility information. 
Screenshots from a typical CalVin session are shown in Figure 2. 
Both tools can be accessed from the VLT Exposure Time Calculators page. The standard version shows 
only those congurations that are oered for the current Call for Proposals. It is updated for each new Call 
for Proposals in order to re
ect the oered VLTI baseline congurations and instrument modes. An \expert" 
version, accessible from the ETC preview pagey, oers an extended interface with many more choices. It supports 
the modes and congurations that are currently not oered. 
http://www.eso.org/observing/etc 
yhttp://www.eso.org/observing/etc/preview.html
Figure 2. Screenshots from a typical CalVin session are shown, with part of the input page shown on the left and the 
output on the right. Colour-coding makes the table easier to follow, and links within each of the entries allow the display 
of tabulated and graphical data. 
4. THE PROBLEM OF AN EXPOSURE TIME CALCULATOR 
In order to plan astronomical observations, one typically uses some sort of software tool to specify the observing 
options and expected conditions, to thus predict the observation quality (signal-to-noise ratio, etc.) and anticipate 
the amount of observing time necessary to achieve the required sensitivity for the scientic target. 
While exposure time calculators (ETCs) are a well-established idea, their use in the world of optical inter- 
ferometry is new. As the VLTI was designed to be run as a standard | rather than specialist | instrument, 
the provision of such tools is crucial to its integration. Furthermore, the calculations are not straight-forward, 
and involve modelling the atmosphere, telescopes, delay lines, instrument, detector, non-physical and combined 
parameters, that cannot be independently measured or calibrated. 
Additionally, it is not simply a matter of calculating the total number of photons, but it is also necessary to 
incorporate the concept of visibility. 
However, to make things even more complex, it was decided to attempt to derive a generic algorithm. In 
other words, one that was capable of simulating the expected performance of all three VLTI instruments. The 
advantage of this is several-fold: 
 By modelling the general case, it helps understand the VLTI better. 
 It forces individual parameters to be properly explored and understood, rather than just hiding them in 
\empirical factors" that emulate the performance characteristics.
 There are many parameters that are common to all three instruments, and this work need not be replicated, 
either in terms of mathematical modelling or by way of additional code. 
 A uniform model will be easier to implement as part of a unied tool for predicting VLTI performance.
(1) 
5. DEFINITIONS 
How does one quantify the performance of a telescope (or telescope interferometer) system? This value must be 
a physical quantity (and not just a Quality Parameter | good, okay, poor, etc.) and must be understandable 
by astronomers. The value must be such that it can be predicted, but also so that it is able to be translated into 
a measurable quantity in the nal observed data; thus, the quality of the observation can be quickly assessed, 
and easily compared to the predictions. In the case of conventional imaging and spectroscopy, this is generally 
characterised by a signal-to-noise ratio. However, this does not suce for the interferometric results of the VLTI. 
The VLTI produces an average of the square of the modulus of the complex coherence factor, which is 
associated with the baseline and position angle of the interferometer elements with respect to the object being 
observed. The square of the coherence factor is measured based on photon counts as the mirror mounted on the 
piezo translator of the VLTI sweeps back and forth through the fringe. 
The VINCI instrument generates four time-sequence signals | two photometric and two interferometric | 
the latter of which are used to derive the square of the coherence factor, 2. Each observation is actually made up 
of many scans of the fringe, typically 500, and so a series of 2 values are determined. Based on the photometric 
signals, only high-quality measurements are accepted by the data reduction pipeline, which are in turn averaged 
and a variance determined, 
2 and 
d2 respectively. The number of scans, n, that was used for these data, 
along with the data themselves, are also recorded in the processed data le. 
The term signal-to-noise ratio is used to refer to the data quality of the photometric channel. The term 
precision is used to convey a measure of the quality of the interferometric data. 
p 
Precision = h2i + hd2ipn ph2i 
ph2iN 
where n is the number of measured scans, and N is the number of scans to which the system is normalised. 
Note that, initially, all quantities are calculated for a single scan of a given exposure time. However, in 
reality, the VLTI takes single observations with several hundred scans, which are averaged to create a single 
measurement. As the exact number of scans is regulated, with rejected scans occurring when low signal-to-noise 
is detected on the photometry chanels, the exact number of scans will 
uctuate according to the conditions. 
6. MODELLING 
The model that has been developed addresses the ve major areas that in
uence the performance of the VLTI. 
These are shown in Figure 3. 
Of course, with a unied model, it is possible to have a unied simulation utility | a single Exposure Time 
Calculator (ETC), to compute the expected signal to noise ratio of the selected instrument. Such a utility acts 
as a planning tool for astronomers, allowing them to test instrument performance against predictions, to explore 
the available parameter space when devising observations and to assist in planning observations to get the best 
results for the allocated time. 
However, there is a nasty cyclic problem. While astronomers need the ETC to make useful observations, the 
useful observations are needed to make a realistic model.
Instruments 
• VINCI 
• MIDI 
• AMBER 
Fringe 
Tracking 
• FINITO 
Atmosphere 
UNIFIED UNIFIED 
VLTI VLTI 
MODEL MODEL 
Adaptive 
Optics 
Telescopes • MACAO 
• 8.2m Unit Telescopes 
• 1.8m Auxiliary Telescopes 
• 0.4m Siderostats 
Figure 3. The ve major components of the VLTI performance model. 
2:5 mc = 2:5 log(210 m 
) 
6.1. Developing a model 
The model that has been developed uses a subset of the VINCI data archive, collected over the period from March 
2001 to November 2002. These data represent 7793 independent measurements. These data are applicable to 
the AMBER instrument, as well as VINCI. However, modications to the model were required for MIDI, due 
to the fact that it has a spectral capability, thus involving channel-based calculations, and that it operates at 
N-band (  10m). 
Initially, the model needs to predict 
d2. 
In order to get an accurate prediction of the visibility, one needs an accurate diameter of the astronomical 
target. Obviously, this is usually not known at the time that the observations are planned. 
The model is also dependent on the projected baseline; something that will change not only due to the 
selection of telescope station positions but also on the topocentric position of the source. As this changes with 
hour angle, it is not always possible to estimate this parameter at the time of observation planning. Furthermore, 
observations will typically wish to observe such that a range of visibilities are measured. This will allow the 
astrophysical quantities to be determined, whether this is the tting of a stellar disk model or the ascertaining 
of binary system parameters. 
Therefore it is necessary, in the presentation of results to users, to indicate model results over the range of 
possible visibilities. That is, the visibility (or, alternatively, the measured coherence factor 
2) is maintained as 
a free parameter and the results are presented as a function of this, or the source-independent quantity correlated 
magnitude, mc, which is dened as: 
where  is the coherence factor and m is the photometric magnitude. 
However, development of the model to match the existing data has proved problematic. In addition to the 
data generally being noisy, the instruments are still being commissioned, and the data were interspersed with 
the remnants of acquisition problems and poor hardware setup and conditions. 
(2)
Figure 4. Predicted and measured visibilities of the VINCI instrument, using the 0.4-metre siderostats. The grey lines 
indicate the actual data, binned in groups of 100 measurements according to correlated magnitude. From top to bottom, 
the worst 10%, mean, median and best 10% of the measurements, within each bin, are shown. In total, 7793 measurements 
were used to compile these data, encompassing observations between 19 March 2001 and 24 November 2002. 
7. UNIFIED MODEL 
The unied model devised is ultimately driven by four eects: 
 The loss of fringe contrast due to the atmosphere (this is what is commonly referred to as the piston noise). 
This source of noise dominates the noise of bright sources. 
 The photometry (both in terms of the system transmission as well as the actual object brightness). This 
dominates the noise observed for faint sources. 
 The thermal background and read-out electronics. 
 The loss of fringe contrast due to the complete telescope system. 
This model was implemented and used to calculate performance predictions for the VINCI instrument. The 
results are shown in Figure 4. 
From the investigations and modelling it became apparent that interferometry presents a new layer of com- 
plications over conventional \photon counting" observation predictions. While these \normal" exposure time 
calculators deal with noise in a linear fashion, the existence of interferometric data within the Fourier domain 
means that the traditional calculations break down, and that the combination of visibility (due to variations in 
the projected baseline length, in combination with the normal variations in brightness between sources) make 
direct comparisons dicult. 
From the analysis it is clear that the primary trade-o in setting the exposure time of an observation, is 
that between the piston noise and the photometric signal-to-noise ratio. Obviously an increase in the exposure
Figure 5. Observations of the star  Psa (20 October, 2002). The points indicate the measured data, and the line is the 
values predicted by the model. 
time will result in improved photometric statistics, but this also means that the longitudinal phase noise due 
to the atmosphere will \blur out" any fringes. Conversely, the reduction of the exposure time will eectively 
\freeze" the fringes, which are combined in the Fourier domain where the phase (piston) noise will no longer 
contribute. However, there is inevitably a point where the degradation in signal-to-noise of the photometry curbs 
this selection. 
As can be seen, there is a large variation in the best and worst results for any given correlated magnitude. 
However, the prediction closely matches the median result. 
We note that the detector noise contributes to, but doesn't dominate, the VINCI data at any given magni- 
tude/visibility regime. However, for fainter sources (or those observed which result in low values of the visibility), 
the detector noise is a signicant fraction of the overall noise component. 
There is a point when the detector noise becomes dominant, but this occurs only at such faint magnitudes 
that the existing data-reduction software of the VLTI instruments rejects such data as being of no worth long 
before this theoretical point is reached. That is why, within the current model and observing parameter space, 
there are only references to the piston-noise and photon-noise dominated scenarios. 
The two main sources of noise at the detector level are the readout noise on the detector and the detected 
background. The latter is not strictly a detector noise as such, but it represents the linear combination of the sky 
background, the instrument emission and the intrinsic thermal noise from the detector itself. These are dicult 
to segregate, and so were modelled together. 
Additionally, a study of a single observation under good conditions was performed in order to verify parts 
of the model. The objective here was to remove the ambiguity of non-uniform calibration, which occurs with 
data from multiple nights; dierent hardware conguration and calibration can cause major shifts in the system 
performance, which are dicult to factor into any model. 
The data chosen were from a single star  Psa, observed on 20 October, 2002.3 Good hour-angle coverage 
provided a large range of baselines and, subsequently a large range of 2. 
The results are shown in Figure 5.
Figure 6. The input and output pages of the prototype VLTI Exposure Time Calculator. 
7.1. MIDI and AMBER 
We currently have insucient data to make a reliable study on MIDI. At the time of writing, MIDI integration 
continues, and it is hoped that a larger database of reliable measurements may be acquired as the instrument 
stabilises and operations become routine. 
As the AMBER system has not yet been commissioned, integration of its performance into the model may only 
be derived from the design documentation. However, the provision for supporting this instrument is incorporated 
into the model, and it is believed that integration of actual performance data should be straight-forward. 
8. EXPOSURE TIME CALCULATOR 
A prototype exposure time calculator has been developed that uses this model to calculate expected observing 
performance. Based on the web-interface of existing ESO observation preparation tools, it presents an initial 
input screen, where users can select the basic parameters of the observation and a results page, which is displayed 
on submission of the input data. Figure 6 shows a typical session. 
The input page allows the specication of the input 
ux distribution (spectral prole and magnitude), the sky 
conditions (airmass, seeing and coherence time) and the interferometer conguration (instrument, instrument 
mode, telescopes, adaptive optics, fringe tracking and exposure/scan-time control). 
The output summarises the input data for verication (and association in the case of printing the page) and 
displays the data as a series of graphs. This is done in preference to tabular or numerical results, as it allows a 
better appreciation of the limiting factors involved in an observation, and allows the user to quickly assess cut-o 
points and regions of the parameter space that would be optimal.
9. CONCLUSION 
With the development of a unied model, it has been possible to generate a generic algorithm to predict the 
performance of the VLTI for each of the three instruments: VINCI, MIDI and AMBER. The model caters for 
dierent telescopes, adaptive optics, fringe-tracking and atmospheric conditions, in addition to the astronomical 
characteristics of the object whose observations are being simulated. 
A prototype, HTML-based simulator has been developed, which acts as an exposure time calculator for the 
VLTI. With the acquisition of a larger database of quality measurements, it will be possible to rene the model 
to generate more accurate results. 
Together with a visibility calculator and calibrator selection tool, the ETC will complete a suite of observa- 
tion preparation tools allows the VLTI to be readily accessed by the astronomical community as a \standard 
observation" system. 
ACKNOWLEDGMENTS 
In preparing the models and software described in this paper, the authors have drawn on the data and work of 
several others. In particular, the authors wish to thank P. Kervella, V. Coude du Foresto, C. Leinert, I. Percheron, 
J. Meisner, A. Glindemann, F. Malbet, A. Richichi, B. Wiseman and M. Wittkowski. 
REFERENCES 
1. Ballester, P., et al., VLT Interferometry in the Data Flow System, this volume. 
2. Coude du Foresto, V., Ridgway, S., Mariotti, J.-M., "Deriving object visibilities from the interferograms 
obtained with a bre stellar interferometer", A&A 121, 379{392, 1997 
3. Davis, J., et al. Calibration Observations of Fomalhaut with the VLTI, A&A, in preparation. 
4. Glindemann, A., et al. "VLT interferometer: a unique instrument for high-resolution astronomy", SPIE 
conference Interferometry in Optical Astronomy SPIE 4006, 2000. 
5. Kervella, P., VINCI Limiting Magnitude, ESO Doc. VLT-TRE-ESO-15810-2867, 2002. 
6. Koehler, B., Flebus, C., 2000, in SPIE Conf. 4006{02, Astronomical Telescopes and Instrumentation 
7. Leinert, C., et al. 10m interferometry on the VLTI with the MIDI instrument: a preview, SPIE, 2000. 
8. Perrin, G., & Malbet, F., 2002a, in EAS Publication Series, Vol. 5, 2002. Observing with the VLTI (Les 
Houches Eurowinter school, Feb.2002), EDP Sciences, in preparation. 
9. Petrov, R. G., et al., AMBER: the near-IR focal instrument for the VLTI, SPIE, 2000.