A High-Dynamic Range Wide-Field of View Survey Telescope

Transcription

INVITED
PAPER
Australian SKA Pathfinder:
A High-Dynamic Range
Wide-Field of View
Survey Telescope
Featuring high-speed sky coverage with a large field of view, the first priority for this
telescope will be better understanding of galaxy formation and evolution.
By David R. DeBoer, Russell G. Gough, John D. Bunton, Tim J. Cornwell,
Ron J. Beresford, Simon Johnston, Ilana J. Feain, Antony E. Schinckel,
Carole A. Jackson, Michael J. Kesteven, Aaron Chippendale, Grant A. Hampson,
John D. O’Sullivan, Stuart G. Hay, Colin E. Jacka, Tony W. Sweetnam,
Michelle C. Storey, Lewis Ball, and Brian J. Boyle
ABSTRACT
|
The Australia SKA Pathfinder (ASKAP) is a new
telescope under development as a world-class high-dynamicrange wide-field-of-view survey instrument. It will utilize focal
plane phased array feeds on the 36 12-m antennas that will
compose the array. The large amounts of data present a huge
computing challenge, and ASKAP will store data products in an
archive after near real-time pipeline processing. This powerful
instrument will be deployed at a new radio-quiet observatory,
the Murchison Radio-astronomy Observatory in the midwest
region of Western Australia, to enable sensitive surveys of the
entire sky to address some of the big questions in contemporary physics. As a pathfinder for the SKA, ASKAP will
demonstrate field of view enhancement and computing/
processing technology as well as the operation of a large-scale
radio array in a remote and radio-quiet region of Australia.
Manuscript received November 25, 2008; revised January 30, 2009.
Current version published July 15, 2009.
The authors are with the Australian Commonwealth Scientific and Industrial Research
Organization, Epping, NSW Australia (e-mail: david.deboer@csiro.au;
russell.gough@csiro.au; John.Bunton@csiro.au; tim.cornwell@csiro.au;
ron.beresford@csiro.au; simon.johnston@csiro.au; ilana.feain@csiro.au;
antony.schinckel@csiro.au; carole.jackson@csiro.au; michael.kesteven@csiro.au;
aaron.chippendale@csiro.au; grant.hampson@csiro.au; john.o’sullivan@csiro.au;
stuart.hay@csiro.au; colin.jacka@csiro.au; tony.sweetnam@csiro.au;
michelle.storey@csiro.au; lewis.ball@csiro.au; brian.boyle@csiro.au).
Digital Object Identifier: 10.1109/JPROC.2009.2016516
0018-9219/$25.00 2009 IEEE
KEYWORDS
|
Correlation; focal plane arrays; interferometer;
radio astronomy
I. INTRODUCTION
The state of astronomical surveys today is analogous to a
population census, which, although with very detailed
knowledge of specific neighborhoods in some of the major
cities, consists of a driving tour of many major streets and a
glance out the window of a plane flying coast-to-coast. To
understand the full context of the universe in which we
live, a much more detailed census of its denizens is
needed. This ability to do wide and deep surveys is the
focus of modern instrumental developments across the full
spectrum of astrophysics and will be essential in answering
major questions in fundamental physics.
Complementing instruments such as the future Very
Large Telescopes in the optical and ALMA in the
millimeter-wave is the Square Kilometer Array (SKA) in
the centimeter-wavelength range1 [1], [2]. The science
drivers for the SKA are complemented by the technological
developments that have burgeoned over the past few
decades, allowing an instrument that can answer some of
these key questions to be built. Though its final form is yet
1
http://www.skatelescope.org.
Vol. 97, No. 8, August 2009 | Proceedings of the IEEE
Authorized licensed use limited to: Univ Carlos III. Downloaded on September 11, 2009 at 06:10 from IEEE Xplore. Restrictions apply.
1507
DeBoer et al.: Australian SKA Pathfinder: A High-Dynamic Range Wide-Field of View Survey Telescope
Fig. 1. The technology menu. The core technology is listed at the top,
with a conceptual Bdrag-bar.[ Each solution undergoes a
cost/performance analysis (a major R&D effort), which feeds into the
system parameter schedule, where a conceptualized figure-of-merit
related to the science output per life-cycle is computed. The
assessment is usually done assuming a fixed total cost.
to be determined, the SKA will leverage the commodity
production of Bcheap and good[ collecting area and the
liberal use of high-speed digital electronics.
The technical solution for the SKA will be influenced
by many factors scientific, technical, and political, but the
driving factor will be the best survey speed per dollar.
Fig. 1 strives to visualize the process: the major
technologies are arrayed along a line from those that get
the survey speed by being able to see a large portion of the
sky at any one time [large field-of-view (FoV)] to those that
throw a great deal of power at a smaller piece of sky but do
not need to dwell at any given spot for long to get the
sensitivity (good point-source sensitivity). Both have
technical merits depending on the particular science and
the frequency range of interest, and groups around the
world are investigating the full range of solutions.
From the initial days of radio astronomy, where
typically the solution was on the very left edge of the
diagram, through what may be called the BMoore’s law
era,[ the most cost-effective solution has evolved to the
right of this diagram as the electronics have become ever
less costly and more powerful. Each technical solution on
the diagram then gets developed and assessed on cost and
technical grounds and finally assessed against a figure-ofmerit of how well it can perform its science mission over
its lifetime.
Complementary activities are being undertaken around
the world in the technical development and assessment of
the different options, which the international community
1508
will assess against the ultimate science performance in an
Bintegrate and down-select[ process to produce the
technical solution for the SKA.
At the Australia Telescope National Facility (ATNF) of
the Commonwealth Science and Industrial Research
Organization (CSIRO), the primary technology being
pursued is field-of-view enhancement of dishes. This
includes the multifeed cluster for existing telescopes (e.g.,
the 13-element multibeam on Parkes) and phased-array
feeds (PAFs) for the large-N, small-D (LNSD) concept.
Specifically for the SKA program, an ATNF project called
the Australian SKA Pathfinder (ASKAP) is developing
phased-array feeds for 12-m antennas to be deployed at a
new radio-quiet site in the Western Australia (WA)
outback. ASKAP will be the fastest spectral-line survey
instrument in the world in its own right and also serve as a
Bfew percent[ SKA pathfinder.2
ASKAP has national and international collaborators to
develop the design and participate in engineering and
science. These include Canada (NRC, University of Calgary,
and University of British Columbia), The Netherlands
(ASTRON), and Germany (MPIfR). ASKAP is also a major
constituent of the EU FP7 PrepSKA program’s activities,
where PrepSKA is a 22.2 million euro, four-year effort
among 20 partners, including the European Union’s
foremost radio astronomy organizations and funding agencies (DEST, NRC, NWO, and STFC) and other major
international associate partners (NSF).
In addition, the ASKAP program is working to develop
the Murchison Radio-astronomy Observatory in remote
WA as the best site for meter/centimeter-wave astronomy.
This includes the enabling infrastructure (connectivity,
power, access, accommodation, etc.) as well as a robust
process to keep the site a pristine radio-quiet reserve.
To satisfy funding constraints and in order to feed
into the SKA process, ASKAP is being developed on an
aggressive time-frame, with many parallel developments.
As shown below, the first antenna is expected around the
end of 2009, and commissioning operations are to start
by the end of 2012. The budget is about AU$100 million
over the period 2007–2012, and the split is about 60%
towards the telescope and 40% to develop the site and its
infrastructure (including the fiber connection to Geraldton
and the support facility located there).
As shown in Fig. 2, the project is proceeding in phases,
beginning with the R&D phase, moving into a phase
yielding the Boolardy Engineering Test Array (BETA),
which will comprise the first six antennas, with prototype
electronics. Then it will proceed into the full ASKAP
system, and then into SKA. The form of participation
within the SKA obviously depends on technology and siting
decisions. The timeline has the design and development
phases delineated from the construction and operational
2
More information may be found at http://www.atnf.csiro.au/
projects/askap.
Proceedings of the IEEE | Vol. 97, No. 8, August 2009
•
Fig. 2. Timeline for ASKAP leading into SKA. The dashed line is the
physical antenna delivery. The phases to its left are for design and
development and the color-matching phases to the right of it are the
implementation and operational phases.
period by the dashed line, which shows the expected
delivery of the antennas up to the full complement of 36.
This paper introduces and describes ASKAP, as well
as the new radio-quiet site being developed for radio
astronomy called the Murchison Radio-astronomy Observatory (MRO). The MRO is Australia’s candidate site for
the core of the SKA, which would have sites spanning the
continent and potentially into New Zealand. Australia is
one of two regions short-listed by the international community to host the SKA, with southern Africa being the
other.
II . S C IE N TI FI C A N D
TECHNICAL PRIORITIES
ASKAP is the latest development in Australia’s continuing
strong role in radio astronomy dating from its earliest
days. The ATNF operates the venerable Parkes 64-m
antenna, the Australia Telescope Compact Array (ATCA), a
single 22-m dish at a higher site in NSW called Mopra, and
science time on the 70-m NASA Deep Space Network
tracking station at Tidbinbilla near Canberra. These telescopes can be operated together as a very long baseline
interferometer (VLBI) array and have been used in conjunction with telescopes around the world in VLBI and
e-VLBI (i.e., VLBI with near-real-time connectivity over
broadband networks) modes.
ATNF has engaged with international partners to
develop a wide and diverse science case for ASKAP. It has
been published in short form in [3] and in longer form in
[4]. The chapters of the science case were written with no
specific configuration nor precise antenna number specified. As a result, the science returns from different
configurations can be assessed in the context of a broader
science prioritization in the period 2010–2015 [5]. As
listed in that document, the order of science priorities for
ASKAP is:
• understanding galaxy formation and gas evolution
in the nearby universe through extragalactic HI
surveys, including near-field cosmology;
• characterization of the radio transient sky through
detection and monitoring (including VLBI) of
transient and variable sources;
determining the evolution, formation and population of galaxies across cosmic time via highresolution confusion-limited continuum surveys;
• exploring the evolution of magnetic fields in
galaxies over cosmic time through polarization
surveys.
This process, in consideration with the overall ASKAP
budget envelope, has established that the configuration
will consist of 36 12-m antennas with PAFs having approximately 30 square degree field of view. Taking the science
priorities into account, we have determined the ground
configuration to consist of 30 antennas arranged within a
circle of diameter 2 km, with a further six antennas
forming a perimeter with a maximum baseline of 6 km.
The configuration takes into account a mask of the site at
the Murchison Radio Observatory.
The standard observing mode will be conducting large
surveys, with only a relatively small fraction of the time for
smaller, targeted observing. ASKAP has developed an
BOpen Skies[ draft user policy,3 and international teams
will scope and specify as well as develop the software tools
and teams to conduct and handle these large surveys.
As mentioned above and implied in its name, ASKAP
has a central role as an SKA Bpathfinder.[ The principal
technical objective is to demonstrate high-dynamic-range
wide field-of-view astronomical imaging. This has many
technical implications; ASKAP must have:
• good sensitivity;
• large field-of-view;
• stable aperture;
• a great deal of digital processing capability;
• efficient and effective processing algorithms;
• efficient and accessible public science archive.
The following sections detail the implementation satisfying these requirements.
III . AS KAP THE T E LE S COPE
ASKAP’s objective is to provide an operational national
facility instrument to trial large field-of-view highdynamic-range technology on one of the earth’s last,
best locations for radio astronomy. Its primary goal is to
conduct very fast and deep HI and continuum surveys of
the observable sky. Though the actual survey speed
depends on many parameters and on the actual application [6], the adopted simplified figure-of-merit is given by
FoM ¼
Ae
TSYS
2
FoV m4 deg2 =K2
(1)
where Ae is the effective area of the entire array, TSYS the
system temperature, and FoV is the processed field-ofview. Obviously other parameters play a role, such as the
3
http://www.atnf.csiro.au/projects/askap/policy.html.
1509
Table 1 ASKAP Specifications and Properties
accessible and processed bandwidths, the antenna number
(for image fidelity), surface brightness sensitivity, and
temporal resolution (for transients). This figure-of-merit
produces a different system optimization from that of a
sensitivity-based figure-of-merit, whereby FoV can be
traded off against sensitivity. Table 1 shows some of the
overall specifications and properties of ASKAP.
A. System Architecture
Fig. 3 gives an overall block diagram of the ASKAP
system. The individual components will be discussed more
fully below. The approach has been very much systemsoriented, and the overall configuration is the result of a
comprehensive and intensive study weighing competing
variables. The adopted configuration is based on the
interplay of the impact of the value of the f =D ratio (where
f is the focal length of the paraboloid and D the diameter)
with performance and cost. Higher f =D values (> 0.6)
have intrinsically better performance with a phased array
feed, however at a much larger cost (energy as well as
dollars) due to the increase in number of receiving
elements needed in the feed itself as well as in the
beamforming process. Lower values ðf =D G 0:45Þ start to
see a rolloff in the performance of the phased-array feed,
one reason being the mutual coupling of the receiving
elements themselves resulting in an intrinsic underillumination of the dish. Folded optics designs, which
would be needed for the larger f =D ratios, are further
constrained by the large size of the phased array feed since
its blockage on the initial primary-to-secondary path
provides a limit to the useful size of the subreflector:
i.e., one cannot arbitrarily decrease the size of the
subreflector to yield less blockage, as the feed blockage
is still present.
These physical and costing constraints have been
analyzed in a reasonably comprehensive costing-sensitivity
tool initially developed at ATNF called SKACost [7] to
yield the current configuration, including the processed
field-of-view of about 30 degrees squared.
In the context of this paper, the term phased-array feed
will refer to the collection of receiving elements at the
focus of the antenna itself, which are then phased and
summed in a beamformer to produce a beam; i.e., it is a
phased array in the focal plane of the antenna. Properly
speaking, a feed would produce an astronomical beam and
should therefore include the beamformer as well.
Each antenna consists of the antenna itself and its drive
and control elements, the dense coupled array, the
digitizer, and the coarse filter-bank. The signals from all
receiving elements are sent back to the central building for
beamforming and imaging. Though more expensive in
capital hardware, this approach brings more complex
equipment in one space rather than being distributed and
also allows flexibility for testing processing scalability for
the SKA. This equates to about 2 Tbps from each antenna
streaming back over the digital fiber-optic network to the
beamformer.
The digital beamformer, correlator, tied-array beamformer, and single-pixel back-ends (e.g., pulsar processors)
are colocated in a screened room at the central building.
The raw correlated image is sent back over the fiber-optic
link to the computing facility in Geraldton, about 350 km
away. The gridded, calibrated images are then stored in the
science archive for use by astronomers.
Fig. 3. ASKAP system diagram. The left-hand side shows the antenna based systems, the middle the site control building based systems,
and the right the off-site control.
1510
B. Technologies
1) Antennas: The ASKAP antenna is a 12-m unshaped
prime-focus dish. The performance characteristics and cost
of the phased-array feed (receiving elements þ beamformer)
set the antenna configuration, which we have dubbed a Bsky
mount.[ The sky mount is an alt-az antenna with a third axis
that spins the entire dish, quadrupod, and feed structure to
fix the parallactic angle on the sky. The arrangement is less
expensive and incurs less risk than developing a wide-skycoverage equatorial antenna option for ASKAP. The three
axis Bsky mount[ antenna also allows significant flexibility
to test issues related to high dynamic range.
An idealized array with no inter-element coupling
would perform well with a paraboloid reflector with an
f =D ratio 0.5. However, the effects of inter-element
coupling reduce the element beamwidth and the beamformer cannot be optimally set to illuminate a system with
such a small f =D ratio. Both prime focus and dual-reflector
options including Cassegrain, Schwarzschild, and shaped
systems were considered.
Though structurally a dual-reflector system would be
preferred, science-per-dollar considerations led to the
combination of an f =D of 0.5 with prime-focus optics.
Simply put, for a fixed field-of-view, the phased-array feed
performance improves with increasing f =D but the cost
also increasesVprimarily in the cost of the beamformer,
which must handle inputs from many more receiving
elements. The structurally rather awkward value of 0.5
yielded the maximum in the cost/performance analysis.
The antenna will be designed, manufactured, and
installed by the 54th Research Institute of the China Electronics Technology Group (known as CETC54), headquartered in Shijiazhuang, Hebei Province, about 200 km
south of Beijing. The design incorporates panels on a spaceframe backup structure, all mounted above the third axis.
2) Phased Array Feed: Key to the success of the ASKAP
project is the PAF that gives the antenna a wide field of
view to increase the speed of astronomical surveysVit is a
critical element needed to achieve the astronomical goals
of ASKAP and SKA wide field imaging. In particular, the
efficiency, receiver noise figure, and imaging fidelity of the
array present demanding challenges.
The phased array feed offers the ability to scan over
wide fields of view. When compared with an array of feed
horns, phased array feeds present a number of advantages.
By choosing the appropriate beamformer element weights,
one can maximize efficiency, sensitivity, or beam quality in
the subsequent beams formed on the sky:
• for maximum efficiency, weights that are the
conjugate of the received signal gain on each port
are used;
• for maximum sensitivity, weights are applied for
optimum efficiency as above but also multiplied by
the inverse of the noise covariance are used;
•
for optimum beam quality, weights can be used
that control the primary beam sidelobes and cross
polarization. That is, we choose weights to fit a
desired, circularly symmetric reflector aperture
illumination.
One can also optimize a combination of the above: for
example, trading off sensitivity for beam quality. These
optimizations may be done on the fly to maximize
performance.
The initial development of the ASKAP phased array feed
has been described in some detail in [8]. The PAF structure
chosen is a dual-polarized connected Bcheckerboard[ array.
The complete structure comprises a Bsandwich[ of printed
circuit board (PCB), foam, and ground plane, forming a
robust and simple-to-manufacture structure. The PCB
consists of a checkerboard array of square conducting
patches on an electrically thin dielectric sheet. In order that
the PAFs have a field of view of approximately 30 square
degrees, the checkerboard array size must be about ten
elements square. This work complements work being done
by others on primarily Vivaldi-based elements, such as at
ASTRON in The Netherlands and the Dominion Radio
Astrophysical Observatory in Canada.
There is a strong interaction between the PAF system
and the low-noise amplifiers. The signals from the PAF are
electrically balanced, and its efficiency depends critically
upon the differential mode and common mode load impedances presented by the low-noise amplifiers. The performance of the low-noise amplifiers is, in turn, strongly
influenced by the driving impedance presented by the focal
plane array. Successful design of the PAF requires
simultaneous optimization of both the PAF elements and
the low-noise amplifiers.
Each element of the PAF consists of two patches
feeding a differential amplifier via lines from the corners.
The other polarization uses the other corners of the
patches. A key feature of the design work is the analysis of
a practical array structure including two-conductor transmission lines that connect the patch corners to low-noise
amplifiers via holes in the ground plane. Analysis of the
Bcheckerboard[ array indicates that the output impedance
of the array is large and that the load presented to the array
by the low-noise amplifiers should also be large. As a selfcomplementary structure, the array itself should present a
differential impedance of 377 . The optimum impedance
depends on the detailed design of the PAF [9] and is about
300 for the Bcheckerboard[ array. Low-noise amplifiers
have been designed to operate with a system impedance of
300 , with a differential input. Preliminary modeling and
noise temperature measurements indicate an amplifier
noise temperature of G 35 K is possible with a 300 differential system impedance.
A prototype array, consisting of 5 4 2 elements,
illustrated in Fig. 4, has been used for the initial development work. This array has differential room temperature amplifiers and has been deployed on the Parkes
1511
Fig. 4. Picture of the prototype installed at the 12-m Parkes
Testbed Facility antenna.
Testbed Facility, a 12-m antenna installed at the Parkes
Observatory. Some recent results will be discussed in the
Conclusion.
3) Receivers: The final roughly ten-element square PAF
will have 192 individual receiving elements, each of which
will be connected to a low-noise amplifier immediately
adjacent to the feed elements, a superheterodyne frequency conversion system, and digitizer. For 36 antennas, this
amounts to almost 7000 receiving chains, and hence cost is
a major driver in achieving the required performance.
The receiver architecture is dependent on the location
of the analog-to-digital converters (ADCs). The nearer the
ADCs are to the PAF elements, the greater the likelihood
of self-generated radio-frequency interference (RFI) pickup from the digitizing hardware; but the signal transmission from the PAF becomes easier to realize, as we can use
digital data transmission. The further the ADC is from the
PAF, the more difficult it becomes to preserve phase
stability, amplitude stability, and the dynamic range of the
RF signal.
The options considered for receivers were the
following.
• Double conversion superheterodyne with local
oscillator (LO) frequencies of about 6 and 4 GHz
with a second intermediate frequency (IF) output
of 420–720 MHz (Fig. 5). The IF signals are sent to
the base of the antenna where the superheterodyne
receiver and ADCs are located. This is the suboctave method selected for BETA.
• Direct conversion (DC) with quadrature IQ outputs. This receiver architecture may have application in the SKA. To preserve the IQ balance, the
ADC must be located close to the mixer, and this
may introduce RFI at the PAF.
• Single up-conversion with an LO of 6 GHz and first
IF bandpass 300 MHz wide at 5 GHz, as depicted
in Fig. 5. The 5 GHz first IF signals are sent to the
base of the antenna where the IF is downconverted and then digitized in the ADC.
• High gain amplification at the LNA for either direct
digitization of the RF signal or transmission of the
RF signal through modulation of an optical carrier
for broadband RF transmission to the receiver
down-converter and ADC’s at the base of antenna.
The high gain required at the PAF, typically on the
order of 80 dB, will introduce intermodulation and
stability issues.
• Solutions that are a combination of the above: for
example, having the down-converters at the focus
with digitizers under the dish surface and digital
fiber links to the beamformer in the antenna
pedestal. RFI could be reduced in this manner.
Fig. 5 indicates the adopted design, with the PAF, lownoise amplifiers, and cable driver amplifiers located at the
focus of the antenna. Signals are transported from the
antenna focus to the pedestal by coaxial cable. The cable
equalizer, frequency conversion system and ADCs are located in the pedestal of the antenna. An important function
of the frequency conversion system is to filter the signals
before sampling so that undesired signals, which could be
aliased or folded into the received band, are adequately
suppressed. The final 300-MHz-wide IF (420–720 MHz) is
digitized at 768 Msamples/s.
DC with quadrature IQ output, that may have application in the SKA, is being investigated. With the recent
advances in the performance of commercially available
quadrature mixer and digitizer chips, a direct conversion
receiver architecture that is capable of easily operating
over even greater input bandwidths is an attractive
Fig. 5. Simplified block diagram of BETA/ASKAP receiver architecture.
1512
alternative to the dual conversion architecture. At present,
this is in early stages of development in both discrete
component assembly and system-on-a-chip (SOC). The
RF-CMOS chip that is being developed [10] covers an
input frequency range of 500–1700 MHz, with an instantaneous IF bandwidth of 500 MHz.
The receiver will contain all active circuitry (bandpass
filter, quadrature mixer, antialiasing filter, digitizer, and
serializer) on one 0.18 m RF-CMOS integrated circuit.
Some functional blocks for the SOC have been successfully
fabricated.
4) Digital Processing: The challenge for digital signal
processing in ASKAP is a compute load of more than 1015
arithmetic operations per second (Bpeta-ops[) and a data
flow that in one stage of the processing addresses 70 Tbit/s.
This must all be achieved in a system that is not connected
to the electricity grid and must therefore generate its own
power. These challenges are well addressed by a fieldprogrammable gate array (FPGA)-based solution. This is
helped by the fact that the operations to be performed
(filtering, beamforming, and correlation) are comparatively simple, are repetitive, do not require recursion, and
can be implemented in fixed-point arithmetic.
A single FPGA can execute more than 5 1011
arithmetic operations per second and provide 40 Gbit/s
of intersystem data connections. A few thousand FPGAs of
this capacity can satisfy ASKAP digital signal-processing
requirements while keeping power consumptions at
acceptable levels. Estimates for the power consumption
have about half of the power dissipation associated with
computation and memory (DRAMs), a quarter in I/O between systems, and the final quarter in ancillary functions
such as the control computers. A drawback of FPGAs compared to other technologies such as CPUs and GPUs is that
there is little on-chip memory per unit of computation.
The arrangement of signal processing for the ASKAP
beamformer and filterbanks is shown in Fig. 6. The ADCs
and coarse filterbanks are at the antenna. This data are
transferred over 192 10GE optical fibers to the central site
where the fine filterbank and beamformer are implemented. This incurs a high transport cost but simplifies
installation and maintenance at the remote Boolardy site
and adds flexibility for future upgrade and SKA scaleability
tests. The beamformer for each antenna has 64 processing
FPGAs distributed across 16 processing boards. For the
beamforming operation, data for all 192 inputs for either
5 or 4 MHz of data must be sent to each FPGA. This is
achieved by transporting data from four A/Ds on four
10 GE links. After decoding at the central site, 16 data
streams of 3.1875 Gb/s are generated from the data, which
are distributed across the backplane of the industry
standard AdvancedTCA (ATCA) chassis. Each board
receives one of these data streams. It then redistributes
it amongst the four processing FPGAs on the board.
There are two main parts for ASKAP imaging: beamforming of the data from the individual phased-array feeds
(the beamformer) and correlating all of the beams from each
antenna (the correlator). The input to the beamformer from
each antenna comprises 96 dual-polarization signals of
300 MHz bandwidth from the PAF, for a total of 192 inputs.
These are digitized and weighted sums of subsets of these
signals are used to generate up to 32 beams. The number of
elements to be summed for each beam depends on the
frequency and the position of the beam on the PAF and the
desired beam quality. The weighting function is also
frequency dependent.
To minimize computational loads, correlator architectures for large-N arrays (e.g., ASKAP and SKA) will
generally be FX correlators [11], where the cross-spectral
power density is calculated after frequency binning. For
ASKAP, this requires the 300 MHz input data to be broken
down into 16 384 frequency channels, followed by a
measurement of the cross-power or correlation in each
individual frequency channel. Correlations are calculated
for each pair of beamformed signals in each channel.
Similarly, beamforming will be done in the frequency
domain. In each frequency channel, the beamformer
weights are approximately constant, and the beamforming
operation reduces to a simple complex multiply and accumulate operation. For this to hold, modelling indicates that
the frequency channels must be at most 1 MHz wide.
The ASKAP specification calls for the input bandwidth
to split into 16 384 frequency channels. Since the FPGAs
have limited memory, above 2000 frequency channels
chip memory is the limiting resource. For a direct 16 384
channel implementation with a single filterbank, the
compute resource would be underutilized by almost an
Fig. 6. ASKAP architecture for the beamformer and filterbanks.
1513
order of magnitude. For a fast Fourier transform (FFT), the
standard technique is therefore to break the processing
into two successive FFTs with an intervening twiddle
factor, and then to perform multiplication and row-column
interchange. The row–column interchange is memory
intensive but can be done in external memory.
When the FFT is used as a filterbank, the channel
response is a sinc function. For ASKAP, this can lead to
radio frequency interference corrupting a significant
number of frequency channels. Also in the correlator,
the use of a simple FFT leads to degraded signal-to-noise
ratio on narrow-band spectral features [12]. Instead, a
polyphase filterbank is used where the FFT is preceded by a
polyphase filterbank [11]. This allows the spectral response
to be chosen arbitrarily, within the constraints imposed by
the length of the polyphase finite impulse response (FIR)
sections.
The filterbank is broken into two stages. The first stage
is the coarse filterbank and is operated in a 32/27
oversampling mode. This is done at the antenna and
allows the needed frequency resolution for PAF calibration. The second stage is the fine filterbank, which is
critically sampled [13] and located at the central site. The
many coarse channels to be processed by the fine
filterbank are stored in DRAM and are read back as long
sequences for a single channel. With a 12-point polyphase
FIR and 64-channel filterbank, 768 data values are needed
before a full data set can be calculated for the FFT pipeline.
No usable data is calculated in this time. To minimize this
loss the data is processed in blocks of 65 536 samples, and
the processing loss is limited to 1%.
As previously mentioned, due to the frequency
dependence of the beamformer weights, beamforming is
done on 1 MHz frequency channels, setting the resolution
of the coarse filterbank. Beamforming will generate on the
order of 32 antenna beams from the 192 PAF elements,
which reduces the processing needed in the fine
filterbank. For the polyphase filterbank, about half the
computation is in the FFT and half in the polyphase FIR.
The FIR calculation has a compute cost that is independent
of the filterbank. Taking all the processing loading into
account, there is a 10% computational overhead for
ASKAP to split the filterbank into two stages.
The correlator receives the channelized data from the
fine filterbank and beamformers and cross-multiplies all
frequency channels in each beam. It has the same signal
distribution issues as the beamformer, and its hardware will
be identical to the beamformer. Data for some subset of the
antenna beams and frequency channels for all 32 antennas
are processed in a single FPGA. The first stage is achieved at
the outputs of the beamformer, where each output carries
data for all beams but a subset of the frequency channels. In
the ATCA chassis, the backplane and FPGA connections
provide the final stages of data cross-connection. The
beamformer has 16-bin ATCA crates, allowing a 16-bit
correlator to be implemented. This simplifies the program1514
ming of the system, as no dynamic gain modifications are
needed before the correlator, and likewise no gain
recalibrations are needed after the correlator.
5) Data and Signal Transport: As shown in Fig. 3, the
ASKAP data and signal transmission comprise many different communication modes:
• analog transmission of the output from each of the
PAF receivers at the antenna focus to the ADC and
coarse filterbank in the pedestal;
• short-haul digital transmission from the antenna to
the centrally located beamformer and correlator;
• long-haul digital transmission of the correlated
ASKAP array output to the science center for imaging;
• generation and distribution of the LO frequencies
at antennas from a distributed master reference;
• provision of 1 GE local area network for monitor
and control of antennas.
a) Antenna IF transmission: Routing the 192 analog
signals down the antenna structure and through three
movable cable wraps on the polarization, azimuth, and
elevation antenna axes (approximately a 30 m length) is a
significant design issue. The considered options include
the following.
• Coaxial cable for each PAF element: bulky for
192 lines of low-loss coax. Gain equalization would
be required. However, relatively thin and low-loss
cables do exist.
• CAT7 cable: high attenuation above a few hundred
megahertz but is a possibility at low IF frequencies
(less than 1000 MHz). Gain equalization would be
required. Return loss is poor.
• RF over fiber: constraints of form factor are eliminated. The noise figure (NF), however, is comparable to the high attenuation in coaxial cable for
a 20 m cable run.
ASKAP will use coaxial cables in the first instance (BETA);
however, RF over fiber will continue to be developed for
possible use for ASKAP and SKA.
b) RF over fiber: The ASKAP design would require
nearly 7000 short-range (typical length 30 m) links.
Although there are commercial vendors for RF over fiber
modules for use in cable television (CATV), antenna
remoting, interfacility communications links, and radio
astronomy (e.g., the ATA [14]), the costs of so many modules as required for ASKAP would be prohibitive. The use
of vertical cavity surface-emitting lasers (VCSELs) in the
850 nm band is considered. These are relatively inexpensive devices (around $30 each in quantities of more
than 1000) designed for digital applications over multimode fiber (MMF). The market push for 10 G Ethernet
interfaces has delivered VCSEL devices that can be
modulated to 7 GHz or higher.
Bench measurements of directly modulated links using
commercial off-the-shelf VCSEL diodes have been made.
The VCSELs were fitted with standard LC-style receptacles.
Intermediate connections using 12-way multiple parallel
optical connectors were also performed to assess the suitability of inexpensive multiway optical quick connects. A
variety of prototype printed circuit cards have been
developed including standard FR4 material and the low
loss Rogers 4003 substrate. Infinicor 50/125 m high
bandwidth MMF was used in the test configurations. This
has a bandwidth distance product of 5 GHz km and
exceeds requirements for a short 20 m link.
c) Dynamic range: The dynamic range of the shortrange VCSEL-based link is primarily determined by the
device linearity (Pout versus If curve) and the relative
intensity noise (RIN) of the VCSEL. Using readily available
Finisar components with an optical output power of 3 dBm
at a bias of 9 mA (threshold 5 mA), a spurious free dynamic
range (SFDR) of 98 dBHz2=3 was obtained in a 300 MHz
bandwidth at a center frequency of 5 GHz. This is an SFDR
of 42 dB in 300 MHz. A gain G ¼ 32 dB and noise figure
NF ¼ 38 dB were measured. This could be improved with a
higher VCSEL slope efficiency (Pout versus If ) characteristic. The link noise floor is RIN limited. The measured RIN
is 140 dB/Hz and was 10 dB better than the quoted
manufacturer specification. The 1 dB compression point
was estimated at P1 dB ¼ 32 dBm and the third-order
intercept IP3 ¼ 20 dBm. Similar results were obtained
with a 300-MHz-wide IF at a center frequency of 600 MHz
using VCSELs designed for third-generation data communications purposes. This is acceptable and commensurate
with 8 bit ADCs. It is expected that 8 bit quantization in any
300 MHz band for the remote radio-quiet ASKAP
environment (Fig. 11) is sufficient.
The VCSEL-based MMF links have exhibited sensitivity
to fiber movement and bending. Modal noise is generated
by the interference between various propagating modes
causing speckle pattern (i.e., amplitude) variation at the
receive photodiode. Using a high power noise diode as a
signal source and an integrating spectrometer (Acqiris
AC240) as a power meter with consecutive 5 ms integration
periods, fluctuations of approximately 0.05 dB (1%) are
measured over a bandpass of 400–700 MHz with gentle
fiber flexing (Fig. 7).
Averaged over a 60-s period, the phase variation should
be less than 0.16 degrees and amplitude stability 0.013 dB
(0.3%) to meet PAF and interferometer specifications.
Larger variations are allowable only if present on timescales much shorter than the integration time.
Similar measurements were performed using distributed feedback and Fabry–Perot (FP) laser devices at 1310
and 1550 nm. Smaller amplitude variations were measured
indicating a bending loss mechanism with fiber flex in
single-mode fiber (SMF). Although uncooled FP lasers are
available at low cost, the use of SMF with 9 m core
requires more precise connectors and optical alignment,
and hence is less desirable.
d) Local oscillator: The receiver conversion will
require low-drift low-phase-noise local oscillators. The LO
Fig. 7. Multimode fiber-optic link amplitude stability with movement.
distribution will be on SMF. Phase drift can be
characterized in the frequency domain by the single
sideband noise spectrum within a 10 Hz offset from the
LO signal. This is normally associated with temperature
coefficient for all components in the local oscillator
distribution, namely, the temperature coefficient of the
fiber and coaxial cable lengths, typically 10 ppm= C. A
round-trip phase measurement system can measure the
change in electrical path length for the local oscillator
signal.
The use of an offset loop [15] provides the measurement of phase at the nominally chosen offset frequency
(e.g., 1 MHz). Experiments with a Stanford Laboratories
SR620 precision time interval counter achieves measurements better than 100 ps at 1 MHz (i.e., 0.036 phase).
The distribution frequencies should be as high as possible
to provide maximum measurement sensitivity to the
phase changes and to minimize the multiplication of
reference local oscillator noise at the antenna.
The use of a phase-locked loop (PLL) at each antenna
will be required to provide a narrow-band cleanup filter
to the RIN dominated local oscillator optical receiver
output at the antenna. The RIN noise is very wide-band
at RF/IF frequencies and appears as BLO leakage,[
perhaps attenuated 30 dB, at the output of mixers used
in the receiver conversion process. An Ball optical[ local
oscillator distribution (with RIN cross-correlating between antennas) would be problematic; however, it
would have the advantage of reduced RF power levels
for the LOs at the PAF and hence less self-generated RFI
issues.
e) Data communications: The short-haul data paths
will be modular systems based on 10 GE small form factor
SFP+ packages (or equivalent 850 nm/1310 nm optical
transceiver) where possible. Inexpensive SFP+ packages
can connect each antenna at 2000 Gbps to the correlator.
The total correlator input rate for 36 antennas is 72 Tbps.
1515
The correlator output rate is approximately 40 Gbps for
long-haul transmission (400 km) to the ASKAP science
center.
The long-haul links could be based on carrier class
terminal equipment for dense wavelength-division multiplexing expanded to terabit per second SKA bandwidths.
Bidirectional capability will enable VLBI antenna connectivity to the ASKAP correlator for potential future eVLBI
observations.
6) Computing: Since ASKAP is intrinsically wide field of
view, it is envisaged mainly to operate as a survey telescope
with limited time available for pointed observations. This
simplifies the computing requirements and has lead to the
adoption of a model based around Bsoftware instruments[
(SIs). An SI is a release of the telescope software developed
and tested to perform a well-determined type of observing,
such as a continuum survey of the entire visible sky. Once
an SI is operating, it will basically continue until the survey
is complete. Different SIs can potentially operate at the
same time, provided the observing is compatible. SIs can
be developed, tested, and deployed incrementally in a
well-controlled way.
The role of an SI must include the data processing since
the data processing must proceed in real time. This is
dictated by the prodigious data rateVin full spectral line
mode in a 6 km configuration, the postcorrelation data rate
is about 4 Gbytes/s. A typical 8-hour observation will
therefore consume about 100 Tbytes of storage. Necessarily, then, our model is to process observations as
they occur. Flagging, calibration, averaging, imaging,
and source cataloging all occur as the observations are
taken.
Another simplification comes from keeping the telescope calibrated at all times. Whenever a calibrator field is
observed, the calculation of calibration parameters occurs
immediately and the relevant numbers are fed back to the
telescope as soon as possible. This approach assumes that
all the important calibration parameters vary sufficiently
slowlyVprobably a very good assumption for moderate
observing frequencies at the baseline lengths involved (less
than 6 km).
The output images to be generated from the processing are:
• continuumVdeconvolved images for all four Stokes
parameters in 256 channels, of size 64 GBytes;
• spectral lineVimages for all four Stokes parameters
(after continuum subtraction), of size 4 TBytes;
• transient detection: images for all four Stokes parameters in 256 channels every 5 s, of size 377 TBytes.
The data from all PAF beams are combined into a single
image, weighting optimally by the sensitivity. The spectral
line image consists of one image plane per channel
observed and the continuum image has one plane per set of
averaged channels (64 full-resolution channels are averaged into one Bcontinuum[ channel).
1516
The processing required must perform the following
steps in a single real-time flow (pipeline processing).
• Receive the data.
• Identify and flag for known and unknown radio
frequency interference and bad data.
• Solve for and apply calibration parameters.
• Average to required spectral resolution.
• Grid the data using convolutional resampling.
• Fourier transform to the image plane.
• Deconvolve point spread function (if warranted).
• Find sources (algorithms to be determined and
developed by the international science teams).
• Archive data products.
After careful analysis, it was determined that no
existing synthesis software packages were available that
could support parallel and distributed processing of the
required scale and sophistication [16]. Hence it was
necessary to write new synthesis processing code in C++,
using existing underlying libraries as much as possible. In
particular, we have relied heavily upon the CASACORE4
libraries derived from the AIPS++ code base. This provides
many useful capabilities such as table handling and
specification and conversion of coordinates and frames.
The code for ASKAP, ASKAPsoft, currently has the
following capabilities:
• construction of complex measurement equations at
the C++ level;
• correction for feed primary beams during gridding,
including correct scaling of each channel with
frequency;
• correction for the noncoplanar baselines effect
using the w projection approach;
• use of multiscale CLEAN deconvolution.
Execution is distributed across multiple computers
using the message passing interface (MPI) framework [17].
MPI is the dominant model for high-performance computing, and excellent implementations are available on all
supercomputers and clusters. This works well for current
testing but is unlikely to be sufficiently robust for real
processing, and therefore alternatives such as workflow or
streaming middleware are being evaluated.
During our development, we will test this package
using both simulated observations and data sets from real
telescopes, including the Australia Telescope Compact
Array and the Very Large Array. Our first simulation of a
full-field ASKAP image was designated CPTest2 (Fig. 8).
We have run this test on both an eight-node Sun v20z
cluster and a CRAY XT3 at the Western Australia
Supercomputer Program in Perth. We were only able to
use about 1% of the final dataVcorresponding to a
snapshot every 150 s over 12 h, 28 channels over 112 MHz,
32 feeds dithered to 128, resulting in an image of 4096 by
4096 pixels. Calculation of the residuals alone took 4400 s
4
http://code.google.com/p/casacore.
antenna separation of 22 m, and a longest separation of
6 km. The origin corresponds to approximately 116.5
east and 26.7 south.
The sensitivity of the configuration at various angular
scales is an important metric for ASKAP. The configuration is designed to optimize sensitivity on angular scales of
3000 (for HI emission surveys; see [20] and [21]) while also
providing good low surface brightness sensitivity (at low
angular resolution) and high-resolution imaging capability
[22]).
Sensitivity as a function of angular resolution is given
in Fig. 10. The naturally weighted beam gives maximum
sensitivity and an angular resolution of just under 2000 .
Gaussian tapering was applied to the data with progressively larger values, resulting in the curve to the right of
the dashed vertical line. In order to get higher resolution
than that available with natural weighting, it is necessary
to Bdown-weight[ the short baselines at the expense of the
longer baselines. This can be achieved through robust
weighting [23], with the results shown to the left of the
dashed vertical line.
Fig. 8. CPTest2: First simulated full-field image from ASKAP, made
using the AProjectWStack algorithm. The imaging used about 1% of the
final dataVsnapshot every 150 s over 12 h, 28 channels over 112 MHz,
32 feeds dithered to 128, resulting in an image of 4096 by 4096 pixels.
Calculation of the residuals alone took 4400 s on an eight-node cluster
of Sun v20z servers. Display range is 1 to þ3 mJy/beam, and the noise
is about 0.140 mJy/beam.
on an eight-node cluster of Sun v20z servers. More
detailed benchmarking has shown that the processing for
spectral line observations in the 2 km core will require
about 5000 cores, and the continuum processing will
require a similar number. Evaluation of the optimum
computer architecture for this processing is continuing.
C. Configuration
The ASKAP configuration is optimised to produce
excellent sensitivity and a good point spread function
(sidelobe levels 2–3%) at an angular resolution of 3000 at 1.
4 GHz [18]. The configuration also provides high survey
speed at an angular resolution of 1000 and good surface
brightness sensitivity at angular resolutions of 60 and
9000 . We expect that this configuration will return
excellent science outcomes for ASKAP for at least the
first five years of its operation.
To achieve this, the locations of 27 antennas were
optimized using AntConfig [19] to produce a Gaussian
distribution of visibilities with a scale of 700 m,
corresponding to a point spread function of 3000 . Three
antennas were then added to the core of the configuration
to provide short spacings (20–100 m) to enhance the low
surface brightness sensitivity. Lastly, six antennas were
arranged to form a Reuleaux triangle with a maximum
separation of 6 km. The layout of the 36 antennas is
shown in Fig. 9. The configuration has a smallest
IV. MURCHISON RADIO-ASTRONOM Y
OB SERVATORY
A site in the remote outback of Western Australia is
being developed as the Murchison Radio-astronomy
Observatory. Due to its very low population density (the
Murchison Shire comprises an area of about 50 000 km2
with a population of about 100 people), the area has very
low intrinsic RFI (Fig. 11). The low population density is
expected to remain, and protections are being put into
Fig. 9. Layout of the 36 antennas of the initial ASKAP configuration.
The circle has a radius of 1 km.
1517
Fig. 10. Sensitivity as a function of the full-width half-maximum of the
point-spread function for the ASKAP configuration compared to
natural weighting (100%). Left of the dashed vertical line applies
Gaussian tapering to the data, and to the right robust weighting was
applied.
place to protect the pristine RFI environment into the
future. These protections cover both intentional and unintentional radiation. In addition to housing ASKAP, other
international projects are deploying or planning to deploy
on the site.
The MRO is within the Boolardy Pastoral Lease, a
3467 km2 (856 833 acre) pastoral station. It is 305 km
cross-country northeast of the town of Geraldton on the
Western Australian mid-north coast, and the closest townships are those of Cue (population 273) and Meekatharra
(population 798), which are 150 and 190 km, respectively,
from the core of the array. The station is WA crown land
leased for pastoral purposes, and CSIRO will hold the lease
and operate the bulk of the property as a pastoral lease
consistent with best radio-quiet practices. The Boolardy
homestead will serve as the accommodation site for the
operation of the telescope. The MRO itself will be excised
Fig. 11. Typical radio-quiet spectrum at the ASKAP site. It shows the
maximum of the median values for a series of scans over several
months.
1518
from the pastoral property and be leased by CSIRO from the
State of Western Australia. Fig. 12 shows the Boolardy
property with an inset of the MRO and a rendering of some
of the dishes on the site.
The ASKAP site is in an area classified as desert
according to the modified Köppen system. The average
annual rainfall at Murchison is 216 mm and the highest
recorded daily maximum is 128 mm. Most rain falls in the
months of January to July. At all times of the year, the
rainfall occurs over short periods of typically 0.5–2 h.
September is particularly dry. The air temperature
extremes from December 1993 to April 2007 were
5–45 C. The site has very low water vapor content in the
airVthe relative humidity averages well below 50%.
The Australian continent has been classified into Bwind
regions[ for the purpose of structural design (from
Australian Standard AS1170.2). The location for the ASKAP
site falls into Wind Region A, which is the most benign
classification possible.
The predominant geological feature of the MRO/
ASKAP site is Archaen granite from the Yilgarn Craton raft
of the continental crust. The generally subdued topographical relief in the area results in slow runoff rates; and,
with the low rainfall, the erosion patterns have produced
very shallow alluvial valleys throughout the region. The
mainly pedogenic soils in the area are typically between
one and 10 m deep and overlie calcerous hardpan and then
saprolytic clays to a further depth of up to 40 m. This thick
saprolytic clay profile is derived from in-situ weathering of
the granite basement.
The degree of seismic activity in the vicinity of the
central region is low, with only one earthquake of significance since 1820. Australia is generally subject to very
low seismic activity.
A. Midwest Radio Quiet Zone
On September 24, 2006, ACMA released a Radiocommunications Assignment and Licensing Instruction,
Coordination of Apparatus Licences within the Midwest
Radio Quiet Zone (RQZ) (RALI MS32). RALI MS32
defines the RQZ as inner restricted zones where new
frequency assignments are not usually permitted (with
exceptions assessed on a case by case basis) and outer
coordination zones where new frequency assignments
require coordination. The frequency span of RALI MS32
and the RQZ is 100 MHz to 25.25 GHz. The circles in
Fig. 12 denote the outer extents of the restricted zone
(150 km for low frequencies) and coordination zone
(260 km for low frequencies).
The RALI MS32, along with the extremely low population levels (for instance, there are no residents within
30 km and only a few within 70 km) and other regulations
(for instance, the Mineral Management Area WA state
regulation for nonlicensed transmissions within 80 km and
a Bsection 19[ declaration embargoing additional mining
in the area) promise to make the MRO an excellent radio-
Fig. 12. Map of Australia showing the MRO with an aerial picture of the site and a rendering of a few dishes (credit: University of
Western Australia). In the middle inset, the concentric circles denote 150 and 260 km radii corresponding to the restricted and
coordination zone extents. The grey shape in the center is the pastoral lease.
quiet site for the long-term. Additional legislative
measures are also being investigated to strengthen the
radio-quiet controls.
B. Other Current Projects
As an excellent RFI-quiet site that is starting to acquire
the infrastructure of a supporting observatory, the MRO is
of interest for use by other instruments. This section
briefly describes the other instruments or experiments that
are currently deploying on site.
1) Murchison Widefield Array (MWA): The MWA [24] is a
joint project between the Massachusetts Institute of
Technology Haystack Observatory and Kavli Institute for
Astrophysics and Space Science, the Harvard-Smithsonian
Center for Astrophysics (CfA), the Indian Raman Research
Institute (RRI), an Australian consortium of universities
and CSIRO to build a radio astronomy array operating in
the frequency range 80–300 MHz. Its main scientific goals
are to detect the Epoch of Reionization, study the
heliosphere/ionosphere, and search for transients.
2) Precision Array to Probe the Epoch of Reionization
(PAPER): PAPER is a collaboration between the University
of California at Berkeley, the U.S. National Radio
Astronomy Observatory (NRAO), and the University of
Virginia [25]. It uses sleeved dipoles on a ground plane and
strives for a clean and stable system to detect the faint
signature of the epoch of reionization. PAPER uses a
phased approach that allows it to fully characterize and
control the systematics in order to minimize such effects
for the full array. PAPER will have a maximum baseline of
about 300 m and may comprise up to 256 or more
elements as the experiment evolves.
3) Cosmic Reionization Experiment (CORE): CORE plans
to use a single well-characterized wide-field log-spiral
pyramidal antenna to measure the global spectral features of red-shifted 21 cm emission/absorption over 100–
200 MHz and thus study the astrophysics during the
epoch of reionization over 6 G z G 13.
V. CONCLUSION
As shown in Fig. 2, the project is still being developed and
deployed. Although, given the innovative nature of the
design, risks are still present, excellent progress is being
made across the technical domains. CSIRO is working with
groups around the world to develop the technology and
techniques for the PAFs. Similarly for the challenging
computing and processing issues that are being facedVthis
also involves many industry partners who are chasing
Moore’s law. The remote nature of the site also clearly
involves challenges.
Currently, a prototype PAF, consisting of 5 4 2
elements, illustrated in Fig. 4, has been installed at the focus
of a 12 m parabolic dish installed at the Parkes Testbed
Facility (PTF). The PTF is adjacent to the Parkes 64-m
antenna at the Observatory. The 5 4 array of dualpolarization elements has 40 ports, and data from all ports
1519
(with interferometric measurements from the 64 m) have
been taken. Initial results were provided in [26], which used
a limited number of ports. These data indicate a system
temperature/antenna efficiency ðTsys ="Þ of about 180 K,
implying about 70 K for the system temperature for this
preliminary room-temperature system. The limited number
of ports implies the low efficiency, which also increases the
spillover temperature. A goal of Tsys =" ¼ 65 K or better is
expected.
Fig. 13(a) shows the port powers (autocorrelations) for
each of the 40 ports as a GPS satellite drifted through the
pointing center of the stationary antenna. The receiver,
with an effective bandwidth of 0.875 MHz, was tuned to
the GPS L2 frequency, which is 1.2276 GHz.
Beams were formed in software as linear combinations of
the recorded port voltages. Fig. 13(b) shows the power
patterns of three aperture fit beams made at different epochs
during the transit of the GPS satellite through the pointing
center of the stationary antenna. These three beams cover 4
on the sky when measured across half-power points. The
beam shape was optimized by iteratively solving for weights
that synthesized an aperture illumination closest to a target,
Gaussian-shaped, illumination pattern.
In addition to weights that optimized the beam shape,
two other sets of weights were derived: to maximize
received power and to maximize signal-to-noise ratio.
Fig. 13(c) and (d) compares beams formed by the three
beamforming methods. Maximum power weights were
Fig. 13. (a). Power patterns for each of the 40 ports measured by letting a GPS satellite drift through the pointing center of the stationary
antenna. Each port power was normalized to its noise power, and the baseline has been removed. (b). Power patterns for three beams using
aperture fit weights that are derived for three points along the trajectory of the GPS satellite. Each beam is a linear combination of the signals
plotted in (a). Each beam power has been normalized to the beam noise power, and the baseline has been removed. These three beams cover 4 on
the sky (c). Power patterns for beams comparing three beamforming methods. These beams are all directed towards the antenna pointing center.
(d). Repeat of (c) on a decibel scale.
1520
derived as the eigenvector of the covariance matrix of the
port voltages with the satellite at the pointing center.
Maximum signal-to-noise weights were derived by combining the maximum power weights with an additional
measurement of the noise covariance matrix when the
satellite was far from the pointing center and not contributing significantly to the port voltages.
Additional testing and prototyping of all components
will continue. The Parkes Testbed Facility will continue to
test PAF and digital processing prototypes in preparation
for the upcoming phases of BETA to ASKAP. The com-
REFERENCES
[1] C. Carilli and S. Rawlings, Eds., Science With
the Square Kilometre Array. Amsterdam,
The Netherlands: Elsevier, 2004.
[2] P. J. Hall, Ed., The Square Kilometre Array:
An Engineering Perspective. Dordrecht,
The Netherlands: Springer, 2005.
[3] S. Johnston, M. Bailes, N. Bartel, C. Baugh,
M. Bietenholz, W. C. Blake, R. Braun,
J. Brown, S. Chatterjee, J. Darling, A. Deller,
R. Dodson, P. G. Edwards, R. Ekers,
S. Ellingsen, I. Feain, B. M. Gaensler,
M. Haverkorn, G. Hobbs, A. Hopkins,
C. Jackson, C. James, G. Joncas, V. Kaspi,
V. Kilborn, B. Koribalski, R. Kothes,
T. L. Landecker, E. Lenc, J. Lovell,
J.-P. Macquart, R. Manchester, D. Matthews,
N. M. McClure-Griffiths, R. Norris, U.-L. Pen,
C. Phillips, C. Power, R. Protheroe, E. Sadler,
B. Schmidt, I. Stairs, L. Staveley-Smith, J. Stil,
R. Taylor, S. Tingay, A. Tzioumis, M. Walker,
J. Wall, and M. Wolleben, BScience with
the Australian square kilometre array
pathfinder,[ Pub. Astron. Soc. Australia,
vol. 24, p. 174, 2007.
[4] S. Johnston, R. Taylor, M. Bailes, N. Bartel,
C. Baugh, M. Bietenholz, C. Blake, R. Braun,
J. Brown, S. Chatterjee, J. Darling, A. Deller,
R. Dodson, P. Edwards, R. Ekers, S. Ellingsen,
I. Feain, B. Gaensler, M. Haverkorn,
G. Hobbs, A. Hopkins, C. Jackson, C. James,
G. Joncas, V. Kaspi, V. Kilborn, B. Koribalski,
R. Kothes, T. Landecker, A. Lenc, J. Lovell,
J.-P. Macquart, R. Manchester, D. Matthews,
N. McClure-Griffiths, R. Norris, U.-L. Pen,
C. Phillips, C. Power, R. Protheroe, E. Sadler,
B. Schmidt, I. Stairs, L. Staveley-Smith, J. Stil,
S. Tingay, A. Tzioumis, M. Walker, J. Wall,
and M. Wolleben, BScience with ASKAP,
the Australian square kilometre array
pathfinder,[ Exper. Astron., vol. 22,
pp. 151–273, 2008.
[5] L. Ball, N. McClure-Griffiths, R. Braun,
P. Edwards, I. Feain, G. Hobbs, and
S. Johnston, 2008. [Online]. Available: http://
www.atnf.csiro.au/observers/planning/
priorities/ATNF_Science_Priorities_v1.0.pdf
puting challenges will continue to be addressed by prototyping scaleable algorithms and hardware. Additional
developments will continue to address scaling ASKAP into
the SKA, and the CSIRO is fully involved in the European
Union Framework 7 BprepSKA[ program and will continue
developments leading towards the SKA. h
Acknowledgment
The authors acknowledge the Wajarri Yamatji people as
the traditional owners of the observatory site.
[6] J. Cordes, BThe SKA as a radio synoptic survey
telescope: Widefield surveys for transients,
pulsars and ETI,[ SKA Memo 97, Jul. 2007.
[7] A. P. Chippendale, T. M. Colegate, and
J. D. O’Sullivan, BSKAcost: A tool for ska cost
and performance estimation,[ SKA Memo 92,
Jun. 2007.
[8] S. G. Hay, J. D. O’Sullivan, J. S. Kot,
C. Granet, A. Grancea, A. R. Forsyth, and
D. H. Hayman, BFocal plane array
development for ASKAP (Australian SKA
Pathfinder),[ in Proc. EUCAP’07, Nov. 2007.
[9] S. G. Hay, J. D. O’Sullivan, J. S. Kot, and
C. J. Granet, BScan impedance behaviour of
some connected arrays with transmission-line
feeds,[ in Proc. EMTS Int. URSI Commission
BVElectromagn. Theory Symp., Ottawa, ON,
Canada, Jul. 26–28, 2007.
[10] S. A. Jackson, BRF design of a wideband
CMOS integrated receiver for phased array
applications,[ Exper. Astron., vol. 17, no. 1–3,
pp. 201–210, 2004.
[11] J. D. Bunton, BSKA correlator advances,[
Exper. Astron., vol. 17, no. 1–3, pp. 251–259,
Jun. 2004.
[12] S. K. Okamura, Y. Chikada, M. Momose, and
S. Iguchi, BFeasibility study of the enhanced
correlator for 3-way ALMA I,[ ALMA
Memo 350, Feb. 2001.
[13] J. D. Bunton, ‘‘Multi-resolution FX correlator,’’
ALMA Memo 447, Feb. 2003.
[14] D. R. DeBoer, R. Ackermann, L. Blitz,
D. Bock, G. Bower, M. Davis, J. Dreher,
G. Engargiola, M. Fleming, G. Keleta,
G. Harp, J. Lugten, J. Tarter, D. Thornton,
N. Wadefalk, S. Weinreb, and W. J. Welch,
Exper. Astron., vol. 17, no. 1–3, pp. 19–34,
2004.
[15] R. J. Beresford, BASKAP photonic
requirements,[ IEEE Microw. Photon., 2008.
[16] T. J. Cornwell, K. Golap, and S. Bhatnagar,
BW projection: A new algorithm for wide field
imaging with radio synthesis arrays,[ in Proc.
Astron. Data Anal. Software Syst. XIV, 2005,
vol. 347, p. 86.
[17] W. Gropp, E. Lusk, and A. Skjellum,
Using MPI, 2nd Edition: Portable Parallel
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
programming With the Message Passing
Interface. Cambridge, MA: MIT Press,
1999, 395 pp.
N. Gupta, S. Johnston, and I. Feain, BASKAP
array configurations: Technical studies,[
ATNF SKA Memo 16, 2008.
M. de Villiers, BInterferometric array layout
design by tomographic projection,[ Astron.
Astrophys., vol. 469, p. 793, 2007.
L. Staveley-Smith, BHI surveys and the xNTD
antenna configuration,[ ATNF SKA
Memo 006, 2006.
L. Staveley-Smith, BResolution and confusion
in ASKAP HI surveys,[ ATNF SKA
Memo 020, 2008.
J. Condon, BASKAP resolution for
continuum surveys,[ ATNF SKA
Memo 015, 2008.
D. Briggs, BHigh fidelity deconvolution of
moderately resolved sources,[ Ph.D.
dissertation, 1995.
J. D. Bowman, D. G. Barnes, F. H. Briggs,
B. E. Corey, M. J. Lynch, N. D. R. Bhat,
R. J. Cappallo, S. S. Doeleman, B. J. Fanous,
D. Herne, J. N. Hewitt, C. Johnston,
J. C. Kasper, J. Kocz, E. Kratzenberg,
C. J. Lonsdale, M. F. Morales, D. Oberoi,
J. E. Salah, B. Stansby, J. Stevens, G. Torr,
R. Wayth, R. L. Webster, and J. S. B. Wyithe,
BField deployment of prototype antenna tiles
for the mileura widefield array low frequency
demonstrator,[ Astron. J., vol. 133, no. 4,
pp. 1505–1518, 2007.
D. C. Backer, A. Parsons, R. Bradley,
C. Parashare, N. Gugliucci, E. Mastrantonio,
D. Herne, M. Lynch, M. Wright,
D. Werthimer, C. Carilli, A. Datta, and
Aguirre, BPAPER: A precision array to
probe the epoch of reionization,[ Bull.
Amer. Astron. Soc., vol. 39, p. 967, 2007.
J. D. O’Sullivan, F. Cooray, C. Granet,
R. Gough, S. Hay, D. Hayman, M. Kesteven,
J. Kot, A. Grancea, and R. Shaw, BPhased
array feed development for the Australian
SKA pathfinder,[ in Proc. Int. Union Radio Sci.
Abstracts 29th General Assembly,
Aug. 2008, p. 38.
1521

A High-Dynamic Range Wide-Field of View Survey Telescope

Transcription

Similar documents

TET HB35T antenna instructions manual

“Salt-Shaker” 900MHz Omni Antenna

Data Challenges for Next-generation Radio Telescopes

NEC Tutorial

EtherHaul™-2200F Datasheet

Owner`s Manual

cobra 142gtl om parts list

Owner`s Manual