PreprintPDF Available

Gaia Radial Velocity Spectrometer

Preprints and early-stage research may not have been peer reviewed yet.

Abstract and Figures

This paper presents the specification, design, and development of the Radial Velocity Spectrometer (RVS) on the European Space Agency's Gaia mission. Starting with the rationale for the full six dimensions of phase space in the dynamical modelling of the Galaxy, the scientific goals and derived top-level instrument requirements are discussed, leading to a brief description of the initial concepts for the instrument. The main part of the paper is a description of the flight RVS, considering the optical design, the focal plane, the detection and acquisition chain, and the as-built performance drivers and critical technical areas. After presenting the pre-launch performance predictions, the paper concludes with the post-launch developments and mitigation strategies, together with a summary of the in-flight performance at the end of commissioning.
Content may be subject to copyright.
arXiv:1804.09369v1 [astro-ph.IM] 25 Apr 2018
Astronomy &Astrophysics
manuscript no. 32763_Cropper c
ESO 2018
April 26, 2018
Radial Velocity Spectrometer
M. Cropper1, D. Katz2, P. Sartoretti2, T. Prusti3, J.H.J. de Bruijne3, F. Chassat4, P. Charvet4, J. Boyadijan4, M.
Perryman5, G. Sarri3, P. Gare3, M. Erdmann3, U. Munari6, T. Zwitter7, M. Wilkinson8, F. Arenou2, A. Vallenari9, A.
Gómez2, P. Panuzzo2, G. Seabroke1, C. Allende Prieto1,10, K. Benson1, O. Marchal2, H. Huckle1, M. Smith1, C.
Dolding1, K. Janßen11, Y. Viala2, R. Blomme12, S. Baker1, S. Boudreault1,13, F. Crifo2, C. Soubiran14, Y. Frémat12, G.
Jasniewicz15, A. Guerrier16, L.P. Guy17, C. Turon2, A. Jean-Antoine-Piccolo18 , F. Thévenin19, M. David20, E.
Gosset21,22, and Y. Damerdji21,23
1Mullard Space Science Laboratory, University College London, Holmbury St Mary, Dorking, Surrey RH5 6NT, UK
2GEPI, Observatoire de Paris, Université PSL, CNRS, 5 Place Jules Janssen, F-92190 Meudon, France
3ESA, European Space Research and Technology Centre (ESTEC), Keplerlaan 1, NL-2201 AG, Noordwijk, The Netherlands
4Airbus Defence and Space, 31 Rue des Cosmonautes, F-31402 Toulouse Cedex France
5Scientific Support Oce, Directorate of Science, European Space Research and Technology Centre (ESA/ESTEC), Keplerlaan 1,
NL-2201AZ, Noordwijk, The Netherlands
6INAF-National Institute of Astrophysics, Osservatorio Astronomico di Padova, Osservatorio Astronomico, I-36012 Asiago (VI),
7Faculty of Mathematics and Physics, University of Ljubljana, Jadranska ulica 19, SLO-1000 Ljubljana, Slovenia
8Department of Physics & Astronomy, University of Leicester, University Road, Leicester, LE1 7RH, UK
9INAF - Padova Observatory, Vicolo dell’Osservatorio 5, I-35122 Padova, Italy
10 Instituto de Astrofísica de Canarias, E-38205 La Laguna, Tenerife, Islas Canarias, Spain
11 Leibniz Institute for Astrophysics Potsdam (AIP), An der Sternwarte 16, D-14482 Potsdam, Germany
12 Royal Observatory of Belgium, Ringlaan 3, B-1180 Brussels, Belgium
13 Max Planck Institut für Sonnensystemforschung, Justus-von-Liebig-Weg 3, D-37077 Göttingen, Germany
14 Laboratoire d’Astrophysique de Bordeaux, Université de Bordeaux, CNRS, B18N, allée Georoy Saint-Hilaire, F-33615 Pessac,
15 Laboratoire Univers et Particules de Montpellier, Université Montpellier, Place Eugène Bataillon, CC72, F-34095 Montpellier
Cedex 05, France
16 Thales Services, 290 Allée du Lac, F-31670 Labège, France
17 Department of Astronomy, University of Geneva, Chemin d’Ecogia 16, CH-1290 Versoix, Switzerland
18 Centre National d’Études Spatiales, 18 Avenue Edouard Belin, F-31400 Toulouse, France
19 Laboratoire Lagrange, Université Nice Sophia-Antipolis, Observatoire de la Côte d’Azur, CNRS, CS 34229, F-06304 Nice Cedex,
20 Universiteit Antwerpen, Onderzoeksgroep Toegepaste Wiskunde, Middelheimlaan 1, B-2020 Antwerpen, Belgium
21 Institut d’Astrophysique et de Géophysique, Université de Liège, 19c, Allée du 6 Août, B-4000 Liège, Belgium
22 F.R.S.-FNRS, Rue d’Egmont 5, B-1000 Brussels, Belgium
23 CRAAG - Centre de Recherche en Astronomie, Astrophysique et Géophysique, Route de l’Observatoire, Bp 63 Bouzareah DZ-
16340 Algiers, Algeria
This paper presents the specification, design, and development of the Radial Velocity Spectrometer (RVS) on the European Space
Agency’s Gaia mission. Starting with the rationale for the full six dimensions of phase space in the dynamical modelling of the
Galaxy, the scientific goals and derived top-level instrument requirements are discussed, leading to a brief description of the initial
concepts for the instrument. The main part of the paper is a description of the flight RVS, considering the optical design, the focal
plane, the detection and acquisition chain, and the as-built performance drivers and critical technical areas. After presenting the pre-
launch performance predictions, the paper concludes with the post-launch developments and mitigation strategies, together with a
summary of the in-flight performance at the end of commissioning.
Key words. Space vehicles: instruments; Instrumentation: spectrographs; Surveys; Techniques: spectroscopic; Techniques: radial
1. Introduction
The Gaia satellite of the European Space Agency (ESA) was
launched on 2013 December 19, arriving at the L2 point a month
later, for a planned five-year mission after the commissioning,
which ended in 2014 July (the mission was extended for a fur-
ther 1.5 years in late 2017). Gaia was conceived as an astro-
metric satellite, extending by orders of magnitude in terms of
distance and accuracy the pioneering results from ESA’s Hip-
parcos satellite. The mission, a collaboration between ESA, in-
dustrial partners, and science institutes in ESA member states,
is described in Gaia Collaboration, Prusti et al. (2016). The first
Article number, page 1 of 19
A&A proofs:
manuscript no. 32763_Cropper
data release was made in 2016 September and is described in
Gaia Collaboration, Brown et al. (2016).
The science return from Hipparcos is very significant (see
Perryman (2009) for a comprehensive overview), but its pay-
load permitted only astrometric and photometric measurements.
Measurement of stellar positions over time produced proper
(transverse) motions and distances, but not a measure of the ve-
locity in the line of sight (the radial velocity). This was recog-
nised as a deficiency at the time; see for example Blaauw in
Torra et al. (1988). Perryman (2009) provides a comprehensive
overview of the proposals made in France and the UK and also
within ESA in the period 1980 – 1987 for new dedicated tele-
scopes and updated instrumentation. None of these was success-
ful. Emphasising that for many studies it is of the greatest im-
portance to have all three space velocities rather than only the
two components of the star’s velocity on the sky, Binney et al.
(1997) noted how few stars in the Hipparcos Input Catalog
(Turon et al. 1992) had radial velocities. A ground-based ESO
Large Programme was instigated to provide radial velocities of
the 60 000 stars in the Hipparcos Input Catalog with spec-
tral type later than F5, but progress was slow. The situation im-
proved only when Nordström et al. (2004) published good mea-
surements for 13 500 F and G dwarfs and Famaey et al. (2005)
published radial velocities for 5 952 K and 739 M giants, but this
was still a small fraction of the total catalogue1. This shortcom-
ing was therefore fully evident at the time when the early Gaia
concepts were being developed, and hence a spectrometer, the
Radial Velocity Spectrometer (RVS), was incorporated into the
payload to avoid such a science loss (Favata & Perryman 1995,
Beyond the radial velocities, this initiative also for the first
time enabled a spectroscopic survey of the entire sky to mea-
sure astrophysical parameters of point sources. Perryman (1995)
emphasised the scientific utility from acquisition of information
complementary to the astrometric measurements in the Gaia
(Lindegren & Perryman 1996) and Roemer (Høg & Lindegren
1994) concepts being developed at that time. In addition to pro-
viding full space motions, he identified the advantages to the
mission of multiple visits for identifying binary systems, and
correction of perspective accelerations, and also the wider bene-
fits of the large-scale determination of elemental chemical abun-
dances that would inform the star formation history and provide
chemical enrichment information, to parallel that from the kine-
matic measurements. The paper highlighted the scale of the task,
given the significant increase in kinematic data in the mission
While Perryman (1995) mainly considered ground-based so-
lutions using multi-fibre spectrographs, it was clear that a ded-
icated instrument in orbit would provide more complete and
uniform complementary information. The initial concept pre-
sented in Favata & Perryman (1995, 1997) was a slitless scan-
ning spectrograph called the Absolute Radial Velocities Instru-
ment (ARVI). This would provide a radial velocity precision of
10 km s1at a limiting magnitude of 17, with 1 km s1for
brighter magnitudes 10 12, to achieve a metallicity determina-
tion precision of 0.1 dex. The instrument would use a separate
optical system to that for the astrometry. Many of the critical as-
pects important in the long-term for the RVS were discussed in
the ARVI papers, including the limiting magnitude, resolution,
1Since that time, several large spectroscopic surveys for galactic sci-
ence have been undertaken, including RAVE (Steinmetz et al. 2006),
APOGEE (Majewski et al. 2017), ESO-Gaia (Gilmore et al. 2012),
LAMOST (Cui et al. 2012), and GALAH (Martell et al. 2017).
bandpass, scanning rate, telemetry- and attitude-control require-
ments, and the wavelength zero point.
Taken together with the photometric measurements planned
in the Gaia mission concept (to provide luminosities and temper-
atures, as well as photometric distances and ages), the change
in emphasis for this next generation of mission should not be
underestimated. Through the acquisition of both kinematic and
astrophysical data, Gaia was developed from an advanced as-
trometric satellite into a complete facility for the study of the
formation and evolution of the Galaxy.
This paper provides a brief overview of the RVS concept
and requirements (Sec. 3 and 4) and then describes the instru-
ment (Sec. 5 to 7) before examining the pre-launch performance
(Sec. 8). Post-launch developments, optimisations, and updated
performance predictions are summarised in Sec. 9 –11.
We distinguish in this paper between pre-launch instrument
parameters – for example as implemented at the critical design
review (CDR) or derived from the ground-based calibrations –
and those post-launch, after which they may have been optimised
for the in-orbit characteristics of the satellite during commis-
sioning. It should be kept in mind that the in-orbit performance
described from Sec. 9 onwards supersedes the pre-launch ex-
pectations, and also that the full end-to-end performance of the
RVS instrument is achieved in conjunction with the full Gaia
data-processing system, described in Sartoretti et al. (2018) and
Katz et al. (2018).
2. Early RVS concepts
Subsequent to Gaia’s adoption in 2000 October initiating the
major industrial activities, ESA in mid-2001 instigated working
groups for the scientific community to contribute to the develop-
ment of the mission. One of these was the RVS Working Group.
In the period 2001 2006, this group examined in detail the sci-
entific requirements of the instrument.
Fundamental considerations included the wavelength range
of the spectrometer, spectral resolution, and the limiting mag-
nitude. Because Gaia would operate in time-delay integration
(TDI) mode, in which the spectra would scan over the focal plane
at the same rate at which the CCD detectors were being read out,
RVS would necessarily be slitless. To minimise the background
light, the wavelength range should be as narrow as possible, con-
sistent with it containing sucient strong spectral lines to pro-
vide radial velocity information, as well as an adequate range
of chemical elements to provide astrophysical information (tem-
peratures, gravities, and metallicities). Spectral regions around
the Mg ii doublet at 440 nm and the Ca ii triplet at 850 nm
were examined, with a 25 nm region centred on the Ca ii triplet
preferred (this spectral domain was originally proposed by U.
Munari, and noted independently by R. Le Poole).
The balance between the radial velocity and astrophysical
information (requiring higher signal-to-noise ratios) for the set-
ting of the resolving power requirements between R=5000 and
20000 was explored, with R=11500 found to be optimal in
providing both adequate spectral resolution while maximising
the radial velocity performance, and taking into account other
factors as well, such as the telemetry budget.
Because the RVS bandpass is narrow compared to that of
the astrometric instrument, so that fewer photons are recorded,
and because the spectral dispersion distributes these over a larger
number of pixels, each of which has associated noise sources, the
limiting magnitude would necessarily be lower. It was therefore
important to consider the scientific drivers carefully and match
Article number, page 2 of 19
M. Cropper et al.: Gaia RVS
the radial velocity accuracy with that of the transverse veloci-
ties for the scientific scenarios of interest. These considerations
set requirements of 3 15 km s1for V=16.5 K-type gi-
ants (Wilkinson et al. 2005). These and other requirements were
consolidated for the spectroscopic requirements in the Mission
Requirements Document for the Implementation Phase of the
programme (Gaia Project Team 2005).
At the start of the implementation phase, the working groups
were shut down and the expertise was transferred to the Gaia
Data Processing and Analysis Consortium (DPAC).
Over the same period, ESA and some European agencies
funded an engineering study led by a consortium constituted
from science institutes, with the aim of complementing the work
of the industrial teams, who were concentrating mainly on the
astrometric performance of Gaia during this competitive ten-
dering phase. Based on the earlier Phase A activities, the pay-
load concept at that point contained a separate telescope (Spec-
tro) for the RVS and medium-band photometry (see for exam-
ple MMS Study Team 1999; Merat et al. 1999; Perryman et al.
2001; Safa et al. 2004). In this complementing study by the RVS
Consortium, the driving performance considerations for the in-
strument design were to maximise the radial velocity precision,
and to minimise the constraints imposed by the telemetry limita-
The signal levels were increased by maximising the field of
view and reducing the focal ratio consistent with optical dis-
tortion and spectral resolution, in order to reduce the scanning
speed over the detector and maximise the exposure duration. The
principal noise source was identified as that arising in the de-
tector from the readout (readout noise), so electron-multiplying
CCDs (also known as L3CCDs) were specified to reduce this to
a minimum. Although a generous fraction of the overall Gaia
telemetry was allocated to the RVS, the length of the spectra
imposed high data rates, and this, with the desirability of two-
dimensional information to separate overlapping spectra opti-
mally, led to a scheme in which data from the CCDs were com-
bined on board in order to remain within the budget. Perfor-
mance predictions and system margins were within budget.
The work during this period was reported in Katz
(2003), Munari et al. (2003), Cropper (2003), Katz (2005),
Cropper et al. (2005b), Cropper et al. (2005a), and especially in
Katz et al. (2004) and Wilkinson et al. (2005).
For the implementation phase in 2006, the selected prime
contractor Astrium (now Airbus Defence and Space) proposed
and implemented a dierent RVS instrument concept, in part us-
ing ideas in Cropper & Mason (2001).
3. Flight instrument concept
The flight RVS design departed from the earlier concepts dis-
cussed briefly in Sec. 2 above by removing the Spectro tele-
scope, and employing, instead, the telescopes for the astromet-
ric instrument. This was motivated by savings in mass, power
(and heat dissipation), improved payload module accommoda-
tion, and cost. The starlight is dispersed by a block of RVS op-
tics, which produces a spectrum that is approximately confocal
with the undispersed beams, and with the same focal ratio. The
optics block also defines the instrument bandpass and corrects
the o-axis characteristics of the beam. The RVS focal plane
is located in the same focal plane array as the astrometric in-
strument. Starlight enters the spectrometer after the astrometric
(and photometric) instruments during normal operations when
the satellite is scanning. There are 12 CCDs in the RVS focal
plane. In order to limit the size of the elements in the optics
block, only four rows of CCDs are employed, instead of seven
in the astrometric focal plane. The instrument uses the SkyMap-
per information from the astrometric field of view. The median
optical spectral resolving power of 10 400 was compliant with
the nominal requirement (Gaia Project Team 2005), and with a
sampling of 3 pixels per resolution element, the window was
1260 pixels long. This layout is shown schematically in Fig. 1.
The RVS is considered to consist of the RVS optics block,
the focal plane, and the dedicated software to place windows
on the focal plane. The instrument and its operation with other
payload elements including the SkyMappers is described briefly
in Gaia Collaboration, Prusti et al. (2016).
Fig. 1. Layout of the optical beams after the beam combiner from the
two telescopes, and the focal plane in Gaia. The scan direction is from
right to left. BAM is the Basic Angle Monitor, and WFS is the Wave-
front Sensor.
The flight design therefore benefits from the larger light
grasp of the telescopes feeding the astrometric instrument, and
with two telescopes, the doubling of the number of observations
of each object, as well as from the removal of an entire optical
system with its separate star trackers. However, the integration
time per CCD is limited to the same as that of the astrometric in-
strument, 4.4 s, which significantly reduces the exposure levels
with respect to the earlier concept (in which the focal ratio was
shorter and the field of view larger), and the smaller number of
rows of CCDs in the RVS focal plane reduces the number of ob-
servations per object. With respect to the performance expected
in the earlier concepts discussed in Katz et al. (2004), and from
a dierent perspective, in Cropper et al. (2004, 2005a), the pro-
jected pre-launch limiting magnitude of the flight design was 1
magnitude poorer owing to the shorter exposure times from the
telescopes, and conventional CCDs, rather than L3CCDs in the
focal plane, with implications for the science case discussed in
Wilkinson et al. (2005). On the other hand, with the experience
of processing in-orbit RVS data, the longer focal length arising
from the use of the same telescope as that for the astrometric in-
strument significantly reduces the spectral overlapping, enhanc-
ing the performance when one or both telescopes scan crowded
regions of the sky.
4. Requirements
Before describing the instrument in more detail, we identify in
Tab. 1 the RVS-specific top level requirements guiding its de-
Article number, page 3 of 19
A&A proofs:
manuscript no. 32763_Cropper
sign. These are extracted from the ESA Mission Requirements
Document (Colangelo 2010), revised to take into account the de-
velopments in Sec. 3 and hence applicable to the as-implemented
instrument. The methodology to be applied to the radial velocity
predictions was specified in de Bruijne et al. (2005a).
The context for some of these requirements is elaborated in
the following subsections. Because of higher scattered light lev-
els encountered in orbit (Sec. 9), some of the considerations dis-
cussed below required reassessment during the commissioning
phase, as described in Sec. 10 and 11.
4.1. Limiting magnitude and wavelength range
The RVS is an atypical spectrometer in that it is slitless while
providing medium resolving power, with constrained exposure
durations. Consequently, at intermediate and faint magnitudes, it
is photon starved and noise dominated. For its role in providing
radial velocities, information from the entire spectrum is con-
densed into a single velocity value through a cross-correlation,
with the radial velocity signature at the faint end emerging only
after adding many transits of the object during the survey. Even
at the end of the mission, the spectra of most stars will be noise
dominated, and will produce only a radial velocity. Simulations
(Katz et al. 2004) showed that final signal-to-noise ratios of 1
per spectral resolution element would nevertheless provide suf-
ficient end-of-mission radial velocity precision. Hence, regard-
less of the instrument design, at its limiting magnitude and after
the noise was minimised, the RVS measurements would have on
average <1 epix1per exposure. To preserve this signal on
average in the presence of noise, it is also essential to provide
sucient levels of digitisation in the detection chain, preferably
to 0.5 e.
At the faint end, the low fluxes in each pixel render the Pois-
son noise from the source negligible compared to the other noise
sources. Narrowing of the bandpass reduces the cosmic back-
ground, which at 850 nm is mostly zodiacal, but also reduces
the kinematic and astrophysical information in each spectrum.
Spanning the Ca ii triplet requires 25 nm. Loss of one of the
triplet lines would substantially reduce the radial velocity sig-
nature, while on the other hand, widening the bandpass mod-
estly would not incorporate strong new spectral lines, would in-
crease the sky background, and would also increase the fraction
of overlapped spectra. Consequently, the required bandpass was
set from 847 to 874 nm. The flux collected in this bandpass is
assigned a Gaia magnitude, GRVS (see Jordi 2014).
The instrumental noise includes components from scattered
light and secondary spectral orders, and from the detectors (read-
out, fixed pattern, and dark noise). Suppression of secondary
spectral orders places requirements on the rejection levels for
wavelengths outside of the bandpass, which drives the optical
coating technology in the instrument. Scattered light may arise
within the instrument itself or within the telescope and payload
module. Before launch, the noise from the scattered light back-
ground was not expected to be a dominant eect, but this was
found not to be the case post-launch. This is discussed in Sec. 9.
In respect of detector noise, the signal at any wavelength is
accumulated over a full column in the CCD in TDI operation,
so the already small photo-response non-uniformity (PRNU) re-
sponsible for the fixed pattern variation requires almost no flat-
fielding correction and is negligible. At the 160K temperatures
envisaged for the CCDs in Gaia, the dark noise is also negligible.
In typical astronomical operation with slow readout rates, CCDs
and their associated video chains can reach readout noise 5 e,
but as the value must be squared, even a readout noise somewhat
below this level will exceed the signal at the faint end by an or-
der of magnitude or more. This is the major noise source in the
instrument, and because it will exceed the cosmic background
by a large factor for almost any instrumental configuration, it is
one of the most stringent drivers of performance.
4.2. Spectral resolving power
For measuring elemental chemical abundances and surface grav-
ity, the spectral lines should be adequately sampled even in the
line cores, and this favours higher spectral resolving power for
brighter stars where the signal-to-noise ratio is sucient for
these measurements to be useful. The spectral resolving power
sets the length of the spectrum, as higher resolving power pro-
duces longer spectra. Longer spectra result in a higher degree
of problematic overlapping and a larger telemetry requirement.
Given the dominance of readout noise in faint spectra, there
was a strong imperative to pack the spectrum into as few pix-
els as possible, hence lower resolving powers are favoured for
the faintest objects; this also reduces the telemetry. To provide
optimal performance over the full magnitude range, a require-
ment was introduced to provide two spectral resolving powers,
high resolution (HR), at the full optical resolution of the spec-
trograph, and low resolution (LR), a resolution-degraded mode
for stars with GRVS >10 in which pixels could be summed at the
detector readout node to reduce the readout noise per sample.
The requirements on the detector readout noise as derived from
the limiting magnitude performance was set at 4.0 efor LR
mode and 6.0 efor HR.
Gaia is required to scan the sky as uniformly as possible, and
to measure the angular distances between stars in the astrometric
fields of view at a range of orientations. The adopted scanning
law is one of forced precession (Gaia Collaboration, Prusti et al.
2016), which results in a sideways (across-scan, AC) displace-
ment (with a sinusoidal dependence) of the star on the spin pe-
riod of the satellite. In order to maintain the spectral resolv-
ing power, the orientation of the dispersion direction was set
along the scanning direction (AL) so that spectral lines were not
broadened by this AC displacement (Cropper & Mason 2001;
Katz et al. 2004). The (lesser) consequence was that the spectra
were broadened with a spatial distribution at a period half that of
the spin period, leading to a variation in signal-to-noise ratio on
the same period.
4.3. Radiation damage
As noted in Sec. 4.1 above, at its limiting magnitude, the RVS in-
strument concept will work with <1 epix 1in each exposure
to achieve its end-of-mission radial velocity precision. Preserv-
ing single electrons in the large number of transfers in the Gaia
CCDs, especially in TDI operation, was unproven, even with-
out in-orbit radiation damage. Trapping sites caused by in-orbit
radiation damage renders preservation of single electrons even
more problematic. Even for brighter objects, electrons released
from traps after a characteristic time delay leave trails because
in TDI operation, the sky moves on from the pixel in which the
electron was trapped. With the orientations discussed in Sec. 4.2,
the trails distort the shape of the spectral lines, inducing a radial
velocity shift and diminishing their contrast. When the charge
recorded in a given pixel reaches the readout register, it encoun-
ters further trapping as it is clocked to the readout node, and
this will distort the spatial profile of the spectrum. The eective-
ness of the trapping is non-linearly dependent on the quantity of
Article number, page 4 of 19
M. Cropper et al.: Gaia RVS
Table 1. Top-level RVS-specific requirements. Additional requirements include the capability for operation in HR and LR mode (see text), at least
Nyquist spectral sampling for HR mode, control over flux rejection levels outside the RVS bandpass, and straylight requirements applicable to the
payload as a whole. Spectral types follow the standard terminology, so that temperature decreases from B to K stars, V in the spectral identifier
denotes dwarfs, and IIIMP denotes metal poor giants. From Colangelo (2010).
Average number of transits over mission 40 for objects V<15
Wavelength range 847 874 nm
Spectral resolving power (HR mode) average 10 500 12 500; 90% 10 000; max 13 500
Spatial resolution 1.8 arcsec to include 90% of flux
Maximum stellar density 36 000 objects degree2
Maximum apparent brightness all spectral types V<6
Minimum apparent brightness B1V V>13
G2V V>17
Radial velocity systematic error after calibration 300 m s1at end of mission
Radial velocity precision of 1 km s1B1V V7
G2V V13
K1III MP V13.5
Radial velocity precision of 15 km s1B1V V12
G2V V16.5
charge being transferred, and on the quantity of charge in pre-
ceding pixels, which will themselves be de-trapping, and hence
on the spectral energy distribution (in the TDI direction) and the
phase of the AC displacement discussed in Sec. 4.2. Each across-
scan element in the spectrum will have a dierent eect, with
less impact on the central brightest elements, and more on the
fainter wings of the spectrum. Minimising, understanding, and
modelling the in-orbit radiation damage was therefore a driving
factor in the RVS design, and additionally, in the RVS data pro-
5. RVS optical design
This section describes the optical design of the combined flight
telescope and the RVS. A full report at the critical design review
stage is in Boyadjian et al. (2010).
5.1. Optical system
The two Gaia telescopes are o-axis three-mirror anastigmats,
with rectangular primary mirrors of 1450 ×500 mm2and a focal
length of 35 000 mm. Their beams are combined at the exit pupil
by beam-combining mirrors, and folded by two flat mirrors to fit
the long focal length into the payload module. They then pass
through the RVS optics (Figs. 1 – 4) to the RVS detector array
(4 ×3 CCDs) (Fig. 5), whose centre is 0.75displaced from
the optical centre of the focal plane in the along-scan direction,
with each telescope producing images displaced by ±0.06in the
across-scan direction (in common with the astrometric field and
the photometers). The rays on the detectors are not telecentric,
and the focal plane is tilted by 7in the focus direction com-
pared to that of the astrometric field. In addition to dispersing the
light at moderate spectral resolving power in (slowly) converg-
ing beams, the RVS optics must also correct optical aberrations
from the o-axis field angles without changing the focal length
significantly, and define the instrument bandpass.
To achieve this, the RVS optics consist of a filter, two plane
prisms, two prismatic lenses ( i.e. lenses cut o-axis) and a
diraction grating between the prisms. The arrangement is ev-
ident within the optical path in Fig. 2 and for the module itself in
Fig 3. The six fused-silica elements are held by thin Invar bipods
within a C-section silicon carbide (SiC) structure that provides
sucient rigidity while allowing access for integration. The ther-
mal behaviour of fused-silica is similar to that of SiC, so that the
design is almost athermal. Stress relief on the optics is achieved
by bonding the bipods to small protrusions on the perimeter of
the optical elements. The alignment tolerances for the RVS mod-
ule within the telescope beams is 0.11 mm and 0.31 milli-
Fig. 2. RVS optical path for one telescope. The primary mirror is at the
lower left, and the RVS focal plane at the upper right, with the RVS
optics immediately preceding it in the light path. Below the RVS optics
is one half of the beam-combining mirror.
5.2. Filter
The RVS nominal 847 874 nm bandpass is defined mainly by
the multi-layer filter on a fused-silica substrate, modified by the
detector quantum eciency and (slightly) by the telescope trans-
mittance. Attention was paid to the out-of-band rejection in order
to minimise the background light in this slitless instrument, es-
pecially considering the wide wavelength range over which the
detectors are sensitive compared to that of the desired bandpass.
Rays from dierent field-of-view points and pupil points pass
Article number, page 5 of 19
A&A proofs:
manuscript no. 32763_Cropper
Fig. 3. (Top) Expanded view of the RVS optical module. Light enters
from the left, initially encountering the bandpass filter. (Bottom) The
RVS optics as realised. The diraction grating is the fourth element
from the left.
through the filter at dierent locations and at dierent angles, so
that there is a slightly dierent bandpass for each CCD and for
the two telescopes.
5.3. Grating
The RVS grating (Erdmann et al. 2010) is an advanced element
using the binary index modulation principle to meet the require-
ments to work in first order with a high eciency (70%) and
low polarisation sensitivity (7%) while complying with the
tight limits on the additional wavefront error it can contribute.
In the context of Gaia, such gratings provide superior eciency
and reduced unwanted orders. Instead of triangular line profiles
produced by ruling the substrate (as in conventional gratings) or
variations in refractive index (as in volume phase holographic
gratings), subwavelength scale repeats of variable width lines
and columns (all of which are at the same height) are used to ap-
proximate the grating pattern (Fig. 4). These can be ion-etched
onto a fused-silica substrate and allow a high level of control of
the grating properties.
5.4. Baffles
Scattered light can arise from solar system sources, bright stars,
and the Galaxy as a whole, and from internal sources. To shadow
the telescope apertures from the Sun (particularly), Earth, and
Moon (the principal external sources), the Gaia sunshield is a
double-layered circular structure with a diameter of 10.2m. The
Fig. 4. (left) Full-scale grating demonstrator. (right) A scanning elec-
tron micrograph of the binary elements in the grating used to produce
the modulation pattern. Two repeats of line and column elements of de-
creasing width are shown here.
second layer is slightly smaller so that it intercepts the diracted
flux from the perimeter of the sunward layer. Within the payload
module, the arrangement of the folded telescope beams limits
the opportunities for baing and requires unconventional tech-
niques. Scattered light at the RVS optical module is suppressed
by diaphragms, one at the entrance and another associated with
the grating, and by an exit bae (Fig. 3). This bae, and further
baes on the focal plane structure, also serve to prevent light
from the RVS (especially unwanted orders) falling on the astro-
metric and photometric focal planes (Fig. 5). Solar system and
stellar sources not in the shadow of the sunshield are not fully
The main source of internally generated scattered light is the
basic angle monitor (BAM), which measures the angle between
the two telescopes. This uses laser light at 850nm (i.e. within
the RVS bandpass) chosen on the grounds of laser power and
stability. Some of the major scattering paths from this source
cannot be baed and were partially mitigated by neutral density
filtering at the lasers.
6. Focal plane
The Gaia focal plane, Fig. 5 (Gaia Collaboration, Prusti et al.
2016; Crowley et al. 2016), consists of 106 CCDs, distributed
across astrometric, photometric, and RVS instruments, as well
as SkyMappers, wavefront sensors, and the BAM detectors, all
supported on a SiC structure with baes and thermal control.
The RVS focal plane of four rows and three strips of CCDs is
the last to be reached in the scan direction, occupying Rows
47, Strips 15 17 (Fig. 6). It is slightly displaced in the
across-scan direction with respect to the rows in the astrometric
focal plane. Each CCD has an associated proximity electronics
module (PEM) behind it, which in turn interfaces to an inter-
connection module and a video processing unit (VPU) on each
row that service the detectors of all instruments on the row. The
RVS PEMs are the same as those for the other instruments, but
operate in RVS-specific modes to support the RVS operation.
The detector structure is passively cooled by the payload mod-
ule environment, and is isolated from an ambient-temperature
structure holding the PEMs and interconnection module by ther-
mal screens. The heat from the warm part of the focal plane is
ejected to space by a radiator.
The detectors in the focal plane detector array are e2v Tech-
nologies CCD91-72, each with 4500 ×1966 rectangular pix-
els of dimension 10 ×30 µm along- and across-scan. In or-
der to maximise the sensitivity at the wavelengths of the RVS
band, the 40 µm deep-depletion variant with red-enhanced anti-
reflective coating is mounted in the RVS focal plane (it is also
Article number, page 6 of 19
M. Cropper et al.: Gaia RVS
Fig. 5. (Top) Gaia focal plane during assembly. The 14 SkyMapper
CCDs are the first two full vertical strips on the left of the SiC structure
and the 12 RVS CCDs on the right (the fifth row of white rectangles
are not CCDs but are used in thermal control). (Bottom) The integrated
focal plane showing the baes and radiator (extended structure on each
side). Again, the RVS CCDs are on the right.
used in the Red Photometer and the BAM). In pixel structure
and readout node, all Gaia CCD91-72 variants are the same.
The RVS CCDs therefore include charge-injection lines that per-
mit an electrical injection of charge into lines in the image area.
These are not used in standard RVS observational sequences as
the lines interfere too commonly with the long spectral win-
dows (see Sec. 7) (100×longer than the astrometric field win-
dows). The CCD91-72 pixel structure contains a supplementary
buried channel that is used to enhance the charge transfer e-
ciency of electrons during the TDI at low flux levels; these elec-
trons are particularly important in the RVS case. It is not clear
(Seabroke et al. 2013) whether these structures were correctly
implemented in the CCDs selected for flight.
As (at the time) the readout noise was expected to be the
dominant noise source, a very significant eort was undertaken
to minimise this in the detection chain. Because of the sub-e
signal levels in the majority of RVS pixels, a fine level of digi-
tisation in the PEM was adopted, 0.55 eper digital unit,
which also has the benefit of reducing the digitisation noise sig-
nificantly. A consequence is that very bright spectra saturate at
the level of the analog-digital converter before saturation lev-
els are reached on the CCDs themselves. Consequently, while
available, the gate structure within the CCDs enabling greater
dynamic range in the astrometric field and the photometers is
not generally used in the RVS, except for some calibrations.
Fig. 6. Nomenclature for the CCDs in the RVS focal plane.
7. Detection and acquisition chain
7.1. Window scheme
Given the telemetry bandwidth, the large focal plane, and the
short eective exposure duration for all CCDs, it is not possi-
ble to transmit to ground all of the sky data. Together with in-
formation from the Red Photometer (Fig. 1), stars within the
RVS magnitude range are selected, and pixels containing their
spectral information are identified by the VPUs to set a window
mode in the CCD readout. Pixels within these windows are read
out normally and stored in the spacecraft mass memory. Other
pixels are generally discarded, except those within virtual object
(VO) windows (described below).
To accommodate the 25 nm of spectrum, the window in the
spectral direction was set at 1260 pixels (12.6 mm) long on the
CCD. During readout, the selected pixels can be summed on the
detector by 3 in the spectral direction to reduce the telemetry,
The un-summed and summed modes are termed high resolu-
tion (HR) and low resolution (LR)2, respectively. In addition,
to reduce the telemetry further, pixels are summed on the detec-
tor in the spatial direction to produce one-dimensional spectra
for stars below a certain RVS flux limit. Nominally, the width
of the window in the spatial dimension is 10 pixels (300µm).
The windowing is then termed Class 0 for two-dimensional win-
dows containing HR spectra of bright stars for GRVS 7, Class
1 for one-dimensional windows containing HR spectra for stars
7<GRVS 10, and Class 2 for one-dimensional LR spectra
fainter than GRVS =10. Because whole lines are transferred into
the readout register during the TDI operation, if any star is su-
ciently bright to be observed in HR, the spectra of other fainter
stars in the same TDI lines would also be observed in HR even
though they might normally be observed in LR. In order to re-
duce the telemetry to ground, such spectra are summed digitally
in the VPU to produce LR spectra (but with the increased read-
out noise from the readout of three individual lines rather than
a single summed line). To complete this organisation, the spec-
tra are divided into 12 subunits in the along-scan direction, called
2This has been modified post-commissioning so that almost all data
have been taken in HR; see Sec. 10.
Article number, page 7 of 19
A&A proofs:
manuscript no. 32763_Cropper
macrosamples, to facilitate the handling of overlapping windows
(see Sec. 7.2 below).
In addition to the reduction in the telemetry bandwidth asso-
ciated with this windowing scheme, the summation of pixels at
the detector (10 for HR and 30 for LR) avoids a readout noise
contribution for every pixel. This is a critical strategy in min-
imising what was expected to be the principal noise source in
the instrument. A further measure associated with the window-
ing is to limit the number of single or summed pixels, referred
to as samples, read in the serial register (the across scan direc-
tion) to a value of 72, which is consistent with being able to
meet the maximum source density requirement of 36 0 00 sources
degree2(Tab. 1). As the unwanted pixels are the large majority,
they can be flushed at a higher rate, and most of the parallel line
transfer period is available to read the desired samples, thus min-
imising the readout noise. In order to maintain the thermal sta-
bility of the detection chain, 72 samples are always read (unused
ones in the overscan region) and discarded if empty.
7.2. Window truncation
The 1.3 arcmin length of the windows in the spectral direction
leads to overlapping of some spectra even in modestly crowded
regions (whether spectra overlap also depends on the orientation
of the satellite in its scanning, so that for some passes, no over-
lapping may occur). Because pixels can be selected only once
at the readout node, they must be assigned to one or the other
of the overlapped spectra. For Class 1 and 2 windows, the over-
lapping spectra are equally split in the spatial direction between
the two windows, and so the eective width to be summed into
the one-dimensional spectrum of one of these overlapped win-
dows is between half and the full nominal width, depending on
the separation between the sources in the across-scan direction
(Fig. 7). Class 0 (i.e. two-dimensional) windows are assigned at
the nominal width and hence take priority over the Class 1 and 2
windows. In the rarer cases of overlapping Class 0 windows, the
pixel values are duplicated in each window by the VPU.
In the spectral direction, overlapped spectra were dealt with
in a hierarchy of overlaps with the window positions adjusted
on a coarse grid (in the spectral direction) of 105 pixels, the
macrosamples mentioned above, so that overlaps can start and
end only at macrosample boundaries. (Spectra from dierent ob-
servations are consequently placed slightly dierently with re-
spect to the window boundaries.) Unless they are aligned in the
scan (spectral) direction, Class 1 and 2 overlapped windows will
include some macrosamples of unblended spectra constructed of
nominal width, spatial width, and some of blended spectra, con-
structed of reduced spatial width, as indicated above. Cases of
triple and higher order overlaps are treated similarly. In more
crowded regions, the spatial width allocated to a source in a one-
dimensional window may change several times from macrosam-
ple to macrosample.
Information from the VPU about the window truncation and
spectral overlapping is also recorded and telemetered for the data
7.3. Calibration faint stars
A subset of stars that would normally be assigned Class 1 and
2 windows because of their magnitude are telemetered with full
two-dimensional information in Class 0 windows. These are the
calibration faint stars which are used to quantify and calibrate the
eects arising from the collapsing on the CCD detectors of the
two-dimensional information to one-dimensional in Class 1 and
2 windows for these fainter objects, and in particular, the spatial
profile of the spectrum. The information in the spectra from these
stars, however, suers from larger readout noise because pixels
are no longer binned before readout.
7.4. Virtual objects
While the windows for celestial sources are generated au-
tonomously by the SkyMappers, it is possible to assign win-
dows by command, although they cannot be assigned explicitly
to positions on the sky (this can, however, be achieved by timed
commands in conjunction with the Gaia scanning law). These
virtual object (VO) windows are used for background monitor-
ing and calibrations, and are of two types: special VOs specify
window pattern sequences to be run outside of observations, for
example during orbit maintenance periods; routine VOs are used
during routine observations, and these windows are propagated
from the SkyMappers to the RVS focal plane position as with
normal windows.
The VO windows can be of all Class 0, 1, or 2, and may
accidentally contain sources. The overlap scheme of VOs with
other windows follows the standard prioritisation above, so a re-
quested VO of Class 0 may override a normal object window of
Class 1. VO patterns are generated for the focal plane as a whole,
but parameters are available specifically for the RVS to enhance
their suitability to its characteristics, in particular its larger win-
7.5. Data priority scheme
Data from the Gaia instruments are stored in the onboard mass
memory unit in file structures that have associated priority levels
for telemetry to the ground, and approximately inversely, for data
deletion (Ecale & Chassat 2010). Brighter objects have higher
telemetry priorities, so that in the RVS, windows of Class 0 have
the highest and Class 2 the lowest priority. The telemetry prior-
ities of the astrometric, photometric, and spectroscopic data are
arranged in an interleaved manner by magnitude. VOs and cali-
bration faint stars have a dierent telemetry priority, higher than
any of the RVS normal object windows, with the VO priority the
higher of the two. This prioritisation takes eect when Gaia is
accumulating more data than the mass memory unit can retain,
generally when the satellite is scanning along the Galactic Plane.
Deleted data are lost, with the consequence that fainter sources
may not be recorded in the archive at every transit. This leads
to a reduced radial velocity performance at the faint end of the
magnitude range.
8. Pre-launch performance predictions
The essential characteristics and performance of Gaia including
the RVS are recorded for several epochs pre- and post-launch in
the Gaia Parameter Database (de Bruijne et al. 2005b). This is
the reference repository for the instrument.
8.1. Instrument throughput and bandpass
Table 2 identifies the predicted overall pre-launch contributions
at CDR to the RVS throughput, including the full optical chain
and detectors. Owing to the excellent transmission of both the
bandpass filter and the grating, this ranged between 0.40 and
0.47. The slope in the response results mainly from the decreas-
Article number, page 8 of 19
M. Cropper et al.: Gaia RVS
Fig. 7. Window overlapping in RVS for Class 1 and 2 windows as implemented pre-launch. The macrosamples are numbered along the bottom, so
that there are 12 macrosamples per spectrum (yellow and orange). The vertical axis is an example pixel number in the spatial direction. Spectral
overlap occurs over macrosamples 6–12, and the window width is apportioned equally (to the nearest integer) over this region. As they are Class
1 and 2 windows, they will be collapsed in the spatial direction at the readout node of the CCDs. In the overlap region, the track of the peak of the
spectrum is shown as dots, while outside of it, the dots have no meaning.
ing detector quantum eciency from 76% at 847nm to 65% at
874nm. The instrument was predicted to detect 1 epix1at
V=15.1 giving a zero point (for which 1 es1is detected in
the full bandpass) at V=21.3. Figure 8 shows the slightly lower
throughput measured for the flight model.
During flight model manufacture, it was found that spatial
mid-frequency errors in the multilayers of the bandpass filter
caused unacceptable performance degradation, and the filter was
remade. However, for this element, the bandpass cutos were
2 nm blueward, also resulting in a slightly narrower-than-
specified instrument bandpass. This resulted in the astrophysi-
cally important Mg ispectral line at 873.6 nm falling on the edge
of the bandpass, degrading the accuracy with which it could be
The wavelength-integrated rejection levels outside of the
bandpass are required to be 10%, and this is met for longer
wavelengths, but marginally exceeded at shorter wavelengths at
some field points, largely because of leakage at wavelengths be-
tween 700 750 nm.
Table 2. Bandpass-averaged optical transmission, followed by the total
photon detection fraction, including both optics and CCDs, for wave-
lengths at the extremes of the RVS bandpass. From Fusero & Chassat
Optical Per element Surfaces Transmission
Mirrors 0.97 6 0.83
Bandpass filter 0.95
Grating 0.80
Prisms & lenses 0.98 8 0.85
Contamination 0.93
Microroughness 0.99
Total optical 0.62
Wavelength Optical CCD TOTAL
847 nm 0.62 0.76 0.47
874 nm 0.62 0.65 0.40
The pre-launch colour conversion between the standard
Johnson-Cousins photometric Vband and GRVS is given in Tab. 3
(Jordi 2014). This table also lists the conversions for the spectral
types used in setting the performance requirements in Tab. 1.
8.2. Spectral resolving power
Figure 9 shows the average spectral resolving power per CCD
for the bandpass centre value and extremes, as well as the distri-
bution of resolving power. This is compliant with the permitted
maximum and minimum levels in Tab. 1: 9% of points are at a
resolving power of <10 000. The median is 11 500. While av-
Fig. 8. RVS flight model bandpass showing the absolute transmittance
with the specified bandpass in vertical red lines (top) for the full optical
chain and detectors, and on a log scale, the out-of-band rejection (bot-
tom). The dierent colours distinguish between the dierent field points
for both telescopes, and hence the slightly dierent angles at which the
rays traverse the filter coatings. From Chassat (2013).
erage spectral resolution values for some CCDs lie above the
10 500 12 500 range, the average is also compliant in total.
The dispersion law can be modelled by a quadratic polyno-
mial in wavelength (Boyadjian 2008), with the linear term being
0.0245 nm pixel1. The quadratic term is small, and there is a
slight spatial variation over the RVS focal plane.
Article number, page 9 of 19
A&A proofs:
manuscript no. 32763_Cropper
Table 3. Pre-launch conversion between Vband and GRVS (Jordi 2014)
and for the spectral types in Tab. 1 used by Fusero & Chassat (2011).
GRVS =V0.06 1.10(VI)
B1V: GRVS =V+0.18
G2V: GRVS =V0.87
Fig. 9. (Top) Average spectral resolving power for wavelengths at the
extremes and centre of the RVS bandpass for each of the 4 CCDs in each
strip and for each telescope. (Bottom) Distribution of optical resolving
power calculated over a grid of 13 000 field points and wavelengths.
From Fusero & Chassat (2011).
8.3. Distortion
Because Gaia operates in a scanning mode, optical distortion
reduces the spectral resolving power and broadens the spatial
profile. The average distortion (for both telescopes) displaces the
light rays as they traverse the CCDs by 0.27 pixels along scan
and 0.11 pixels across scan.
8.4. Noise performance
The principle contributors to the noise in RVS spectra (in addi-
tion to the Poisson noise of the spectra themselves) as expected
pre-launch were the CCD readout noise, the external cosmic
background, and the leakage from the BAM lasers that oper ate in
the RVS band. An average readout noise of 3.7efor the full de-
tection chain was measured at the payload module thermal vac-
uum testing, the most significant noise source when considered
against the Poisson noise of the expected background flux of 0.6
eand BAM laser leakage of 0.3 e.
8.5. Bias non-uniformity
During the flight model testing programme, it was found that
the electronic bias levels (the electronic signal corresponding to
zero optical flux) in the PEMs were not constant, but varied in
response to perturbations in the readout pattern in the serial reg-
ister. One of the perturbations is the transition between the rapid
flushing of unwanted pixels at MHz rates, and the slower reading
of desired pixels at kHz rates; another is the pause in reading out
the serial register at the time at which the parallel phases of the
CCD are clocked for the TDI operation. There are four of these
per pixel and hence four ‘glitches’ per serial readout. With these
perturbations, the bias level drops sharply before recovering.
The eect on Gaia as a whole is discussed fully in
Hambly et al. (2018), but it is discussed briefly here because
of its particular impact for the RVS. It arises for two reasons:
firstly in order to achieve a fine digitisation of the signal of
0.55 e, the overall gain of the detection chain is unusually
high, a factor 6 higher than elsewhere in the payload; and,
secondly, excursions of 70 e(in the case of the flushes) can
exceed the typical signal levels by two orders of magnitude.
Allende Prieto & Cropper (2009) analysed the impact of the bias
non-uniformity on the radial velocity performance, and also on
the fidelity of recovery of astrophysical parameters from the
spectra. They found that the increase in radial velocity error,
even after the application of a simple correction proposed by
Fusero (2009) in which the boundaries of macrosamples were
aligned (because the flushing patterns remain constant within a
macrosample) was in excess of 10% for stars at the faint limit.
The impact of the spectral line distortions on the derivation of
astrophysical parameters was more significant, with variations
dependent on the spectral type and radial velocity of the star.
Fig. 10. Example of the calibration (red lines) of the bias non-
uniformity (black dots) after a flush. Three groups of datapoints are
evident: the lowest, corresponding to the maximum bias excursion, is
the first pixel after the transition from flush to read; the middle and
upper are the second and third pixels, respectively. The excursion de-
pends strongly on how many flushes precede it (the horizontal axis),
with the eect saturating after 200 flushed pixels. The upwards ex-
cursion above the calibration line after 400 pixels is, in this particular
case, the glitch caused by the pause in the serial readout.
Article number, page 10 of 19
M. Cropper et al.: Gaia RVS
While steps were taken to improve the stability of the
hardware, options were limited on programmatic grounds. The
adopted approach was to calibrate the eect, and an extensive
campaign of laboratory measurements was executed. Figure 10
shows an example of the excursion after a flush, and it includes
the eect of a glitch. It is evident that the time constant for re-
covery from the flushes is 1 2 serial clock periods. While the
glitches occur always at the same points for each TDI advance,
flushes occur randomly relative to these, depending on the place-
ment of the source windows and VOs. Analytic models were
found to be eective predictors of behaviour, with the distribu-
tion of the residuals after correction at the 4–5eFWHM
level. These required parameters to be specific to each CCD-
PEM pair and operating mode (HR, LR). The stability of the
parameters in time and amplitude were not established in this
campaign, which remained a concern until launch.
8.6. Radiation damage
Ions in the solar wind and from sources in the wider Universe
that impact on the detectors cause displacement damage, or
"traps", in the Si lattice. These can impede the transfer of elec-
trons from pixel to pixel in both the image area (parallel) and
the readout register (serial). As an electron trapped in a dam-
age site is released only after some time, its position on the sky
will be recorded incorrectly, the scanning and readout having
proceeded. This radiation damage has significant eects for the
RVS: electrons in traps with long release times may be released
only after the spectrum has entirely passed the trapping pixel,
reducing the total counts in the spectrum; traps with interme-
diate release times modify the spectral energy distribution, in
particular by removing the flux at the leading edge of the spec-
trum; and the traps with shorter release times modify the spectral
line shapes and reduce the line amplitudes and equivalent widths
(Allende Prieto 2009). The consequence is shown in Fig. 11,
where it is also evident that the damage eects are relatively
greater for fainter flux levels. The predicted impact on the radial
velocity calibration, the equivalent width, and the charge loss is
shown in Fig. 12.
Fig. 11. Radiation damage eects to spectra as measured from labora-
tory testing using a spectral mask approximating a G2V star for flux
levels corresponding to magnitude V=10 (left) and V=15.7(right).
Spectra shown in black are the reference with no radiation damage,
and those shown in red are the spectra after a radiation fluence of 109
p+cm2(10 MeV equivalent energy), the calculated fluence at end-of-
mission for a launch date of 2012 December (Fusero & Chassat 2011).
Noise levels are not equivalent in these plots, as a dierent number of
individual spectra are combined in the two cases.
Fig. 12. Eect of radiation damage on RVS outputs as a function of
magnitude (GRVS V0.87 for G2V stars) for a fluence of 109p+
cm2(10 MeV equivalent). (Top) Trapping and delayed release biases
the shape of the lines, so that the velocity cross-correlation can be af-
fected by 1.5 pix (13 km s1) for the faintest stars. (Centre) Spectral
lines are filled in by the released traps, reducing the equivalent width
by 30%, while (bottom) a total charge fraction exceeding 10% is lost
from the spectrum.
A substantial test and characterisation programme was car-
ried out under the auspices of the Gaia Radiation Calibration
Working Group and the Gaia DPAC in order to achieve an un-
derstanding of the radiation-induced eects. Models were de-
veloped to counter it, especially the charge distortion model de-
scribed in Short et al. (2013), which was suciently computa-
tionally fast to be used in forward-modelling approaches. These
investigations demonstrated conclusively that for the expected
levels of radiation damage, single photoelectrons would survive
the 4500 line transfers to reach the readout register. On the other
hand, the low backgrounds expected in Gaia, and especially in
the RVS, were found to create particular susceptibility. Low-
intensity diuse optical background sources to flood the RVS
focal plane were considered, but the consequent photon Pois-
son noise outweighed the positive eects of the improved charge
transfer. Additionally, although the CCDs and PEMs were con-
figured to permit an electrical injection of charge into the image-
area pixels that the TDI would sweep through the image area,
filling traps, given the frequency of these by comparison with
the length of the RVS spectra, they were not expected to be
used except in specific diagnostic circumstances. In the event,
the higher scattered light backgrounds encountered post launch
(Sec. 9) provide some amelioration of this aspect of the radiation
The eects in the readout register (serial) are complex, partly
because the charge is transferred both at MHz rates (flushing un-
wanted data) and then at kHz rates (for reading), and hence is
subject to traps with release times on both timescales. The MHz-
rate traps dominate. The across-scan line spread function width
increases with magnitude (so that it is worse for fainter stars) and
also increases with the number of transfers (the distance from the
readout node). In ground-testing, the red variant CCD91-72 was
Article number, page 11 of 19
A&A proofs:
manuscript no. 32763_Cropper
found to have more intrinsic traps (i.e. already present at manu-
facture and not induced by radiation damage) than the other Gaia
CCD variants. The eect of these serial register traps is shown in
Fig. 13, and it is significant for the flux levels expected in RVS
spectra, with flux loss from the window and asymmetries in the
profile that must be taken into account in overlapping windows
(Sec. 8.7).
Fig. 13. Red variant CCD91-72 across-scan line-spread function for
sources with 20 e(top) and 1.5 e(bottom) per HR sample (i.e. in-
tegrated over the line profile) corresponding to V=11.9 and 14.7,
respectively, as a function of number of serial register transfers. This
profile assumes the nominal optical performance, but no across-scan
broadening from distortion or the forced precession in the scan law.
From Chassat (2009).
8.7. Effects of window collapse to one dimension
The collapse of the two-dimensional to one-dimensional Class 1
and 2 windows at the detector level has little impact on spectra
that are not overlapped by other spectra. Appropriate modelling
of the spatial (across-scan) profile allows the fraction lost from
the window to be corrected, and the calibration faint stars pro-
vide information on this profile down to the limiting magnitude
of the instrument. However, overlapped spectra require special
treatment in the ground data processing (called ‘deblending’) to
assign the flux detected in each (truncated) window correctly to
each spectrum. The success with which this can be done will de-
pend on a number of factors, primarily the magnitudes of each
star, their separation, the amount of spatial (across-scan) broad-
ening resulting from the forced precession of the scanning law,
and the optical distortion at that field point. In addition, radiation
damage in the serial register changes the across-scan line spread
function as a function of radiation fluence and of source magni-
tude (Sec. 8.6), so this will need to be taken into account as the
mission progresses.
Pre-launch predictions of the fraction of spectral overlaps are
shown in Fig. 14. This indicated that the average probability of
overlap is 22% (including both telescopes). The number of mul-
tiple overlaps is non-negligible, and this requires further treat-
ment steps in the data processing. In each case, blended spec-
tra will be more noisy after separation because of the imperfect
deblending. However, because of the dierent path of the scan
at each observation of a source, spectra that overlap at some
epochs may not overlap at others, and it is possible to use the
non-overlapped spectra and information from the photometer to
assist in the deblending.
Fig. 14. Predicted frequency of overlapped and multiply overlapped
spectra as a function of source density for each telescope. From
Allende Prieto (2008).
The lack of two-dimensional information within and around
the window requires the faint source background to be modelled.
Given the low flux levels, and because the sources are dierently
spatially arranged with respect to the selected source for each
epoch of observation, this approach can be successful. However,
in densely crowded regions, sources with GRVS that would nor-
mally be assigned a window may not be, because the maximum
number of assignable windows (72 per CCD) may be exceeded.
These sources will be recorded as background. In this case, the
background spectra may be moderately bright, and there may be
no epochs at which a source spectrum is not overlapped by one
source or another, so that a more specific treatment of the source
data may be required.
8.8. Dead time
The Gaia scan law (Gaia Collaboration, Prusti et al. 2016) pro-
vides for more than the requisite average number of 40 transits
across the RVS focal plane (Tab. 1). However, for a number
of reasons, some transits may not be recorded. These include
nominal orbital maintenance operations; inadequate resources
for placement of the windows at the detection chain level in
high-density regions; deletion in the onboard memory as a re-
sult of inadequate capacity, particularly when both telescopes
are scanning the Galactic plane; and data transmission losses.
All of these eects were modelled by Fusero & Chassat (2011)
Article number, page 12 of 19
M. Cropper et al.: Gaia RVS
to ensure that the average number of transits and radial velocity
accuracies were met despite this dead time.
The derived dead-time fractions are shown in Tab. 4. The
dead time increases for fainter stars because of the GRVS prioriti-
sation assigned to windows and to the transmission of data. This
creates a distribution (beyond that inherent from the scan law) in
the number of transits recorded on the ground from star to star,
and therefore in the expected end-of-mission radial velocity de-
rived at each GRVS. This is particularly the case at the faint end,
where the dead-time fraction exceeds 0.4.
8.9. Radial velocity precision
The radial velocity precision at CDR was predicted using a
model taking into account the expected distribution of transits,
dead time, spectral type, radiation damage eects, Poisson and
readout noise, internal and external backgrounds, digitisation
noise, across-scan collapse to one dimension for HR, and addi-
tionally, along-scan binning by 3 for LR. Spectra for each source
were summed for all scans and cross-correlated with the same
template as was used to generate the spectrum. The computed
value was increased by 20% in accordance with de Bruijne et al.
(2005a). The resulting performance is shown in Tab. 4.
Table 4. Pre-launch predicted radial velocity precision for the specified
stellar types in Tab. 1 for end of mission. The precision does not include
calibration residuals. Dead time combines the likelihood of all of the
eects in Sec. 8.8. From Fusero & Chassat (2011).
Spectrum Magnitude Dead Mode RV precision
Vband time (km s1)
B1V 7 0.14 HR 0.6
G2V 13 0.34 LR 0.6
K1III MP 13 .5 0.34 LR 0.6
B1V 12 0.34 LR 8.5
G2V 16.5 0.42 LR 12.8
K1III MP 17 0.42 LR 13.3
The radial velocity precision requirements from Tab. 1 were
required to be 1 and 15 km s1for the brighter and fainter
stars, respectively, in the upper and lower halves of Tab. 4. These
predictions indicated that the key RVS radial velocity perfor-
mance should be met with margin. It should be noted, however,
that these predictions did not include the significant eects en-
countered in orbit, as discussed in the next section.
9. Post-launch developments
On 2013 December 19, Gaia was launched on an accurate tra-
jectory to its orbit insertion at L2. During this initial phase, the
sunshield deployed and communications were established, and
the payload underwent a controlled cool-down to its operational
temperature (in the case of the RVS focal plane, 163K) during
the transfer to L2. After spacecraft checkout, the payload was
activated, and the first images were received on 2014 January
4. These trailed images from the non-spinning spacecraft indi-
cated that the focus was approximately correct. The spacecraft
spin rate was synchronised to the CCD readout rate as set by
the onboard atomic clocks, and the focus of the two telescopes
was optimised using the astrometric focal plane. The first RVS
science data were received on 2014 January 17. Commissioning
of the RVS then proceeded until the In-orbit Commissioning Re-
view on 2014 July 18, using analyses for the initial data products
tailored to the exploratory nature of these activities.
Fig. 15. Early RVS spectrum: Class 0 window (12.6×0.3 mm on the
CCD) of the GRVS =6.2 star with transit_ID=4635432253571710_1
from telescope 1 in FPA Strip 15, Row 4, observed on 2014 January 23.
In this raw image, the bias level has not been subtracted and the aspect
ratio of the window is widened in the across-scan direction for better
visibility. The grey-scale is linear in digital units (0.55 e).
Figure 15 shows one of the first RVS spectra. This is a Class
0 window preserving two-dimensional information. The focus is
good even prior to optimisation, as is evident from the sharp ab-
sorption lines. In this window the spectrum is well centred in the
10-pixel width, with a small spectral tilt, but an adjustment in the
spectral (along-scan) direction is required. Figure 16 shows the
first public3spectrum from RVS, of the V=6.67 K5 star HIP
86564. Lines of Fe and Ti are visible, and most prominently, the
Ca triplet is clearly evident, with the instrument spectral band-
pass centred on the Ca triplet. The quality of the data in this
single 4.4s exposure is striking. Figure 16 also shows ground-
based data with a high signal-to-noise ratio for comparison, and
it is evident that even spectral features with low equivalent width
in the two spectra are in common.
Fig. 16. (Top) The first public spectrum from RVS, of the V=6.67
K5 star HIP 86564 identifying the major spectral features. The elec-
tronic bias has been subtracted and a wavelength calibration applied.
This single CCD exposure has a signal-to-noise ratio 125. (Bottom)
A spectrum in the RVS spectral range of the same star taken with the
NARVAL spectrograph at Observatoire Pic du Midi, convolved to the
same spectral resolving power as the RVS spectrum.
Article number, page 13 of 19
A&A proofs:
manuscript no. 32763_Cropper
Figure 17 shows sample RVS spectra4for spectral types B2
to M6. The dominant lines in hot stars are the Paschen series,
with the Ca triplet dominating in spectral types F – K, and TiO
molecular bands for M stars. The richness of astrophysical diag-
nostics in this short wavelength range is evident, vindicating the
recommendations of the RVS Working Group (Sec. 2).
Broadly speaking, the in-orbit RVS spectra showed the
characteristics expected pre-launch (Fusero & Chassat 2011;
Chassat & Ecale 2014). Nevertheless, it had become clear dur-
ing investigations of the impact on the RVS of the scattered light
from the BAM laser almost immediately after receipt of the first
RVS science data that unexpected variations were evident in the
flux levels in the (nominally empty) VO windows.
9.1. Scattered light
Figure 18 shows the background light variation early in the mis-
sion. A pattern repeating on the Gaia 6hr spin period is evi-
dent. Further analysis based on the correlation of the flux levels
with the satellite spin phase established that this variation was
caused by scattered light from both the Sun (the broader fea-
tures broadly common to all rows) and the bright stars/planets
and Galactic Plane (sharper features most prominent in Row 7 at
the top of the focal plane). The contribution from these sources
changes relative to that of the Sun (and hence spin phase) as dif-
ferent sources are viewed and as the satellite spin axis precesses
while the solar aspect angle is held constant at 45.
The origins of the increased scattered sunlight lev-
els were traced to fibres at the perimeter of the flexi-
ble segments of the Gaia sunshield (Faye & Chassat 2014;
Gaia Collaboration, Prusti et al. 2016), while those from bright
stars/planets and the Galaxy disk arose from unexpected unbaf-
fled optical paths in the payload module (Sec. 5.4). The scat-
tered light background can be more than an order of magnitude
higher than the expected <0.8 epixel1per 4.4 s exposure,
with solar scattered light values reaching 20 30 epixel1per
exposure at some spin phases. At the top of Row 7, the scat-
tered light level can exceed 103epixel1per exposure, and
may even reach detector saturation level. This development had
significant implications for the RVS: as noted in Sec. 4.1, control
of noise sources is critical in reaching the required radial veloc-
ity performance at lower flux levels where the instrument was
to operate with expected background signals of <1 epixel1
per exposure. For a significant fraction of the spin period, the
Poisson noise on the background now dominates the noise bud-
get, exceeding that of the readout noise. In addition, exceptional
care would be required for the background subtraction in order
not to introduce biases into the velocity measurement for faint
stars, and in brighter stars, for the measurement of line equiva-
lent widths used for the determination of atmospheric parameters
and individual abundances.
9.2. Other new non-conformances
The other significant non-conformances noted during the com-
missioning period such as the basic angle instability and the con-
tamination buildup of water ice were less problematic for the
RVS. The former aects the zero point of the wavelength scale,
which is derived from the astrometric measurements, but at a
low level compared to the end-of-mission systematic radial ve-
locity error. The latter aects the RVS throughput, but less so
than it aects the astrometric and photometric payload, owing
to the operation of the RVS at far red wavelengths. The con-
tamination is eectively removed by a de-contamination pro-
cedure (during the commissioning, there were three of these;
see Gaia Collaboration, Prusti et al. 2016). However, the thermal
perturbations to the payload module as a result of the decon-
tamination procedures require time to return to stability (adding
to the dead time) and a refocussing, with consequent eects on
the across-scan line-spread function and the spectral resolving
power calibrations, as well as the throughput.
10. Mitigations
When the origin and nature of the contaminating scattered light
(Sec. 9.1) was understood, detailed simulations were carried out
to evaluate the impact (Katz et al. 2014a), and the Gaia onboard
software was modified to optimise the RVS operation in the new
environmental conditions as follows:
1. the spectral sampling was optimised given the new noise bal-
2. the instrument limiting magnitude was reduced (brightened),
taking into consideration the degraded signal-to-noise ratios;
3. the RVS windows were enlarged in order to measure the in-
stantaneous straylight level.
To elaborate, the LR mode, i.e. the along-scan summing of
the spectra by groups of 3 pixels (Sec. 7.1), had been imple-
mented to reduce the read-out noise by 3 as well as to minimise
the telemetry budget by a factor 3. With the scattered light sig-
nificantly larger than the readout noise, the utility of the on-chip
summing to minimise the electronic noise is reduced. Moreover,
the LR mode has drawbacks: the spectral resolution element is
sampled with only one pixel; it requires a separate and extensive
set of calibrations; and the frequent switches between LR and
HR modes appear to produce cross-talk with other CCDs. On
2014 July 10–17, shortly before the start of the nominal mission,
the onboard software was modified to record all spectra in HR
Recording all (even faint) spectra in HR mode increased the
RVS telemetry rate outside of the allocation, and compensatory
measures were required. With the strong increase of the scattered
light, the faintest spectra recorded near the instrument limiting
magnitude contained almost no information, even when com-
bined together at the end of the mission. To avoid using teleme-
try bandwidth for these spectra, the limiting magnitude was de-
creased from GRVS 17 to GRVS 16.5 on 2014 June 12, then to
GRVS 16.2 on 2014 July 10, prior to the in-orbit commission-
ing review. This partly mitigated the telemetry increase resulting
from the recording of all stars in HR mode. GRVS =16.2 was ap-
propriate for the average scattered light level, but because of the
large-amplitude fluctuations over the six-hour spin period of the
satellite and the variations over the RVS focal plane (Fig. 18), a
further optimisation took place in 2015 June when the limiting
magnitude was adapted to the level of the instantaneous stray-
light in each VPU, varying from GRVS =15.316.2, following
the straylight pattern.
Pre-launch, it was planned to measure the scattered light,
then expected to arise from the laser in the BAM, using virtual
object windows. While the BAM is relatively stable in time, al-
lowing its scattered light contribution to be accumulated over
long periods, the scattered light from the Sun and bright stars
has strong variations over the six-hour spin period, long-term
seasonal variations, peaks from bright stars, and strong local gra-
dients at the top of CCD Row 7 (Fig. 18). The calibration of the
Article number, page 14 of 19
M. Cropper et al.: Gaia RVS
Fig. 17. RVS spectra of 6 Hipparcos stars from spectral type B2 to M6, with an identification of the major spectral lines in the RVS bandpass. In
addition, although not evident in these spectra, diuse interstellar bands are recorded at 862nm.
post-launch scattered light therefore requires many more free pa-
rameters than expected pre-launch. An increase in frequency of
background measurement was required, and in 2015 June, the
RVS windows were therefore enlarged from 1260 to 1296 pix-
els by adding 3 pixels to each of the 12 RVS macro-samples in
a spectrum. In a slitless spectrometer, the beginning and end of
the window receive no source photons (as these wavelengths lie
outside of the bandpass filter) but record the scattered light back-
Other measures were also considered, including a modifi-
cation of the nominal width of the windows depending on the
source magnitude and instantaneous background level; the fol-
lowing of spectral tilts (Sec. 11.2) by adapting the window
boundary half way along the spectrum; and an enhancement of
the prioritisation of window overlaps. These enhancements were
implemented onboard and commissioned, but were found not to
enhance the performance significantly while at the same time in-
troducing additional complexity, and therefore they are currently
not used.
In addition to the onboard software changes, modifications
were made in the RVS data processing chain in order to calibrate
and subtract the scattered light (Sartoretti et al. 2018).
11. In-orbit characteristics and performance
This section describes the in-orbit characteristics of the RVS,
generally superseding the predictions made in earlier sections.
Most of these characteristics are as they were known at the time
of the In-Orbit Commissioning Review at the end of the com-
missioning (2014 July) and covering Gaia spin periods 308 –
1048. However, included are relevant post-optimisation values
(2015 April) that take into account the mitigations in Sec. 10
introduced to improve the performance in the presence of the
higher-than-anticipated scattered light background. The longer-
term instrument parameter trends as derived from calibrations
within the data processing are available for Data Release 2 in
Sartoretti et al. (2018) and Katz et al. (2018).
Table 5 summarises the overall instrumental parameters.
More details are provided in Panuzzo et al. (2015) and in the
Gaia Parameter Database.
11.1. Focus and spectral resolving power
The best focus search was carried out in several stages with a
final position for the commissioning identified during Gaia spin
periods 662–682. The along-scan (spectral) resolving power and
sampling averaged over each RVS CCD across the bandpass
as measured over Gaia spin periods 680–697 (2014 April 16
– 21) are given in Tab. 6. These were measured from a cross-
correlation of Fe lines with a binary mask (Panuzzo et al. 2015).
An alternative analysis using high-resolution ground-based spec-
tra for comparison (Katz et al. 2014b) resulted in an estimation
of a slightly lower spectral resolving power (10%). Both these
analyses yielded values that are lower than predicted in Fig. 9,
but within specification. Alignment between RVS and the astro-
metric field is suciently accurate that requirements can be met
with the optimal astrometric field focus; no compromise inter-
mediate position is required. In HR mode the optical resolution
is fully sampled at slightly more than Nyquist (the early spectra
taken in LR during the commissioning phase are significantly
As noted in Sec. 9, refocussing is carried out after each de-
contamination, so the values in Tab. 6 change slightly each time
this occurs.
Article number, page 15 of 19
A&A proofs:
manuscript no. 32763_Cropper
Fig. 18. Scattered light measurements early in the mission. The colour
scale of the plots shows background level (for each VO, the median
number of electrons per pixel per 4.4 s exposure) in logarithmic units
as a function of onboard mission time in units of Gaia spin periods on
the horizontal axis. The total duration is 12 hr, covering Gaia spin
periods 363 and 364 on 2014 January 30. The 4 rows of RVS CCDs
are abutted, so that the plots show the across-scan dependence. There
is a separate plot for each of the 3 CCD strips, each with an (exagger-
ated) displacement to denote their 4.4 s and 8.8 s later positions in the
The across-scan (spatial) line-spread functions arrived at dur-
ing the final best-focus search average 3.5 and 2.8 pix FWHM
(Panuzzo et al. 2015) over the field of view for Telescope 1 and
Telescope 2, respectively.During observations, this is broadened
by both the transverse motion induced by the scanning law and
the optical distortion in the field of view (Panuzzo et al. 2014).
This causes the line-spread function width to vary from 2.25 –
2.6 pix in the ecliptic scanning law (where the transverse motion
Table 5. Summary of useful RVS parameters.
Exposure 4.4167 s
Typical number of transits 41, each over 3 CCDs
Image scale 169.7 µm arcsec1nominal
Pixel scale along scan 10 µm
0.0589 arcsec nominal
across scan 30 µm
0.1767 arcsec nominal
Window size along scan 1260 pix until 2015 June
12 macrosamples of 105 pix
1296 pix after 2015 June
12 macrosamples of 108 pix
across scan 10 pix except in overlaps
Wavelength range 845.0 – 872.5 nm (FWHM)
845.5 – 872.0 nm (at 90%)
Mean dispersion at 847nm 0.0244 nm pix1
8.51 km s1pix1
at 873nm 0.0246 nm pix1
8.58 km s1pix1
Telescope 1, Row 4 bottom 0.02440 nm pix1
Row 7 top 0.02460 nm pix1
Telescope 2, Row 4 bottom 0.02438 nm pix1
Row 7 top 0.02461 nm pix1
Digitisation 0.539 – 0.595 e(dig. unit)1
is small) in the case of CCD Row 4 Strip 15 in Telescope 1. In
the nominal scanning law, the range broadens to 2.25 – 4.2 pix.
11.2. Spectral tilt
The RVS spectra are slightly tilted with respect to the CCD win-
dow boundaries. This contributes to flux loss from the ends of
the spectra if they are not correctly centred in the window. The
across-scan dierence between the central position of the first
and last macrosamples ranges from 3 – 4 pixels, with the most
positive tilts for Row 4 and the most negative for Row 7, and the
change in tilt following a linear relationship. During Gaia spin
periods 400 – 437, there was an oset of 2 across-scan pixels
between the spectra from the two telescopes.
11.3. Readout noise
Commissioning-phase readout noise measured from the pre-scan
pixels in the CCDs averaged 3.1 e. This was slightly better than
that measured in the thermal vacuum testing, where the corre-
sponding value was 3.7 e. DPAC measurements found slightly
higher values (higher by 0.20.3 e) (Katz et al. 2014a). The
low noise levels vindicated the attention paid to minimising this
noise source, which is still the dominant source during the low-
background phases of the spin period.
The readout noise for each CCD is given in Tab. 7.
11.4. Bias non-uniformity
The bias non-uniformity was calibrated from a set of special
VOs. When applied back to the set, residuals were in the range
0.52 σ1.09efor all 12 detector chains, showing an ex-
cellent level of correction (Fig. 19). Several special VO patterns
were taken over a period of weeks during commissioning to ex-
Article number, page 16 of 19
M. Cropper et al.: Gaia RVS
Table 6. Mean spectral resolving power and number of pixels per optical resolution element for each detector and each field of view. From
Panuzzo et al. (2015).
Telescope 1 Telescope 2
Row Strip Resolving power Resolution element (pix) Resolving power Resolution element (pix)
4 15 12 587 2.798 ±0.009 12 065 2.923 ±0.009
4 16 12 361 2.849 ±0.009 12 106 2.912 ±0.008
4 17 12 240 2.876 ±0.009 11 954 2.948 ±0.009
5 15 12 159 2.891 ±0.008 11 600 3.032 ±0.010
5 16 12 430 2.827 ±0.007 11 809 2.978 ±0.009
5 17 12 085 2.908 ±0.008 11 861 2.965 ±0.009
6 15 12 021 2.919 ±0.007 11 523 3.045 ±0.010
6 16 12 132 2.891 ±0.007 11 447 3.066 ±0.011
6 17 12 148 2.888 ±0.006 11 901 2.948 ±0.009
7 15 12 117 2.890 ±0.007 11 078 3.160 ±0.010
7 16 11 885 2.946 ±0.007 10 983 3.188 ±0.010
7 17 11 525 3.038 ±0.008 11 377 3.077 ±0.009
Table 7. Measured post-launch detector chain noise (CCD readout
noise and PEM noise, including digitisation) in efor the 12 RVS de-
tectors in HR mode. From Chassat & Ecale (2014).
Row 4 Row 5 Row 6 Row 7
Strip 15 3.1 3.1 3.0 3.0
Strip 16 3.0 3.4 3.1 3.0
Strip 17 2.9 3.4 3.1 3.2
amine the stability of the eect; this was the first time that this
had been possible. During this time, the eect of gate activation
was also examined; gates ensure that there are negligible num-
bers of photon events recorded in the VO. With the exception of
one detector chain (Row 6, Strip 17), the coecients were found
to vary only slightly over the intervening period. When applied
to routine VOs, the residuals were higher, at the 2 3elevel;
this and the existence of outlier points indicated that some im-
provement was required.
11.5. Radial velocity performance
The key performance criteria for the RVS have from the outset
been the radial velocity precisions in Tab. 1, and it is instructive
to compare the in-flight performance to these requirements. The
commissioning phase included a fortnight of performance veri-
fication (Gaia spin periods 772 – 829). To maximise the number
of repeated observations to achieve an assessment of the end-
of-mission performance, the ecliptic pole scanning law was se-
The upper panel of Fig. 20 shows the distribution of residual
radial velocities for single focal plane transits (three CCD strips)
compared to ground-based standards (Soubiran et al. 2018).
These stars were bright stars in the range 5 GRVS 10, taken
in HR mode. The mean of the residuals is 330 m s1, almost
consistent with the end-of-mission requirement (Tab. 1) of 300
m s1after only this limited period, and non-standard data pro-
cessing. In addition, at this level, the mean is disturbed by some
errors from the standards themselves.
In respect of the radial velocity precision, the residuals in
the upper plot of Fig. 20 are for single transits rather than the
average 41 expected at the end of mission. The dispersion in the
fit of 2.48 km s1indicates that this should be easily met, but the
Fig. 19. Bias non-uniformity residuals for Strip 16, Rows 4 – 6 for a
set of special VOs taken on 2014 April 26. The calibration parameters
derived from the VO set have been applied to the same set. Residuals in
this case are at the level of 1eor less.
distribution is dominated by stars brighter than that specified for
1 km s1in Tab. 1 (for example V>13 in the case of a G2V
star), so further analysis is required. See Katz et al. (2018) for a
report in the Gaia DR2 dataset.
The spectra from stars with 40 transits were combined and
compared to fainter ground-based validation stars (Frémat et al.
2017) to examine the radial velocity precision as a function
of magnitude. This is shown preliminarily in the lower plot of
Fig. 20. In this magnitude range during commissioning, only LR
mode spectra were available. The number of available standards
is small, but the results indicate that the radial velocity precision
decreases after V=15 (the systematic oset is an artefact of
the LR processing). A more detailed analysis of early mission
Article number, page 17 of 19
A&A proofs:
manuscript no. 32763_Cropper
Fig. 20. (Top) Radial velocity residuals by comparison with ground-
based validation stars for single transits of the RVS focal plane (3
CCDs). (Bottom) Radial velocity performance for stars with 40 tran-
sits as a function of magnitude, compared with ground-based standards.
These spectra were taken in LR mode using the ecliptic pole scanning
law. The systematic oset of 19.6 km s1is an artefact of a 2-pixel o-
set for the LR wavelength scale with respect to HR in the processing,
and should be ignored. The cause of the large residual for one star of
V15 in this preliminary processing is unknown.
data (Seabroke et al 2018, in prep) indicates, however, that for
G to K stars in HR mode, the 15 km s1precision is reached at
aVlimiting magnitude 15.816.5, indicating (in the absence
of significant radiation damage) a 0.5 magnitude shortfall of
the original requirement (Tab. 1) in the 6000 – 4500K tempera-
ture range because of the higher scattered light background. This
is 0.3 – 0.6 magnitude better than the revised predictions5at the
end of the commissioning, but at this early stage does not include
any eects of in-orbit radiation damage.
11.6. Radiation damage status
Because of conservative assumptions and low solar activity lev-
els during solar cycle 24 (Gaia was launched at the peak of
this cycle), radiation damage to the Gaia CCDs has been well
within the 109p+cm210 MeV equivalent fluence that was de-
signed for, with predicted end-of-mission values 10% of this
level (Crowley et al. 2016). Degradation in RVS performance as
a result of radiation eects has been less than expected, aided
also by the higher background levels from scattered light. In the
short period from launch to the end of commissioning, no signif-
icant degradation was identified.
12. Conclusion
This paper has described the RVS on Gaia, starting with the ra-
tionale for the inclusion of a spectroscopic instrument on a pri-
marily astrometric mission. This has had the benefit of extending
the mission from one that measures the dynamics in the Galaxy
into a comprehensive facility for the wide-ranging investigation
of the Galaxy structure and evolution.
The RVS is not a typical spectrometer. Exposure times are
set by the scanning requirements, resulting in extremely low
signal-to-noise ratios in the spectra. This requires exceptional
attention to the noise sources, driving all aspects of the design,
from the throughput, selected bandpass, bandwidth, and spectral
resolution to the noise performance and stability of the detec-
tion chain. Preservation of the information at the single photo-
electron level and in the presence of radiation damage was re-
quired to be proven. The high data rates arising from relatively
long spectra and short exposure durations required innovative
and elaborate schemes to permit the information to be teleme-
tered; these in turn had implications for the detection chain sta-
bility and noise. The important considerations driving the design
of the RVS and its stages of development have been described
here, together with the expected and in-orbit performance and
mitigations taken to optimise this with the higher scattered light
background both in the instrument and in the data reduction soft-
The data release policy6for Gaia RVS envisages the progres-
sive release of increasingly fainter source data. This is because
fainter sources require a sucient number of transits to reach the
signal-to-noise ratios in the accumulated spectrum necessary to
achieve the specified radial velocity accuracy. It is also the case
that an increasingly careful and elaborate approach is required to
process data for stars at or near the limiting magnitude. Given the
rapid increase in the Galactic distances probed by RVS measure-
ments with increasing magnitude and the consequent rapid in-
crease in the number of stars, the RVS scientific resource will be
enhanced commensurably if the instrumental eects described
here are calibrated and processed at a detailed level. In particu-
lar, attention is required on the bias non-uniformity, the scattered
light and faint source background subtraction, correction for the
radiation damage, the deblending, and the optimal combination
of spectra (the ultimate fine corrections for the deblending and
the radiation damage eects rely on a priori knowledge of the
radial velocity itself). From the extensive understanding of the
instrument, gained both from the pre-launch analyses and tests
from the in-orbit performance, it is clear what is required. The
processing steps taken for the first RVS data release in Gaia Data
Release 2 are described in Sartoretti et al. (2018) and Katz et al.
(2018), and further enhancements will be described with later
The RVS is an exceptional resource for stellar and Galactic
science. The scale of the survey is unparalleled and already ex-
ceeds by an order of magnitude the number of spectra recorded
in previous surveys, with the expectation of more as the survey
progresses. In terms of radial velocities, its scale and advantage
is even greater by more than two orders of magnitude. While in-
creased scattered light levels have reduced the precision of the
radial velocities at fainter magnitudes, this is less than initially
feared, and the mission extension will go some way to recover
the initially expected radial velocity performance, especially if
careful weighting of low background spectra with higher signal-
to-noise ratio is implemented in the data processing.
Article number, page 18 of 19
M. Cropper et al.: Gaia RVS
Acknowledgements. We wish to acknowledge the role of Airbus Defence &
Space for their central role in the development of the Gaia RVS, and the Gaia
Project Team at ESA for their support of the RVS instrument within the Gaia
payload. This work has made use of results from the ESA space mission Gaia,
the data from which were processed by the Gaia Data Processing and Analy-
sis Consortium (DPAC). Many of the authors are members of the DPAC, and
their work has been supported by the following funding agencies: the United
Kingdom Science and Technology Facilities Council and the United Kingdom
Space Agency; the Belgian Federal Science Policy Oce (BELSPO) through
various Programme de Développement d’Expériences Scientifiques (PRODEX)
grants; the French Centre National de la Recherche Scientifique (CNRS), the
Centre National d’Etudes Spatiales (CNES), the L’Agence Nationale de la
Recherche, the Région Aquitaine, the Université de Bordeaux, the Utinam In-
stitute of the Université de Franche-Comté, and the Institut des Sciences de
l’Univers (INSU); the German Aerospace Agency (Deutsches Zentrum für Luft-
und Raumfahrt e.V., DLR); the Italian Agenzia Spaziale Italiana (ASI) and the
Italian Istituto Nazionale di Astrofisica (INAF); the Slovenian Research Agency
(research core funding No. P1-0188); the Swiss State Secretariat for Education,
Research, and Innovation through the ESA PRODEX programme, the Mesures
d’Accompagnement, the Swiss Activités Nationales Complémentaires, and the
Swiss National Science Foundation.
Allende Prieto, C. 2008, Deblending RVS Spectra, GAIA-C6-SP-MSSL-CAP-
003, Tech. rep., Mullard Space Science Laboratory, University College Lon-
Allende Prieto, C. 2009, A preliminary analysis of the RVS tests in the 3rd
Astrium Radiation Campaign, GAIA-C6-SP-MSSL-CAP-005, Tech. rep.,
Mullard Space Science Laboratory, University College London
Allende Prieto, C. & Cropper, M. 2009, The Impact of the CCD-PEM Oset-
Instability Anomaly on RVS Spectra, GAIA-C6-SP-MSSL-CAP-004, Tech.
rep., Mullard Space Science Laboratory, University College London
Binney, J. J., Dehnen, W., Houk, N., Murray, C. A., & Penston, M. J. 1997, in
ESA Special Publication, Vol. 402, Hipparcos - Venice ’97, ed. R. M. Bon-
net, E. Høg, P.L. Bernacca, L. Emiliani, A. Blaauw, C. Turon, J. Kovalevsky,
L. Lindegren, H. Hassan, M. Bouard, B. Strim, D. Heger, M. A. C. Perry-
man, & L. Woltjer, 473–478
Boyadjian, J. 2008, Gaia Instruments Optical Performances Delivery Explana-
tory Note, GAIA.ASF.TCN.PLM.00108), Tech. rep., EADS Astrium/ESA
Boyadjian, J., Arsalane, H., & Piérot, D. 2010, Gaia Payload Mod-
ule optical design and performances at Critical Design Review,
GAIA.ASF.TCN.PLM.00068, Tech. rep., EADS Astrium/ESA
Chassat, F. 2009, Serial Register CTI, report at Gaia Calibration Working Group
No. 4, Tech. rep., EADS Astrium/ESA
Chassat, F. 2013, RVS spectral performances assessment,
GAIA.ASF.TCN.PLM.00694, Tech. rep., EADS Astrium/ESA
Chassat, F. & Ecale, E. 2014, Gaia In-orbit performance status at IOCR,
GAIA.ASF.RP.SAT.00850, Tech. rep., EADS Astrium/ESA
Colangelo, G. 2010, Gaia Mission Requirements Document, GAIA.EST-RD-
00553 issue 3.0, Tech. rep., ESA
Cropper, M. 2003, in Astronomical Society of the Pacific Conference Series, Vol.
298, GAIA Spectroscopy: Science and Technology, ed. U. Munari, 75
Cropper, M., Bingham, R., Cole, R., et al. 2005a, Design and Performance Re-
port for Gaia-RVS, MSSL/Gaia-RVS/TN/018.02, Tech. rep., Mullard Space
Science Laboratory, University College London/ESA
Cropper, M., Katz, D., Holland, A., et al. 2005b, in ESA Special Publication,
Vol. 576, The Three-Dimensional Universe with Gaia, ed. C. Turon, K. S.
O’Flaherty, & M. A. C. Perryman, 351
Cropper, M., Katz, D., Zwitter, T., & Munari, U. 2004, The RVS Performance
Capability, MSSL/Gaia-RVS/TN/015.02, Tech. rep., Mullard Space Science
Laboratory, University College London/ESA
Cropper, M. & Mason, K. O. 2001, Considerations on GAIA system design,
with particular reference to the Spectro instrument/RVS-MSC-001, Tech.
rep., University College London
Crowley, C., Kohley, R., Hambly, N. C., et al. 2016, A&A, 595, A6
Cui, X.-Q., Zhao, Y.-H., Chu, Y.-Q., et al. 2012, Research in Astronomy and
Astrophysics, 12, 1197
de Bruijne, J., Perryman, M.,Lindegren, L., et al. 2005a, Gaia astrometric, photo-
metric, and radial-velocity performance assessment methodologies to be used
by the industrial system-level teams, GAIA-JDB-022, Tech. rep., ESA
de Bruijne, J. H. J., Lammers, U., & Perryman, M. A. C. 2005b, in ESA Spe-
cial Publication, Vol. 576, The Three-Dimensional Universe with Gaia, ed.
C. Turon, K. S. O’Flaherty, & M. A. C. Perryman, 67
Ecale, E. & Chassat, F. 2010, VPU and PDHU default file allocation,
GAIA.ASF.TCN.SAT.00302, Tech. rep., EADS Astrium/ESA
Erdmann, M., Kley, E. B., & Zeitner, U. 2010, in International Conference on
Space Optics, Rhodes Island, Greece, ed. E. Armandillo, N. Karafolas, &
B. Cugny
Famaey, B., Jorissen, A., Luri, X., et al. 2005, A&A, 430, 165
Favata, F. & Perryman, M. A. C. 1995, in ESA Special Publication, Vol. 379,
Future Possibilities for Astrometry in Space, ed. M. A. C. Perryman & F. van
Leeuwen, 153
Favata, F. & Perryman, M. A. C. 1997, in ESA Special Publication, Vol. 402,
Hipparcos - Venice ’97, ed. R. M. Bonnet, E. Høg, P. L. Bernacca, L. Emiliani,
A. Blaauw, C. Turon, J. Kovalevsky, L. Lindegren, H. Hassan, M. Bouard,
B. Strim, D. Heger, M. A. C. Perryman, & L. Woltjer, 771–776
Faye, F. & Chassat, F. 2014, Gaia in orbit spacecraft open anomaly status at
IOCR, GAIA-ASF-RP-SAT-00843, Tech. rep., EADS Astrium/ESA
Frémat, Y., Altmann, M., Pancino, E., et al. 2017, A&A, 597, A10
Fusero, F. 2009, PEM response stability anomaly: On-ground RVS post-
processing for performance recover, GAIA.ASF.TCN.PLM.363, Tech. rep.,
EADS Astrium/ESA
Fusero, F. & Chassat, F. 2011, Science Performance Budget Report,
GAIA.ASF.RP.SAT.00005 Issue 5.0, Tech. rep., EADS Astrium/ESA
Gaia Collaboration, Brown, Brown, A. G. A., Vallenari, A., et al. 2016, A&A,
595, A2
Gaia Collaboration, Prusti, Prusti, T., de Bruijne, J. H. J., et al. 2016, A&A, 595,
Gaia Project Team. 2005, Gaia Mission Requirements Document, GAIA.EST-
RD-00553 issue 1.0, Tech. rep., ESA
Gilmore, G., Randich, S., Asplund, M., et al. 2012, The Messenger, 147, 25
Hambly, N., Cropper, M., Boudreault, S., & et al. 2018, A&A, DPACP-29, this
Høg, E. & Lindegren, L. 1994, in Galactic and Solar System Optical Astrometry,
ed. L. V. Morrison & G. F. Gilmore, 246
Jordi, C. 2014, Photometric relationships between Gaia photometry and ex-
isting photometric systems, GAIA-C5-TN-UB-CJ-041-9, Tech. rep., ICC-
IEEC, University of Barcelona
Katz, D. 2003, in Astronomical Society of the Pacific Conference Series, Vol.
298, GAIA Spectroscopy: Science and Technology, ed. U. Munari, 65
Katz, D. 2005, in ESA Special Publication, Vol. 576, The Three-Dimensional
Universe with Gaia, ed. C. Turon, K. S. O’Flaherty, & M. A. C. Perryman, 51
Katz, D., Cropper, M., Sartoretti, P., et al. 2014a, Recommendations for the re-
vision of the RVS operational mode, GAIA-C6-TN-OPM-DK-021-1, Tech.
rep., Observatoire de Paris Meudon
Katz, D., Marchal, O., Soubiran, C., et al. 2014b, A calibration of the RVS resolv-
ing power by global spectrum fitting, GAIA-C6-TN-OPM-DK-022-1, Tech.
rep., Observatoire de Paris Meudon
Katz, D., Munari, U., Cropper, M., et al. 2004, MNRAS, 354, 1223
Katz, D., Sartoretti, P., Cropper, M., & CU6 et al. 2018, A&A, DPACP-54, this
Lindegren, L. & Perryman, M. A. C. 1996, A&AS, 116, 579
Majewski, S. R., Schiavon, R. P., Frinchaboy, P. M., et al. 2017, AJ, 154, 94
Martell, S. L., Sharma, S., Buder, S., et al. 2017, MNRAS, 465, 3203
Merat, P., Safa, F., Camus, J. P., Pace, O., & Perryman, M. A. C. 1999, Baltic
Astronomy, 8, 1
MMS Study Team. 1999, GAIA Concept & Technology Study Final Report,
GAIA/MMS/TN/037.97 Issue 02, Tech. rep., Matra Marconi Space/ESA
Munari, U., Zwitter, T., Katz, D., & Cropper, M. 2003, in Astronomical Society
of the Pacific Conference Series, Vol. 298, GAIA Spectroscopy: Science and
Technology, ed. U. Munari, 275
Nordström, B., Mayor, M., Andersen, J., et al. 2004, A&A, 418, 989
Panuzzo, P., Benson, K., Cropper, M., et al. 2014, Analysis of the optical AC
LSF of the RVS from the commissioning data, GAIA-C6-TN-OPM-PPA-008,
Tech. rep., Observatoire de Paris Meudon
Panuzzo, P., Benson, K., Cropper, M., et al. 2015, Summary of RVS Payload Ex-
perts analysis of Commissioning data, GAIA-C6-TN-OPM-PPA-006, Tech.
rep., Observatoire de Paris Meudon
Perryman, M. 2009, Astronomical Applications of Astrometry: Ten Years of Ex-
ploitation of the Hipparcos Satellite Data (Cambridge University Press)
Perryman, M. A. C. 1995, in IAU Symposium, Vol. 166, Astronomical and As-
trophysical Objectives of Sub-Milliarcsecond Optical Astrometry, ed. E. Hog
& P. K. Seidelmann, 211
Perryman, M. A. C., de Boer, K. S., Gilmore, G., et al. 2001, A&A, 369, 339
Safa, F., Charvet, P., & Chassat, F. 2004, in ESA Special Publication, Vol. 554,
5th International Conference on Space Optics, ed. B. Warmbein, 291–296
Sartoretti, P., Katz, D., Cropper, M., & CU6 et al. 2018, A&A, DPACP-44, this
Seabroke, G. M., Prod’homme, T., Murray, N. J., et al. 2013, MNRAS, 430, 3155
Short, A., Crowley, C., de Bruijne, J. H. J., & Prod’homme, T. 2013, MNRAS,
430, 3078
Soubiran, C., Jasniewicz, G., Crifo, F., & et al. 2018, A&A, DPACP-29, this
Steinmetz, M., Zwitter, T., Siebert, A., et al. 2006, AJ, 132, 1645
Torra, J., Turon, C., & Blaauw, A. 1988, The European astrometry satellite HIP-
PARCOS. Scientific aspects of the Input Catalogue preparation II. Proceed-
ings of a colloquium held at Sitges, Catalonia, Spain, 25 - 29 January 1988,
Tech. rep.
Turon, C., Crézé, M., Egret, D., et al., eds. 1992, ESA Special Publication, Vol.
1136, The HIPPARCOS input catalogue
Wilkinson, M. I., Vallenari, A., Turon, C., et al. 2005, MNRAS, 359, 1306
Article number, page 19 of 19
ResearchGate has not been able to resolve any citations for this publication.
  • Allende Prieto
Allende Prieto, C. 2008, Deblending RVS Spectra, GAIA-C6-SP-MSSL-CAP-003, Tech. rep., Mullard Space Science Laboratory, University College London
Gaia Instruments Optical Performances Delivery Explanatory Note
  • J Boyadjian
Boyadjian, J. 2008, Gaia Instruments Optical Performances Delivery Explanatory Note, GAIA.ASF.TCN.PLM.00108), Tech. rep., EADS Astrium/ESA
Gaia Payload Module optical design and performances at Critical Design Review
  • J Boyadjian
  • H Arsalane
  • D Piérot
Boyadjian, J., Arsalane, H., & Piérot, D. 2010, Gaia Payload Module optical design and performances at Critical Design Review, GAIA.ASF.TCN.PLM.00068, Tech. rep., EADS Astrium/ESA
Gaia Mission Requirements Document
  • G Colangelo
Colangelo, G. 2010, Gaia Mission Requirements Document, GAIA.EST-RD-00553 issue 3.0, Tech. rep., ESA
Design and Performance Report for Gaia-RVS
  • M Cropper
  • R Bingham
  • R Cole
Cropper, M., Bingham, R., Cole, R., et al. 2005a, Design and Performance Report for Gaia-RVS, MSSL/Gaia-RVS/TN/018.02, Tech. rep., Mullard Space Science Laboratory, University College London/ESA
The RVS Performance Capability
  • M Cropper
  • D Katz
  • T Zwitter
  • U Munari
Cropper, M., Katz, D., Zwitter, T., & Munari, U. 2004, The RVS Performance Capability, MSSL/Gaia-RVS/TN/015.02, Tech. rep., Mullard Space Science Laboratory, University College London/ESA
  • X.-Q Cui
  • Y.-H Zhao
  • Y.-Q Chu
Cui, X.-Q., Zhao, Y.-H., Chu, Y.-Q., et al. 2012, Research in Astronomy and Astrophysics, 12, 1197
Gaia astrometric, photometric, and radial-velocity performance assessment methodologies to be used by the industrial system-level teams
  • J De Bruijne
  • M Perryman
  • L Lindegren
de Bruijne, J., Perryman, M., Lindegren, L., et al. 2005a, Gaia astrometric, photometric, and radial-velocity performance assessment methodologies to be used by the industrial system-level teams, GAIA-JDB-022, Tech. rep., ESA de Bruijne, J. H. J., Lammers, U., & Perryman, M. A. C. 2005b, in ESA Special Publication, Vol. 576, The Three-Dimensional Universe with Gaia, ed. C. Turon, K. S. O'Flaherty, & M. A. C. Perryman, 67
VPU and PDHU default file allocation
  • E Ecale
  • F Chassat
Ecale, E. & Chassat, F. 2010, VPU and PDHU default file allocation, GAIA.ASF.TCN.SAT.00302, Tech. rep., EADS Astrium/ESA