SKA HI end2end simulation

Abstract

The current status of the HI simulation efforts is presented, in which a self consistent simulation path is described and basic equations to calculate array sensitivities are given. There is a summary of the SKA Design Study (SKADS) sky simulation and a method for implementing it into the array simulator is presented. A short overview of HI sensitivity requirements is discussed and expected results for a simulated HI survey are presented. Comment: 7 pages, 6 figues, need skads2009.cls file to latex

Full text

Page 1

arXiv:1002.0502v1 [astro-ph.IM] 2 Feb 2010
Widefield Science and Technology for the SKA
SKADS Conference 2009
S.A. Torchinsky, A. van Ardenne, T. van den Brink-Havinga, A. van Es, A.J. Faulkner (eds.)
4-6 November 2009, Chˆ ateau de Limelette, Belgium
SKA HI end2end simulation
H.-R. Kl¨ ockner1, R. Auld2, I. Heywood1, D. Obreschkow1, F. Levrier3and S. Rawlings1
1Oxford Astrophysics, Denys Wilkinson Building, Keble Road, Oxford, OX1 3RH, United Kingdom⋆
2Cardiff University, Queens Building, The Parade, Cardiff, CF24 3AA, United Kindom∗
3LERMA/LRA -UMR 8112 - Ecole Normale Sup´ erieure, 24 rue Lhomond, 75231 Paris CEDEX 05, France∗
Abstract. The current status of the HI simulation efforts is presented, in which a self consistent simulation path
is described and basic equations to calculate array sensitivities are given. There is a summary of the SKA Design
Study (SKADS) sky simulation and a method for implementing it into the array simulator is presented. A short
overview of HI sensitivity requirements is discussed and expected results for a simulated HI survey are presented.
1. Introduction
One of the key-science goals of the SKA is to detect
most of the neutral hydrogen (HI; 1420.4 MHz in the
rest frame) content of galaxies out to cosmological red-
shifts of z ∼ 1 (for further details see e.g. Science with
the Square Kilometre Array eds. Carilli and Rawlings or
Cosmology, Galaxy Formation and Astroparticle Physics
on the Pathway to the SKA eds. Kl¨ ockner, Rawlings,
Jarvis and Taylor). The technical parameters that de-
termine the performance of the SKA have been identi-
fied previously during the SKADS program and are de-
fined in the Benchmark scenario (Alexander et al. 2007).
At the moment the SKA design includes three distinct
telescope technologies in order to cover the required fre-
quency range, i.e. between a few hundreds of MHz to
10−25 GHz. The parameter space one needs to cover to as-
sess the performance of the low-frequency sparse dipole ar-
ray, the mid-frequency aperature array (AA), or the high-
frequency dish array is enormous, and this is impossible to
accomplish via a single telescope simulation. To get a basic
understanding of the array parameters and their influence
on the quality and sensitivity of the final images, one can
parameterise each of the different antenna types in terms
of a dish equivalent. For a dish array, two basic param-
eters are the system equivalent flux density (SEFD) and
the image sensitivity (∆I). Respectively, these are given
by
SEFD =Tsys2k
ηaA
,
(1)
where Tsys is the system temperature [K], k is the
Boltzmann constant, A is the collecting area [m2], and
ηais the aperture efficiency, and (for a naturally weighted
image) the image sensitivity can be calculated via:
∆I =
SEFD
?N(N − 1)NStokest∆ν,
(2)
⋆This work was supported by the European Commission
Framework Program 6, Project SKADS, Square Kilometre
Array Design Studies (SKADS), contract no 011938.
where N is the number of Antennas, NStokes is the
number of Stokes parameters, t integration time [s], and
∆ν is the bandwidth [Hz] (Wrobel and Walker 1999;
a sensitivitycalculator can
astro.physics.ox.ac.uk/∼hrk/ARRAY−EXPOSURE.html).
In order to increase the image sensitivity both equa-
tions dictate that the system temperature and the effec-
tive area are crucial to the telescope performance. However
in the case of continuum emission that is spectrally well-
behaved (spectral index of around zero), the image sen-
sitivity can be increased by trading the SEFD for in-
creased bandwidth. For spectral line observations, where
the bandwidth is tailored to the width of the expected
signal, this cannot be done. Bandpass stability also plays
a crucial role in the image quality in this regime.
In this article we describe an end-to-end (e2e) HI sim-
ulation plan that is focused on exploring a more manage-
able parameter space defined by the system temperature,
effective area, and the spatial configuration of the array.
be foundat www-
2. Simulation
Any aperture synthesis telescope acts as a spatial fre-
quency filter and an analytic determination of the fi-
nal image quality is in most cases impossible. Therefore
the purpose of any interferometry simulation is to de-
fine a practical sensitivity limit with respect to the the-
oretical estimates. Generally one would like to have a
full array simulation that simulates the complete signal
path from the astronomical source up to and including
the receiver electronics. Such an approach is very com-
plicated, computationally expensive and for the purposes
of a full SKA simulation completely impractical. The e2e
simulation can be broken down into individual compo-
nents each of which is treated as a standalone simula-
tion. If one is interested in the electromagnetic properties
(e.g. the directional gain and its stability) of individual
dish designs one needs to take into account the telescope
structure and make use of a full electromagnetic sim-
ulation (e.g. Holler et al. 2008). Similar simulations are
needed if one is interested in the performance of an aper-

Page 2

Kl¨ ockner, Auld, Heywood, Obreschkow, Levrier, & Rawlings: SKA HI end2end simulation
Fig.1. Overview of a full simulation path.
ature array and its primary beam pattern (e.g. OSKAR;
www.oerc.ox.ac.uk/research/oskar). So far little has been
done to simulated the electronic path from the receiver
to the correlator, but that is certainly in the scope of the
PrepSKA (www.jb.man.ac.uk/prepska) program. Finally,
after the correlation stage the data quality and array per-
formance can be evaluated by the simulating the response
of individual baselines, or by simulating and assessing the
final image.
Radio astronomy software packages which are cur-
rently available and offer simulation functionality (e.g.
AIPS, CASA, MeqTrees) generally focus on generating a
model visibility set given a model sky and a set of param-
eters which describe the observation. Such functionality
has emerged naturally, due to the self-calibration process
relying on the generation of model visibilities, however the
sophistication of the simulations offered by packages has
generally evolved beyond this fundamental role, particu-
larly in the case of MeqTrees.
An overview of the developed e2e simulation path is
shown in Fig. 1. The grey box shows the steps we take
to generate a model sky. In our case this is based on
the SKADS Simulated Skies (S3), which are a series of
databases describing the properties of a range of simu-
lated astrophysical objects. The databases themselves are
discussed further in the next section. Specific subsets of
these databases can be retrieved and processed, and finally
converted into either a 2-D image or a 3-D datacube (the
third axis being frequency) with optional Gaussian noise1.
1A suite of Python-based routines with user-friendly GUIs,
collectively known as the S3-Tools allows access, manipulation
and imaging of the SKADS Simulated Skies. These tools have
been developed by F. Levrier and a general overview can be
found in this volume (Levrier, 2009 in these proceedings).
These represent idealised radio skies which can then be fed
into a telescope simulation package. The array simulator
is represented in Fig. 1 by the black box, which will take
the model sky image and generate a model data set based
on this. The final step is the analysis of this end product.
In the case of the e2e HI simulation the analysis step in-
volves running both the ‘idealised’ and ‘observed’ HI dat-
acubes through a source finder algorithm and comparing
the two resulting catalogues. This comparison can then be
used to benchmark a specific telescope design or observing
strategy. In addition, the ability of the S3-tools to produce
maps with Gaussian noise allows to test the source finding
algorithm and provide measures for completeness studies.
2.1. The sky simulation
Here we present a short summary of the sky simulations
that can be used within the array simulator. Further de-
tails on the simulations together with a webform which
can be used to query the databases is available on the
Oxford S3webpage2. The full suite of simulations is pre-
sented using two distinct products. Properties of individ-
ual extragalactic objects and of Galactic pulsars are stored
in databases (SEX, SAX, PUL), whereas morphologically
complex structures such as the Global Sky Model (GSM)
and the signals of the Epoch of Reionization (EOR) are
available as images.
The GSM is the radio foreground of our Galaxy
which has been modeled by the radio and (sub)-
millimeter emission between 10 MHz to 100 GHz.
In addition to the diffuse Galactic emission it also
includes emission from individual point sources e.g.
supernovaremnants (de Oliveira-Costa et al. 2008;
space.mit.edu/home/angelica/gsm).
The EOR images display the HI line signal of
the Intergalactic medium (IGM) during the Epoch of
Reionization. This simulation covers the redshift range be-
tween z = 5.6 and z = 23.6. In addition to the ionization
field, the effect of inhomogeneous heating of the IGM by
X-rays and the Lyman-α radiation field are taken into
account. The simulations have been produced in a cubic
simulation box with a side length of sbox = 100 Mpc/h
and a particle mass resolution of 3×106/h solar masses
(Santos et al. 2008).
The semi-empirical simulation of extragalactic sources
(SEX) describes the radio continuum emission in a sky
area of 20×20 deg2out to a cosmological redshift of z
= 20. As the name suggests, the sources were drawn
from observed (or extrapolated) luminosity functions
and grafted onto an underlying dark matter density
field with biases which reflect their measured large-scale
clustering. This approach puts an emphasis on modelling
the large-scale cosmological distribution of radio sources
rather than the internal structure of individual galaxies.
2s-cubed.physics.ox.ac.uk

Page 3

Kl¨ ockner, Auld, Heywood, Obreschkow, Levrier, & Rawlings: SKA HI end2end simulation
Five source types of radio sources have been included in
the simulation:
- Radio-quiet AGN [1 core; 36,132,566 sources]
- Radio-loud AGN of the FRI class [1 core + 2 lobes;
23,853,132 sources]
- Radio-loud AGN of the FRII class [1 core + 2 lobes +
2 hot-spots; 2,345 sources]
- Quiescent star-forming galaxies [1 disk; 207,814,522
sources]
- Starbursting galaxies [1 disk; 7,267,382 sources]
For each of the source types, the database provides the
radio fluxes at observer frequencies at 151 MHz, 610
MHz, 1.4 GHz, 4.86 GHz and 18 GHz, down to flux
density limits of 10 nJy. Intermediate frequencies can
be determined by using the S3-tools. In addition to the
continuum emission this simulation provides a rough es-
timate of the HI mass of the starbursting and star-
forming galaxies (Willman et al. 2008). A secound ver-
sion of these simulations extending the simulated proper-
ties into the far-infrared (principally for comparison with
data from the Herschel satellite) has now been completed
(Willman et al. 2010).
The semi-analytic simulation (SAX) provides the prop-
erties of neutral atomic (HI) and molecular (H2) hydrogen
in galaxies and associated radio and sub-millimeter emis-
sion lines, i.e. the HI-line and various CO transition lines.
This simulation relies on the Millennium simulation of cos-
mic structure (Springel et al. 2005), which reliably recov-
ers comoving length scales from 10 kpc to several hundred
Mpc and galaxies with cold hydrogen masses (HI+H2)
above 108M⊙.
There are two versions of the SAX database, reflecting
the different versionsof the Millennium simulation: the full
Millennium simulation (sbox = 500/h Mpc; ∼ 685 Mpc)
and the smaller test version, called the Milli-Millennium
simulation (sbox = 62.5/h Mpc; ∼ 85.6 Mpc) where sbox
defines the diameter of the simulated box. Both of the
SAX databases have been produced by constructing a
mock observing cone from the corresponding simulation
box. The opening angle or the field of view (FoV) of these
simulations therefore depends on the values of the maxi-
mum redshift one requires (zmax), e.g. for a redshift of 1
the FoV of the simulation would be 12×12 deg2. More in-
formation about the FoV of the simulation can be obtained
via the simulation page. Currently, radio continuum data
is not available, although efforts are being made to add
this information (Obreschkow et al. 2009a; Obreschkow et
al. 2009b; Obreschkow et al. 2009c)
S3-Tools is used to build mock radio maps or cubes in
which the radio emission of the extragalactic radio sources
could be combined with the diffuse radio emission of the
GSM or EOR.
2.2. Array simulation
The developed array simulator3
sical”AIPS
Kettenis et al. 2006), the Python interface to AIPS.
The core function of the array simulator makes use of
the AIPS task UVCON. The basic input of this task is
a list of antenna locations, together with properties of
each antenna, such as diameter, system temperatures and
aperture efficiencies. Additional inputs are the total time
of observation, the integration time per visibility and the
input sky model which also defines the pointing position
on the sky. The output is a standard UV-FITS data file
in which the visibilities correspond to the input model
with added Gaussian noise appropriate for the specified
antenna characteristics. Dirty or deconvolved images can
be produced by invoking the task IMAGR. If the input
model consists of a cube instead of a 2-D image then two
different simulation paths can be used to generate the
“observed sky”. If the output map is to be a continuum
image formed at the central observing frequency, the
visibilities are produced per frequency step and are finally
merged into a single visibility set. If instead the output
should be a 3-D datacube, the simulation produces a
unique visibility set at each frequency step, each of
which is then imaged independently. Each plane is finally
combined into a cube by using the task MCUBE. For cases
where the duration of the observation equals or exceeds
that necessary for a full UV coverage, the visibilities are
generated for a full synthesis and the noise scaled down
accordingly. This cuts down on both processing overheads
and the size of the resulting UV data files.
This style of simulation is well suited to the investi-
gation of the completeness of surveys as well as to the
understanding of the imaging capabilities when observing
individual galaxies. In particular one can investigate the
quality of snapshot observations for different array lay-
outs. There are, however, several AIPS–based limitations
which have a direct influence on the questions we can
ask, and more complicated simulations are needed to ad-
dress these. Currently the number of array elements which
AIPS can handle is limited to 255 (E. Greisen 2009, pri-
vate communication), which is sufficient for most of the
current arrays but is not enough for a full simulation of,
for example, the LOFAR array at the level of individual
dipole elements. One can circumvent this to some extent
by combining several elements into a single station. This
is relevant since LOFAR does indeed perform correlations
between the beamformed data from each station, and not
between individual dipoles. This is a model which the SKA
is likely to follow in order to minimize the data stream.
Two properties of a real interferometric array which
are absent from this simulation framework are primary
beam effects and bandpass variability. These are two fac-
tors which significantly affect the dynamic range and im-
is based on “clas-
(Greisen 1990,andParselTongue
3A copy of this simulator can be downloaded via www-
astro.physics.ox.ac.uk/∼hrk

Page 4

Kl¨ ockner, Auld, Heywood, Obreschkow, Levrier, & Rawlings: SKA HI end2end simulation
age quality of a real observation, particularly the former
in the case of an aperture array.
Despite the various limitations we feel that AIPS is
likely to be the most tested software package and our
approach provides a basic simulation pipeline which can
provide a reliable check for more complicated simulation
efforts (e.g. within MeqTrees or CASA).
In the redshift range around z = 1 the SKA’s map-
ping speed has the potential to be revolutionized by mid-
frequency aperture arrays (AA). As mentioned before, the
simulation software is limited to 255 antennas and it is
necessary to combine the aperture arrays into individual
stations. The following equation shows how to calculate
the total area of an aperture array, assuming Hertzian
dipoles, and therefore the equivalent dish diameter of an
AA station:
AAA= NtileNdipole
3
8π(c
ν)2Nstokes,
(3)
where Ntileis the number of tiles and Ndipoleis the num-
ber of dipoles per tile. Assuming a square kilometre of
collecting area the diameter of a SKA AA station would
be 56 m. Note that an aperture efficiency of 80% and a to-
tal of 255 stations has been assumed for a S3default SKA
realization (S3
real). This value only corresponds to obser-
vations towards the zenith, and the effective station area
varies with elevation. However simulations to investigate
the performance of such a simplified AA are still valu-
able because the layout of the array will directly affect
the synthesized beam and therefore the imaging capabil-
ity. Figure 3 shows a cut through the dirty beam pattern
showing a high sidelobe pattern which can be minimized
by varying the station distribution within the array con-
figuration (e.g. AntConfig can be used for this purpose;
www.kat.ac.za/public/wiki/AntConfig).
The current design of the aperture array has two lay-
outs that have the same core, but differ in the outer regions
(R. Bolton 2009, private communication). The core has a
radius of 2.5 km and has 165 randomly placed stations
which are separated at least by 96 m.
– The “concentrated” layout has 72 stations beyond the
core placed in 5 spiral arms out to 10 km (radius).
Beyond this 13 stations are placed on the same spiral
arms out to 180 km (S3
realC).
– The “not-concentrated” layout has 85 stations beyond
the core logarithmically placed in 5 spiral arms out to
180 km (S3
realNC).
Figure 2 shows the UV coverage of a test simulation of
1 hour duration. For displaying purposes each visibility
has an integration time of 10 minutes. The high density
of visibilities and the filled central core will provide high
sensitivity and image fidelity when observing diffuse HI
emission. However deconvolution of such structure will be
difficult because of the dirty beam pattern shown in Fig. 3.
The broad sidelobe pattern will add ambiguities during
attempts to recover diffuse, extended HI emission.
Fig.2. The UV coverage of the “concentrated” aperture-
array layout (S3
realC). The simulation is based on 1 hr in-
tegration with an integration time of 10 minutes per vis-
ibility. Note that the 10 minutes integration per visibility
is for displaying purposes only.
BEAM 947.625 MHZ CH11.IBM001.1
JY/BEAM
DECLINATION (J2000)
-00 00 30 1500 00 00 1530
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Fig.3. Cut through the synthesized beam (S3
to the high density of stations in the core area, the syn-
thesizes beam profile plateaus around 30 arcsec and shows
strong sidelobes at a level of ∼10% .
realC). Due
A final limitation of the simulation pipeline arises at
the imaging stage. The images that can be handled by
AIPS are limited to 8096 pixels per dimension. For the cur-
rent SKA layout the maximum baseline length of 360 km
would provide sub-arcsec angular resolution at frequencies
higher than 200 MHz. Such resolution limits the field-of-
view that can be simulated, and producing a simulation
covering several square degrees requires the sky to be di-
vided into sub-patches. For example, the spatial resolution

Page 5

Kl¨ ockner, Auld, Heywood, Obreschkow, Levrier, & Rawlings: SKA HI end2end simulation
0
1
23
45
redshift
?10
?8
?6
?4
?2
0
log10(flux density [Jy])
1010.6Msun
M
?
HI
106.8Msun
Fig.4. Flux density of HI masses with a fixed line width
of 164 kms−1versus redshift. The three lines show the en-
tire HI mass range of the HIPASS survey. Note for detec-
tion arguments one needs to assume a channel resolution
similar to the HI line width, otherwise the sensitivity is
reduced via Eq. 2.
of the array configuration at 700 MHz is 0.3 arcsec. If we
Nyquist-sample the sky then these images need to have a
pixel resolution of about 0.1 arcsec per pixel. Taking the
AIPS limitation into account, such images would cover a
sky area of 0.67×0.67 deg2and thus to simulate a 4 deg2
field 9 sub-images are necessary.
2.3. The HI e2e simulation
The anticipated HI simulation will make use of the SAX
simulated sky. These simulations do not include a physi-
cal model of the associated continuum emission. However,
to evaluate the influence of continuum emission to the HI
simulation a mock continuum component may be added
to the line emission (using the task UVMOD). The HI emis-
sion of each galaxy has been pasted into the model cube
by using S3-tools, selecting the “Oxford” HI templates in
the map making tool. The following equation is given to
provide a general understanding of the detectability of HI.
The basic relationship between HI mass and HI line flux
density is defined by
MHI=2.36 × 105
(z + 1)
D2
L
π
?2ln(2)Sp∆V,
(4)
where DL is the luminosity distance [Mpc], assuming a
Gaussian line profile where Sp is the peak flux density
[Jy] and ∆V is the full line width measured at half-
maximum [FWHM; kms−1]. To investigate how much of
the HI mass function one can trace with the SKA, the
expected HI flux density is shown in Fig. 4. The HI flux
density has been calculated by using average values over
the HIPASS sample for the HI line width (FWHM; 164
kms−1) and for the HI masses (range between 2·107—
Table 1. Simulation parameters at various redshifts.
Bandwidth corresponding to a fixed redshift interval of
∆z = 0.1. The fraction of the HIPASS volume to the co-
moving volume of 1 square degree and ∆z = 0.1 (e.g. at
redshift 1 such volume would correspond to 0.0008 Gpc3
deg−2). The FoV of the SAX sky simulations (values
are obtained from s-cubed.physics.ox.ac.uk). Expected HI
sources per square degree for a flux limit of 3 µJy (as-
suming a rms of 1 µJy). For example to investigate the
HIPASS volume at redshift 1 a 4×4 deg2field must be
simulated and need to be split into 36 individual sub-
array-simulations to handle the AIPS image limitation.
redshift bandwidth
HIPASSvol
co−volume
[deg2]
SAXsky
[deg2]
SAXsources
[deg−2][MHz]
0.5
1
1.5
2
59
34
22
15
37
16
12
11
21.4 × 21.4
12.0 × 12.0
9 × 9
7.5 × 7.5
15136
7136
1897
816
6·1010M⊙) (Zwaan et al. 2003, Zwaan et al. 2005). Note
that in Fig. 4 to detect the HI mass at specific redshift one
assumes that the entire HI signal is confined within one
channel. For a freely chosen 1-sigma rms noise of 1 µJy
one could detect M∗HI galaxies at the 3-sigma level out
to a redshift of 1 (Note that a rms noise of 1 µJy would
correspond to an integration time of 36 hours, assuming a
channel width of 250 kHz and a Tsysof 50 K.). This is the
anticipated aim of the SKA and the HI simulations need to
be able to investigate such a cosmic volume. Furthermore,
if there is little evolution in the HI mass function one
could expect to detect the most massive HI galaxies up to
redshifts of about 4.
The FoV of the Millenium simulation box corresponds
to an area of 116×116deg2at redshift 0.1, which shrinks to
5.4×5.4 deg2at redshift 4. At low redshifts a HI simulation
would help to investigate potential systematic errors in a
measurement of the faint end of the HI mass function.
Such a simulation would require around 30276 individual
simulation runs, based on the field-of-view limitations per-
facet, as discussed above. This is a computationally very
expensive exercise, and is only worthwhile once the SKA
array configuration has been finalised. For investigating
and optimising array configurations it makes more sense
to perform smaller scale simulations which still allow us
to quantify the performance of the array.
We propose such a HI simulation here, which will
partly match the co-moving volume of the HIPASS
dataset. This requires a smaller number of individual sim-
ulation runs, and it is thus possible to re-run the simula-
tion several times.
HIPASS is a blind HI radio survey of the sky at de-
clinations southwards of 25 degrees. However to calculate
the co-moving volume we use the initially presented cat-
alogue covering the hole southern hemisphere. The sur-

Page 6

Kl¨ ockner, Auld, Heywood, Obreschkow, Levrier, & Rawlings: SKA HI end2end simulation
vey covers a sky area of 21314 deg2, a redshift range
of z = 0.001–0.042 (corresponding to a co-moving vol-
ume of 0.013 Gpc3), and has a sensitivity limit approx-
imately corresponding to a peak flux density of 0.05 Jy
(see Zwaan et al. 2003 for a full description of the sam-
ple completeness). This survey yields a total number of
4315 sources, all of which are subsequently identified with
optical galaxies (Doyle et al. 2005).
Using the SAX simulation to mimic the HIPASS sur-
vey no extrapolation of the sky simulation is needed be-
cause the simulated volume is large enough to cover the
entire HIPASS volume. In fact this can been shown by
comparing the HIPASS co-moving distance of 171 Mpc
(z=0.042 and using the cosmological parameter of the
Millenium simulation) which is smaller than the radius
of the simulation box, i.e. 500/2 Mpc/h ∼ 342 Mpc.
Using the online query of the SAX simulation and
defining the redshift and the sensitivity specifications4
of the HIPASS catalogue one obtain 4545 sources. This
means that the SAX simulation predicts that the HIPASS
catalog contains 4545 sources, which matches the observed
number of 4315 sources within 5 %. This difference can
be partially attributed to the fact that HIPASS yields a
continuously varying completeness function rather than a
strict peak flux limit. Assuming no flux limits the SAX
simulation predicts 772120 source within the HIPASS
volume, i.e. 180-times more sources than picked up by
HIPASS. However Table 1 shows the expected number
counts assuming a minimum peak flux density of 3 µJy.
The simulation still contain enough source to address e.g.
the study of the faint end of the HI mass function and to
make statistical significant predictions out to high cosmo-
logical distances.
Figure 5 shows the HI intensity map of an input cube.
No continuum emission has been added and the individ-
ual galaxies are unresolved. The corresponding “observed
sky” is shown in Fig. 6. In order to analyse these images
the automated source finder, Duchamp5will be used to
generate “observed” source catalogues for comparison to
the input catalogues. An important part of this analy-
sis will be to produce idealised input images with purely
Gaussian noise of an equivalent level to cross–check the
reliability and completeness of the source finding software
in the presence of the image artifacts introduced by an
interferometer.
3. Conclusions
A full e2e simulation path has been developed. The ar-
ray simulator makes use of images or datacubes based
on the S3catalogues to simulate an observation. In the
simulations no errors due to calibration or telescope hard-
ware have been introduced, but a simplistic treatment of
4SAX query input: select count(*), from galaxies−line
wherezapparent between
hiintflux*hilumpeak>0.05
5www.atnf.csiro.au/people/Matthew.Whiting/Duchamp
0.001 and0.042 and
Fig.5. Example of an HI input sky. The cube has 11 spec-
tral channels covering a frequency range of 11×62.5 kHz.
For illustration purposes the image shows the channel av-
eraged line emission (using SQASH) and in addition the
averaged image has been convolved with a 0.3 arcsec
Gaussian beam (using CONVL).
Fig.6. Example of a simulated HI sky. The rms of this im-
age including sources is of the order of 8 µJy. The cleaned
image (using 200 clean components) displays the channel
averaged line emission.
gain and phase errors could be included at later stages.
The analysis of the simulated data sets (either images
or cubes) will make use of automated source finder soft-
ware. In addition to this, a more sophisticated analysis
of the simulated data is possible because the simulator
produces visibility data as well as images. For example
one could investigate the sensitivity of an array to dif-

Page 7

Kl¨ ockner, Auld, Heywood, Obreschkow, Levrier, & Rawlings: SKA HI end2end simulation
fuse emission by using the UV-gap (∆U
niques (Vir Lal et al. 2009) or by analysing the statistics
of Fourier phases (Levrier et al. 2006).
The proposed HI simulations match the requirements
for studying the capability of the SKA aperture array
when imaging HI structures in nearby galaxies. Such sim-
ulations will also investigate the impact of different tele-
scope designs on the proposed SKA blind HI surveys. The
input HI sky will have an equivalent co-moving volume to
that of the HIPASS survey, and this relatively small vol-
ume makes it feasible to re-run the simulations with differ-
ent parameters within a reasonable time. With this setup
we are able to analyse the “observed skies” and study the
influence that the antenna layout has on a blind HI galaxy
survey by investigating following points:
U) analysis tech-
- completeness (peak flux density)
- positional accuracy
- redshift determination
- necessity of subtracting continuum emission and our
ability to do so
Acknowledgements. HRK would like to thank Rosie Bolton for
the two array configuration files.
References
Alexander P., et al., 2007, SKA Memo, 93
de Oliveira-Costa A., Tegmark M., Gaensler B.M., Jonas J.,
Landecker T.L., Reich P., 2008, MNRAS, 388, 247
Doyle M.T., 2005, MNRAS, 361, 34
Holler C.M., Hills R.E., Jones M.E., Grainge K., Kaneko T.,
2008, MNRAS, 384, 1207
Greisen E., 1990, Acquisition, Processing and Archiving of
Astronomical Images, eds. Longo G. and Sedmak G., 125
Kettenis M., van Langevelde H.J., Reynolds C., Cotton B.,
2006, ASPC, 351, 497
Levrier F., Falgarone E., Viallefound F., 2006, A&A,456, 205
Obreschkow D., Croton D., DeLucia G., Khochfar S., Rawlings
S., 2009a, ApJ, 698, 1467
Obreschkow D., Rawlings S., 2009b, ApJ, 696, 129
Obreschkow D., Kl¨ ockner H.-R., Heywood I., Levrier F.,
Rawlings S., 2009c, ApJ, 703, 1890
Santos M., Amblard A., Pritchard J., et al., 2008, ApJ, 689, 1
Springel V., White S.D.M., Jenkins A., et al., 2005, Nature,
435, 629
Vir Lal D., Lobanov A.P., Jimenez-Monferrer S., 2009, SKA
Memo, 107
Wilman R.J., et al., 2008, MNRAS, 388, 1335
Wilman R.J., Jarvis M.J., Mauch T., Rawlings S., Hickey S.,
2010, MNRAS, submitted
Wrobel J.M. and Walker R.C., 1999, Synthesis Imaging in
Radio Astronomy II, ASP Conference Series, 180, 171
Zwaan M.A., et al., 2003, AJ, 125, 2842
Zwaan M.A., Meyer M.J., Stavely-Smith L., Webster R.L.,
2005, MNRAS, 359, L30