Low-Frequency Aperture Array (LFAA)

Transcription

Technical Description
Low Frequency Aperture Array
Reference
:
AADC-TEL.LFAA.SE.MGT-AADC-PL-002
Issue - Revision
:
1-0
Date
:
2013-06-08
Author(s)
:
A.J. Faulkner
Checked
:
J.G. bij de Vaate
Approval
:
Authorisation
:
J.G. bij de Vaate
Customer Approval
:
N.A.
M. Gerbers
M. Gerbers
-
Reference
Issue – Revision
Date
: AADC-TEL.LFAA.SE.MGT-AADC-PL-002
Page 1 of 56
: 1-0
: 2013-06-08
RESTRICTED
© 2013 AADC Consortium
This document shall only be reproduced for the agreed purpose for which it has been supplied.
Record of Changes
Issue
Rev
Date
Change
Affected Area
Author
0
A
2013-05-06
Initial version
Entire document
A. J. Faulkner
0
B
2013-05-13
Updated version comments
from Marchel Gerbers, Peter
Hall and Jan Geralt Bij de Vaate
Entire Document
A. J. Faulkner
0
C
2013-05-17
Updated version after
discussion on document review
and input from other people
Entire Document
A. J. Faulkner
Entire Document
A. J. Faulkner
0
D
2013-06-02
Updated version after
discussion on document review
and input following Consortium
meeting
0
E
2013-06-04
Updates from Peter Hall
Entire Document
A. J. Faulkner
0
F
2013-06-04
Station processing mod
Section 5.1.2.1
A. J. Faulkner
0
G
2013-06-08
Template and distribution list
Distribution list
M. Gerbers
1
0
2013-06-09
Formal release
Entire document
A. J. Faulkner
Reference
Issue – Revision
Date
Page 2 of 56
: 1-0
: 2013-06-08
RESTRICTED
Table of Contents
1
Introduction ................................................................................................................................................... 7
1.1
1.2
2
Applicable and Reference Documents............................................................................................................. 8
2.1
2.2
3
Applicable Documents........................................................................................................................... 8
Reference Documents ........................................................................................................................... 8
LFAA description ............................................................................................................................................ 9
3.1
4
Scope .................................................................................................................................................... 7
Document Structure.............................................................................................................................. 7
LFAA Concept.......................................................................................................................................10
LFAA science considerations ..........................................................................................................................12
4.1
EoR Science..........................................................................................................................................13
4.2
High Band science ................................................................................................................................14
4.2.1 Pulsars .............................................................................................................................................15
4.2.2 High band imaging ...........................................................................................................................15
4.2.3 Observing flexibility..........................................................................................................................16
4.3
Section references ...............................................................................................................................16
5
Detailed technical description .......................................................................................................................17
5.1
Implementation options .......................................................................................................................19
5.1.1 LFAA design choices already made ...................................................................................................19
5.1.2 Ongoing design options for Stage 1 decisions ...................................................................................20
5.1.3 Existing experience ..........................................................................................................................21
5.1.4 LFAA design presented in this document ..........................................................................................23
5.2
Antenna and LNA .................................................................................................................................23
5.3
Power for the LNA and antenna receiver ..............................................................................................25
5.4
Receiver ...............................................................................................................................................26
5.4.1 Antenna front-end ...........................................................................................................................26
5.4.2 Signal Transport ...............................................................................................................................26
5.4.3 Analogue processing in the bunker...................................................................................................27
5.5
Digital Signal Processing .......................................................................................................................27
5.5.1 Tile processing .................................................................................................................................27
5.5.2 Station beamforming .......................................................................................................................29
5.6
Implementation of bunker receiver and tile processor ..........................................................................29
5.6.1 Tile Processing .................................................................................................................................31
5.6.2 Tile output data rate ........................................................................................................................32
5.7
Station processing ................................................................................................................................32
5.8
Processing Rack - construction .............................................................................................................33
5.9
Monitoring and Control ........................................................................................................................35
5.10
Bunker system and SKA implementation ..............................................................................................36
5.10.1
Integration with LFAA correlator and post processing...................................................................36
5.11
Transient buffering...............................................................................................................................37
5.11.1
Element level data storage...........................................................................................................37
5.11.2
Tile level data storage ..................................................................................................................37
5.11.3
Station beam storage ...................................................................................................................37
Reference
Page 3 of 56
Issue – Revision
: 1-0
Date
: 2013-06-08
RESTRICTED
5.11.4
Dedicated storage nodes .............................................................................................................38
6
Risk mitigation ..............................................................................................................................................38
7
Cost estimation .............................................................................................................................................39
7.1
Relative costing between Baseline and Proposed Designs .....................................................................42
7.2
Cost discussion.....................................................................................................................................42
7.3
Consortia interfaces and cost allocations ..............................................................................................43
7.3.1 Buildings ..........................................................................................................................................43
7.3.2 Power distribution ...........................................................................................................................43
7.3.3 Clock distribution .............................................................................................................................44
7.3.4 Telescope manager and TM data transport ......................................................................................44
7.3.5 Data transport to the correlator .......................................................................................................44
7.3.6 Data for correlation..........................................................................................................................44
8
Scaling to SKA Phase 2 ...................................................................................................................................45
8.1
Bunker design issues ............................................................................................................................47
8.1.1 Receiver and tile processor cards .....................................................................................................47
8.1.2 Tile signal Processing........................................................................................................................48
8.1.3 Station beamforming and system data rates.....................................................................................48
8.2
Configuration .......................................................................................................................................49
8.3
Extending the frequency range for SKA2 ...............................................................................................50
Appendix A.
Possible station configuration for SKA1 Core ................................................................................52
Appendix B.
Tile processing and communication requirements........................................................................52
Appendix C.
Bunker power requirement estimation ........................................................................................53
List of tables
Table 1: Outline specifications for SKA-low for the Baseline Design and the Proposed Design ................................18
Table 2: Outline comparison of technologies that could be used for LFAA subsystems, for investigation and downselect during Stage 1 Preconstruction. The example approach adopted in this proposal is shown shaded.......22
Table 3: Costing for Baseline and Proposed Design for LFAA (part 1) ......................................................................39
Table 4: Costing for Baseline and Proposed design for LFAA (part 2 .......................................................................41
Table 5: Outline predicted specification of LFAA, as a simple extension of Phase 1, in SKA2 compared to proposed
SKA1 .............................................................................................................................................................46
Table 6: Approximate tile processing requirements ...............................................................................................53
Table 7: Estimated bunker power requirements ....................................................................................................54
List of figures
Figure 1: Computer-generated visualisation of SKA-low core showing the log periodic antenna prototype. Credit:
Swinburne Astronomy Productions/ICRAR/U. Cambridge/ASTRON ................................................................10
Figure 2 SKA-low layout showing a single processing bunker fed with signals from every individual antenna .........11
Figure 3: The sensitivity of SKA-low over frequency. This is as expected for the Baseline Design, limited to 350MHz
and the sensitivity at higher frequencies is also shown for the Proposed Design, limited to ~650MHz. Note
that there is still good sensitivity even to a scan angle of 60°. ........................................................................14
Reference
Issue – Revision
Date
Page 4 of 56
: 1-0
: 2013-06-08
RESTRICTED
Figure 4: Work structure and signal flow for SKA-low ............................................................................................17
Figure 5: SKA-low test array on the SKA site ..........................................................................................................24
Figure 6: SKA-low processing for the Baseline design.............................................................................................28
Figure 7: Processing for the SKA-low Proposed Design...........................................................................................29
Figure 8: Outline layout of Receiver and Tile Processor. The industry standard SPF module houses the analogue
components for pairs of channels ..................................................................................................................30
Figure 9: Station beamforming accumulation approach (See Figure 11 for illustration of Aisle switch) ...................33
Figure 10: Outline LFAA rack, handling a total of 1024 antennas, using 8 shelves or 8 tile processors linked via two
36-port data switches. The rack shown has the second level, aisle switch ......................................................34
Figure 11: Outline of overall interconnect. Essential features are wide bandwidth into the correlator and the
Monitoring and Control system .....................................................................................................................35
Figure 12: Station beams in a tile beam. Stepped beamforming for off-centre beams on the right. ........................48
Figure 13: Enhanced configuration architecture for SKA2 ......................................................................................50
Reference
Issue – Revision
Date
Page 5 of 56
: 1-0
: 2013-06-08
RESTRICTED
Definition and Acronyms
Acronym
Explanation
AADC
Aperture Array Design and Construction
AD-n
nth document in the list of Applicable Documents
AIV
Assembly Integration and Verification
CE
Conformité Européenne
CIDL
Configuration Item and Document List
DRB
Delivery Review Board
EIDP
End Item Data Package
ENG
Engineering
HW
Hardware
ICD
Interface Control Document
LFAA
Low Frequency Aperture Array
MOM
Minutes of Meeting
PA
Product Assurance
QA
Quality Assurance
RD-n
nth document in the list of Reference Documents
RFP
Request for Proposal
SKA
Square Kilometre Array (www.skatelescope.org)
SKA-low
The low frequency system for the SKA. Includes, LFAA, associated correlator and post processing
SKAO
Square Kilometre Array Office
SOW
Statement of Work
SW
Software
TBC
To be continued
TBD
To be determined
TBS
To be supplied
TRB
Test Review Board
WBS
Work Breakdown Structure
WP
Work Package
Reference
Issue – Revision
Date
Page 6 of 56
: 1-0
: 2013-06-08
RESTRICTED
1
Introduction
1.1
Scope
This is the Technical Description for the Low Frequency Aperture Array, a major part of SKA-low, for
implementation as Phase 1 of the SKA. This covers the Baseline Design as described in the RfP
documentation and a Proposed Design with significantly enhanced performance. It is in response to the
Request for Proposal from the SKA Office dated 11 March 2012.
This description should be read in conjunction with the rest of the AADC consortium response to the
LFAA Element design.
1.2
Document Structure
•
•
•
•
•
•
•
•
Section 1 introduces the purpose and scope of this document.
Section 2 lists the applicable and reference documents.
Section 3 describes the SKA-low concept.
Section 4 presents the SKA-low science with particular reference to the Proposed Design.
Section 5 provides a detailed description of the Low frequency Aperture Array by sub-system.
Section 6 discusses the principal risks, mitigations and consequences.
Section 7 discusses LFAA costing.
Section 8 considers the requirements of scalability to SKA Phase 2.
Reference
Issue – Revision
Date
Page 7 of 56
: 1-0
: 2013-06-08
RESTRICTED
2
Applicable and Reference Documents
2.1
Applicable Documents
Id
Title
Code
Issue
AD-1
SKA Request for Proposals
SKA-TEL.OFF.RFP-SKO-RFP-001
1
AD-2
Statement of Work for the Study, Prototyping and
Design of an SKA Element
SKA-TEL.OFF.SOW-SKO-SOW-001
1
AD-3
Statement of Work for the Study, Prototyping and
Preliminary Design of an SKA Advanced
Instrumentation Programme Technology
SKA-TEL.OFF.AIP-SKO-SOW-001
1
AD-4
SKA Pre construction Top Level WBS
SKA-TEL.OFF.WBS-SKO-WBS-001
1
AD-5
SKA-1 System Baseline design
SKA-TEL-SKO-DD-001
AD-6
The Square Kilometre Array Design Reference
Mission: SKA Phase 1
SCI-020.010.020-DRM-002
3
AD-7
SKA System Engineering Management Plan
SKA.TEL.SE-SKAO-MP-001
1
AD-8
The Square Kilometre Array Intellectual Property
Policy
SKA IP Policy
1.3
Draft
AD-9
Draft Consortium Agreement
PD/SKA.26-4
Draft
AD-10
Document Requirements Description
SKA-TEL.SE-SKO-DRD-001
1
AD-11
SKA Document Management Plan
SKA-TEL.OFF.MGT-SKO-MP-001
1
AD-12
SKA Product Assurance and Safety Plan
SKA-OFF.PAQA-SKO-QP-001
1
AD-13
Change Management Procedure
SKA-TEL.SE.CONF-SKO-PR-001
1
AD-14
SKA Interface Management Plan
SKA-TEL.SE.INTERF-SKO-MP-001
1
2.2
Reference Documents
Id
Title
Code
Issue
RD-1
LFAA Delivered Items and Document List
AADC-TEL.LFAA.SE.MGT-AADC-PL002
1.0
RD-2
LFAA Management Plan
AADC-TEL.LFAA.SE.MGT-AADC-PL-0021
1.0
RD-3
RD-4
Reference
Issue – Revision
Date
Page 8 of 56
: 1-0
: 2013-06-08
RESTRICTED
3
LFAA description
The low frequency aperture array, LFAA, covers the lowest frequency band for the SKA, from 50MHz up
to 350MHz in the Baseline Design or optionally up to >650MHz at a small additional cost of ~2%. It is an
aperture array consisting of 262,144 wide bandwidth antennas of a single design. The configuration is
very close packed with 75% of the antennas within a 2km diameter core and the remaining collecting
area situated on three spiral arms, extending out to a radius of 50km and enabling higher spatial
resolution observations.
The overall system is organised as logical stations constituting correlatable entities. The Baseline Design
specifies 911 stations of 35m diameter; our "Proposed Design" offers selectable station sizes, as
appropriate to the science with a commensurately varying number of stations.
Reference
Issue – Revision
Date
Page 9 of 56
: 1-0
: 2013-06-08
RESTRICTED
Figure 1: Computer-generated visualisation of SKA-low core showing the log periodic antenna prototype.
Credit: Swinburne Astronomy Productions/ICRAR/U. Cambridge/ASTRON
The principal science specified for SKA-low is the Epoch of Re-ionisation experiment and the Baseline
Design has been optimised for the requirements of this experiment. However, there is considerable
additional science to be addressed with SKA-low; an overview is given in Section 4. To include this
science and provide improved performance for the EoR experiment, the AADC consortium proposes an
extended design that could reduce the overall cost of SKA Phase 1 (aperture arrays plus dishes).
3.1
LFAA Concept
While alternative implementations of the LFAA can be designed, the rest of this description focuses on a
representative implementation to justify our assessment of the viability and costs of the LFAA.
Alternative sub-system implementations are discussed in Section 5.1, some of which will be investigated
during Stage 1 before the final design is tested in AAVS1.
Reference
Issue – Revision
Date
Page 10 of 56
: 1-0
: 2013-06-08
RESTRICTED
The LFAA will be built using log-periodic antennas, which work well over a very large frequency range,
providing a good impedance match to the low noise amplifier. An artist’s impression of part of the core
of the LFAA is shown in Figure 1. Here, the antenna is a log-periodic optimised for the close antenna
spacing required, dual polarisation performance and low cost. Each antenna houses analogue
electronics: low noise amplifier, gain with filtering and communications driver. The signal from each
polarisation of each element is transported to a signal processing site ("bunker") where it is digitised and
processed into beams by combining with other antennas constituting a logical station. Each antenna
requires a small amount of power that can be supplied by local solar power for individual antennas or
groups of antennas, or by a power network using copper wires. This is further discussed in Section 5.3.
Our representative, practical implementation for the LFAA uses these antennas with a combination of
solar power and RFoF. Figure 2 shows an overall LFAA layout which highlights that the system only
requires a single processing facility for all the stations, making a conceptually simple arrangement that is
highly flexible, upgradeable and easily maintained. An important consideration for SKA implementation
is the opportunity to locate the SKA-low correlator in the same processing bunker as the LFAA
beamforming hardware. The architecture shown in Figure 2 is a low-cost evolution of the Baseline
Design, as it redefines the digital signal transport system as digital links across a single bunker, avoiding
the need to use long distance links from individual stations.
Figure 2 SKA-low layout showing a single processing bunker fed with signals from every individual antenna
Local solar power (including cost-effective energy storage), maybe even as one power source per
antenna, provides considerable benefit for the system, not least of which is potentially lower whole-oflife operating costs, since there will be no ongoing requirement to externally supply power to the
antennas, which require about 1 watt each for electronics and optical drive (250kW total). The local
solar power supply can have very low RFI and an integrated design, which can be assembled prior to
deployment, would give a very consistent EM performance for the antenna. The combination of RFoF
Reference
Issue – Revision
Date
Page 11 of 56
: 1-0
: 2013-06-08
RESTRICTED
and local solar power means that there is no galvanic connection throughout the array, potentially
important across an all-electronic, highly distributed telescope. In addition, the infrastructure
installation costs may be considerably reduced, with only optical fibres to install and maintain.
Within the processing bunker the analogue signals are digitised, channelised in frequency and calibrated
into “tiles” of 16 antennas; tile beams are formed using the available bandwidth. The tile beams are
combined into “station” beams of the selected station size and finally passed for correlation. With all
the antenna signals coming to a single bunker it is straightforward to form stations of programmable
size by combining beams from the required number of unit tiles. The stations may overlap for better
beam performance or even support more than one station size concurrently.
4
LFAA science considerations
The Baseline Design for the LFAA recognized the importance of Epoch of Reionization (EoR) science for
SKA1 and proposed a design tailored to this application. The AADC Consortium proposal likewise
recognizes the need to optimise EoR science and offers an extended design, the Proposed Design,
identical to the Baseline Design in terms of sensitivity and angular resolution, with enhanced EoR
performance and significantly wider scientific application. In particular, the AADC Consortium design
gives:
a) greater flexibility in EoR observing techniques and calibration methods, leading to an enhanced
likelihood of a robust EoR detection and a consequentially a large SKA1 scientific impact;
b) greater scientific utility of the SKA-low via:
i.
enhanced frequency coverage,
ii.
simultaneous, parallel observing programs enabled by collecting area re-use (multiple
station beams giving widely-separated fields-of-view), and
iii.
better matching of the instrument performance to specific experiments through
selectable station size.
c) considerably enhanced SKA1 pulsar observation capability for both search and timing of a large
number of objects, to reveal the detailed physics of matter in extreme conditions;
d) access to other leading-edge time-domain science, such as cosmic transients, via highly flexible
signal aggregation and processing architectures, including transients buffers, opening the door
to observations of the most energetic processes in the Universe; and
e) effective demonstration of new calibration and imaging techniques via SKA1 continuum and
spectral line science, enabling much greater use of cost-effective sparse aperture array
technology and representing an important step in demonstrating scalability for SKA Phase2.
While proposing the enhanced LFAA frequency coverage, we keep our focus for all science - EoR and
other - below 350 MHz. Much of the extensive pathfinder science cases, including the comprehensive
cases outlined for LOFAR [4-1] and MWA [4-2], are directly applicable to the LFAA and we note the
exciting experiments being pursued by these telescopes in areas as diverse as very low-frequency pulsar
studies, coronal mass ejections and space weather, advanced ionospheric modelling, and tracking of
space junk.
Reference
Issue – Revision
Date
Page 12 of 56
: 1-0
: 2013-06-08
RESTRICTED
4.1
EoR Science
The need for flexibility in EoR science arises from the currently incomplete state of pathfinding programs
in instruments such as LOFAR, MWA and PAPER. Statistical estimates of the EoR rely on the reduction of
noise to detect the faint (milli-kelvin) cosmological signal. Broadly, noise falls into three categories:
radiometric noise due to sensitivity limitations; spatial sample noise due to sky-coverage limitations; and
cosmic variance remaining after the whole sky has been observed. Most SKA pathfinder EoR
instruments are thermal noise dominated on most spatial scales, and spatial sample variance is a
secondary concern. For the LFAA, exceptional sensitivity will give low radiometric noise, and statistical
measurements are expected to be dominated by spatial sample variance. Spatial sample variance is
reduced by independent observations of different regions of the sky (incoherent averaging of power),
while thermal noise is reduced most effectively by sampling identical sky (coherent averaging of
visibilities). The balance between coherent and incoherent addition of information determines the
balance of thermal and sample variance noise to the overall error budget; this balance is still being
explored.
Small stations (such as in the Baseline Design) yield large angular FoV per beam, providing the sky area
necessary for reducing spatial sample variance. Equivalently, they provide a compact beam in Fourier
space, reducing the coherence of measurements made from differing baselines. This effect is countered
by the increase in number of baselines from a large-N array (for constant total array collecting area).
Conversely, larger stations sample a smaller sky area, with less ability to reduce spatial sample variance.
Large station Fourier beams are less compact, yielding greater coherence from similar baselines, but
fewer independent bins for incoherent addition. The information lost from averaging the sky signal over
a larger station can be recovered effectively using two schemes:
a) drift scans to sample a larger fraction of the sky, reducing spatial coherence of measurements
and improving sample variance reduction - this option used alone comes with the need for extra
observing time per EoR field, the information coming from a single, small, station beam
generated by the larger station; and
b) multiple station beams to observe simultaneously multiple fields across the same frequency
range, recovering and exceeding the equivalent beam of the smaller station - this gives a
substantial reduction of spatial sample variance from the addition of incoherent modes across
all station beams.
In a recent analysis [4-3], which extended earlier work [4-4] examining observing strategy on EoR
detection, the calibratability of the LFAA Baseline Design and designs using 50m and 100m diameter
stations was examined. Smaller stations perform less favourably when radiometric noise is dominant
but in the presence of noise dominated by background sources, all stations produce comparable
measurement precision when one beam is formed. By extension, if larger stations with widelyseparable, independent FoVs (many station beams) are used to give additional area sampling,
calibration precision will not suffer but the EoR signal estimate will be improved via the reduction of
spatial sample variance flowing from simultaneous independent measurements.
The LFAA can have a large processed bandwidth and duplicating bandwidth for an EoR experiment
across multiple FoVs and station beams as in the AADC Consortium Proposed Design gives an instrument
which is robust in the face of current uncertainty concerning thermal noise and spatial sample variance
in practical experiments. The Proposed Design also offers almost the ultimate risk mitigation in that the
Reference
Issue – Revision
Date
Page 13 of 56
: 1-0
: 2013-06-08
RESTRICTED
central LFAA collecting area can be beamformed into stations of selectable sizes. Assuming an
acceptable intra-core antenna configuration can be found - highly likely given the density of antennas in
the core - it will be possible to conduct experiments with a range of station sizes and FoV options.
As a related point, and since the Baseline Design document mentioned mosaicking, we note that
combining multiple station beams (and more widely-separated FoVs) from an aperture array differs in
principle from the classical, dish-based operation; the LFAA beams will be formed simultaneously via the
same receptors and signal paths, potentially significantly improving the wide-field imaging capability for
a range of SKA-low science, including EoR experiments.
4.2
High Band science
Figure 3: The sensitivity of SKA-low over frequency. This is as expected for the Baseline Design, limited to 350MHz
and the sensitivity at higher frequencies is also shown for the Proposed Design, limited to ~650MHz. Note that
there is still good sensitivity even to a scan angle of 60°.
While the array packing considerations in the AADC Consortium Proposed Design are the same as those
in the Baseline Design (i.e., EoR driven), the Proposed Design offers a high-band (350-650 MHz)
capability available for the substantial periods when EoR (or other low-band) observations are not
feasible or not required. The LFAA will be sparse in the high-band but sensitivities across this band, as
shown in Figure 3, are an improvement over the SKA-mid dish array at the bottom of the extended band
and comparable at the top of the band. This, combined with multi-beaming capability, gives
considerable additional SKA1 science potential. The principal driver for the high-band extension is timedomain science, including cosmic radio transients and pulsars.
In the field of transients trail-blazing discoveries [4-5]–[4-7] offer tantalizing prospects of a new view of
the Universe and promise unparalled insight into hitherto hidden baryonic matter; new event-rate
formalisms [4-8] link telescope architectures and resulting windows on the Universe; and recent
Reference
Issue – Revision
Date
Page 14 of 56
: 1-0
: 2013-06-08
RESTRICTED
rigorous analyses [4-9] show that the LFAA is highly efficient as a transients detector, especially above
350 MHz. These factors, and the extremely low impulsive RFI of the LFAA site, are important
motivations for the extended frequency range.
4.2.1
Pulsars
Pulsars are identified as an important part of the SKA1 science mission [AD-6] although the LFAA pulsar
observing capabilities have not previously been considered in the DRM extensively. However, the
extended frequency range and multi-beaming capabilities of the Proposed Design are particularly
attractive for pulsar observations. Pulsars are particularly steep-spectrum sources (flux proportional to
wavelength to power 1.8), but precision measurements at low-frequency are adversely affected by
interstellar scattering (which scales as wavelength to the fourth power). Nonetheless, the LFAA’s large
instantaneous sensitivity and FoV are very well suited to both pulsar surveys and the regular monitoring
of the largest possible sample of known radio pulsars. Compared with SKA-mid pulsar surveys of the
Galactic plane, where observing frequencies around 1.2GHz are optimal, the LFAA at ~500-600MHz is
better matched for surveys at Galactic latitudes above five degrees, where the effects of scattering and
sky background are significantly less severe and there is considerably more sky area to cover.
The LFAA can provide an efficient all-sky pulsar survey, including multiple passes to search for
intermittent sources. Recent results [4-10]–[4-13] have shown that a significant fraction of the radioemitting population of neutron stars are only sporadically observable – making a strong case for multipass surveys and large total on-sky time (i.e., longer dwell time can sometimes trump instantaneous
sensitivity). Large total on-sky time is also critical for probing transient events with low event rates,
similar to those of core collapse supernovae and gamma-ray bursts [4-14]–[4-16].
Regular monitoring of the largest possible sample of known radio pulsars is critical for studying their
physics and for performing basic timing of SKA pulsar discoveries. With thousands of expected SKA
pulsar discoveries, follow-up timing will be an observational challenge. Nonetheless, it is a challenge
that has to be met if we want to reap the scientific harvest of these discoveries. High-cadence
monitoring of known sources is very likely to reveal new surprises about the physics of pulsar
magnetospheres. The last years have underlined the richness of the observable phenomena [4-17]–[419] and how understanding these is potentially critical for making the best use of pulsars as precision
astronomical clocks. Lastly, high-cadence pulsar monitoring with SKA-Low will provide an unparalleled
interstellar medium weather report, which can be used, for example, to map the Galactic magnetic field
and to correct dispersion measure variations in the high-precision pulsar timing data collected with SKAMid at frequencies above 1 GHz.
4.2.2
High band imaging
Results in aperture array calibration and imaging emerging from LOFAR and MWA lead us to expect that
the LFAA will be a formidable imaging telescope, even at its highest frequencies. The SKA Design
Reference Mission and Baseline Design documents identify a number of science applications in addition
to the EoR, including HI-line absorption against continuum sources. The AADC Consortium LFAA design
will verify the capacity of the instrument to access a wide range of this science, illustrating
unambiguously the imaging performance of sparse aperture arrays and potentially pointing the way to
the use of these cost-effective receptors in a an SKA2 frequency domain hitherto assumed to require
dishes or dense arrays.
Reference
Issue – Revision
Date
Page 15 of 56
: 1-0
: 2013-06-08
RESTRICTED
4.2.3
Observing flexibility
As noted in the next Section, there are a vast number of trade-offs possible in assigning beams,
bandwidths and number of stations within experiments. Data rates across the LFAA and total SKA-low
processing system will be the subject of investigation during the design process but, from the
astronomer's viewpoint, even simple architectures yield great flexibility. For example, the same LFAA
architecture supporting a 911 station, single FoV, EoR experiment with 30 MHz of bandwidth could
comfortably support a 200 station, 32 FoV version of the same experiment. Similarly, for transient
experiments, the LFAA processing can handle the entire high band, with several FoVs. The LFAA
processing is of course only part of the story, with final spectral and time resolutions set by the
SKA1-low central signal processing, and software and data processing capability. Nevertheless, the
flexibility of the LFAA architecture augurs well for the science potential of SKA1.
4.3
Section references
[4-1] van Haarlem, M. et al., "LOFAR: the LOw Frequency ARray", Astron. Astrophys (submitted),
http://arxiv.org/abs/1305.3550v2, May 2013.
[4-2] Tingay, S. et al., "The Murchison Widefield Array: the Square Kilometre Array precursor at low
radio frequencies, Publications of the Astron. Soc. Australia, vol. 30, e007, Jan. 2013.
[4-3] Trott, C. "Impact of array design on calibration of SKA-low", Astrophysical Journal (in preparation),
2013.
[4-4] Wijnholds S. J. and van der Veen, A., "Multisource self-calibration for sensor arrays", IEEE
Transactions on Signal Processing, vol. 57, issue 9, pp. 3512-3522, 2009
[4-5] D. R. Lorimer et al., “A bright millisecond radio burst of extragalactic origin,” Science, vol. 318, no.
5851, pp. 777–780, 2007.
[4-6] E. F. Keane et al., “On the origin of a highly dispersed coherent radio burst,” Monthly Notices of the
RAS, vol. 425, pp. L71–L75, Sept. 2012.
[4-7] Thornton, D. et al., “Discovery of a fast radio burst population at cosmological distances”,
submitted for publication
[4-8] J.-P. Macquart, “Detection rates for surveys for fast transients with next generation radio arrays,”
Astrophysical Journal, vol. 734, p. 20, June 2011.
[4-9] T. M. Colegate and N. Clarke, “Searching for Fast Radio Transients with SKA Phase 1,” Publications
of the Astron. Soc. of Australia, vol. 28, pp. 299–316, Nov. 2011.
[4-10] M. A. McLaughlin et al., “Transient radio bursts from rotating neutron stars,” Nature, vol. 439,
pp. 817–820, 2006.
[4-11] M. Kramer, A. G. Lyne, J. T. O’Brien, C. A. Jordan, and D. R. Lorimer, “A Periodically Active Pulsar
Giving Insight into Magnetospheric Physics,” Science, vol. 312, pp. 549–551, Apr. 2006.
[4-12] J. S. Deneva et al., “Arecibo pulsar survey using ALFA: Probing radio pulsar intermittency and
transents,” Astrophysical Journal, vol. 703, pp. 2259– 2274, Oct. 2009.
[4-13] E. F. Keane et al., “Rotating Radio Transients: new discoveries, timing solutions and musings,”
Monthly Notices of the RAS, vol. 415, pp. 3065–3080, Aug. 2011.
Reference
Issue – Revision
Date
Page 16 of 56
: 1-0
: 2013-06-08
RESTRICTED
[4-14] J. van Leeuwen and B. W. Stappers, “Finding pulsars with LOFAR,” Astronomy and Astrophysics,
vol. 509, Jan. 2010.
[4-15] A. P. V. Siemion et al., “The Allen Telescope Array ﬂy’s eye survey for fast radio transients,”
Astrophysical Journal, vol. 744, p. 109, Jan. 2012.
[4-16] P. J. Hall et al., “Towards SKA studies of the radio transient universe,” in Resolving the Sky - Radio
Interferometry: Past, Present and Future PoS(RTS2012)042, (Manchester, UK), Apr. 2012.
[4-17] M. Kramer, A. G. Lyne, J. T. O’Brien, C. A. Jordan, and D. R. Lorimer, “A Periodically Active Pulsar
Giving Insight into Magnetospheric Physics,” Science, vol. 312, pp. 549–551, Apr. 2006.
[4-18] A. Lyne, G. Hobbs, M. Kramer, I. Stairs, and B. Stappers, “Switched Magnetospheric Regulation of
Pulsar Spin-Down,” Science, vol. 329, pp. 408, July 2010.
[4-19] W. Hermsen et al., “Synchronous X-ray and Radio Mode Switches: A Rapid Global Transformation
of the Pulsar Magnetosphere,” Science, vol. 339, pp. 436, Jan. 2013.
5
Detailed technical description
The overall work flow of the AADC Consortium is shown in Figure 4; this is closely mapped to the
required blocks of the SKA-low system and illustrates the signal flow. The sections below describe the
sub-systems in detail.
Figure 4: Work structure and signal flow for SKA-low
To set the basis of the description, the specifications for both the Baseline Design and the Proposed
Design are summarized in more detail in Table 1. The main difference between the Baseline Design and
the Proposed Design is a significantly higher top frequency, two bands of operation, capability of higher
station data rates (for multiple beams) and the ability to have programmable station sizes.
Reference
Issue – Revision
Date
Page 17 of 56
: 1-0
: 2013-06-08
RESTRICTED
Table 1: Outline specifications for SKA-low for the Baseline Design and the Proposed Design
Parameter
Baseline Design
Proposed Design
Number of
antenna
262,144
262,144
1
1
The full frequency range will be covered by a single
element type, as with the Baseline Design. This is for
lowest cost.
Frequency –
low
50MHz
50MHz
Push the low frequency as low as reasonable without
losing performance at 50MHz for future science
capability.
Frequency –
high
350 MHz
650 MHz
The increased frequency limit is accommodated by
the element design. The LFAA operates well at these
frequencies and avoids the dish system having a
large, expensive LF feed and a very large subreflector, while maintaining contiguous frequency
coverage.
1.35m
(λ/2 at 111MHz)
1.35m
(λ/2 at 111MHz)
35m
20-100m
The “station” diameter, particularly in core, can be
varied to suit the experiment. This is managed by the
station processing. The variable station diameter
should be able to mitigate central processing
requirements for many of the experiments.
2 – linear
2 – linear
Essential to have a dual polarisation system
1
2
The full available bandwidth is divided into 2, which
handle mutually exclusive science cases. This is
similar to switching feeds on a dish.
Max
instantaneous
Bandwidth
250-300MHz
335MHz &
300MHz
Because the full frequency range is covered in two
bands, the output bandwidth is essentially the same
as the Baseline Design but is reused through using a
switched-in first alias on the digitisation.
Data rate into
Correlator/
Beamformer
10Gb/s
per 35m Station
10Tb/s total
≥10Tb/s total
The performance of the SKA-low depends on the
total data rate to the central processing. This can be
increased as required.
Data
flexibility
1 beam/station
Flexibly assigned
to beams within
a band
A great benefit of an AA is that data output can be
assigned to arbitrary beamlets in arbitrary directions
up to the total designed data rate. Then each
experiment can be optimised, or concurrent
experiments run.
Types of
antenna
Element
separation
Station
diameter
Polarisations
Number of
bands
Reference
Issue – Revision
Date
Comments
250,000 antennas is for the core only.
911 stations of 289 antenna = 263,279.
Rounded to 262,144 (218)
The spacing of 1.35m is taken from the Baseline
Design. This may not be optimal for the science and
needs checking early in Stage 1.
Page 18 of 56
: 1-0
: 2013-06-08
RESTRICTED
Sample
resolution
5.1
8-bit
4 or 8-bit
capability
Many experiments will operate effectively with 4-bit
data, hence doubling the total bandwidth for the
same data rate. This needs to be agreed with the
correlator and post processing groups.
Implementation options
The technologies used to implement the LFAA are subject to choice: generally with pros and cons for
each approach. In effect the Baseline Design document has already made some decisions, which are
supported by the AADC e.g. single antenna type, configuration of the array etc. Early down-select of
implementation options as soon as a decision becomes clear is an important mechanism to both focus
development on the final solution and save design resources.
5.1.1
LFAA design choices already made
The AADC has also made further design decisions as part of this proposal being put forward. In summary
the following decisions have been made:
5.1.1.1
Single element: log-periodic antenna
The LFAA for Phase 1 of the SKA will only use a single element type to cover the full frequency range.
The element type chosen is a log-periodic antenna that was suggested in the Baseline Design document.
There will be one element since:
i.
the log-periodic will readily cover the specified frequency range, and
ii.
a large proportion of the cost of the LFAA is in the antenna, LNA and associated infrastructure
making incremental cost for using two antenna types substantial.
Consequently, a single antenna type is a lower cost solution.
5.1.1.2
All beamforming in the digital domain
In principle, beamforming can be performed using analogue or digital electronics or a combination of
both. The choice has been made to entirely use digital beamforming. The reasons are:
i.
The beam performance requirements of the LFAA will need the precision associated with digital
beamforming;
ii.
Analogue beamforming for such large antenna spacing as the LFAA will require large and
expensive switchable analogue delay lines. Digital beamforming in the timescale of SKA Phase 1
is likely to be more cost effective; and
iii.
Digital beamforming is much more flexible in the forming of beams on arbitrary positions on the
sky of different bandwidths, leading to a better scientific return.
5.1.1.3
Digitisation in the bunker
The location for digitisation of the input signals can either be near or at the antenna or in the bunker.
Signal transport from the antenna to the bunker is then either digital or analogue. While digitisation at
the antenna is ostensibly a good solution, there are some significant downsides:
Reference
Issue – Revision
Date
Page 19 of 56
: 1-0
: 2013-06-08
RESTRICTED
i.
Digitisation near the antenna requires a clock signal that is phase stable enough to not impact
the beamforming precision. This implies transporting a clock signal to every antenna, which is
difficult, expensive and may cause RFI;
ii.
Digital electronics generate RFI that needs to be shielded. This is a risk, as this approach would
involve many RFI tight enclosures throughout the LFAA. This is both expensive and potentially
creates considerable self-generated interference;
iii.
Digitisers draw significant power, which adds to the power distribution issues at the antenna;
and
iv.
Due to the wide distribution of components, it is very difficult to maintain the digitisation
systems and upgrade to higher-performance components e.g. for wider bandwidth or higher
resolution.
Since long-range analogue signal transport is available, and indeed is planned for ASKAP, these factors
can all be mitigated by keeping the electronics near the antenna as simple as possible and transporting a
wideband analogue signal to a bunker for digitisation and processing.
5.1.2
Ongoing design options for Stage 1 decisions
The detailed design of the LFAA has further alternatives for the different subsystems, which are not so
clear-cut. These need to be considered and reviewed during Stage 1 of Preconstruction. It is likely that
some of these can be down-selected early in the development. By SRR/PDR there is expected to be a
single design for the LFAA, which will be further developed during Stage 2 and physically tested
wherever possible on AAVS1. The list of significant ongoing technical options are listed in Table 2, it can
be seen that there are pros and cons for each decision and some are obviously linked e.g. a single
bunker is only possible with low-cost, long range analogue communications.
Further comments on options from Table 2:
5.1.2.1
Station processing
The Baseline Design requires relatively low data rates per station, which while specified as one beam,
can be split up into multiple beams of narrower bandwidth. The implementation in this representative
design uses a “daisy chain” approach whereby the tile beams are progressively accumulated by the tiles
themselves, using data steering through switch network. This is described in more detail in section 5.7.
This is elegant, cheap and uses conventional COTS components for the data steering and unless it does
not operate efficiently or if there is a lower cost, easily realisable solution then this will probably be the
chosen approach.
Alternative station processor implementations will be considered in stage 1 to find if there are lower
cost solutions that meet the agreed specification. For example, if much larger station data rates are
required which significantly exceed the capacity of a reasonable switch network then the “daisy-chain”
approach may not be efficient. The principle alternative is to provide a second level of hierarchical
beamforming in a station processor. This takes the output of all the tiles for a station and accumulates
them to form station beams, many from each tile beam. If the tile processor data output rate is to be
less than the station data output rate then hierarchical beamforming needs to be used as described in
the upgrades for SKA2 in 8.1.3.
Reference
Issue – Revision
Date
Page 20 of 56
: 1-0
: 2013-06-08
RESTRICTED
5.1.3
Existing experience
The knowledge that is being gained from a large AA installation in Europe, LOFAR, and the smaller
instrument based on the SKA site, MWA, is invaluable. Also, ASKAP will illustrate the use of RFoF. Several
AADC members are key proponents or leaders of these programmes and the pathfinder instruments are
allowing detailed evaluations of technology choices, science applications and a plethora of real-world
considerations; lessons learned will be applied to the AADC developments.
Reference
Issue – Revision
Date
Page 21 of 56
: 1-0
: 2013-06-08
RESTRICTED
Table 2: Outline comparison of technologies that could be used for LFAA subsystems, for investigation and downselect during Stage 1 Preconstruction. The example approach adopted in this proposal is shown shaded.
Sub-system
Analogue signal
transport
Approach
Pros
Cons
Coaxial copper links
Well proven on multiple RA systems.
Short range – 100’s metres.
Can also be used to transport power
Frequency dependent transmission.
Material is expensive and attractive to
thieves.
Requires multiple processing bunkers.
Analogue RF over
Fibre, RFoF
In use on existing phased arrays.
Requires low cost, low power laser.
Long range – 10’s km.
Antenna power must be supplied
separately.
Individual optical strands are cheap.
Wide, flat frequency response.
Each polarisation on
a separate channel
Antenna signal
transmission
Guaranteed signal separation.
Requires two cables/fibres, per antenna.
Simple integration.
Maybe higher cost.
No risk of RFI from a local oscillator, LO.
Requires two analogue channels & ADCs
in the bunker.
Multiplex both pols
into one channel
electronically (see
Section 5.4.2)
Halves the number of analogue links.
Risk of crosstalk between polarisations.
Can use a single, fast ADC per antenna at
reduced system cost.
Requires LO at antenna, so may get RFI
issues.
Optically multiplex
both channels onto
one fibre
Halves the number of analogue links.
More expensive lasers and combiners.
Guaranteed separation.
Requires two analogue channels & ADCs
in the bunker.
Line power
delivered via wire
network
Low risk.
DSP needs to correct upper channel
frequencies.
No risk of RFI from LO.
Easily implemented with coaxial links.
Risk of RFI, ESD and lightning issues with
expensive protection systems.
Substantial additional infrastructure.
Operational cost for antenna power.
Antenna Power
Local Solar power
to single or multiple
antennas
Reduced global RFI, ESD and lightning
issues.
Unit cost (offset by infrastructure/
antenna savings)
Simplified infrastructure.
Potential technical issues e.g. battery life,
local power supply RFI.
Potential operational cost savings
Assemble w/antenna before deployment
Green power usage.
Single bunker
Lower cost.
Flexibility in LFAA configuration.
Processing
Facilities
Ease of maintenance.
Efficiency degradation with time e.g.
solar cell lifetime, debris accumulation.
Requires long range, low cost analogue
signal transport.
Potential single-point of failure.
Ease of system upgrade.
Could accommodate correlator as well.
Reference
Issue – Revision
Date
Page 22 of 56
: 1-0
: 2013-06-08
RESTRICTED
Sub-system
Approach
Pros
Cons
Multiple bunkers
Less cost in analogue signal transport.
Higher total cost for bunkers
Can use copper analogue signal
transport.
Higher risk of RFI.
ASIC based
Fewer LFAA single-points of failure.
More power, clock and communication
networks required.
External digital backhaul required.
Very low power for processing.
High development costs.
No maintenance.
Long lead time.
Low part costs.
Inflexible and hard to upgrade.
Requires other technologies to perform
complex tasks.
Tile processing
FPGA based
Low power.
Expensive piece parts.
Programmable, can be upgraded.
Hard to programme.
Low NRE.
Processor based
Easily programmable.
Higher power.
Very flexible.
Possible insufficient communications
bandwidth.
Low development costs and NRE.
Ability to follow COTS developments.
“Daisy chain” tile
processing using
switches for data
steering
Station
processing
All complex development on tile
processor board.
Station data output rate limited to tile
processor output data rate.
Use COTS for data steering.
Cannot use hierarchical beamforming.
Very flexible.
Low cost.
Dedicated station
processor
Can supports high station output data
rates.
Higher cost to develop and deploy.
Less processing on tile processors.
Less flexibility on data routing for beam
formation e.g. overlapping stations.
Can use hierarchical beamforming.
5.1.4
Requires specific hardware development.
LFAA design presented in this document
In this document a Representative Implementation is described and costed as it is not practical to cost
all variants at this point; much more detailed costing work will be undertaken in Stage 1 of Preconstruction. While the shaded approaches in Table 2 presents a reasonable set of choices, for final
implementation, the architecture and options for LFAA will developed during Stage 1, with a discussion
on the selection made for Stage 2 to be presented at SRR/PDR.
5.2
Antenna and LNA
The antenna sub-system consists of the antenna, analogue gain and filtering components feeding the
communications link; all, ideally, will be solar powered.
The current Consortium prototype antenna is a log-periodic element, shown in Figure 5, as part of a test
array at the MRO site. This antenna design provides a very wide frequency range, determined only by
Reference
Issue – Revision
Date
Page 23 of 56
: 1-0
: 2013-06-08
RESTRICTED
the range of sizes of the dipole arms, with a good match to the LNA. The antenna is constructed using
bent wire, which has been shown to give excellent performance in our testing to date. The
manufacturing technique leads to a minimal use of material and is inherently low cost. Prototypes are
easily manufactured on CNC wirebenders, with a clear route to SKA volume manufacturing. Although
the choice of materials will be finalised in the design phase, stainless steel has survived very well in the
Australian site conditions and represents a low risk, but at slightly higher cost than mild steel. A
groundplane may not be required as it supports only the lowest frequency of operation, where the
system in any case operates in a very high sky noise regime. At all the other frequencies the log-periodic
provides its own “groundplane” via the lower frequency dipole arms. However, an integrated
groundplane can be considered to give system predictability in different environmental conditions.
Figure 5: SKA-low test array on the SKA site
The electronics are housed at the apex of the antenna and will contain the LNA, analogue components,
RFoF driver and connector. This is protected from the weather conditions with a cover and the printed
circuit boards themselves are protected either with conformal coatings or potted, to be confirmed
during Pre-Construction Stage 1 development.
The combination of antenna element and LNA define the noise performance of the system. Many tests
on current prototype of all types have been conducted; the results verify that design simulations are
Reference
Issue – Revision
Date
Page 24 of 56
: 1-0
: 2013-06-08
RESTRICTED
accurate and that the noise performance over the entire band of the antenna is very good. We are
therefore confident in developing mechanical improvements to the antenna design and slightly
extending its frequency range for the Proposed Design.
More quantitatively, the prototype antenna element has been shown to work well to 600MHz, and a
single extra dipole arm element and redistribution of dipole arm sizes will extend operation to
>650MHz. There are no additional signal processing costs since the system would use a switched
analogue filters in the bunker to select baseband or the digitisation first Nyquist band. There is further
discussion in Section 5.5.1.
5.3
Power for the LNA and antenna receiver
There are two alternatives for powering the antenna electronics: copper-based distribution of power
from a central source or solar power. While copper wire is clearly able to be implemented, it suffers
from a number of issues:
•
The material and installation are expensive;
•
It is susceptible to interference from its associated electronics, unless there is careful design of
the antenna and bunker systems;
•
The electromagnetic properties of the antenna can be affected; and
•
In the event of a lightning strike, or an electro-static discharge, copper links can distribute
damaging voltages across various systems.
Following an examination of projections for the cost of various power solutions in Australia, the AADC
plan is to develop a solar power solution, which can be installed for each antenna or for small blocks of
antennas. The putative solar solution will be compared for cost, performance and maintainability with a
”normal” copper power network across the LFAA system, allowing an informed decision to be made at
SRR/PDR. Assuming that solar power proves viable and cost-effective then the ideal solution is to
integrate the design and installation of the solar power unit with the antenna structure. This guarantees
that performance conforms to the simulation and test results, and enables the power unit to be raised
off the ground to avoid issues with flash flooding.
There are implementation issues to solve, including:
•
Ensuring that the lifetime of the solar unit, particularly the battery to support operation
overnight, meets with the maintenance cost and resources budgeted;
•
Suppressing RFI from the solar unit to acceptable levels (e.g. use linear or low frequency
regulation and charging for the battery); and
•
Verification of efficient operation in the presence of long-term dust or other material.
A major benefit is that the "fuel" component of the operational cost of powering the antennas is zero,
so increasing the power used by the LNAs and amplifiers at the antenna only results, to first order, in
more up-front capital expense for the solar power unit.
Reference
Issue – Revision
Date
Page 25 of 56
: 1-0
: 2013-06-08
RESTRICTED
5.4
Receiver
The receiver development is defined, as shown in Figure 4, to cover the analogue signal from the LNA at
the antenna and delivery to the ADC in the processing bunker. This consists of three principal
components:
•
•
•
Gain and filtering at the antenna;
Signal transport using RFoF components; and
Analogue gain, filtering and pre-whitening in the processing bunker pre-ADC.
The actual installation of the fibre cables will be specified by the LFAA Local Infrastructure task and
ultimately undertaken as part of SKA infrastructure site works.
5.4.1
Antenna front-end
This development will be in close collaboration with the Antenna and LNA work, indeed the
implementation is likely to be on single circuit boards one for each polarisation channel (as they are with
the prototype antenna).
It is unlikely that the implementation will use custom analogue chips since the moderate volume and
long development timeline would not justify the development cost for the LFAA. The only justification
for customization may be for power reduction purposes, and then only on the basis of a trade-off
between the cost of solar power versus that of an ASIC. The requirement is for sufficient gain at the
antenna to ensure that the system temperature is not increased due to any losses in the signal
transport, as well as preserving good dynamic range headroom. Filtering is required, particularly a low
frequency cut off to remove large RFI (such as over the horizon short wave radio) signals below 50MHz.
There are no dedicated control signals for the antenna and associated electronics; this would require a
communications link from the bunker to the antenna which would be expensive and is not considered
necessary. The health and signal condition of the antenna sub-system can be gauged from the signal
strength and spectral analysis of the data signal received at the tile processor.
5.4.2
Signal Transport
The planned signal transport is analogue RF over optical fibre, RFoF. This is widely used commercially,
e.g. duplicating cellular phone base-stations using specialist lasers developed for analogue
communications; and is also planned for ASKAP. These are low cost and can transmit the signal up to
10km with the lowest cost laser and >50km using a higher performance device.
In our LFAA design solution, most lasers are low cost and will be integrated with the remainder of the
gain chain. For simplicity and lowest risk we have chosen to use separate lasers and fibres for each
polarisation from every element, however; it is conceivable to electronically multiplex both polarisations
onto one fibre by heterodyning one polarisation up to a band of frequencies higher than the top
frequency used. This can be done with a reasonably stable, but free running local oscillator, LO, at each
antenna. The two polarisations can be recovered digitally and errors in the LO corrected following the
spectral filter.
Phase stability of a fibre solution, particularly for the longer baselines, may become an issue. This can be
due to temperature and environmental effects. Instability may be severe enough, even at these low
frequencies, for the calibration schemes not to be able to get within one RF cycle. The proposed solution
Reference
Issue – Revision
Date
Page 26 of 56
: 1-0
: 2013-06-08
RESTRICTED
for the few stations that may be affected is to use the phase and time transfer system (being developed
by the SADT consortium) on one fibre strand in a bundle of 200-500 fibres. This effectively “measures”
the optical length of a fibre and enables calibration to be successful. The costs are not substantial on just
the long baselines.
The fibre bundles will be armoured and specified for simply laying on the surface of the site or in
conduit. It is known that trenching costs are very high on the MRO site. Each cable of many fibres needs
to be split near the antennas such that two fibre strands are delivered to individual antennas. This is an
important development and discussions with the fibre cable manufacturers/suppliers to pre-construct
this system prior to deployment.
The optical receiver is a simple, cheap PIN diode integrated with the receiver analogue board in the
processing bunker.
Since the communications are all specifically point-to-point, there is no requirement for a bunker patch
panel system, with the fibre bundles being taken directly within the bunker to the racks housing the
receiver systems.
5.4.3
Analogue processing in the bunker
The full bandwidth analogue signal is presented at the processing bunker, and then bandpass filtered,
pre-whitened and fed to the ADC. The analogue receiver board and digital processing boards will be
organised to process a complete “tile” of antennas – probably 16 x 2-polarisation channels.
For the Baseline Design only one 50-350MHz filter would be used (Figure 6). For the Proposed Design,
two switchable filters will be available: 50-375MHz and 375-650MHz (Figure 7), using the baseband and
first alias of the ADC to provide separate, non-commensal low and high band operation.
Any control signals for the analogue processing system will be supplied by the tile processor e.g.
selection of band and any required gain control signals.
5.5
Digital Signal Processing
As can be seen in Figure 4, Figure 6 & Figure 7, the signal processing requirement consists of two distinct
parts: a tile processing system and the station beamformer. The tile system is fed from the analogue
receiver on a per-tile basis. This is described below:
5.5.1
Tile processing
The tile processing cards can handle 32 incoming channels, consisting of 16 dual polarisation antennas.
The role of the tile processors is to:
•
•
•
•
•
Digitise the analogue signal using 8-bit high speed digitisers, which are available and able to
accommodate the likely and predicted RFI levels.
Channelise the bandwidth into relatively narrow 1MHz channels for a phase shift controlled
delay approximation.
Calibrate the signal as a function of frequency to compensate for bandpass errors, gain and
phase errors etc.
Apply beamformer weights.
Aggregate all 16 antenna signals into a tile beam or beams.
Reference
Issue – Revision
Date
Page 27 of 56
: 1-0
: 2013-06-08
RESTRICTED
These functions are described in the following subsections.
A potential practical implementation of the tile processor card plus the bunker receiver card is
illustrated in Figure 8. The system is mounted in a 4U high rack mounted shelf with a simple midplane.
5.5.1.1
Digitisers
The analogue to digital converters (ADCs) operate up to ~800MS/s, and feed the processing system.
For the Baseline Design (Figure 6), each signal covers the full bandwidth from 50-350MHz.
In the Proposed Design (Figure 7), the input band is switched between 50-375MHz and 375-650MHz.
The ADC converts either at baseband for the lower band or using the first alias for the higher band. The
ADC analogue input must be able to effectively sample and hold to the highest signal frequencies. To
ensure that there is no discontinuity in LFAA frequency coverage, while also ensuring that there is no
aliasing from the other band, the ADC sampling frequency is also switched: 800MS/s for the low band
and 700MS/s for the high band.
Figure 6: SKA-low processing for the Baseline design
5.5.1.2
Channelisation and calibration
The digitised signals are passed to a processing device, probably FPGA-based, which splits the
bandwidth into relatively narrow channels of ~1MHz. This gives a complex sample stream for each
channel that can be corrected for amplitude and phase errors. The beamformer delay and phase
weights are applied to each channel for each antenna ready for beamforming.
5.5.1.3
Tile beamforming
The frequency channels required for the station beams are selected, creating the bandwidth required
for the science experiment. Also, channels with gross RFI, for example, might be excluded. The selected
Reference
Issue – Revision
Date
Page 28 of 56
: 1-0
: 2013-06-08
RESTRICTED
channels are all summed individually for every antenna and passed to the output data stream. Each tile
will use a distributed beamformer approach and the output data rate from each tile is identical to the
full station output data rate, assuming no overlapping stations. With a large number of stations each
with a relatively low data rate in the Baseline Design, this is the best implementation, since it can
provide the best possible quality for the beams.
Figure 7: Processing for the SKA-low Proposed Design
5.5.2
Station beamforming
The data output of the tile processors need to be accumulated for station beamforming. The function of
the station beamformer is now very simple: it sums corresponding time & frequency samples from all
the tiles, as the tile beamformer has provided all the alignment and gain adjustment required on the
data streams within the packets.
This function represents the aggregation of 17 tiles in the Baseline Design, or a variable number of tiles
for selectable station sizes as put forward in the Proposed Design (e.g., for 250 stations there would be
~64 tiles being summed).
5.6
Implementation of bunker receiver and tile processor
There is considerable work to be done on the detailed design of the receiver and tile processing.
However, it is clear that the number of elements handled per tile processor directly affects the scale of
the installation in the Bunker. On our initial analysis, it is reasonable to assume that each tile processor
will handle 16 antennas i.e., 32 signal paths for both polarisations. This assessment is from road
mapping the processing and communication ability of an FPGA likely to be available in the 2018
timeframe. The expectation is that the most capable FPGAs at that time will have the processing ability
to handle all 32 channels on one chip. The actual implementation may use two devices for lower cost
and additional I/O capability, which contrasts with the 16 devices, needed using the present UNIBOARD
system with 2010-generation components.
Reference
Issue – Revision
Date
Page 29 of 56
: 1-0
: 2013-06-08
RESTRICTED
Figure 8: Outline layout of Receiver and Tile Processor. The industry standard SPF module houses the analogue
components for pairs of channels
Notwithstanding this advance, the construction of the receiver and tile is expected to be similar to the
proven UNIBOARD layout, the primary difference being to move the ADC components onto the
processing board to ease the complexity of the high speed digital interconnect. An outline design is
shown in Figure 8. Particular care will need to be taken to minimise digital noise on the analogue signal,
hence all the analogue inputs are housed in an industry standard SPF form factor carrier which
integrates the laser receiver with the gain and filter components on a small internal board. Normally SPF
modules, picture inset in Figure 8, house Tx and Rx channels in this case there are two Rx channels
installed. Note that the modules are on both sides of the receiver carrier board to increase packing
density. The main features of the design are:
•
The overall construction is a double sided rack with a central midplane board or connector
system.
•
All the analogue fibres connect to a “receiver board” which then:
o Converts the optical signal to an electrical signal;
o Provides analogue gain and any necessary equalisation to compensate for any
frequency dependence in the fibre link;
o Provides Nyquist filtering and low end cut off prior to the ADC. The signal should be
conditioned appropriately for the ADC; and
o Transmits the analogue signal to the processor board through a back-to-back connector.
•
The central mid-plane provides common services and the necessary physical sockets to
accommodate the receiver board and the tile processor. The principal features are:
o Physical mounting for the interconnect for the receiver and processor cards;
Reference
Issue – Revision
Date
Page 30 of 56
: 1-0
: 2013-06-08
RESTRICTED
o
o
o
o
o
o
•
5.6.1
Support straight through back-to-back connectors to carry 32 differential analogue
signals and control signals for each receiver and tile processor pair;
Power distribution with separate circuits for the analogue system and the digital
processing;
Any control signals required for the receiver board will be supplied by the tile processor,
which is controlled through the Monitoring and Control network via the
communications link;
Clock distribution if this is carried as an electrical signal. This may be an optical signal
connected directly to the tile processor;
Tile processor identification. This is to provide positional identification such that every
tile processor can identify its position within the system; and
For the Proposed Design: a signal from the tile processor identifying the low or high
band filters.
The tile processor connects to the midplane, and performs the digitisation and beamforming
functions for the tile. The main characteristics are:
o Plug into the midplane board to pick up the 32 analogue signal channels, power,
position identification and, possibly, the synchronisation clock;
o The analogue signals are taken to an array of 32 ADC channels which are directly linked
into the FPGA(s) for processing;
o On the Proposed Design, the tile processor receives configuration information over the
monitoring and control network and can configure the system for low or high-band
operation by switching the sample clock signal rate for the ADCs and provides the signal
for its associated receiver board for selecting the appropriate Nyquist filters; and
o The beamformed signal is presented at a high-speed optical link on the outside edge of
the card for connecting to the station processing switch.
Tile Processing
The signal processing on the tile is a beamforming function that has been well proven on multiple
installations e.g. LOFAR and APERTIF.
The tile processor performs the tile beam formation and is part of the Station processing chain, see 5.7
below.
Essentially, a simplified list of the processing tasks on the tile is:
1.
2.
3.
4.
5.
6.
7.
8.
Take each signal channel and put it through a polyphase filter for spectral separation into
approximately 1024 frequency blocks;
Apply amplitude and phase calibration coefficients to each active frequency block;
Perform any polarisation corrections to pairs of signals from each antenna;
Select the frequency band(s) required for the observation;
Apply beamformer weight(s) required to each frequency block;
Sum the blocks for each polarisation from all the antennas on the tile into tile beams;
Sum incoming partial station beams with the tile beams; and
Transmit the new, partial station beam via a switch to the next tile; or if completed, station
beams to the correlator.
Reference
Issue – Revision
Date
Page 31 of 56
: 1-0
: 2013-06-08
RESTRICTED
5.6.2
Tile output data rate
The tile data output link will be 40Gb/s or greater Ethernet or 56Gb/s or greater Infiniband. This is
significantly more than the minimum 10Gb/s data rate required for the Baseline Design. These
standards are available today and are cost effective at the volumes required for LFAA; it may well be
advantageous to use the next speed increment in the standards. Importantly it means that:
•
•
•
•
•
5.7
The packets of data can be time stamped to ensure synchronisation;
Better usage can be made of the antennas in the core, by overlapping and each tile contributing
to multiple beams to provide station apodisation;
The management and control signals can use the same data network;
With SKA-low configured (using the Proposed Design) with fewer, larger stations for specific
experiments then there is capacity to maintain the same aggregate data rate into the correlator
as the Baseline Design (~9Tb/s), thus maintaining the survey speed specification; and
There is little risk of packet collision on the output data.
Station processing
The beamforming at the station level is performed as a distributed beamformer, whereby the
time/phase delays required for the station beam are applied to each individual antenna by the tile
processor and the rest of the processing system simply sums all of the antennas in a hierarchical fashion.
The advantages of this system are that every station beam is precisely calculated and has minimum
errors and artefacts; also, it is relatively simple to implement. The disadvantage is that the data rate
throughout the system at each partial beamforming stage is the same, or greater, than the station
output data rate. Hence, relatively fast switches are required for controlling the configuration,
illustrated in Figure 9.
The station processing is performed cumulatively by the tile processors, or if it is available, a modified
data switch, which can route input datastreams to a single summing point which then outputs a
completed station beam.
Reference
Issue – Revision
Date
Page 32 of 56
: 1-0
: 2013-06-08
RESTRICTED
Figure 9: Station beamforming accumulation approach (See Figure 11 for illustration of Aisle switch)
The data routing for the station is shown in Figure 11. Here it is assumed that the racks are organised as
16 aisles of 16 racks each. The station processing requirements are to enable flexible linking of adjacent
tiles, so each half rack of tile processors is connected via 36-port switch, all the half rack switches are
linked to an aisle switch, which enables routing throughout the aisle. The aisle switches are “daisy
chained to link across to the nearest aisles. By organising the antennas by location almost any
configuration of stations should be feasible.
5.8
Processing Rack - construction
An individual processing rack consists of “shelves” of receivers + associated tile processors back-to-back
with a mid-plane, mounted in a 19” rack. Each shelf will be 4U high and can accommodate 8 tile
processing systems. These are reasonably spaced (~50mm) to provide good airflow. If testing shows
they can be closer packed, then there will be fewer racks in the bunker. Also, the rack can accommodate
the necessary data switches for data steering. An illustration of a rack is shown in Figure 10. The makeup
of the rack is:
•
•
•
•
8 x Rx + tile processor of 4U each
2 or 3 x data switch of 1U each
Power supplies estimate 4U
Space for additional cooling / spare
32U
3U
4U
3U
The total of 42U is a standard, low cost 19 inch rack format.
Each shelf can handle 128 antennas. For 256 antennas per station (slight modification to the Baseline
Design) then 2 shelves can handle the entire tile processing for an individual station. Of course, using a
switch to link the tiles that form a station means that any reasonable configuration of station can be
implemented. Each rack can handle 1,024 antennas.
Reference
Issue – Revision
Date
Page 33 of 56
: 1-0
: 2013-06-08
RESTRICTED
There are always two 36-port data switches to link all the tile processors into stations. Some racks also
accommodate a 3rd switch which acts as a second tier linking up to 32 switches together. Consequently
there are 16 racks fitted with the 2nd tier switch.
The result is that for the Baseline Design of 262,144 antennas that there will be 256 racks.
Figure 10: Outline LFAA rack, handling a total of 1024 antennas, using 8 shelves or 8 tile processors linked via two
36-port data switches. The rack shown has the second level, aisle switch
Reference
Issue – Revision
Date
Page 34 of 56
: 1-0
: 2013-06-08
RESTRICTED
Figure 11: Outline of overall interconnect. Essential features are wide bandwidth into the correlator and the
Monitoring and Control system
The precise topology of this system is the subject of study in Stage 1.
5.9
Monitoring and Control
As well as the main data path through the tile processors to the correlator, the whole of the LFAA needs
to be controlled and monitored. This requires an overlay data network to all the controllable subsystems. By using a 40Gb Ethernet or 56Gb Infiniband data network, there is spare capacity to use the
same network as the data path system. This approach reduces cost and makes installation more
straightforward. Referring to Figure 11 it can be seen that the monitoring, control and calibration data is
routed to and from the tile processor via the rack switches, the aisle switches and concatenated in a
final switch linking all the aisles.
The LFAA monitoring and control is centred on a number of dedicated servers that are part of the
network, they provide:
•
•
Configuration control to set up the requirements of the observation;
Calculation of all the relevant coefficients for the beamforming on all the tiles;
Reference
Issue – Revision
Date
Page 35 of 56
: 1-0
: 2013-06-08
RESTRICTED
•
•
•
•
•
Regular updates of the coefficients in order to steer the beams on a fixed point on the sky;
Reception of calibration information for the hardware in the arrays and distribution of
appropriate calibration coefficients;
Narrow band correlations of the elements to calibrate astronomically or with an artificial source;
Checking and reporting of the health of the system by determining which tiles/antennas are
faulty; reporting and accommodating the missing sub-systems in the beamforming system; and
Communication with the overall Telescope Management system.
5.10 Bunker system and SKA implementation
The whole of the processing for LFAA is planned to be in a single, large bunker. This provides all the
power, cooling and RFI shielding for the system plus connection to the rest of the SKA interfaces: data
output to the correlator, telescope manager and the synchronisation clock for the digitisers.
In this implementation, illustrated in Figure 11, there are 256 racks, dependant on the precise number
of antennas used. These are expected to be organised as 16 racks to an aisle and 16 aisles. Each rack has
its own power supply, from a distribution of power around the bunker. Cooling is likely to be provided
by a closed-cycle water-cooled door although the power requirements per rack are relatively modest at
around 7kW. Clock reference signals will originate from the SKA system clock and be distributed to
individual racks; this is likely to be via an optical network, at least to the rack. Studies in Stage 1 will
determine the most effective distribution system, which may be in collaboration with the SADT
consortium.
The analogue optical fibres are located along the aisles, racks and tile processors as a point-to-point
system from the antennas. The rack and aisle switches are distributed among the main processing racks.
An additional rack houses the monitor, control and calibration servers, plus the master clock reception
and distribution systems.
The total required processing load is the same, regardless of whether it is done near the stations or
brought into one very large bunker. However, centralised processing is cost effective as it offers:
•
•
•
•
•
•
•
5.10.1
Ease of maintenance;
Lower cost for the facility;
Ease of interconnect;
Single power and cooling systems;
Simple clock distribution;
Streamlined integration with the LFAA correlator; and
The opportunity to house the master maser clock.
Integration with LFAA correlator and post processing
With all the processing for the LFAA in a single facility, close to the core of the SKA, it would seem a
sensible, cost-effective extension to the concept to integrate the LFAA correlator in the same facility.
This would restrict digital communication to within the building, removing the need for a wide area
network for this part of the system.
A software correlator for SKA-low using even 2016-era devices can be housed in <30 racks, which is a
small additional space over the processing for LFAA. This should be readily able to be accommodated.
Reference
Issue – Revision
Date
Page 36 of 56
: 1-0
: 2013-06-08
RESTRICTED
Also, the central maser time standard for the SKA-low can be housed in this central building, making the
distribution of the clock for all the LFAA stations very efficient.
There is more investigation needed around the scale and power requirements of the post processing
system but the practicality of integrating the post processing system and correlator should be
investigated since it avoids transmission of large quantities of data over a long distance and gives a
major opportunity for integrating the whole processing chain and try new, novel processing approaches
in readiness for SKA2.
5.11 Transient buffering
A potential feature that is only alluded to in the Baseline Design and is not a Baseline requirement is the
ability to buffer the incoming signal data and use a trigger to freeze the buffer, and then look back at a
previous astronomical event. This is useful, for example, in transient detection and verification. All of
these buffers would be expected to be circular buffers, whereby only a fixed period prior to the present
time is available. While this is not costed at this stage it is worth noting some implementation options to
be explored during Stage 1 pre-construction.
5.11.1
Element level data storage
This is the most flexible, but also the most expensive implementation. This option implies considerable
memory storage at each of the tile processors. The tile processor streams a selected set of element data
to a high capacity memory system on the tile processor. The amount of data (or length of time the
buffer can look back) can be reduced by selecting only relatively narrow bandwidths, reducing the
digitisation depth or summing the polarisations.
The advantage of storage at this level is that beams can be reconstituted anywhere within the scan
angle of the array, > ±45°.
The disadvantage is that it is expensive on memory and processing and may be memory bandwidth
limited to store 16 elements worth of data.
5.11.2
Tile level data storage
Here the tile beam(s) are stored, as above, on the tile processor. This is a much reduced data set and can
therefore be much cheaper to store.
This would enable a station beam to be formed within the tile beam by post processing, perhaps to
more accurately identify the position of a transient.
This is cheaper in memory and could be used as an observation-time-selected alternative to the element
level storage.
5.11.3
Station beam storage
Storing station beams is again much less expensive in memory terms than the foregoing approaches.
This storage could be at the tile processor or at the input to the correlator.
The correlator storage approach is flexible, particularly for a software correlator where there is a normal
CPU or multicore processor to process the data; a lot of storage could be made available on these
machines. The reprocessing could vary between beamforming and correlation.
Reference
Issue – Revision
Date
Page 37 of 56
: 1-0
: 2013-06-08
RESTRICTED
5.11.4
Dedicated storage nodes
Using data routing switches, which can duplicate data on multiple ports if required, data could be routed
to dedicated transient buffer boards or processors for storage/analysis.
6
Risk mitigation
This design of the LFAA makes a few predictions on the availability, performance and cost of some
subsystems for the LFAA. There are risk mitigation strategies to cover the possibility of not meeting
these predictions, albeit at some combination of reduced performance, increased cost or more difficult
maintenance.
There follows the principal risks for the LFAA, these do not include obvious “higher component cost”,
“longer to develop”, “insufficient resources” etc.:
Risk
Likelihood/
impact
Mitigation
Consequences
1.
The analogue fibre transport
including laser is not low cost
enough/low performance/too
much power
Low/
high
Use copper based analogue signal
transport.
Due to the relatively short range of the copper
links, the system architecture will need to be
modified to use local, small bunkers near to
the stations with the correlator in a central
facility.
2.
The analogue fibre range is
not as far as predicted.
Medium/
low
Use some local
digitisation/processing bunkers near
the long baseline stations
The result is fairly minimal, provided that the
bulk of the core and immediate stations can
be accommodated. The design is not as
flexible and will require some wide area digital
communications.
3.
The analogue receiver cannot
be made as small as predicted
Low/
medium
Use larger receivers.
The system will require more racks of
equipment leading to a larger, more expensive
facility
4.
The processing or
communications performance
of the 2018 devices is not up
to the requirement of the tile
processors
Low/
medium
Use the available processing devices
There will likely be increased cost and size of
the system
5.
The ADCs are more
expensive/higher power than
predicted
Medium/
low
Use the ADCs that are available
The cost/power of the LFAA will be higher
than predicted, so may need to be reduced in
scale
6.
Solar power does not give the
power required/short
life/battery technology
inadequate
Medium/
medium
Use a network of copper cables
powered from the central power
supplies and upgrade lightning/RFI
measures.
The use of copper cables is likely to result in a
higher cost and susceptibility to lightening and
RFI. Potentially higher RFI emissions
7.
The data communication
switches have lower
performance/unpredicted
delays
Low/
low
Use different switches or
compensate in the processing
The switches are already considerably overspecified, so unlikely to suffer low
performance. The consequences though
would be a longer development time and
maybe more resources for the tile processor
8.
Power consumption too high
Medium/
medium
Further development or more likely
to accept the available components
Running costs are higher until SKA Phase 2 is
commissioned.
Reference
Issue – Revision
Date
Page 38 of 56
: 1-0
: 2013-06-08
RESTRICTED
7
Cost estimation
Table 3: Costing for Baseline and Proposed Design for LFAA (part 1)
Reference
Issue – Revision
Date
Page 39 of 56
: 1-0
: 2013-06-08
RESTRICTED
1
System & sub-system
#/system #/SKA
Cost
Each
Baseline Proposed
€
€
%age
Sub-system
Baseline Proposed
€
€
Totals:
Sub-Totals/SKA
Baseline
Proposed
€k
€k
Comments
of
Total
135,312 137,819
Sub total SKA-low
SKA.TEL.LFAA.AL
Antenna & LNA
Antenna arms
Antenna other mechanical
Housing for electronics
LNA & electronic comp'ts
Connectors to Arms
Conformal coating
Subtotal
SKA.TEL.LFAA.RE
Receiver
Antenna end:
Ant. Circuit boards
Amplifier & filters
Power supplies
Shielding
Laser control
Optical lasers
Optical connectors
Subtotal
Bunker end (32 channels):
Bunker circuit board
Optical connector
PIN receiver diode
Amplifier
Filters & switch
Shielded boxes
Box cct board
Output connector
"Glue"
Subtotal
SKA.TEL.LFAA.SP
Signal processing
ADC & Beamforming (32 chans)
Input connector
Circuit board
ADC per channel
Processor FPGA
Comms out 40GigE
Clock distribution
"Glue"
Subtotal
135,312
262,144
4
1
1
2
4
2
9.00
20.00
3.00
9.00
1.00
0.50
9.50
20.00
3.00
9.00
1.00
0.50
36
20
3
18
4
1
82
38
20
3
18
4
1
84
21,496
22,020
16.0%
6
10
2
1
2
16
4
41
6
10
2
1
2
16
4
41
10,748
10,748
7.8%
20
32
64
96
160
16
32
10
20
450
20
32
64
96
256
16
32
10
20
546
7,373
8,946
6.5%
10
50
1,280
800
30
50
20
2,240
10
50
1,280
800
30
75
20
2,265
36,700
37,110
26.9%
360
3,000
3,360
360
3,000
3,360
1,720
1,720
1.2%
1,080
3,000
4,080
1,080
3,000
4,080
65
65
0.0%
262,144
2
2
2
2
2
2
2
3.00
5.00
1.00
0.50
1.00
8.00
2.00
3.00
5.00
1.00
0.50
1.00
8.00
2.00
16,384
1
32
32
32
32
16
16
1
1
20.00
1.00
2.00
3.00
5.00
1.00
2.00
10.00
20.00
20.00
1.00
2.00
3.00
8.00
1.00
2.00
10.00
20.00
16,384
1
1
32
2
1
1
1
Rack Switches (2/Tile rack)
Input cables
36 port switch
Subtotal
36
1
Aisle Switches (1/16 racks)
Input cables
36 port switch
Subtotal
36
1
Reference
Issue – Revision
Date
137,819 100.0%
10.00
50.00
40.00
400.00
30.00
50.00
20.00
10.00
50.00
40.00
400.00
30.00
75.00
20.00
512
10.00
3,000.00
10.00
3,000.00
16
30.00
3,000.00
30.00
3,000.00
Page 40 of 56
: 1-0
: 2013-06-08
RESTRICTED
Table 4: Costing for Baseline and Proposed design for LFAA (part 2
2
System & sub-system
#/system #/SKA
Cost
Each
Baseline Proposed
€
€
%age
Sub-system
Baseline Proposed
€
€
Sub-Totals/SKA
Baseline
Proposed
€k
€k
Comments
of
Total
SKA.TEL.LFAA.LINFR
Desert infrastructure
Solar power units:
Solar panel
Battery
Control electronics
Housing
Subtotal
Fibre bundles (192 strands):
Fibre separation
Core bundles:
Av length (km)
Station bundles:
Av length (km)
Bunker:
Infrastructure
Building
Internal power dist
Subtotal
Tile Racks:
Rack+backplane
Power supplies
Subtotal
Monitoring and Control
Rack
Input cables
36 port switch
Servers
Subtotal
Installation
Antenna installation
Assembly at site - mins
Deployment - mins
Subtotal
Fibre Deployment
Cables
Per cable term'n - mins
Kilometres
Laying per km
262,144
1
1
1
1
20.00
30.00
5.00
10.00
20
30
5
10
65
20
30
5
10
65
17,039
17,039
12.4%
2,731
1
2,596
1.6 per km
135
20
40.00
40.00
40
40
109
109
0.1%
3,000.00
3,000.00
4,800
4,800
12,460
12,460
9.0%
3,000.00
3,000.00
60,000
60,000
8,093
8,093
5.9%
6,000,000
100,000
6,000,000
100,000
6,000,000 6,000,000
100,000
100,000
6,100,000 6,100,000
6,100
6,100
4.4%
1
1
1
256
1
1
4,000
3,000
4,000
3,000
4,000
3,000
7,000
4,000
3,000
7,000
1,792
1,792
1.3%
2,000
1,080
3,000
9,000
15,080
2,000
1,080
3,000
9,000
15,080
15
15
0.0%
20
10
30
20
10
30
7,864
7,864
5.7%
1
1
36
1
3
2,000
30
3,000
3,000
2,000
30
3,000
3,000
262,144
20
10
1.00
1.00
1.00
1.00
2,731
1000
1.00
1.00
1,000
1,000
2,731
2,731
2.0%
1.00
1.00
30
30
206
206
0.1%
1.00
1.00
840
840
1
1
0.0%
Total # of racks
2 person day/rack
1.00
1.00
400,000
400,000
400
400
0.3%
4 person years
1.00
1.00
400,000
400,000
400
400
0.3%
5 min per fibre
6,851
30
Bunker (Construction included above)
Racks
Build & load a rack - mins
840
Commissioning
Labour
400,000
Verification
Labour
400,000
Reference
Issue – Revision
Date
20.00
30.00
5.00
10.00
1
1
1
4 person years
Page 41 of 56
: 1-0
: 2013-06-08
RESTRICTED
The estimated LFAA costings for systems which meet the Baseline Design and Proposed Design
requirements are shown in Table 3 and Table 4. These are largely broken down in the blocks as shown in
Figure 4. This implementation is our current assessment of the best implementation for the LFAA in the
context of the SKA. There are some implications on the apparent costs for some of the other element
consortia which are discussed in Section 7.3 below.
The costs are clearly not final or definitive, but represent our current best estimate, with a probable
error of approximately ±20% overall. Not included are:
•
Profit margins for companies to construct and test the various sub-systems and components;
the actual costs need to be carefully negotiated, at the time of procurement there are excellent
opportunities for reducing costs. Typical margins would be 10-20% for most of these systems,
leading to an anticipated uplift of ~€15m;
•
Equipment transport costs; and
•
Commissioning, test and operating equipment.
The cost of the array in this implementation is:
7.1
•
Baseline Design:
€148M
•
Proposed Design:
€151M
Relative costing between Baseline and Proposed Designs
The basic architecture required for implementation of the Baseline Design is identical to the Proposed
Design. The Proposed Design makes two principal modifications over the Baseline Design:
1. Enhance the frequency range of the log-periodic antenna to accommodate higher frequencies.
There is an additional small dipole arm element to extend the frequency range – this is the
smallest dipole arm element and is very low cost. The analogue amplification chain and RFoF
signal transport needs to handle the wider frequency range, which will require some additional
design work, but this is unlikely to add any significant cost since, even in the Baseline Design,
the components are typically used at well under their maximum frequency; and
2. Additional Nyquist bandpass filtering and switching for the two bands implemented. The costs
for this can be clearly seen in Table 3.
The use of identical signal processing and signal delivery means that the cost increment is small at ~2%.
There are potential major savings for the dish element, avoiding the very low frequency requirement on
SKA-mid.
7.2
Cost discussion
The costs (without profit) in the Tables show that 24% are in the antenna plus electronics; 35% in
bunker based analogue and digital signal processing; 27% is in power and signal transport infrastructure;
6% in Bunker infrastructure and 8% in installation and commissioning. This seems a reasonable
apportionment of resources. The upgradeable bunker based processing systems are about a third of the
total.
Reference
Issue – Revision
Date
Page 42 of 56
: 1-0
: 2013-06-08
RESTRICTED
Much of the external antenna and infrastructure can be used in SKA Phase 2. By upgrading the
processing additional performance and cost savings can be achieved.
7.3
Consortia interfaces and cost allocations
The LFAA will interface to:
•
SKA.TEL.INFRA Infrastructure
o Power distribution and reticulation
o Buildings and roads
•
SKA.TEL.SADT Signal and data transport
o Clock and frequency distribution
o Data transport to the correlator
o Data transport to and from Telescope Manager
•
SKA.TEL.CSP Central signal processor
o Data for correlation
•
SKA.TEL.MGR telescope manager
o Control and monitoring network information
There is no fundamental difference in the architecture of the LFAA in the context of SKA1, however,
there are savings being made over expectations for other consortia:
7.3.1
Buildings
The cost of all the buildings in the SKA is the responsibility of the Infrastructure consortium. We have
proposed a single building and included likely costs. This is bound to be much cheaper than having
multiple buildings as was expected. The buildings also include all the cooling systems necessary, which is
now much cheaper as one large facility. These are also the responsibility of the Infrastructure
consortium.
The value of these potential savings needs to be assessed by the Infrastructure Consortium, and the
appropriate cost "credit" given to the AADC when assessing and comparing LFAA solutions.
7.3.2
Power distribution
It is assumed that power is delivered by SKA.TEL.INFRA to the major installations of the LFAA. If the AA
station processing and bunker power requirements are distributed, as was anticipated in the Baseline
Design document, then the costs of reticulation are inevitably higher than delivering power to only one
very large bunker (no additional reticulation is necessary with the proposed local antenna solar power).
If, however, solar power proves not to be cost effective, then power will need to be reticulated within
the LFAA by the Infrastructure Consortium.
Further, if the same facility houses the LFAA correlator (and possibly post processing) then the overall
SKA system costs are further reduced and SKA.TEL.INFRA in particular.
The value of these potential savings needs to be assessed by the Infrastructure Consortium, and the
appropriate cost "credit" given to the AADC when assessing and comparing LFAA solutions.
Reference
Issue – Revision
Date
Page 43 of 56
: 1-0
: 2013-06-08
RESTRICTED
7.3.3
Clock distribution
As with power distribution it was anticipated in the Baseline Design that clock distribution would be to
each of the processing bunkers at every AA station, then received and distributed amongst the station
processing. It is clearly cheaper to deliver a single clock signal to a central bunker; indeed the LFAA
processing bunker could house the master clock maser so there is no wide area distribution...
The amount of these potential savings needs to be assessed by SADT consortium, and the appropriate
credit given to the AADC.
7.3.4
Telescope manager and TM data transport
Again, the expectation in the Baseline Design was to deliver the data from central facility to every AA
station bunker. There is a saving if there is only one central bunker. Further, the TM systems for the
LFAA could be housed in the same central bunker. It is certainly the case that TM is not required to send
control and monitoring information to multiple locations and one can expect the LFAA systems to
distribute and collect the information locally.
The amount of these potential savings needs to be assessed by SADT and TM consortia, and the
appropriate credit given to the AADC.
7.3.5
Data transport to the correlator
The digital data backhaul systems are similarly simplified using an architecture with a single bunker. At
the very least, there is only one central bunker to transport the data from for a correlator facility if they
are in separate buildings. However, there would appear to be an opportunity to house both the
beamforming equipment for LFAA and its associated correlator in the same facility, reducing the
backhaul costs radically.
The amount of these potential savings needs to be assessed by SADT consortium, and the appropriate
credit given to the AADC.
7.3.6
Data for correlation
This interface is a data definition interface with little opportunity for cost benefit.
Reference
Issue – Revision
Date
Page 44 of 56
: 1-0
: 2013-06-08
RESTRICTED
8
Scaling to SKA Phase 2
The fundamental architecture of the Proposed LFAA SKA1 makes scaling to SKA Phase 2 completely
feasible and Phase 1 can readily be seen as a “testbed” for SKA Phase 2 full implementation.
Table 5 below lays out the broad specification anticipated for LFAA in Phase 2, derived from the full SKA
DRM, and, for comparison, the Proposed Design specification as described in this document.
Reference
Issue – Revision
Date
Page 45 of 56
: 1-0
: 2013-06-08
RESTRICTED
Table 5: Outline predicted specification of LFAA, as a simple extension of Phase 1, in SKA2 compared to proposed
SKA1
Parameter
SKA2 Predicted
SKA1 Proposed
Number of
antennas
3-4 million
~256,000
1
1
Frequency –
low
<50MHz
50MHz
Frequency –
high
TBD
650 MHz
The performance of relatively sparse antennas at the
high-end frequencies will determined with SKA1. The
upper frequency of SKA2 will be based on the
capabilities of all parts of SKA2, taking much reduced
processing cost in Phase 2 of the SKA into account.
1.35m
(λ/2 at 111MHz)
1.35m
The spacing of 1.35m can be adjusted if required. The
SKA2 design will be scientifically justified
Station
diameter
20-200m
20-100m
With this many antennas, small stations may be hard to
process due to the very large number of stations
formed – even just in the core. However, flexibility in
station size will be retained
Polarisations
2 – linear
2 – linear
Essential to have a dual polarisation system
Number of
bands
1
2
Max. inst.
Bandwidth
600MHz
335MHz &
300MHz
Data rate
2.5 Pb/s (1015)
≥10Tb/s total
This is a very great change between SKA1 and SKA2 and
leads to the extreme performance capability
Data
flexibility
Completely
flexibly assigned
Flexibly
assigned within
a band
Data output can be assigned to arbitrary beamlets in
arbitrary directions up to the total designed data rate.
Then each experiment can be optimised, or concurrent
experiments run.
4 or 8-bit
4 or 8-bit
capability
Many experiments will operate effectively with 4-bit
data, hence doubling the total bandwidth for the same
data rate.
Types of
element
Element
separation
Sample
resolution
Comments
This is a big change between SKA1 and SKA2. A
dramatic increase in sensitivity.
The full frequency range will be covered by a single
element type. This will have been demonstrated in
SKA1.
It is conceivable that the minimum frequency will move
down for SKA2
SKA2 processing will be capable of digitising and
processing the full available bandwidth, both for
scientific benefit and reducing implementation cost
See above
The considerations for making the scaling up to SKA2 are:
Reference
Issue – Revision
Date
Page 46 of 56
: 1-0
: 2013-06-08
RESTRICTED
8.1
Bunker design issues
The main question when dealing with ~4 million antennas rather than 250,000 is how large and
complicated does the bunker become? In SKA1 the plan is to process 1024 antennas per rack. If,
somewhat arbitrarily, the reasonable limit for the number of racks in the bunker is 1000 – 1500, then
that determines the density of processing required. This means that for a total of 4 million antennas the
target for each rack to support is 4096 antennas.
8.1.1
Receiver and tile processor cards
Given that the spacing and number of receiver and tile processing cards should remain fixed, then this
requires each tile processor to support 64 antennas, compared to 16 proposed in SKA1. To achieve this
density the following design improvements are targeted:
8.1.1.1
Multiplex (electronically) both polarisations from an antenna onto one fibre.
This reduces the number of fibres by 50%, which makes a big difference to the handling and density of
the communications links. Hence, the density of the fibre on the rack front panels has only increased by
a factor of 2 over the SKA1, this appears achievable.
By reducing the number of fibres, the number of analogue gain and filter stages has been reduced by
50%, saving considerable board area.
8.1.1.2
Custom analogue devices
In order to reach the packing density required for the analogue chain it may be viable and cost effective
to have custom devices manufactured for both the antenna and bunker ends of the analogue link. The
volumes required more readily justify the investment and there will be sufficient time to plan and test
such devices.
8.1.1.3
Digitise both bands simultaneously
While the digitisation bandwidth for SKA1 is half the total analogue bandwidth for lower cost, it would
be anticipated that the performance and cost of ADCs will have improved substantially by the time of
deployment on SKA2 in ~10 years time. Consequently, it is not only a performance improvement to
digitise the whole 600+ MHz available per polarisation, but also it is cheaper since both channels can be
digitised simultaneously. There is only one fixed Nyquist filter to encompass both polarisation channels,
again a size and cost saving.
8.1.1.4
Dense, fast ADCs
The requirement at the front end of the digital signal processing is multi-ADCs per chip. They will need
to be 8-bit resolution with a sample rate of >3GS/s. To accommodate the very high data transfer rate a
suitable device could be mounted on a module with a processing device capable of processing all 64
antennas, or 128 signal channels after splitting the multiplexed signals. See below for discussion on
processing devices.
To keep inside the envelope of a standard 42U high rack, which will probably need to accommodate
extra data switches over SKA1, it will be necessary to reduce the height of the Receiver+Tile Processor to
Reference
Issue – Revision
Date
Page 47 of 56
: 1-0
: 2013-06-08
RESTRICTED
from 4U to 3.5U or even 3U. The limitation is likely to be the packing density of the analogue fibre
connections, although these may be moulded into multi-way plugs.
8.1.2
Tile signal Processing
There is little doubt that the processing ability of devices will match or exceed the requirements of a 64
antenna tile processing card; there will be sufficient bandwidth on a module. Indeed the drive to
Exascale processing for many of the large ICT manufacturers requires these levels of performance and,
by the time SKA2 is being deployed, Exascale processors will exist and as will the drive will be to reduce
power and costs. Apart from processing, a key aspect for Exascale processing is the existence of on-chip
very high-speed local network connections. This will be essential to meet the total data rates into the
correlator.
8.1.3
Station beamforming and system data rates
The required total data rate into the correlator (which determines array survey speed for a given
sensitivity) is very high, earlier concepts considered 10Tb/s for each of 250 stations. For the purposes of
this analysis if it assumed that 1024 stations are required then the data rate per station drops to
2.5Tb/s. While the on-chip beamforming for the tile is likely to be able to be performed at this data
rate, it is difficult to envisage being able to switch this much data between tiles in precisely the same
approach as in SKA1. The consequence is that a hierarchical beamformer will probably need to be
developed. In this scheme the tile beams are combined in a separate station beamformer and multiple
beams are produced within a tile beam.
For the first stage of hierarchical beamforming and all tiles observe identical areas of sky for subsequent
station beamforming. The tile beams determine a number of areas of sky that the AA is observing for
subsequent station processing. The bandwidth between the tile processors and the station processors
will be a key determinant of the quality of the station beams. The station beamformer forms the
required number of beams within each tile beam. The number of good quality beams that can be
formed within a tile beam is to be determined, but if for example 10 station beams are formed inside
each tile beam, then the data rate for the tile is about 250Gb/s. This is a feasible data rate on the project
implementation timescale.
Station beams
Central Ô
perfectÕbeam
Tile beam
Figure 12: Station beams in a tile beam. Stepped beamforming for off-centre beams on the right.
Reference
Issue – Revision
Date
Page 48 of 56
: 1-0
: 2013-06-08
RESTRICTED
While the station design needs to have a hierarchical structure due to the communication overhead, as
with most reduction techniques there will be a level of compromise involved. By processing antennas as
tiles, limited numbers of tile beams are used to create a large numbers of station beams. This structure
leads to errors in the beamforming which increase the further off-centre from a tile beam it is. This is
illustrated in Figure 12. The “perfect” station beam requires linearly increasing delays with distance
across the whole array which is illustrated in Figure 12; in the hierarchical system each tile has the same
local delay slope, these are then put together by the station processor to form station beams. Of course,
the station beam that has precisely the same delay slope as the tiles can produce a perfect beam; this
will be the central station beam with a tile beam shown in Figure 12 left hand illustration. However, for
station beams off-centre from the tile beam a series of steps in the element delays across the array is
created shown in the right hand illustration. The consequence of these regular steps is to produce
substantial sidelobes and reduced beam sensitivity.
The implementation of the hierarchical beamforming will need some consideration. It could restrict the
flexibility in station size due to the amount of data switching involved. For fixed station size then a set
number of tiles feed a single station beamformer which creates all the necessary beams for transmission
to the SKA-low correlator.
8.2
Configuration
The current specifications in the DRM for SKA2 suggest that baselines from the core will be up to a
maximum of ~200km; this may be extended to cover confusion or provide improved resolution. This is
beyond the range of analogue transmission over fibre. So, with a strongly core concentrated layout, as
would be expected, the vast majority of links can be linked directly to the processing bunker, however,
the longest baselines will now need remote digitisation and transmission of beams directly to the
correlator. This is illustrated in Figure 13.
Reference
Issue – Revision
Date
Page 49 of 56
: 1-0
: 2013-06-08
RESTRICTED
Figure 13: Enhanced configuration architecture for SKA2
8.3
Extending the frequency range for SKA2
The fundamental concept and performance of LFAA for SKA Phase 2 makes it practical to consider
extending the system to use two antenna types to give a dual-band array. The log-periodic antenna
discussed in this document would continue to be used covering the frequency range specified. However,
the processing infrastructure coupled with the ability to move analogue signals considerable distances
implies that a second, higher frequency antenna operating from (say) <500MHz to >1000MHz , maybe
1.4GHz could use the same processing system.
Assuming the new upper frequency antenna element to be a derivative of the current element, many
performance lessons will be learned from SKA1. For example, the upper band of SKA2 would also be a
sparse array, losing sensitivity at the higher frequencies, but still having a very large survey speed
capability. Conversely, the important lower part of its band will have a high effective area, giving the
large survey speeds required at those frequencies.
The LFAA processing system can be dual-ported to cope with the dual-band array and provide sufficient
channels for the upper frequencies that will likely require more antennas. Alternatively the system can
be replicated. The higher sampling frequency described in Section 8.1.1.4 above can be selected for
most benefit from this system.
Assuming that RFoF is cost effective then the two arrays can be separated appropriately.
Reference
Issue – Revision
Date
Page 50 of 56
: 1-0
: 2013-06-08
RESTRICTED
Using the same processing system for both antenna types in the same bunker also implies that the
correlator and post processing systems can also be reused.
Reference
Issue – Revision
Date
Page 51 of 56
: 1-0
: 2013-06-08
RESTRICTED
Appendix A. Possible station configuration for SKA1 Core
There is a discussion on improving beam performance in the SKA-low core. The following is an extract
from a paper by Keith Grainge. The configuration of processing described above supports this structure,
albeit requiring each tile to contribute to a maximum of three beams. This may limit the total bandwidth
of the station beam, however, if it is found to be important to support the full bandwidth using this
improvement then a higher bandwidth switch network is required – either faster switches (which will
probably be available at construction time) or an overlay network of switches giving a total bandwidth
>60Gb/s for each tile processor.
The Paper is appended to this description.
Appendix B. Tile processing and communication requirements
The cost and power requirements of the tile processing board are strongly dependent upon the relative
performance of the FPGAs available for construction and processing requirements for 32 signal
channels. In section 5.5.1 the number of FPGAs required per tile processor was estimated from scaling
from the proven UNIBOARD system using 2010 vintage FPGAs. Here, the requirements on the tile FPGAs
are estimated and compared to the known roadmaps.
The requirements on the FPGAs are to:
1.
Provide sufficient processing performance
2.
Provide links to 32 ADC channels
3.
Provide control signals for the receiver electronics (if required)
4.
Provide high speed ≥40Gb/s Ethernet or ≥56Gb/s Infiniband links for the output and control
data
An estimate of the processing required per tile processor is shown in Table 6. This is the processing
before accounting for any inefficiency. As can be seen the processing is ~2 TMACs/s.
Reference
Issue – Revision
Date
Page 52 of 56
: 1-0
: 2013-06-08
RESTRICTED
Table 6: Approximate tile processing requirements
Spectral Filter/signal channel
Sm
n
ntap
Sampling rate:
No. of spectralchannels
No. FIR taps
PPF
PRch
PPF filter processing
Processing rate
Total spectral filter processing
sch Signal channels
PTsf Total spectral proc.
Beamforming
BFch
Scplx
CM
PTbf
Channels to b/form
Complex sample rate
Complex multiply
Total b/form proc
Total processing
800 MS/s
1024 250kHz channels
5
76,800 MACs
5n(log(n)+ntap)
60 GMACs/s PPF.Sm/n
32
1,920 GMAC/s PRch*sch
34
400 MS/s
4 MACs
54.4 GMAC/s
2 of 16 signals + accumulation
Sm/2
MACs per complex accumulation
BFch*Scplx*CM
1,920 GMAC/s PTsf+PTbf
The devices used on UNIBOARD are ~0.5TMAC/s and UNIBOARD has 8 FPGAs. These are can process 16
signal channels. This would indicate that there is some inefficiency in a real implementation that might
be a factor of 2. So, sufficient FPGAs with a processing ability >4TMACs should handle 16 dual
polarisation antennas, however, having more processing available would provide more headroom.
The projected capability of 28nm FPGAs in the 2015 timeframe is ~3.5TMACs/s. There is likely to be
another generation available before construction of SKA1. Hence, two devices at a fairly well negotiated
price of €400 in ~2018 is a reasonable price/performance point.
Communication ability is also vital to the performance. The requirement for linking to 32 ADCs each
sampling at 800MS/s with 8-bits per sample is an aggregate approaching 250Gb/s. This is readily
accommodated by projected 28nm FPGAs.
The design decisions on one or two FPGAs of medium or largest sizes at the time will be on a
price/performance/ease of implementation balance. The devices will certainly have the required
capability.
Appendix C. Bunker power requirement estimation
A rough estimate of the bunker power requirements is shown in Table 7. The total power is ~2MW,
assuming that ~0.5MW is supplied by solar power for the antennas.
Reference
Issue – Revision
Date
Page 53 of 56
: 1-0
: 2013-06-08
RESTRICTED
Table 7: Estimated bunker power requirements
Reference
Issue – Revision
Date
Page 54 of 56
: 1-0
: 2013-06-08
RESTRICTED
Distribution List
Group
Other
Michiel van Haarlem (ASTRON)
Jan Geralt Bij de Vaate (ASTRON)
Andre van Es (ASTRON)
Marchel Gerbers (ASTRON)
Andre Gunst (LFAA)
Ilse van Bemmel (ASTRON)
Andrew Faulkner (University of Cambridge)
Eloy de Lera Acedo (University of Cambridge)
Nima Razavi-Ghods (University of Cambridge)
Peter Hall (ICRAR)
Tom Booler (ICRAR)
Jader Monari (INAF)
Davide Fierro (INAF)
Jonathan Hargreaves (JIVE)
Arpad Szomoru (JIVE)
Kris Zarb Adami (University of Oxford/University of Malta)
Mike Jones (University of Oxford)
David Zhang (University of Manchester)
Wanging Wu (KLAASA)
Rui Cao (KLAASA)
Ping Chen (KLAASA)
Mathias Hoeft (GLOW)
Reference
Issue – Revision
Date
Page 56 of 56
: 1-0
: 2013-06-08
RESTRICTED
1
Possible station configuration for SKA1 Core
Keith Grainge
Within the core region the elements will be randomly scattered
subject to the constraints of a minimum distance between elements,
dmin , and that the desired array filling factor, ff ill be achieved. The
station centres can then be placed in a hexagonal close packed configuration as shown in Figure 1. The scale of hexagonal close packed
Figure 1: Hexagonal close packed configuration for station centres. The black
circles have a radius, rf ull , equal to half the distance between adjacent station
centres and all the elements within this area receive their full weight√
in the beam
formation process. The green filled circles have radius equal to 3rf ull and
indicated the full extent of the area over which elements will contribute to this
particular station beam. Red arrows shows one of the baselines between adjacent
stations which will be discarded before imaging. The shortest legitimate baseline
occurs between stations whose green circles are adjacent to each other and do
not share any elements and the blue line shows as example of these. The black
baselines are both examples of legitimate baselines — note that both of the
baselines between one station and a pair of adjacent stations are legitimate,
despite the fact that the baseline between these two is discarded.
configuration is defined by the radius of the circles of which it is comprised, called rf ull for reasons that will become apparent shortly.
1
Stations will be formed from more elements than just those within
a radius of rf ull of the station centre, in fact all those elements with
√
3rf ull will contribute. However, those within rf ull will be added
with full weight during beam formation while those beyond will be
apodised with a Hanning window function (see Figure 2) i.e.
r ≤ rf ull
w=1
w=
1
2
1 + cos
π (r − rf ull )
√
3 − 1 rf ull
!!
rf ull ≤r ≤
√
3rf ull
(1)
(2)
(3)
Figure 2: Hanning tapered radial station weighting function (see eqn 1).
Therefore the signals from individual elements will be included
in multiple different
√ station beams (see Figure 3); on average they
will appear in 3 3/2.
2
Figure 3: Diagram showing the degreee of overlap between stations and that
individual elements will therefore be included in multiple different station beams
The possibility therefore exists that visibilities could be formed
between stations which incorporates elements in common which
would give rise to autocorrelation type terms appearing in the correlations; this is undesirable. Therefore any visibilities
formed from
√
stations where the baseline length is less than 2 3rf ull must be discarded — examples of these baselines appear in red and only occur
between adjacent stations in the hexagonal close packed configuration. Approximately 3N of the N (N − 1)/2 baselines between the
N stations are discarded. The shortest legitimate baseline occurs
between stations whose green circles are adjacent to each other and
do not share any elements. However, both of the baselines between
one station and a pair of adjacent stations are legitimate, despite
the fact that the baseline between these two is discarded.
1.1
Benefits
• The tapering function greatly improves the station sidelobes,
which will therefore suppress bright sources far from the field
centre and so in turn improve imaging and calibration, see Figure 4.
3
• Signals from all elements in the core are used without having
to leave gaps between circles or having a tessalating station
shape (e.g. square or hexagonal which would give rise to strong
sidelobes in some directons.
Figure 4: Plot which shows on the left the beam pattern for a (filled) station
using the weighting function in eqn 1; and on the right a uniformly weighted
circular station.
1.2
Disadvantages
• The beamformer will be more complicated: element signals will
be required by more than one station beamformer and the
√ total
number of data points to be summed will increase by 3 3/2.
1.3
Noise considerations
Compared to a telescope with circular stations of radius rf ull , the
proposed configuration has much the same number of baselines (apart
from those that are discarded) but a significantly larger collecting
area per station; so it is tempting to conclude that this implies an
improvement in the sensitivity, but this is not the case. The reason for this is that the visibilities formed from baselines such as the
black ones those shown in Figure 1 are√significantly correlated and
so the noise level will not decrease as nbaselines ; this is discussed
further in Section 1.4. There will in fact be a degradation in the
maximally achievable sensitivity resulting from the fact that some
elements receive more weight than others due to their position. The
extent to which this is a problem can be judged from the summed
4
weights 1 ; see Figure 5.
Figure 5: Plot showing the sum weights for a hexagonally close packed configuration of stations using the individual station weights according to Eqn 1. The
regions of very high weight arise from regions where 3 stations contribute with
just short of full individual weight.
1.4
Visibility correlations
As stated above, visibilities from baselines such as the black ones
those shown in Figure 1 are significantly correlated. However, this is
not necessarily a problem. It is always the case that for two visibilities, where there is overlap between their uv positions convolved by
the aperture illumination function, will be correlated, because they
are sampling a common region of the underlying uv-plane. This
correlation between the signals must be accounted for in the covariance matrix when analysing the data in the uv-plane, but the
thermal noise is usually assumed to be diagonal. However, for these
“black” baselines, there will also be a correlated component to their
noises, which must included in the covariance matrix. This idea is
1 I am not entirely sure that this is the correct metric as opposed to, say, the sum of squared
weights
5
further illustrated in Figure 6. The signal from each station can be
considered to be the sum of the voltages from each of its comprising
elements. Therefore the visibility formed by the correlation of the
two stations is the sum of all the possible products between the two
stations. The visibility is therefore a weighted sum of the values in
the uv-plane within a patch defined by the convolution of the two
station areas (i.e. the aperture illumination function). The “black”
baselines will therefore share some of these product components between individual elements. The result will be that these components
will be upweighted with respect to the rest. This is therefore equivalent in standard inteferometry to arbitrarily upweighting certain
visibilities in the uv-plane and will therefore lead to a sub-optimal
signal-to-noise but will not introduce biases.
Figure 6: The lefthand plot shows 2 stations and some of the individual baselines
that could be formed from correlating elements within each. These baselines are
then also shown in the righthand plot in the uv-plane together with the extent of
the aperture illumination function for the correlation of the two stations. The
redarrow shows the baseline between the centres of the two stations and the
redpoint show the corresponding uv position in the aperture plane.
2
2.1
Perturbations on this original idea
Randomised station positions
Using a hexagonal close packed configuration gives rise to a large
number of redundant baselines, which will be poor for filling of the
uv-plane and hence for imaging. Also, since these redundant baselines will have somewhat different beam patterns due to the random
distribution of elements, the level of redundancy is unlikely to be
6
useful for calibation. Therefore a certain amount of reduction in
filling factor and a randomisation of the station centres may be
valuable.
2.2
Different apodisation functions
Figure 5 shows that some elements are significantly upweighted with
respect the mean which will degrade the overall sensitivity of the
telescope. Randomising the station centres may alleviate this problem to a certain extent. However, adopting a different weighting
function should also be considered; plots for a tapered gaussian
weight function of the same extent are shown in Figure 7. The results look attractive and certainly imply that an improved weighting
function should be sought.
7
Figure 7: Plots for a tapered gaussian station weighting function. From top
left: the radial weighting function itself; the associated beam response; the sum
weights for a hexagonally close packed configuration of stations.
Another possible way forward would be to extend the extent of
the tapered region for each station, for example having a Hanning
window out to say 2rf ull . The disadvantages of implementing this
are:
• Increased complexity for the beam former.
• Increased number of baselines that must be rejected since the
component stations share elements.
One final possibility that should be considered is whether one
needs a “guard area” between adjacent stations whose signals are
8
correlated, since there will certainly be cross-coupling between adjacent elements leading to auto-correlation terms appearing in the
visibilities (even if these are significantly downweighted).
2.3
Apodisation through thinning
Apodisation can be achieved either through downweighting of outlying elements in an element or through reducing the density of
the elements as a function of radius (i.e. thinning) and weighting
them all equally. Downweighting of outlying elements will produce
the better beam pattern, but outside the core it will be wasteful of
element collecting area, so thinning may be a good comprimise; it
will certainly give a better beam pattern than a uniformly weighted,
uniformly dense, circular station.
3
Autocorrelation
Throughtout this document I have assumed that introducing autocorrelation type terms into the visibilities will be disasterous. While
I believe this assertion to be true, the impact of autocorrelations
should be quantified.
9

Low-Frequency Aperture Array (LFAA)

Transcription

Similar documents

Corningware Factory Store

Marabella - La Mirada

INTERCER AMIC DIGITAL TECHNOLOG Y

Corningware Factory Store - ECLIPSE INTERLOCKING TILE SYSTEM

TET HB35T antenna instructions manual

13105 W CALAVAR RD - Single Property Websites

Exquisite Cocina

Featured in urbanite Magazine

ArtMud -‐ Pieces make a Whole

The Spaces Between the Notes