A supercomputer-based software correlator at the

1
A supercomputer-based software correlator at the
Swinburne University of Technology Tingay, S.J.
and Deller, A. stingay_at_astro.swin.edu.au
adeller_at_astro.swin.edu.au
Background The Centre for Astrophysics and
Supercomputing at the Swinburne University of
Technology has been funded under the Major
National Research Facilities (MNRF) program to
investigate applications of baseband processing
in radio astronomy, as a demonstration of
techniques for the Square Kilometre Array
(SKA). Baseband data are the sampled, digitised
output of radio telescopes that are
auto-correlated to form single dish spectra or
cross-correlated for interferometry.
Traditionally these data have been processed
using application-specific integrated circuits
(ASIC). We have departed from tradition and have
developed parallelised software to correlate
baseband data using supercomputers and/or
commodity clusters to correlate baseband
data. Computing resources The main computing
resources at our disposal are a 300 node Beowulf
cluster (Fig 1a) at Swinburnes main campus a 32
node cluster (Fig 1b) at the Parkes radio
telescope a Cray XD-1 at the University of
Western Australia (Fig 1c) and a 16 node cluster
(Fig 1d) at the Australia Telescope National
Facility (ATCA).
Extensive development and testing of these
facilities has taken place over the last 3 years.
A major milestone of the project was the
successful correlation of a VLBI experiment that
involved 9 different radio telescopes across 4
different continents (Australia, Africa, Asia,
and North America). These telescopes recorded
data in 3 different disk-based formats (LBADR in
Australia see Chris Phillips poster at this
meeting Mark5 in South Africa and in the USA
K5 in Japan). The data were correlated together
at the Swinburne University of Technology
supercomputer and reduced/imaged using standard
techniques in existing software packages such as
AIPS, AIPS, DIFMAP, and MIRIAD (Fig 3a
Horiuchi et al. 2006a, 2006b, in
preparation). Since this milestone, spectral
line data at 22 GHz have been correlated (Fig 3b
Horiuchi et al. 2006c, in preparation) and pulsar
data for scintillation studies (Fig 3c Brisken
et al. 2006, in preparation). In May 2006, the
first true PI-based projects from the ATNF time
assignment process were correlated at Swinburne
and delivered to the PIs (projects V193a Norris
et al. 2006, in preparation v195a Bains et al.
v190a,b,c Deller et al. v188a Lenc et
al.). This signals the start of a transition
period that will see all Australian VLBI
observations correlated in software at Swinburne
within the next 12 months. The MNRF-funded
program at Swinburne has therefore fulfilled all
of its goals and has provided (within the 5 year
MNRF timeframe and budget) a very significant
enhancement of Australian VLBI facilities, that
is now open for all ATNF users. Further, the
MNRF project has been a stepping off point for a
number of Australian Research Council projects at
the Discovery, LIEF, and SRS level in 2005/2006.
Other projects have now been proposed to the ARC
on the basis of the continued success of the MNRF
project. Funding has also been secured through
the Australian National Collaborative Research
Initiative Scheme (NCRIS) to adapt the software
correlator for geodetic VLBI work, to support a
new Australian 3-station geodesy network that
will participate in IVS activities.
Fig 1 (left to right) a) 300 node cluster b) 32
node cluster c) Cray XD-1 d) 16 node cluster.
Interferometry The software correlator developed
at the Swinburne University of Technology is now
operated as part of the Australian VLBI array.
This constitutes a major upgrade for this
instrument, increasing the maximum recorded
bandwidth to 1 Gbps, from 128 Mbps. These
facilities are now available to all VLBI
users http//www.atnf.csiro.au/vlbi
Fig 3 (left to right) a) First VLBI image from
correlator J0003-066 b) High spectral resolution
water maser data c) Complex secondary spectrum
(FFT of dynamic spectrum) for pulsar.