Design of an Aperture Phased Array System for the SKA
A. J. Faulkner1, P. Alexander2, M. E. Jones3, R. Bolton2, A. van Ardenne4, S. Torchinsky5
1
Jodrell Bank Observatory, The University of Manchester, Macclesfield, Cheshire, SK11 9DL, UK.
[email protected]
2
3
University of Cambridge, Cavendish Laboratory, Cambridge CB3 0HE, UK.
[email protected];
[email protected]
Astrophysics, Denys Wilkinson Building, Keble Road, Oxford, OX1 3RH, UK.
[email protected]
4
5
ASTRON, P.O. Box 2– 7990 AA Dwingeloo, The Netherlands.
[email protected]
Observatoire de Paris, Carte du Ciel, 61 avenue de l’Observatoire, 75014 Paris.
[email protected]
Abstract
Aperture phased arrays operating up to 1 GHz are highly flexible collector systems, uniquely capable of performing
an HI survey of a billion galaxies and observations of the Epoch of Reionisation, two of the Key Science
Programmes for the Square Kilometer Array, SKA. Such high performance arrays are only becoming feasible in the
timescales of the SKA due to technological advances in array design, low noise amplifier implementations, and
increased processing and communications speed. In this paper we discuss the scientific benefits of an aperture
phased array system operating from 70 MHz to 1,000 MHz, using a mixture of sparse and dense arrays and show
their implementation as part of the SKA is achievable and highly desirable.
1.
Introduction
A radio telescope with close to a square kilometre of collecting area used to survey red-shifted hydrogen emission to
the earliest cosmological epochs has long been an aspiration of radio astronomers, see e.g. Wilkinson [1]. The
preliminary specification of the Square Kilometre Array, SKA, as described by Schilizzi et al. [2], is expanded in
frequency coverage and angular resolution, over the original concept, to cover many other scientific observation
requirements, see Carilli et al. [3]. This instrument can be considered to be a physics instrument exploring
fundamental scientific questions. High observation speed is a critical requirement for hydrogen surveys, to keep the
time required to a few years. This level of SKA performance would be unaffordable if it was built as a typical radio
telescope, however, the progress of high performance data processing, wide-bandwidth communications and lownoise, ambient temperature amplifiers makes the prospect of a telescope using almost entirely digital signal
processing a real possibility.
The SKA Design Studies, SKADS [4], a European Community (EC) Framework Programme 6 project, has the task
of producing a costed SKA design. Most of the technical work is in the development of a 300 – 1,000 MHz, aperture
phased array capable of meeting the performance requirements of the SKA. While this work will evolve until the
end of SKADS in June 2009, early design and cost modelling has been published (Alexander et al. [5] and Bolton et
al. [6]). This shows that by using the benefits of the aperture array at a low RFI SKA site and fully integrating it
with high frequency dishes, it is possible to design a practical SKA. The development of high frequency aperture
arrays will continue up to the start of SKA construction in an EC programme called PrepSKA (prepare for SKA).
2.
Aperture phased array background
An aperture array, AA, is a large number of small, fixed antenna elements plus receiver chains which can be
arranged in a regular or random pattern on the ground. A beam is formed and steered by combining all the received
signals after appropriate time delays for phase alignment, this can be repeated concurrently many times to create
many simultaneous independent beams, yielding a very large total Field of View, FoV. The number of useful beams
produced, and hence total FoV, is essentially limited by signal processing, data communications and computing
capacity. Aperture arrays can readily operate at low frequencies with large effective areas, unlike dish based
systems, where performance falls off when the dish diameter is only a few wavelengths. AAs using substantial
digital processing systems are an inherently very flexible collector technology, since the system can ‘trade’ FoV,
bandwidth and number of bits per sample, consequently the performance can be matched to that required by the
experiment. It is also possible to tailor the processed FoV as a function of frequency: for example, for the HI
experiment an FoV increasing substantially faster than the λ2 (as for a single feed dish) is key to obtaining a
reasonable survey speed.
Inherently, there are two basic configurations of AA, close packed (dense) and sparse, which are discussed in Braun
& van Cappellen [7]:
•
A dense array samples the incoming wavefront at least at the Nyquist rate by having elements spaced ≤ λ/2.
As the frequency reduces the array oversamples the wavefront resulting in the Aeff remaining roughly
constant. The benefit of this fully sampled system is that there can be very tight control of the beams, with no
array artefacts introduced. This type of array has the highest dynamic range capability of AAs.
•
A sparse array, as its name implies, has elements spaced further apart than λ/2. In the limit, each element can
act independently and provide an element level Aeff which scales as λ2. This is of great benefit, particularly at
frequencies below ~500MHz where sky noise becomes dominant. The increasing Aeff increases the sensitivity
and hence survey speed with increasing redshift (as does a single pixel feed on a dish), which helps
counteract the decreasing flux density from the sources. In an interferometer such as the SKA, it is likely that
a sparse array will be the preferred solution at frequencies <500 MHz.
3.
SKA array selection
As part of a costed design, SKADS has proposed an AA system for use in the SKA. It consists of two collecting
arrays: AA-lo a sparse array using wideband feeds, probably log-periodic, which operate from 70MHz to 450MHz
and a dense array AA-hi which operates between 300MHz and 1GHz, close packed up to ~800MHz. Between
800MHz and 1GHz the Aeff falls off since the array is becoming sparse. This is to minimise the total element count
and, hence, cost and power requirements.
The
system
characteristics
are
shown in Figure 1. The
Sparse AA-lo
blue lines show the
effective areas of the
Fully sampled AA-hi
arrays. The sparse AAlo can be seen to have
100
increasing an Aeff as the
Aeff
Tsky
frequency is reduced
Becoming sparse
which
is
almost
above fAA
offsetting the increasing
sky noise, Tsky, derived
from Medellin [8], also
10000
10
shown in Figure 1. The
AA-hi has a roughly
5000
constant Aeff from its
Aeff/Tsys
lower
operating
frequency of 300MHz
up to where the array
1
1000
stops Nyquist sampling
100
500
1000
3000
the incoming signal at
Frequency (MHz)
800MHz.
Above
Figure 1: Characteristics of an SKA capable aperture array system
800MHz the Aeff for
AA-hi falls as λ2. Both arrays can be used in the overlap frequency range from 300 – 450 MHz: this has the effect of
roughly maintaining the total system sensitivity where the sky noise is starting to increase. The resulting sensitivity
Aeff/Tsys, is shown in the bottom orange line.
Aeff / Tsys (m2 / K)
Sky Brightness Temperature (K)
1000
4.
Scientific Rationale
A key science driver for the SKA, indeed the original concept for the SKA, is an HI survey for galaxy evolution and
dark energy. This experiment detects the faint emission at 1.421 GHz from neutral hydrogen. Due to Doppler
frequency shift with increasing distance requires coverage from below 1400 MHz to <500MHz.
Detecting
redshifted HI becomes increasingly challenging as we move out in z to ~ 3-4. To complete a survey to a specified
galaxy mass limit requires an increasing survey speed with increasing z. Spectral requirements and baselines for
these experiments are relatively modest, this is a detection experiment in survey mode, so we only need to
adequately sample the HI line to avoid confusion in position-velocity space. As discussed above this can be
implemented with an AA, the survey can be completed in a matter of years, compared to decades for any other
technology.
The parameter that has been largely unexplored in the radio spectrum is the time domain for short, transient signals.
Ideally, we would like to monitor as much of the sky as possible for unexpected events. Whenever this has been
improved before new phenomena have been discovered, e.g. pulsars. There are many configurations that could be
investigated, a large FoV and relatively low sensitivity, scanning with higher sensitivity, various frequency ranges
etc. The AA can uniquely be configured for these ‘exploration of the unknown’ surveys.
5.
Implementation Considerations
The SKA configuration is still the subject of considerable study and simulation, however the general layout is
anticipated to have about 50% of the collecting area in a concentrated core of ~5 km diameter and the rest of the
collectors organised as stations spread logarithmically along maybe five spiral arms. The wide field collectors will
have restricted maximum baselines of ~180 km. There is a trade-off between size and number of the arrays to give
the total collecting area. To give excellent
u-v plane coverage we anticipate using ~250
arrays, each AA-hi having a diameter of
~60 m and the AA-lo diameter of ~200 m.
Here we will concentrate on the design of
the AA-hi, illustrated in Figure 2, which is
the most critical in terms of total number of
elements to be integrated and the
bandwidths involved. Each array consists of
75,000 dual polarisation elements, making
150,000 receiver chains. The design of the
elements and the details of the overall array
are the subject of ongoing research. There
Figure 2: Cutaway of high-frequency AA
are a number of different element types
being considered: Vivaldi, Bunny Ear Comb Array, Munk-Checkerboard etc, while these designs are being
optimised, we already have suitable solutions available. A major cost driver is the element spacing, currently
expected to be ~21cm. The receiver noise is critical to the system sensitivity. Due to the very high number of low
noise amplifiers (LNAs) dispersed over a wide area, it is not considered viable to cryogenically cool them, although
it will be necessary to stabilise their temperature. Hence, technologies are being explored to develop high
performance, room temperature devices which can be matched over a very wide fractional bandwidth to the antenna
elements. Substantial progress is being made in this area and we anticipate overall Tsys at 800MHz to be ≤ 50 K for
Phase 1 SKA and ≤ 37 K for Phase 2.
Georgina Harris
Figure 3 shows the outline system design of the AAs. In essence, the receiving elements are positioned as required
with local matching and amplification, the received signal is then carried at baseband over copper links using
commodity Category-7 (CAT7) networking cables to screened processing areas. Within the screened areas all the
digitisation and beamforming is carried out. A key constraint is to keep all digital signals within their screened
environments, and have only analogue signals close to the array itself. This is primarily for self-induced RFI
mitigation, but it also means that the complete received signal is available to the processing system, which enables
an upgrade path. Further, the complex electronics are mounted in conventional racking systems, which can be watercooled for good temperature stability, leading to lower power requirements and improved reliability.
The processing is in a two
Next Proc. Bunker
n x Optical
stage structure, the first stage
fibres per 2
Bunker-n
RFI Barrier
stage processor
processors perform the initial
Station
digitisation and beamforming
1 Stage Processors
Processor 1
on ‘tiles’ of 256 dual
Mid P1
polarisation elements, the
Mid P2
AA-hi
number of beams, their
Station
Processor 2
Mid Py
frequencies plus bandwidth
0.3-1.0GHz Analog links
and bits per sample can be
CAT7
configured at the observation
AA-lo
time. The constraint is the
Internal
Station
Digital links
Processor X
total data rate of the internal
Box
digital links. Identical beams
O-E
Low P1
Phase
Standard &
from all the tiles in each array
Distribution
are then combined in the
O-E
Low Pz
Control
To Central
‘Station Processors’ to produce 500MHz Analog links
processors
Control
the required station beams. To
10Gb Digital
Digital fibre links
fibre links
achieve the total FoV needed
Prev. Proc. Bunker
there are many hundreds of
individual
station
beams. Figure 3: AA system diagram
Every first stage processor has a link to all the Station Processors. The Station Processors each link to wide area
communication fibres directly to the central correlator. Again, the overall array performance is constrained by the
total bandwidth to the correlator, which can be flexibly allocated for an observation. The signal processing, which
will be using integer arithmetic for efficiency in power and silicon, requires a capability estimated for an AA-hi
array of ~10 PetaMACs (1016 multiply-accumulate instructions per second). It has become clear that this
performance level is achievable for the start of SKA Phase 1 in 2012 using either a dedicated ASIC solution, or
massively multi-core integer processors. This latter processing solution is very attractive for flexibility and the
implementation of novel algorithms.
nd
st
…
To Correlator
.
.
.
.
.
...
..
Phase transfer
over fibre
Very high dynamic range observations are essential to the SKA. This AA implementation gives an unblocked
aperture with a very flexible processing system capable of applying arbitrarily accurate and fine calibration
adjustments. The research is into measuring the required coefficients and uniquely the AA can continuously observe
multiple astronomical calibration sources during observations for on-line adjustments.
6.
Conclusions
Using the base of rapidly advancing digitisation and processing technologies and improving the other components
required, the design and build of a very high performance aperture array system can be realistically be achieved in
the timescale of the SKA. The availability of SKA-performance aperture arrays will enable important scientific
experiments to be carried out which will lead to significant discoveries in fundamental physics.
7.
References
1. P.Wilkinson, "The Hydrogen Array" IAU Colloquium 131, ASP Conference Series,
Vol. 19, 1991, T.J. Cornwell and R.A. Perley (eds.), pages 428-432
2. Schilizzi R. et al., “Preliminary Specifications for the Square Kilometre Array”, 2007, www.skatelescope.org
3. Carilli C. et al., "Science with the Square Kilometre Array", New Astronomy Reviews, Vol.48, Elsevier,
December 2004
4. “Square Kilometre Array Design Studies, SKADS”, www.skads-eu.org
5. Alexander P. et al., “SKA Memo 93: SKADS Benchmark Scenario Design and Costing”, 2007,
www.skatelescope.org
6. Bolton, R. et al., “ SKADS Benchmark Scenario Design and Costing - 2”, in press
7. Braun R. & Cappellen W. , “SKA Memo 86: Aperture Arrays for the SKA: Dense or Sparse?”, 2006,
www.skatelescope.org
8. Medellin G.C. “SKA Memo 95: Antenna Noise Temperature Calculation”, 2007, www.skatelescope.org