Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more ➡
Standard view
Full view
of .
Add note
Save to My Library
Sync to mobile
Look up keyword
Like this
0 of .
Results for:
No results containing your search query
P. 1


Ratings: (0)|Views: 4,848|Likes:

More info:

Published by: European Southern Observatory on May 08, 2009
Copyright:Attribution Non-commercial


Read on Scribd mobile: iPhone, iPad and Android.
See More
See less


The Messenger
No. 131 \u2013 March 2008
The Messenger 131 \u2013 March 2008
Telescopes and Instrumentation
Advanced Calibration Techniques \ue001or Astronomical

Paul Bristow1
Florian Kerber1
Michael R. Rosa2,3

2 Space Telescope European
Coordinating Facility, ESO
3 A\ue001\ue000liated to the Science Operations and
Data Systems Division, Science Depart-
ment, European Space Agency

ESO\u2019s Calibration and Model Support
Group is involved in a variety o\ue001 activi-
ties related to the calibration and physi-
cal description o\ue001 instruments, with the
objective o\ue001 supporting the reduction
o\ue001 science data and \ue001acilitating opera-
tions. Here we describe the construc-
tion, optimisation and application to sci-
enti\ue000c data reduction o\ue001 physical in-
strument models. Such models have
been implemented \ue001or the HST STIS
spectrograph and \ue001orm an integral part
o\ue001 the data reduction pipelines \ue001or
CRIRES and X-shooter. These models
are supported by validated physical
data o\ue001 the instrumental components
and calibration re\ue001erence data.

The li\ue001e cycle o\ue001 an instrument can be
described as \ue001ollows:
1. Science Requirements
2. Optical Design (Code V/Zemax)
3. Engineering Expertise

4. Testing and Commissioning
5. Operation and Data Flow
6. Calibration o\ue001 Instrument
7. Scienti\ue000c Data and Archive

Experience shows that it is di\ue001\ue000cult to
ensure that the know-how and expertise
that went into designing and building

the instrument (steps 1\u20133) is brought to \ue001ull use in the instrument calibration and scienti\ue000c operations (steps 6 and 7).

A case in point is the wavelength calibra-
tion, in which well-understood physics is
employed to design a spectrograph with
an optimal \ue001ormat while during operations
the dispersion solution is then derived
over and over again in a purely empirical
manner by, \ue001or example, \ue000tting polynomi-
als to a sparse calibration line spectrum.

Most o\ue001 the computations involve rotation
matrices to represent the change o\ue001 ori-
entation o\ue001 the optical ray at the sur\ue001aces
o\ue001 the components. For example, the
matrix representation o\ue001 the orderm
trans\ue001ormation per\ue001ormed by an echelle
grating with constantsE at o\ue001\ue001-blaze
angleq, operates on a 4D vector with

components (l, x, y, z) representing a ray
o\ue001 wavelengthl. Hereq andsEar e
amongst the physical model parameters
\ue001or this instrument.

Hence there is a complete set o\ue001 param-
eters that describe the passage o\ue001 a
photon through the spectrograph. These
pa rameters are physical quantities (an-
gles, distances, temperatures, etc.) and
describe the actual status o\ue001 compo-
nents. They can always be adjusted to
match the observed behaviour o\ue001 the
instrument or to predict the e\ue001\ue001ects o\ue001 tilt-
ing/modi\ue001ying a component. For exam -
ple, adjusting the camera \ue001ocal length will
change the scale on the detector.


The model parameter set can be opti-
mised to refect the per\ue001ormance o\ue001
the operational instrument with suitable
calibration data, in a similar way that a
polynomial dispersion solution would be
\ue000t. The di\ue001\ue001erence is that the parameters
optimised here have physical meaning
and represent the actual con\ue000guration o\ue001
the instrument. There are essentially two
scenarios in which one needs to per\ue001orm
the optimisation.

Be\ue001ore the instrument is actually built, the
only parameters available are those \ue001rom
the instrument design. Inevitably, once
the instrument has been built, it will di\ue001\ue001er
\ue001rom the design predictions, so it is nec-

essary to establish the true values. This

may also be the case a\ue001ter a major main-
tenance intervention, upgrade to the
instrument or even an earthquake, result-
ing in a physical change in the instru-
ment. In this situation a comprehensive
and uni\ue001orm set o\ue001 robustly identi\ue000ed
calibration \ue001eatures \ue001rom dedicated cali-
bration exposures is required. The core
model \ue001unction is then iteratively called
\ue001or the identi\ue000ed calibration wavelengths
and the results o\ue001 each iteration are com-
pared with the centroids \ue001or these wave-

One way to ensure that the engineering
data propagates \ue001rom instrument building
to operations is to capture all the engi-

neering in\ue001ormation in a physical model-
based description o\ue001 the instrument.
This model accompanies the instrument

throughout its li\ue001e cycle and is used to
drive the science data reduction pipeline.
In our concept the model is combined

with validated physical data o\ue001 the instru- mental components and calibration re\ue001er- ence data.

Implementation and application o\ue001 an
instrument physical model

Our approach comprises an instrument-
speci\ue000c model kernel and associated
so\ue001tware to optimise the model parame-

ters and to apply the model\u2019s predictive
power to the calibration o\ue001 science data.
Model kernel

First o\ue001 all a streamlined model o\ue001 the dis-
persive optics, that enables a rapid eval-
uation o\ue001 where any photon entering the
instrument arrives on the detector array,
is constructed. Though based upon the
optical design, it is no substitute \ue001or the
\ue001ully-fedged optical (e.g. Zemax/Code V)
models developed by the designers.
Clearly this model kernel is speci\ue000c to
each instrument, but the \ue001ollowing
sub-components and associated param-
eters are typical:
\u2013 Entrance slit and collimator

\u2013 Relative position and orientation o\ue001
the slit
\u2013 Focal length o\ue001 collimator

\u2013 Pre-disperser (e.g. Prism)
\u2013 Orientation o\ue001 entrance sur\ue001ace
\u2013 Orientation o\ue001 exit sur\ue001ace

\u2013 Temperature
\u2013 Re\ue001ractive index as a \ue001unction o\ue001
wavelength and temperature

\u2013 Main disperser (e.g. refection grating)
\u2013 Orientation
\u2013 Grating constant

\u2013 Camera and detector array
\u2013 Focal length o\ue001 \ue001ocusing optics
\u2013 Orientation o\ue001 detectors
\u2013 Relative positions o\ue001 detectors
\u2013 Dimensions o\ue001 pixel grid

We \ue001ollow the prescription o\ue001 Ballester and
Rosa (1997) in constructing this model.
The Messenger 131 \u2013 March 2008
lengths as measured in the calibration
data. We employ the Taygeta (Carter 2001)
implementation o\ue001 the Simulated Anneal-

ing technique to continually adjust, in a
statistically sound manner,a l l o\ue001 the mod-
el parameters until the best match be -
tween predicted and measured centroids
is \ue001ound. Figure 1 is a schematic repre-
sentation o\ue001 this procedure.

In the case o\ue001 an instrument such as
CRIRES which has multiple modes de -
\ue000ned by the orientations o\ue001 optical com-
ponents (and there\ue001ore by parameters

in the physical model), we are able to

optimise the parameter set \ue001or multiple
modes simultaneously by assigning a
unique value to each o\ue001 the changing
parameters on the basis o\ue001 all data col-
lected \ue001or the corresponding mode. We
can then characterise the parameters
associated with the moving components
that determine the mode.

Most spectrographs have some moving
components that allow selection o\ue001 a
given wavelength range. Since there are
physical limits to the repeatability and
accuracy o\ue001 these mechanisms, it is use-

\ue001ul to be able to \ue000ne tune the model to

match the per\ue001ormance o\ue001 the instrument
at the time o\ue001 a given observation. More-
over, even without human intervention,
instruments develop mal\ue001unctions such
as a dri\ue001t in wavelength zero points that

are not well understood initially. Other

a\ue001\ue001ects such as thermal or gravitational
fexure occur at some level during rou-
tine operations and also subtly a\ue001\ue001ect the
exact details o\ue001 the instrument optics. In
such cases it is clear that there must
be some deviation \ue001rom the initial param-
eter calibration that was done with data
acquired in the absence o\ue001 these e\ue001\ue001ects
(or in the presence o\ue001 another alignment
o\ue001 these e\ue001\ue001ects).

For these reasons we have developed the
capability to re-optimise speci\ue000c parame-
ters, using either automatically identi\ue000ed
wavelength \ue001eatures in contemporaneous
calibration exposures or wavelength
standards speci\ue000ed by users (e.g. known
sky lines seen in science exposures).

These are used in a similar way to that
depicted in Figure 1, except that only the
known changing parameters, or param-
eters suspected o\ue001 causing the spurious
dri\ue001ts, are optimised. Moreover, one

can choose to optimise more parameters
when more data points are available.

We have recently achieved the \ue001ull auto-
mation o\ue001 this process \ue001or CRIRES. The
procedure is illustrated by Figure 2. First
the model is used to trace the locus on
the detector o\ue001 a given entrance slit posi-
tion. A 1D spectrum is then extracted
\ue001rom a Th-Ar hollow cathode lamp (HCL)
\ue001ull slit exposure along this locus and
bright \ue001eatures are identi\ue000ed. Using the
baseline physical model parameter con-
\ue000guration, we predict the positions o\ue001
wavelength \ue001eatures along this locus (red
crosses in Figure 2). A crucial point here
is that we only consider wavelength
\ue001eatures that we know will be well isolated

(see \u201cOptimising Calibration Systems\u201d
below) in order to avoid the possibility o\ue001
\ue001alse matches. The signi\ue000cant o\ue001\ue001set
between the red crosses and the corre-
sponding \ue001eatures identi\ue000ed in the data
(magenta circles) is due to a shi\ue001t in spec-
tral \ue001ormat that has occurred in CRIRES
between the acquisition o\ue001 the calibration
data used to determine the baseline
model parameter con\ue000guration and the
epoch o\ue001 this data. Hence known wave-
lengths are reliably matched to measured
positions in the new data and the model
parameters can be re-optimised to match
the actual per\ue001ormance o\ue001 CRIRES. As

Figure 1: Schematic representation o\ue001 the optimi-
sation process \ue001or instrument physical models.

Optical materials data
e.g. prism refractive
index (T,\u03bb)

Initial \u2018first guess\u2019
configuration file

Spectral atlas for
calibration source,
e.g. high quality Th-Ar
hollow cathode data

Measured feature
centroids in detector
pixel coordinates
from calibration
source exposure

Simulated annealing
Simulated annealing
Change configuration

Compare lists and
compute metric that
describes how well
they match

Physical model


Predicted centroids of spectral features in detector pixel


You're Reading a Free Preview

/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->