Australian Square Kilometre Array Pathfinder: I. System Description
A. W. Hotan, J. D. Bunton, A. P. Chippendale, M. Whiting, J. Tuthill, V. A. Moss, D. McConnell, S. W. Amy, M. T. Huynh, J. R. Allison, C. S. Anderson, K. W. Bannister, E. Bastholm, R. Beresford, D. C.-J. Bock, R. Bolton, J. M. Chapman, K. Chow, J. D. Collier, F. R. Cooray, T. J. Cornwell, P. J. Diamond, P. G. Edwards, I. J. Feain, T. M. O. Franzen, D. George, N. Gupta, G. A. Hampson, L. Harvey-Smith, D. B. Hayman, I. Heywood, C. Jacka, C. A. Jackson, S. Jackson, K. Jeganathan, S. Johnston, M. Kesteven, D. Kleiner, B. S. Koribalski, K. Lee-Waddell, E. Lenc, E. S. Lensson, S. Mackay, E. K. Mahony, N. M. McClure-Griffiths, R. McConigley, P. Mirtschin, A. K. Ng, R. P. Norris, S. E. Pearce, C. Phillips, M. A. Pilawa, W. Raja, J. E. Reynolds, P. Roberts, D. N. Roxby, E. M. Sadler, M. Shields, A. E. T. Schinckel, P. Serra, R. D. Shaw, T. Sweetnam, E. R. Troup, A. Tzioumis, M. A. Voronkov, T. Westmeier
PPublications of the Astronomical Society of Australia (PASA)doi: 10.1017/pas.2021.xxx.
Australian Square Kilometre Array Pathfinder: I.System Description
A.W. Hotan , J.D. Bunton , A.P. Chippendale , M. Whiting , J. Tuthill , V.A. Moss , D. McConnell ,S.W. Amy , M.T. Huynh , J.R. Allison , , C.S. Anderson , , K.W. Bannister , E. Bastholm ,R. Beresford , D.C.-J. Bock , R. Bolton , J.M. Chapman , K. Chow , J.D. Collier , , F.R. Cooray ,T.J. Cornwell , , P.J. Diamond , , P.G. Edwards , I.J. Feain , T.M.O. Franzen , D. George ,N. Gupta , , G.A. Hampson , L. Harvey-Smith , , , D.B. Hayman , I. Heywood , , , C. Jacka ,C.A. Jackson , , S. Jackson , K. Jeganathan , S. Johnston , M. Kesteven , D. Kleiner ,B.S. Koribalski , K. Lee-Waddell , E. Lenc , E.S. Lensson , S. Mackay , E.K. Mahony ,N.M. McClure-Griffiths , R. McConigley , P. Mirtschin , A.K. Ng , R.P. Norris , , S.E. Pearce ,C. Phillips , M.A. Pilawa , W. Raja , J.E. Reynolds , P. Roberts , D.N. Roxby , E.M. Sadler , ,M. Shields , A.E.T. Schinckel , P. Serra , , R.D. Shaw , T. Sweetnam , E.R. Troup , A. Tzioumis ,M.A. Voronkov , T. Westmeier CSIRO Astronomy and Space Science, PO Box 1130, Bentley, WA 6102, Australia CSIRO Astronomy and Space Science, PO Box 76, Epping NSW 1710, Australia Sydney Institute for Astronomy, School of Physics A28, University of Sydney, Sydney, NSW 2006, Australia CSIRO Astronomy and Space Science, PO Box 2102, Geraldton, WA 6530, Australia The Inter-University Institute for Data Intensive Astronomy (IDIA), Department of Astronomy, University of Cape Town, Private BagX3, Rondebosch, 7701, South Africa School of Science, Western Sydney University, Locked Bag 1797, Penrith, NSW 2751, Australia Tim Cornwell Consulting, 17 Elgan Crescent, Sandbach, CW11 1LD, United Kingdom SKA Organisation, Jodrell Bank, Lower Withington, Cheshire, SK11 9FT, United Kingdom Inter-University Centre for Astronomy and Astrophysics, Post Bag 4, Ganeshkhind, Pune 411007, India Deans Unit, Faculty of Science, Dalton Building F12 UNSW Sydney, NSW 2052, Australia Sub-Dept. of Astrophysics, Department of Physics, University of Oxford, Denys Wilkinson Building, Keble Rd., Oxford, OX1 3RH,United Kingdom Rhodes University, PO Box 94, Makhanda (Grahamstown) 6140, Eastern Cape, South Africa INAF - Osservatorio Astronomico di Cagliari, via della Scienza 5, 09047 Selargius, CA, Italy CSIRO Astronomy and Space Science, PO Box 276, Parkes, NSW 2870, Australia Research School of Astronomy & Astrophysics, The Australian National University, Canberra, ACT 2601, Australia CSIRO Astronomy and Space Science, 1828 Yarrie Lake Rd, Narrabri, NSW 2390, Australia ICRAR, M468, The University of Western Australia, 35 Stirling Highway, Crawley WA 6009, Australia CSIRO Astronomy and Space Science, PO Box 2225, Ellenbrook WA 6069, Australia ASTRON, the Netherlands Institute for Radio Astronomy, Oude Hoogeveensedijk 4, 7991 PD Dwingeloo, the Netherlands Jansky fellow of the National Radio Astronomy Observatory, NRAO, 1003 Lopezville Rd, Socorro, NM 87801 USA
CSIRO’s ASKAP telescope was built and brought into operation by a large team working over a number of years. The authorship of thispaper and its ordering reflects the effort invested by team members in preparing the paper, in addition to their contributions towardsthe construction and commissioning of ASKAP itself.
Abstract
In this paper we describe the system design and capabilities of the Australian Square Kilometre ArrayPathfinder (ASKAP) radio telescope at the conclusion of its construction project and commencementof science operations. ASKAP is one of the first radio telescopes to deploy phased array feed (PAF)technology on a large scale, giving it an instantaneous field of view that covers 31 deg at 800 MHz. Asa two-dimensional array of 36 ×
12 m antennas, with baselines ranging from 22 m to 6 km, ASKAP alsohas excellent snapshot imaging capability and 10 arcsecond resolution. This, combined with 288 MHzof instantaneous bandwidth and a unique third axis of rotation on each antenna, gives ASKAP thecapability to create high dynamic range images of large sky areas very quickly. It is an excellenttelescope for surveys between 700 MHz and 1800 MHz and is expected to facilitate great advances inour understanding of galaxy formation, cosmology and radio transients while opening new parameterspace for discovery of the unknown.
Keywords:
Radio interferometers – Wide-field telescopes a r X i v : . [ a s t r o - ph . I M ] F e b Hotan et al.
ASKAP is a Square Kilometre Array (SKA ) precursortelescope located at Australia’s SKA site, the MurchisonRadio-astronomy Observatory (MRO). This radio tele-scope was developed using new phased array feed (PAF)technology to achieve high survey speed by observingwith a wide instantaneous field of view. The design con-cept is described in more detail by DeBoer et al. (2009)and the science goals in Johnston et al. (2007). Figure 1shows a photograph of ASKAP’s core taken in 2018.After more than ten years of design, prototyping,construction and commissioning, ASKAP became fullyoperational in 2019. During construction, we developeda second generation of electronics (Hampson et al.,2012) based on lessons learnt from prototype hardware(Schinckel et al., 2011; Hotan et al., 2014; McConnellet al., 2016), along with many iterations of firmware andsoftware. Thousands of components have been installedand connected together to form one of the most compli-cated and powerful radio astronomy signal processorsever developed.The first interferometry between PAF-equipped anten-nas at the MRO was conducted using a prototype systemknown as the Boolardy engineering test array (BETA,Hotan et al., 2014). This consisted of 6 antennas fittedwith first-generation receiver systems (Schinckel et al.,2011). Beyond teaching us how to improve ASKAP’ssystem design, BETA contributed to astrophysical re-search (e.g., Serra et al., 2015b; Harvey-Smith et al.,2016; Hobbs et al., 2016; Heywood et al., 2016; Allisonet al., 2017; Moss et al., 2017) including the discoveryof neutral hydrogen in a young radio galaxy at redshift z = 0 .
44 through an absorption-line search (Allisonet al., 2015). BETA also demonstrated real-time spatialradio-frequency interference (RFI) mitigation (Hellbourget al., 2016).A diverse early science program conducted on a subsetof 12 to 18 antennas fitted with the second-generationPAF systems also produced a wide range of results,showcasing the utility of a wide-field radio telescope atgigahertz frequencies. See Section 15 for details.In 2019 we commenced a program of pilot surveys withthe full array and an all-sky survey known as the RapidASKAP Continuum Survey (RACS ), demonstratingthe full science capability of the telescope for the firsttime.With up to 36 beams per antenna and 36 antennas,ASKAP produces a torrent of raw data (approximately100 Tbit / s). This is correlated and averaged at the obser-vatory, producing an output visibility data stream of upto 2 . / s that is sent to the Pawsey SupercomputingCentre in Perth, some 600 km south of the telescope, https://pawsey.org.au Figure 1.
CSIRO’s Australian Square Kilometre Array Pathfinder(ASKAP) telescope. for image processing. This gives the research communitya hint of the challenges to come in the era of the SKA.This paper describes the technical details of ASKAPand documents key performance metrics based on com-missioning data. Future papers in the series will de-scribe the image processing software and performancemetrics in more detail. We describe ASKAP’s design(Section 1.1) followed by information about the observa-tory site (Section 2), an overview of the system and itscomponents (Section 3), and more detailed informationabout key subsystems (Sections 4 to 12). We report onmeasured performance metrics in Section 13, the siteradio frequency environment in Section 14, the telescopeoperations model in Section 15 and future upgrade op-tions in Section 16.
The primary design goal of ASKAP was to maximisesurvey speed, the rate at which the telescope can ob-serve a given area of sky to a certain sensitivity limit(Johnston et al., 2007). Compared to existing telescopes,it was clear that high survey speed could be achieved bybuilding an array of many small antennas to keep theprimary beam size large, provide sensitivity over multi-ple spatial scales, and achieve good surface brightnesssensitivity. Such an array would have many baselines (an-tenna pairs) and would therefore produce a very largeamount of data and cause the computational cost ofimaging to dominate array design.ASKAP uses PAF receivers to achieve a 31 deg in-stantaneous field of view at 800 MHz. PAFs capturemore information from each antenna and can be a morecost effective way to increase survey speed than using amuch larger number of smaller antennas with cryogenicsingle-pixel feeds (Chippendale et al., 2007).The survey speed of a radio telescope array also de-pends on bandwidth (Bunton, 2003a) and the precise SKAP System Description Table 1
Key parameters of the ASKAP telescope.
Number of antennas 36Antenna diameter 12 mFocal ratio f /D . . Maximum baseline 6 kmAngular resolution 10 at 1 GHzObserving frequency 0 . . . .
58 kHzEffective system temperature 75 KSensitivity 54 m / KDual-polarisation beams 36Field of view a (800 MHz) 31 deg Field of view a (1700 MHz) 15 deg Survey speed b (800 MHz) 91 400 m deg K − Survey speed b (1700 MHz) 44 200 m deg K − Calculated via (15) b Calculated via (14)nature of the survey parameters (Johnston & Grey, 2006;Johnston et al., 2007), but can be broadly representedby the figure of merit (Bunton, 2003a; DeBoer et al.,2009; Bunton & Hay, 2010)SS = (cid:18) A e T sys (cid:19) Ω FoV (1)where A e is the total effective area of all antennas, T sys is the system equivalent noise temperature, and Ω FoV is the instantaneous (processed) field of view. Effectivearea is related to the physical area A of the antennasvia A e = ηA and η is the antenna efficiency.Table 1 summarises key indicative performance param-eters of the ASKAP array using (1) to measure surveyspeed and A e /T sys to measure sensitivity. A nominalmeasured value of 75 K is used for the effective systemtemperature T sys /η . With 36 dual-polarisation beams,ASKAP achieves a 31 deg field of view at 800 MHz butthis reduces to 15 deg at 1700 MHz (McConnell, 2017b).Field of view and survey speed may be increased at thehigh end of the band in the future by processing morethan 36 beams. See Section 13 for more details on thedefinition and measurement of sensitivity, field of view,and the resulting survey speed.ASKAP’s large field of view and increased surveyspeed came at the expense of sensitivity since coolingeach PAF to cryogenic temperatures was not economicalon the scale required at the time of design. This resultedin a beamformed T sys /η in the range 60 K to 80 K overmost of the frequency range, or a system equivalent fluxdensity (SEFD) of approximately 1800 Jy for a singleantenna (see Section 13). Table 2
RMS noise measured in pilot survey phase I data.RMS noise per beam is given as the minimum and averageover all observations in CASDA, then the minimum scaledto a standard 1 h and 288 MHz (for continuum, above theline) or 18 . ν ∆ ν t int Min Avg Scaled(MHz) (MHz) (µJy) (µJy) (µJy)944 288 10 h 24 37 74864 288 12 m 202 420 901368 144 8 h 40 70 801665 9 8 h 186 582 95856 18 . . T sys is dominated by low-noise amplifier(LNA) noise, in part because the LNA uses a transistor(ATF-35143, circa 2006) that is now quite old. Sensitivitycould be improved significantly in the future by updatingthe room-temperature LNA design with new transistors(Shaw & Hay, 2015; Weinreb & Shi, 2019) or by scalingup the manufacturability and affordability of cryogenicPAF technology like that under development for theParkes 64 m telescope (Dunning et al., 2016, 2019).Pilot survey observations (see Section 15) have testedmany different observing modes, providing direct ex-perience of the sensitivity that can be achieved withinpractical constraints. In many cases, we approach thethermal noise limit. However, for some observing modes(particularly with short integration times) other factorssuch as deconvolution errors contribute to an elevatednoise floor. The values given in Table 2 should providea realistic basis for planning observations, and may im-prove as updates are made to the telescope. The dataused to compile Table 2 were obtained using differentbeam arrangements and processing strategies, so somevariation is expected on top of the intrinsic spectralbehaviour shown in Figure 22.Importantly, the measured noise can be increasedabove the theoretical value by weighting the data dif-ferently to natural weighting. ASKAP uses a precondi-tioning approach with robust weighting. The robustnessparameter has a similar effect to the description in Briggs(1995), ranging from -2.0 (uniform weighting) to 2.0 (nat-ural weighting). The robustness values typically used forpilot survey processing were: 0.0 for continuum imagingat low frequencies, -0.5 for continuum imaging in themid-band and 0.5 for spectral imaging in the low andmid-bands. The noise increases by factors of approxi-mately 2.5, 1.5 and 1.2 for robustness values of -0.5, 0.0and 0.5 respectively.Most of ASKAP’s observing time will be dedicatedto large-scale survey projects and the observatory willprovide science-ready data products via a public archive Hotan et al.
Figure 2.
Diagram showing the relative location and size ofthe Murchison Radio-astronomy Observatory, with the final insetshowing ASKAP antennas as blue dots and service tracks as whitelines. (see Section 12). ASKAP’s deep survey projects are ex-pected to discover millions of new radio sources in thesouthern sky (Johnston et al., 2007). The large instanta-neous field of view also opens up new parameter spacefor the study of transient sources. ASKAP will excelat wide-area, high cadence surveys for slow transients;wide-area surveys for spectral line emission and absorp-tion; rapid, wide-field searches for gravitational-wavecounterparts; and the discovery and localisation of fasttransients via 1 ms cadence autocorrelations and a deepvoltage buffer (up to 14 s).
ASKAP is located on a remote site in Western Australia,specifically established as the Australian radio quietzone (WA) for the SKA and its precursors (Wilson et al.,2013, 2016). The total power in RFI signals over theASKAP band at this site is typically more than an orderof magnitude less than ASKAP’s system noise power(Chippendale & Wormnes, 2013). Legislation regulatesthe use of radio transmitters within 260 km of the site(ACMA, 2014), helping to ensure that the environmentwill remain as favourable as possible for radio astron-omy into the future. Further protection is afforded bycarefully testing and, if necessary, shielding all electronicequipment installed at the site (Beresford & Li, 2013).The observatory site (see Figure 2) is roughly 315 kmnorth-east of Geraldton, the largest town in the region. Itis operated by the Australian Commonwealth Scientificand Industrial Research Organisation (CSIRO) and hoststhe Murchison Widefield Array (MWA) (Tingay et al.,2013; Wayth et al., 2018) and EDGES (Bowman et al.,2018) telescopes in addition to ASKAP. In future, thelow-frequency component of the SKA will be established nearby.In order to prevent ASKAP’s infrastructure from cre-ating RFI, all processing hardware is contained withina dual-shielded central control building (Abeywickremaet al., 2013) that attenuates any radio emission by160 dB. Signal processing and computational systemsare housed inside the second layer of shielding, whileoffice and workshop space are inside the first layer. Thebuilding provides working space for maintenance crewsbut no on-site accommodation. The number of peopleon site is kept to a minimum and crews are housed inthe Boolardy station accommodation complex, roughly40 km from the telescope.
Figure 3 shows the major components of the telescopeand highlights how data flows from the antennas throughto the image archive. Each subsystem is described inmore detail in the sections below.ASKAP consists of 36 paraboloidal reflector antennas,each 12 m in diameter with a chequerboard PAF at theprimary focus. Within each PAF, the radio frequencysignals received by 188 active feed elements (94 perpolarisation) are converted into analogue optical signals.The optical signal from each element is transmitted overits own dedicated optical fibre back to the central controlbuilding. See Section 5 for details.Inside the control building, each PAF element signal isdigitised and then passed through a coarse oversampledpolyphase filter bank with 1 MHz channel spacing. Thisgenerates 640 or 768 channels depending on which ofthree frequency bands is being used. The digital receiversystem selects 384 of these channels and sends themvia digital optical communications links and a passiveoptical cross-connect to the beamformers.The beamformers compute the weighted sum (see Sec-tion 9) across PAF elements for each frequency channeland antenna, forming 36 dual-polarisation beams over336 MHz. The beamformed signals then pass into a finefilter bank that divides each 1 MHz channel into 54 chan-nels of 18.5 kHz width, for a total of 18 144 fine channels.Bandwidth can be exchanged for higher frequency res-olution by up to a factor of 32 using zoom modes. SeeSection 7 for details. The beamformer also generates1 ms power averages of the signal for each 1 MHz chan-nel of each beam. This is processed on site to search forfast transients.The fine channels are sent via digital optical linksand an optical cross-connect to the correlator, whichcomputes and accumulates the visibilities for each base-line. The correlator has the capacity to process 15 552fine channels for a maximum bandwidth of 288 MHz.These visibilities are sent directly to the Pawsey Super-computing Centre in the city of Perth via a long-haulunderground optical fibre network, with no buffering at
SKAP System Description
MeasurementSet files (version 2,Kemball & Wieringa, 2000) in parallel.The visibility data are imaged offline in batch pro-cessing mode using a custom software package calledASKAPsoft (Guzman et al., 2019). The Pawsey Su-percomputing Centre currently maintains a dedicatedcompute cluster called
Galaxy for the purpose of imagingASKAP science data. This system is due to be replacedwith a more powerful supercomputer in 2021. See Sec-tion 11 for details.Data products are made available to astronomers viaa public archive, described in Section 12. As shownin Figure 3, the archive can store calibrated visibilitydata (averaged in frequency to reduce disk space usage),mosaicked image cubes (but typically not individualbeam images) and source catalogues.
ASKAP’s control system is built on the Experimen-tal Physics Instrument Control System (EPICS) li-brary . Each major hardware subsystem has an associ-ated EPICS input/output controller (IOC) which is asoftware application that issues commands and gathersmonitoring information via network transactions. Sev-eral servers located at the MRO run these IOCs andthere is a dedicated monitoring and control network inaddition to the office network and astronomical datanetwork. The design of the control system is describedin Guzman & Humphreys (2010).The digital signal processing hardware was designedwith a common command and control interface thattranslates instructions from the network into firmwareregisters on field-programmable gate arrays (FPGAs)throughout the system. A common software library waswritten to support the hardware layer and this libraryis used by the EPICS IOCs. Various other commodityelectronics devices (such as drive controllers and powersupplies) are also used, requiring different network pro-tocols at the IOC level. Due to its remote location, the MRO is not connectedto a mains power grid. The site is supplied by a hybridsolar-diesel power station purpose-built with sufficientRFI shielding to protect the observatory from stray radioemissions. The power station is located approximately6 km from the core of the telescope to further aid in RFImitigation. The solar component of the power stationcan supply 1.6 MW (enough to run the entire site during https://epics.anl.gov the day) and is linked to a battery bank with 2.5 MWhcapacity. Several redundant diesel generators are themain power source at night.Power is transmitted via underground high-voltagecable to the central control building. Distribution to theantennas is on six independently-switched undergroundtracks that roughly follow the surface roads and powerseveral antennas each. A transformer at each antennapad steps the high-voltage input down to three-phase415 V for the equipment in the antenna pedestal. ASKAP’s digital signal processing hardware in the cen-tral control building consumes 280 kW of power andrequires a commensurate amount of cooling. This isachieved by circulating chilled water to each electron-ics cabinet in the central building, where internal heatexchange units extract heat from the air. The coolingsystem vents this waste heat into a geothermal exchangesystem. When the air temperature is low a water-to-airheat exchanger can be put in series, extending the lifeof the geothermal system. The entire cooling systemwas designed to minimise radio frequency emissions, asdescribed in Abeywickrema et al. (2013).
The network between Perth and the MRO consists ofapproximately 900 km of optical fibre using carrier-gradetransmission equipment. The network was designed andis managed jointly by CSIRO and AARNet (the Aus-tralian Academic and Research Network). Using densewavelength division multiplexing, the system can sup-port up to 80 channels on a single fibre pair. Over thisdistance, optical amplification is required every 80 km to100 km. Only two of these channels are used for produc-tion services, each lit by a 100 Gbit/s transponder splitout into 10 ×
10 Gbit / s services. The first runs directto the Pawsey Supercomputing Centre in Perth with4 ×
10 Gbit / s used for ASKAP data and one 10 Gbit/sservice to carry MWA data. The second 100 Gbit / s linkprovides services between Perth, Geraldton and theMRO such as CSIRO internal network connectivity, IPtelephony, and monitor and control traffic.There are additional transponders to support SKAdevelopment activities at Curtin University and for long-haul network performance testing undertaken by AAR-Net and CSIRO. Currently, there is 500 Gbit/s of litcapacity between Perth and the MRO using four of the80 available channels. There is also a parallel networkconsisting of two 1 Gbit/s links directly patched to fibrepairs between Geraldton and the MRO. This providesnetwork services at the various locations en route andacts as a backup network should the transmission systemfail, however it does not protect against a fibre cable Hotan et al.
Figure 3.
Overview of key ASKAP subsystems and data flow. cut.The network at the MRO itself is a dual-core design.One core handles all the telescope data and connectsthe digitisers and beamformers. The second core handlesthe monitoring and control of the telescope as well asthe general purpose network at the MRO. The dual-coreallows for separation of the different systems providingbetter control and resilience. The data network is large,the core router has over 2400 network ports.The second core makes extensive use of virtual localarea networks (VLANs) to partition traffic. This avoidstelescope monitor and control commands being mixedwith unrelated network traffic. VLANs provide differentnetwork connections at relevant locations across theMRO.The conditions at the MRO present some challengesto network equipment; the antenna pedestals are notcooled, so industrial switches are used, providing greaterreliability and enabling connectivity for the varied rangeof equipment that has been installed at the observatory.The flexible network design allows for services to beprovided to several clients located at the observatory,helping to broaden the research taking place at the site.
ASKAP’s antennas were designed, constructed and in-stalled by the 54th Research Institute of China Elec-tronics Technology Group Corporation (CETC54) toCSIRO’s specification (CSIRO, 2008). The antennas are 12 m diameter unshaped prime-focus paraboloidal reflec-tors with a focal ratio f /D of 0.5. They have a solidsurface, specified up to 10 GHz, and are mounted ona novel “sky-mount” that can roll the entire reflector,quadrupod and feed structure about the optical axis(DeBoer et al., 2009). Slew rates of 1 deg / s in elevationand 3 deg / s in azimuth and roll mean that ASKAP canslew to and track an arbitrary position on the sky withinone minute of request.Continuous rise-to-set tracking of most sources is pos-sible above the antenna elevation limit of 15°. From theobservatory’s position at 26 .
7° S latitude, the telescopecan observe sources between −
90° and +48° declination,although with decreasing time above the horizon forsources closer to the northern limit. The PAF’s widefield of view extends this reach by another 2 . The antenna structure consists of a steel pedestal andsupport frame, topped with solid panels made of non-metallic honeycomb sandwiched between aluminiumsheets. The feed is located at the prime focus.Due to the small size of the reflectors and the com-paratively large size of the PAF, special considerationwas given to the strength and rigidity of the prime-focussupport structure. The robust feed legs, specified tosupport a 200 kg PAF receiver, block 4 % of the totalcollecting area with the symmetric antenna design used.However, the symmetric design produces more uniform
SKAP System Description While it would be possible to track parallactic an-gle by continuously updating the electronic beamformerweights or rotating just the feed, the third axis exchangesa small amount of mechanical complexity for greatlyreduced computational complexity at the time of cal-ibration and imaging, since it maintains the angularrelationship of the feed elements with respect to thesupport structure.Unlike the azimuth axis, the antenna polarisation axis(also known as the roll axis) does not have overlappingrange limits and can only be driven slightly less than ± . The ASKAP antennas were specified to operate underconditions similar to other radio telescopes. The anten-nas will stow if the peak wind speed exceeds 45 km / hduring the last 30 min. We find that very little observingtime is lost as a result.Strong winds at the MRO are usually associated withoncoming storm fronts. These can cause the wind speedto rise extremely rapidly, sometimes faster than the timeit takes the antennas to stow. On a few occasions, thishas led to drive errors during the stow procedure aswind loading exceeded the safe operating range of themotors. To prevent circumstances such as this, we have The third antenna axis was first suggested by Dr. Peter Dewd-ney and colleagues from the Dominion Radio Astrophysical Ob-servatory in Canada.
Figure 4.
Location of each ASKAP antenna plotted relative toAntenna 1. A circle of radius 1 km is drawn for scale. The outersix antennas form a Reuleaux triangle and provide approximately10 resolution. The dense cluster of antennas in the core providesexcellent surface brightness sensitivity. implemented a storm stow system that uses satellitemeteorological data and lightning detectors to sense ap-proaching storm fronts and stow the antennas in advanceof their arrival (Indermuehle et al., 2018a).Not all nearby storms lead to high winds at the tele-scope, but associated lightning activity usually producesextensive radio frequency interference. False positivestow triggers are therefore not a major concern as as-tronomy data would be impacted by a storm in anycase. Of the 36 ASKAP antennas, 27 were placed to providea Gaussian distribution of spatial scales with a pointspread function of 30 (Gupta et al., 2008). Three addi-tional antennas were added to the core of the array toincrease surface brightness sensitivity and another sixantennas were added on longer baselines (up to 6 km) ina Reuleaux triangle (see Figure 4) to provide improvedresolution (approximately 10 ) for compact sources. An-tennas are plotted relative to ASKAP antenna 1 which islocated at 26 .
697 000 722 5° S, 116 .
631 424 286 1° E andan elevation of 361 m.ASKAP has excellent instantaneous ( u, v ) coverage(Gupta et al., 2008), making snapshot surveys and equa-torial imaging possible.
Hotan et al.
The first production prototype PAFs for ASKAP usedcopper coaxial cables for signal transport, with digiti-sation inside the antenna pedestal (Hotan et al., 2014).6 of these units were constructed for assessment in thefield. Aperture array measurements (Schinckel et al.,2011; Chippendale et al., 2014), tests at the focus of a12 m antenna located at the Parkes observatory (DeBoeret al., 2009; Chippendale et al., 2010; Sarkissian et al.,2017) and the BETA array at the MRO (Hotan et al.,2014) revealed several issues with the Mk I PAF. Theseincluded effective system temperatures in excess of 150 K(more than twice the design requirement) in the upperhalf of the frequency band (McConnell et al., 2016) andthe need to fully remove the PAF from the focus of theantenna to perform internal maintenance.In order to better assess the performance of prototypePAF designs, Chippendale et al. (2014, 2015, 2016a)developed an aperture-array method of determining thenoise temperature of a PAF beam using measurementsof a microwave absorber, the radio sky, and broadbandnoise transmitted from a reference antenna. This was avaluable step in the testing process as it helped validateelectromagnetic modelling of the PAF that in turn en-abled improvements in sensitivity (Shaw et al., 2012).Aperture-array testing was also practical as it allowedrapid testing of PAF prototypes on the ground insteadof at the focus of an antenna.During these early tests, research into RF over Fi-bre showed that the PAF signal could be transportedto the central site for processing rather than digitisingthe signals at the antenna. A major redesign was initi-ated to incorporate experience from the Mk I tests andtake advantage of new technological developments. TheASKAP Design Enhancements (ADE) project developeda Mk II system with lower overall cost, improved main-tainability and effective system temperature less than80 K across most of the band. LNAs could be changedwithout removing the PAF from the dish and sensitiveelectronics were relocated from the antenna pedestal tothe climate-controlled central building. All 36 ASKAPantennas are now fitted with Mk II PAFs and digitalsystems.
The ASKAP PAFs fill a region of the antenna’s focalplane with small receptors that consist of flat, squareconductive patches (see Figure 5) printed on a circuit-board substrate to form a planar connected-array an-tenna (Hay et al., 2007; Hay & O’Sullivan, 2008). Thedistance between each patch is 90 mm. A row of patchescan be considered as a row of bow-tie antennas that areconnected edge to edge, forming a linearly connected
Figure 5.
Photograph of a Mk II PAF installed on one of theASKAP antennas at the MRO. The chequerboard surface is visible,along with the composite case and air vents for the cooling system.Power and optical cables attach via several bulkheads on the sideof the case. array. The feed point at the centre of each “bow-tie”is differential and both orthogonal linear polarisationsare available due to the two-dimensional nature of thegrid. Each element has a radiation pattern that wouldover-illuminate the dish surface, but the beamformingprocess combines these elements to create an efficientillumination pattern for a given direction on the sky.The precise geometry of the chequerboard surface wasdetermined through electromagnetic simulation (Hayet al., 2011). Matching the simulation results to exper-imental data from the Mk I system was a key step toensure that design modifications could be assessed priorto manufacturing. Co-design of the chequerboard arrayand its LNAs was critical for achieving high sensitivity.An initial impedance target for the LNA and match-ing network design was derived by optimising for maxi-mum beamformed sensitivity over the field of view (Hay,2010). Slight modifications were made to the shape ofthe surface elements on the Mk II design to improve theimpedance match with the LNAs in the upper observingband (1400 MHz to 1800 MHz) and therefore reduce thesystem temperature.The surface panels are bonded to a ground planevia several centimetres of non-conductive honeycombbacking which provides structural rigidity as well asinsulation. Twin-wire feed lines are used to connectthe corners of each “bow-tie” element to LNAs housedunderneath the ground plane. A conformal coating isapplied to the outward-facing surface for protection fromdust and weather.
The density of electronics inside the PAF and the some-times harsh environmental conditions at the focus of the
SKAP System Description
Each pair of LNAs is housed in a self-contained RFshielded enclosure known as a domino. The domino de-sign was developed for the Mk II PAF to improve shield-ing of the components with the highest gain, and also toimprove modularity and maintainability. The dominosare bolted directly to the ground plane to provide con-ductive heat transfer and can be removed individuallyfor maintenance.Each domino has three layers that contain differentialLNAs, filters and electro-optical converters stacked inorder of distance from the ground plane. A photographof the interior of each layer is shown in Figures 6, 7and 8, and a transparent view of the assembled stack isshown in Figure 9. The signal path through the dominolayers is shown in the upper-left of Figure 11 in theboxes marked LNA, FILTER and TRANSMITTER.
Figure 6.
Photograph of the low-noise amplifier layer in thedomino, referred to elsewhere as LNA.
The top of each domino provides connections for twofibre optic cables that carry the analogue RF output,and for a copper ribbon cable that supplies DC powerand supports monitoring and control within the PAF.Due to the proximity of the LNAs, front-end monitoringhas a very high risk of generating RFI. ASKAP’s PAFmonitoring system was designed to be low-impact, with
Figure 7.
Photograph of the filter layer in the domino, referredto elsewhere as FILTER.
Figure 8.
Photograph of the optical transmitter layer in thedomino, referred to elsewhere as TRANSMITTER. no continuous clock signal and RFI filters on all con-nections to wires within the enclosure. The monitoringsystem is based on the Serial Peripheral Interface (SPI)bus and is shown in the lower-right part of Figure 11.Transmission of data packets may generate RFI, butmost monitoring is disabled during an observation, withshort updates between scheduling blocks to maintain arecord of the system health. Control of each PAF is man-aged by a front-end controller (FEC) unit that convertsinstructions from an EPICS IOC into optical signals forthe PAF and returns monitoring data to the softwarelayer.
The Mk I PAFs employed water cooling inside theirenclosures, using a heat exchange unit in the PAF andfans to circulate cool air throughout the electronics. Al-though this provided good cooling capacity, it requiredhousing a water chiller at each antenna pedestal (withassociated RFI issues) and pumping water up to thefocus of the antenna. In addition, the possibility of con-densation around the chilled water inlet or water leakageinside the PAF enclosure was a significant concern.The Mk II PAFs use heat-pipes embedded in theground plane to transport heat to the edge of the en-closure. Further heat pipes connect to thermo-electriccoolers (TECs, shown in Figure 11 and seen around the0
Hotan et al.
Figure 9.
Diagram showing the assembled domino with trans-parent walls. The LNA card is closest to the bottom of the stack,followed by the filter card in the middle and the optical transmitterat the top. edge of the PAF in Figure 5) that transport the heatto the outside of the case where a ducted fan blowsambient air over heat-sink fins. This system has proveneffective at keeping the internal temperature as much as15 °C below ambient. It can be regulated via the voltageapplied to each TEC. A servo loop keeps the internaltemperature at a steady 25 °C set point (which can bechanged in software if necessary).The PAFs were designed to operate in ambient tem-peratures up to 45 °C. This threshold is rarely exceededexcept during the most extreme summer conditions. Thecooling system is capable of maintaining safe internaloperating temperatures in all conditions experiencedto date. However, it is not always possible to keep theinternal temperature fixed. The optimal internal temper-ature of 25 °C is exceeded when ambient temperaturesrise above 35 °C, which is quite common during summer.The PAF is fitted with an automatic safety shutdownthat is triggered when the internal air temperature ex-ceeds 50 °C.The exact cooling capacity is slightly different foreach unit. When the cooling capacity is exceeded, theinternal temperature of the PAF begins to track theexternal ambient temperature, with a constant offset ofapproximately 10 °C. This can cause amplitude calibra-tion errors during astronomical data processing as thegain of the analogue electronics depends on their operat-ing temperature. The impact of this on image quality hasyet to be determined and it may be possible to correctusing time-dependent self-calibration, a temperature-based lookup table, or the on-dish calibration (ODC)system described in the next section.
Figure 10.
Diagram showing key components and signal pathsfor ASKAP’s on-dish calibration (ODC) system
For a dish with a single feed, amplitude calibration isoften obtained by placing a switched noise source inthe feed that adds a known amount of power to bothpolarisations when activated. To determine the systemamplitude gain, the power induced by the noise source issynchronously demodulated and measured with respectto the background.Individually injecting a calibration signal into eachPAF element is nontrivial due to the number of activeelements. Instead, a common noise signal is radiatedfrom the dish vertex to the PAF, illuminating all ele-ments as uniformly as possible. The calibration signal istypically set to inject around 15 K equivalent noise tem-perature into the PAF, but this can be adjusted withinthe range 100 mK to 200 K as required. We expect toreduce the calibration signal level as we gain experiencein its use. The calibration noise is currently on continu-ously during all observations, but it is largely cancelledby beamforming with maximum signal-to-noise ratiobeamformer weights resulting in less than 1 % increaseto beamformed system temperature (Chippendale et al.,2018).Using a radiated calibration signal has the disadvan-tage that it can leak into neighbouring antennas, butthe signal power is such that the correlated contributionin a nearby antenna is less than that due to leakage ofamplifier noise. Such common noise contributions couldbe further mitigated via Walsh modulation of the phaseswitches incorporated into each domino module, howeverthis has not been found necessary.Processing one or two calibration signals is a smallengineering overhead compared to the 188 signals gener-ated by each PAF. This allows the injected calibrationsignal to be digitised and correlated with the signalsfrom all PAF elements, yielding an amplitude and phasecalibration for each port. Corrections are then madeto the beamformer weights to maintain stable beampatterns and sensitivity as described in Section 9.6.The on-dish calibration system consists of severalmodules shown in Figure 10. There is one Cicada moduleper antenna located in the central building. During
SKAP System Description G = 4 dBi) located at the vertexof the 12 m dish, which points at the PAF and is orientedso that its linear polarisation is at 45° to that of thePAF elements. The illumination of the PAF is close touniform but there is a path length difference of up to21 mm from the central elements to the edge. To a firstapproximation the signal injected into all PAF elementsis the same.The Cicada also generates an optical copy of the noisesignal. Half the power is sent directly to the correspond-ing antenna’s digital receiver and half is sent to theantenna on the same multi-core fibre cable as the PAFsignals. This copy is reflected by a Faraday mirror inthe antenna pedestal and returns to the Cicada, thencontinues into the digital receiver. In the digital receiverboth optical signals are processed in the same way as thePAF signals, becoming inputs to the array covariancematrix and a higher-sensitivity calibration correlatorthat forms all correlation products between a nominatedreference signal and the PAF signals.The direct signal from the Cicada gives the best esti-mates of gain and phase variation. The signal returnedfrom the Faraday mirror allows the optical path to theLPDA to be calibrated via a round-trip phase measure-ment. This allows the phase variation of the full analogueoptical path to be characterised. The system also pro-vides a way to inject a common calibration signal intothe PAFs on all dishes, though this is not routinely used.Although the main purpose of the ODC system isto measure complex gain and provide updates to thebeamformer weights, it can also be used to improve sys-tem diagnostics in areas of RF continuity, RF distortionassessment and filter-bank frequency integrity. The maincontribution of the ODC at present is to stop beamsfrom being completely destroyed by phase slopes corre-sponding to unpredictable delay changes that occur witheach full reset of the analog-to-digital converters (ADCs).With further research it is hoped to use the ODC totemporally stabilise PAF beam (voltage) patterns tobetter than 1 % at the half power points (Hayman et al.,2010). In the BETA version of the ASKAP PAF (Schinckelet al., 2011; Hotan et al., 2014) all analogue signal pro-cessing and digitisation occurred in the antenna pedestal.At the time of the BETA design, photonics for radiofrequency over fibre (RFoF) signal transport appeared expensive and sufficiently unproven to deploy on thescale of a 6768 element array for radio astronomy.The BETA system comprised two shielded cabinetspacked full of equipment and occupying most of thespace in each antenna pedestal, along with a chillerplant located on the concrete antenna foundation padto keep both the cabinets and PAF cooled. The chillersrequired continuous maintenance and the pedestal elec-tronics were not easily accessible due to the confinedspace. Containment of radiated emissions was less thansatisfactory due to the single layer of shielding and thelarge number of filtered conductive connections to theoutside world. Even so, BETA served well to verify basicPAF performance and signal processing algorithms andwas a pivotal step to the more advanced Mk II design.For the Mk II version of ASKAP described here, theprice of high linearity uncooled distributed feedback(DFB) laser diodes had fallen. Investigations of per-formance of low-cost RFoF analogue signal transportusing these DFB components were undertaken with fi-bre spans up to 6 km. It was found that, with suitableprecautions, a directly modulated DFB laser solutioncould meet astronomy requirements (Beresford et al.,2017a). The cost of the optical transmitter was reducedto one hundred dollars rather than several hundred dol-lars for cooled laser devices or thousands of dollars forhigh quality externally modulated lasers traditionallyused in broadband applications.This allowed all analogue electronics, except the PAFitself, to be moved to the central building, which elim-inated self RFI problems, lowered cost and improvedaccessibility and maintainability of the analogue signalprocessing hardware.
Each domino houses the low-noise amplifiers (LNAs)for a pair of PAF elements (Figure 6) and there are 94dominoes in each PAF. ASKAP’s LNA design forms anactive balun by connecting the twin-wire feed line of oneconnected-array element on the PAF surface directly intotwo single-ended LNAs with a passive output-side balun(Shaw et al., 2012). A passive input-side balun wouldhave introduced too much loss. The differential inputcomplicates the measurement of LNA noise temperature,so an innovative method was developed by Shaw et al.(2012). The measured LNA minimum noise temperature T min of the current design is 30 K in the middle of thefrequency range (Shaw & Hay, 2015). After the LNA, the signal encounters selectable anti-aliasing filters (Beresford et al., 2017b) as shown inFigure 7. These filters define four observing bands sum-marised in Table 3 and Figure 14. The fourth band2
Hotan et al.
Figure 11.
The analogue signal path for a single PAF element from the feed to the ADC in the digital receiver. After the chequerboardfeed elements, the signal encounters a low noise amplifier (LNA), followed by switchable bandpass filters (BPF) and power amplifiers(PA). The transmitter module provides an option to introduce phase switching ( φ SW ), after which the signal is used to modulate alaser diode for transmission over single mode optical fibre (SMF) from the antennas to the control building. After analogue to digitalconversion (ADC), the digital signal processing (DSP) stage begins. PAF monitoring and control (M / C) is distributed to all elementsvia a system based on the serial peripheral interface (SPI) standard. Cooling inside the PAF is provided by a series of 8 thermoelectriccooler (TEC) modules. Direct current (DC) and extremely low voltage alternating current (ELVAC) powers the various subsystemsinside the PAF. The on-dish calibration system consists of a log-periodic dipole antenna (LPDA) mounted at the dish vertex andconnected to an optical receive module (ROAR) fed by a noise source (CICADA) located in the front end control (FEC) module. Thiscan be driven synchronously using binary atomic time (BAT) events loaded via a gigabit ethernet (1Ge) interface. was included for assessment of the low-frequency perfor-mance of the PAF and has not been used for astronomy.
After filtering, the signal is converted from electricalto optical for transmission back to the central building.The optical RFoF transmitter (Beresford et al., 2017c,b;Hampson et al., 2012) is shown in Figure 8. The RFsignal is directly added to the laser diode DC bias cur-rent and modulates the optical power. An average powercontrol loop ensures constant optical level, compensatingfor component ageing. The low loss in the single-modefibre allows the signal to be transported several kilome-tres. Each antenna connects to the central processingbuilding via underground fibre-optic cables containing216 cores per antenna, 188 of which carry RFoF PAFsignals. Three additional fibres are used for the RFoFODC system described in Section 5.6.
Table 3
ASKAP receiver bands.
Band Sampled Sample clock 1 MHzRF band frequency channels(MHz) (MHz)1 700–1200 1280 6402 840–1440 1536 7683 1400–1800 1280 6404 600–700 1536 768
All observing bands are suboctave as they are in thesecond or third Nyquist band of the analogue-to-digitalconverter (ADC). This mitigates the second-order har-monic distortion of the RFoF link. Radio astronomysignal paths are designed to respond linearly to theirinput and this was an important consideration for the op-tical link as well. For systems operating close to linearity,small departures are often characterised using a measureknown as the input or output intercept point (IIP orOIP) for various polynomial orders. The ASKAP RFoFlink has an OIP1 of −
10 dBm, an OIP2 of 8 . −
47 dBm (0 dBc),relative intensity noise (RIN) and shot noise define theoutput noise floor at −
72 dBm ( −
25 dBc).For narrow-band RFI equal in power to the systemnoise at the RFoF output (0 dBc) the second-order distor-tion is −
53 dBc and the third is −
102 dBc. The strongestnarrow-band RFI is due to aviation Automatic Depen-dent Surveillance-Broadcast (ADS-B) transmitters, lo-cated on aircraft flying over the observatory at highaltitude. This can be as high as 29 dBc for short periodsof time (minutes) if the aircraft passes overhead.Placement of half the anti-aliasing filter after the RFoFlink eliminates the severe in-band second-order responsefrom the RFoF link as well as out of band relativeintensity noise and shot noise (Beresford et al., 2017c).This allows the system to handle an extra 17 dB of RFI(third-order intermodulation at −
53 dBc, see Figure 12).
SKAP System Description Figure 12.
Distortion performance, measured as the spurious-freedynamic range (SFDR), of the RF over fibre link.
The other half of the anti-aliasing filter preceding theRFoF link removes out of band RFI that could corruptthe link. Adjustment of the attenuator in the dominotransmitter (see Figure 11) can reduce the signal levelon the RFoF link to increase the compression headroom,but this comes at the cost of increased additive noise.With nominal settings, the RFoF link adds 1 . . . The conversion from optical back to electrical signalsoccurs in the digital receiver (see Section 7.1). Furtheramplification increases the electrical signals to the levelneeded to drive the analog-to-digital converter (ADC) inthe digital receiver ( −
35 dBm). The ADC has a specifiedspurious-free dynamic range (SFDR) of approximately61 dB. The RF signal chain has an SFDR of approxi-mately 48 dB. Marginal degradation of SFDR occurs forlonger fibre spans due to the optical loss of 0 .
35 dB / km.The longest fibre span in the array is 6 km. The next stage in the ASKAP signal chain consists ofa distributed, custom digital signal processing (DSP)system built on field-programmable gate array (FPGA)technology (Hampson et al., 2012). This system is re-sponsible for the following key processing steps:1. sampling the PAF analogue RF signals;2. channelising the band with 1 MHz coarse resolution;3. selecting channels for further processing;4. beamforming on the selected coarse channels;5. applying time-varying coarse delays to align wave-fronts with an error of at most half a sample at
Figure 13.
The Dragonfly digital receiver module, designed byCSIRO specifically for ASKAP. There are 12 per antenna.
ASKAP’s digital receiver (Brown et al., 2014) is imple-mented with 12 Dragonfly modules (see Figure 13) foreach antenna, each processing 16 optical signals. The192 signals include 188 from each PAF, two calibrationsignals, and two spares for future applications such asRFI mitigation.The ADC devices used for ASKAP are National Semi-conductor ADC12D1600 parts that have 12-bit resolu-tion with an effective number of bits (ENOB) of 9 and afull-power analogue bandwidth of 2 . Hotan et al. clock is generated in the Dragonfly synthesiser moduleas described in section 7.6.The ASKAP digital receiver has a bandpass samplingarchitecture. It directly samples the PAF signals at RF(Figure 11) without the need for analogue frequencyconversion. The full 700 MHz to 1800 MHz frequencyrange of the PAF is covered by three selectable andoverlapping sampling bands (see Table 3 in Section 6).Each sampling band is designed to match a Nyquistzone of the ADC. Two sampling frequencies, 1280 MHzand 1536 MHz, are used to sample at RF with an in-stantaneous bandwidth of 640 MHz or 768 MHz. Fig-ure 14 illustrates how the sampling bands correspond tosampling frequency, how they overlap, and how muchusable RF band is left after filtering. The lower andupper observing bands are located in the second andthird Nyquist zones respectively of the ADC running at1280 MHz, while band 2 is located in the second Nyquistzone of the ADC running at 1536 MHz which overlapsthe adjacent bands at the other sampling frequency. Thesampling frequency and corresponding FPGA firmwareare configured at the start of an observation.An overview of the signal path following the ADC, fora single PAF element, is shown in Figure 15. Also shown,in unshaded boxes, are a number of other ancillary signalstatistics and monitoring modules. In operation, theADC histogram is particularly useful as it shows whetherthe signal is Gaussian and helps to set the signal levelinto the ADC.The sampled PAF element voltages feed directly intoa polyphase filter bank (PFB) with an oversamplingratio of 32/27 (Tuthill et al., 2012). The PFB formschannels with a sample rate of 1 .
185 MHz and 1 MHzspacing. As the channels have a noise bandwidth of1 MHz and a channel spacing of 1 MHz they will be re-ferred to as 1 MHz channels. The oversampling providessubstantial gains in sub-band fidelity and downstreamprocessing capabilities with only a modest increase insystem complexity (Bunton, 2003b; Tuthill et al., 2012).The channel-dependent frequency rotation within eachoutput channel that is introduced by oversampling iscompensated for within the PFB implementation so thatthe output 1 MHz channels are all correctly DC-centredbefore being forwarded on to the beamformer subsystem(Tuthill et al., 2015).The PFB produces either 768 or 640 1 MHz channelsacross the selected observing band (Figure 14), depend-ing on the sampling frequency. In both cases the channelsample rate is the same and the channel centre frequencyis an integer multiple of 1 MHz. The output of the PFBfor each channel is the complex-valued analytic envelopeof the full-band signal component within that sub-band,i.e. the single-sided spectrum in that sub-band, trans-lated to DC (Rice, 1982). The equivalent operations togenerate the output of the PFB for each channel area multiplication of the input by e πift , where f is the channel centre frequency, followed by low pass filteringand decimation to the output sample rate.Of the total number of output channels from thePFB, only 384 channels are selected for transport to thebeamformer subsystem for further processing. These aretransported as eight data streams each with 48 channels.However, cost constraints and computing limitationsmean that only seven of these data streams are beam-formed and only six are correlated. This limits the totalbandwidth available for imaging (see Section 11) to288 MHz.The channel selection process is very flexible and per-mits contiguous or non-contiguous groups of channels tobe selected and also multiple copies of the same channelsto be selected and transported to different downstreamend-points. The complex task of managing channel selec-tion is done by the Telescope Operating System (TOS)software. At present, only contiguous bands are imple-mented to simplify visibility data storage and imaging.In future, we plan to support two or more split bandsfor purposes such as avoiding satellite interference andobserving widely separated spectral lines.Bit growth is permitted through the PFB to ensurethat the DSP system meets dynamic range and noisefloor requirements, so the output number representationfor the coarse 1 MHz channels is 16-bit real and 16-bitimaginary two’s complement (signed) integers. The re-sulting aggregate data rate at the output of the ASKAPdigital receiver and PFB is approximately 100 Tbit / s.This is efficiently transported to the beamformer usinga custom point-to-point packetised streaming protocolthat has very low overheads. The physical transportmedium consists of 10 368 multi-mode optical fibres in864 12-fibre ribbons. All fibres in a ribbon route to dif-ferent locations and the fibre reordering is achieved byusing custom passive optical cross-connects. Output from the digital receiver for a single antenna isprocessed by seven Redback modules (Hampson et al.,2014) shown in Figure 16. Each receives 48 MHz of band-width from all PAF elements for a single antenna and cangenerate 36 independent dual-polarisation beams. Withseven modules, 336 MHz out of 384 MHz available fromthe digital receiver is processed. Wiring for the eighthmodule is in place but cost constraints meant it wasnot installed. Each Redback has six processing FPGAs,which are fully interconnected electrically. The intercon-nect is used to distribute data so that each FPGA has8 MHz of data from all 192 digitised ports (188 PAFelements + two calibration signals + two spare ports).The signal path through the beamformer is shown inFigure 17. The beamformer engine produces up to 72independent beams on the sky, each a linear combinationof M digitised port voltages x = [ x , x , ..., x M ] T . The SKAP System Description Frequency (MHz)
Fs = 1,280MS/s Fs = 1,536MS/s1st Nyquist zone,Fs = 1,280MS/s 2nd Nyquist zone,Fs = 1,280MS/s 3rd Nyquist zone,Fs = 1,280MS/s1st Nyquist zone,Fs = 1,536MS/s 2nd Nyquist zone,Fs = 1,536MS/s700MHz - 1,200MHz,Fs = 1,280MS/s, Inverted840MHz - 1,440MHz,Fs = 1,536MS/s, Inverted 1,400MHz - 1,800MHz,Fs = 1,280MS/s, Not inverted
Figure 14.
ASKAP digital system sampling bands. For convenience, we refer to these as bands 1, 2 and 3 in left-to-right order elsewherein this document. The labels indicate whether the frequency channel order is inverted or not, with respect to the natural ascending order.
Figure 15.
The signal path for a single PAF element through the ASKAP digital receiver ADC and PFB. Hotan et al.
Figure 16.
Redback module used for beamformer and correlator.Field programmable gate arrays (FPGAs) used for digital signalprocessing are hidden beneath cooling fans. Banks of dynamicrandom access memory (DRAM) are visible beside each FPGA. beamforming operation is defined by y ( n ) = w H x ( n ) (2)where w H is the conjugate transpose of the beamformerweights and x ( n ) is the vector of port voltages for asingle frequency channel at time sample n . The beam-former weights are complex valued and may be variedindependently for each 1 MHz channel of each beam.The weights are represented as 14-bit real and 14-bitimaginary two’s complement (signed) integers.The number of ports M that are weighted into eachbeam can be as many as all 192 ports from the digitalreceiver for one ASKAP antenna, but it reduces to 60ports as the number of beams increases to the full com-plement of 72 single polarisation beams. The reductionof input ports with beams is summarised in Table 4 andis required due to hardware resource limitations in thebeamformer. The selection of M ports from 192 canbe specified arbitrarily for each 1 MHz channel of eachbeam.The system is typically configured to produce 36 dual-polarised beams. The beamformed voltages for each1 MHz channel pass into a second polyphase filter bank,which performs a final fine frequency channelisation. Fornormal operation the frequency resolution is 18 . Table 4
Number of ports per polarisation per beam forvarious numbers of dual-polarisation beams.
Dual-polarisation beams Ports per beam M R = D ( x − µ x ) ( x − µ x ) H E (3)where x = [ x , x , ..., x K ] T is a vector of K array ele-ment voltages assumed to be stationary discrete-timestochastic processes, µ x is the element-wise mean valueof x , and h·i denotes expectation. For the ASKAP signalmodel the x i are the frequency-channelised PAF elementvoltages, which are complex-valued random time-serieswhose statistics are assumed to be stationary with zeromean over the beam calibration interval. Under theseconditions, the elements of R in (3) are adequately ap-proximated for each 1 MHz channel by the time-averageˆ r i,j = 1 N N X n =1 x i ( n ) x ∗ j ( n ) (4)where x i ( n ) is the n th voltage sample of the i th inputport of K = 192 and x ∗ is the complex conjugate of x .With this definition, the ACM estimate ˆ R is a 192 × r i,j is the correlation between x i and x j . This is computed for all 1 MHz frequencychannels.The full computation of the ACM has a higher com-pute load than the final correlator for each antenna.This load is reduced by using every fourth time sampleand computing ACMs for 42 × ×
42 MHz = 336 MHz of processed bandwidth. RawACM data products are transported to conventionalcompute hardware where they are used to calculate the
SKAP System Description Figure 17.
Major components of the signal path through the ASKAP Beamformer beamformer weights offline as described in Section 9.The beamformer weights are then uploaded back to thebeamformer block, which is shown below the ACM blockin Figure 17.Depicted above the ACM in Figure 17 is the calibra-tion correlator. This calculates the correlation betweena reference calibration signal and all PAF element out-puts. This is mathematically equivalent to computingone row of the ACM, but in the calibration correlatorit is done for a specific reference input without time orfrequency decimation. This achieves the best possibleSNR for tracking PAF element gains. The calibrationsystem is described in more detail in Section 5.6.The beam voltages output from the beamformer blockare fed into two separate processing modules:1. a deep “corner-turn” buffer; and2. a transient ring buffer.Both of these modules share a large double data rate(DDR) memory resource for temporarily buffering largeamounts of beam voltage data. The transient module im-plements a ring buffer that can be frozen by an externaltrigger and is described in Section 7.3.The primary signal flow for ASKAP, however, is thecorner-turn buffer that implements a streaming datatranspose, supplying large blocks of sample data for thesame beam and frequency channel to the FFB, whichis operated in block-mode for efficiency. The FFB is asecond-stage frequency channeliser implemented as acritically-sampled polyphase filter bank on the 1 MHzchannels which have a sample rate of 1 .
185 MHz. Thismodule can be configured at the start of an observationto provide a standard observing frequency resolution, orone of five “zoom” resolutions, according to Table 5.Note that since each of the 42 FPGAs in a singlePAF beamformer sub-system processes the 8 × Table 5
ASKAP frequency zoom modes. In all cases thenumber of frequency channels correlated is 15 552 for thecurrent hardware deployment.
Mode FFB Channel Correlatednumber output width Bandwidthchannels (kHz) (MHz)1 64 18 . . . . . . As shown in Figure 17, a secondary beamformer output issent to the fast-transient detector. This operates with the1 MHz channels and has an adjustable time resolutionthat is typically set to 864 µs or 1 . Hotan et al. ring buffer in the beamformer, telling it to freeze thecontents of the ring buffer (which stores an adjustableamount of raw voltage data from all beams, correspond-ing to the last 0 . . Distant astronomical sources remain effectively station-ary in the sky, while the Earth (and therefore the tele-scope) rotates once per day. This rotation means thatthe projected baseline vector between any pair of an-tennas in the array is constantly changing. Interferencefringes, therefore, move during an observation, whichwould decorrelate the signal from the interferometerif not compensated for. The process of adjusting thedelay and phase of each antenna to keep interferencefringes stationary is known as delay tracking and fringestopping.Since ASKAP is a multi-beam instrument, we inde-pendently stop fringes in the nominal pointing directionof each beam. This is straightforward when using theantenna polarisation axis to keep beams pointing at afixed position on the sky. We also offer the option tostop fringes in the boresight direction of the antenna forall beams, which can be useful for experiments such asholography where polarisation axis tracking is not used.The ASKAP digital hardware implements:1. a pure time delay to a precision of 0 .
84 µs appliedacross all beams and 1 MHz coarse frequency chan-nels; and2. a time-dependent and frequency-dependent phasecorrection applied to each fine frequency channel.A pure “coarse” time delay (with precision of 0 .
84 µs)is implemented between the beamformer and the FFBby offsetting the read pointer into the corner-turn sam-ple buffer for the FFB frame input. The coarse delayis the same for all beams and 1 MHz coarse frequencychannels for a single PAF system, but is typically differ-ent between antennas in the array to allow for a coarsecompensation of the difference in arrival time (up to20 µs) of a wavefront at different antennas. During anobservation, the required coarse delay is computed re-peatedly and set in the hardware at times synchronisedwith signals from the timing and synchronisation subsys-tem (see Section 7.6). Changes occur at the start of eachdata frame and do not introduce any discontinuities.The second “fine” stage of fringe stopping occurs inthe fringe rotator (Figure 17), which uses complex multi-pliers to rotate the phase of each fine frequency channel time-series as they come out of the FFB. The rota-tion phase is calculated from coefficients loaded via thecontrol system software and permits a unique phasecorrection to be applied at each of the four levels:1. coarse 1 MHz channels;2. dual-polarised beams;3. fine frequency channels; and4. particular time samples.The fringe rotator module exploits the fact that therotation phase, for any beam, is linearly dependent onfrequency. Furthermore, the module parameters are up-dated often enough that the time-variation of phase isadequately approximated with a piecewise-linear func-tion. Under these conditions, the rotator can calculatethe required phase, for a beam, from a unique startingphase for each coarse frequency channel and two incre-ment parameters. These parameters are the change inphase with time increment and change in phase withfrequency increment.For each coarse channel and beam combination, thephase correction applied for an individual time sample∆ t units after the update epoch and ∆ f fine frequencychannels from a reference channel, is given by φ (∆ t, ∆ f ) = φ O + φ T ∆ t + φ F ∆ f (5)where φ O is an initial phase offset value and φ T and φ F are the time slope and frequency slope values respec-tively.The array-wide synchronous event that updates thecoarse time delay has an appropriate processing delayapplied before being used to update new fringe rotatorparameters φ O , φ T and φ F so that the coarse delay andfringe parameters are consistent for each update interval.The smallest incremental change that can be madeto the fringe stopping parameters at the fine controlstage (corresponding to 18.5 kHz channel bandwidth) is0 .
206 ns in delay, 0 .
044 deg in phase and 0 . / s inphase rate. The fine-level delay slope across the bandis implemented as a combination of delay steps andmatching phase offsets with the increments given above.At the output of the FFB, the beamformer subsystemproduces 2 M − ×
64 fine frequency channels for each1 MHz channel of each PAF beam, where M is the zoommode number in Table 5 and 336 × /
27 in the coarse filter bank,each 1 MHz of processed bandwidth is contained withinthe centre 2 M − × × /
32 = 2 M − ×
54 fine channels.The redundant 2 M − × (64 −
54) = 10 × M − finechannels at the band edges of each 1 MHz sub-bandare discarded before further processing. The number ofchannels processed is independent of zoom mode. SKAP System Description The correlator subsystem is the final stage in the FPGA-based signal processing chain at the MRO central site.It takes in fine frequency channelised beam voltage time-domain data, streaming from the 36 antenna beamformersubsystems, and generates the raw visibilities (cross-correlations) for the final science data processing stage.The data are carried from the beamformer to the cor-relator via multimode optical fibre and are transportedusing a custom, unidirectional streaming protocol formaximum link efficiency.The outputs from the beamformer Redback modulesconnect to the correlator subsystem via several 12-fibreribbon cables. A single correlator subsystem (also knownas a block) processes 48 MHz of bandwidth and cur-rently six of a possible eight blocks are installed. Eachcorrelator block consists of 12 Redback modules (pro-grammed for correlator functionality) and passive opticalcross-distribution networks. The numbers in this sectionrepresent the installed capacity.The accumulation operation in computing the visibil-ities results in a large reduction in the data rate suchthat the correlator output can be sent as Ethernet dataover conventional network infrastructure to the remotehigh-performance computing facility in Perth, which isapproximately 600 km from the telescope site.The aggregate data arriving at the correlator inputconsists of 15 552 fine channels for 72 beams from eachof the 36 antennas. To provide the capacity for po-larimetry, these are configured as 36 dual-polarisationbeams (with a linear basis). For each fine channel of eachdual-polarisation beam, the correlator computes the cor-relations between all 36 antennas for both polarisations,resulting in (2 ×
36 + 1) × (2 × / ,
628 prod-ucts. This is full Stokes cross-correlation for each of 630baselines and full Stokes autocorrelation for each of 36antennas. The total correlator output in one integrationperiod, for all 36 dual-polarisation beams and all finechannels, is 2628 × ×
15 552 = 1 471 343 616 complex-valued visibilities in the current 6-block configurationwith 15 552 fine channels.The complex-valued visibilities are streamed from thecorrelator as pairs (real/imaginary) of 32-bit floating-point numbers. The data streaming cycle from the corre-lator is locked to the integrate-and-dump cycle which, inturn, is controlled through the timing event subsystem(see Section 7.6). For the minimum allowed integrationinterval of 5 s, the total output data rate from the correla-tor is 1 471 343 616 × × / (8 bit / Byte × ≈ . / s.See Section 8 for more information on system-level con-straints.Data from the correlator are transmitted as user data-gram protocol (UDP) packets, each containing up to657 complex floating-point numbers corresponding toa single spectral channel, beam and polarisation. Parti- tioning the data in this way makes it easy to scale upthe size of the telescope, but means the data must bereordered before further processing. Each UDP packetcontains a total of 5256 bytes of raw data, to which anadditional 48 bytes of metadata are added, including thereference time of the accumulation cycle. The metadataexpands the total amount of data leaving the correlatorhardware by approximately 1 %. For a synthesis array radio telescope, the accurate distri-bution of a stable clock signal to multiple endpoints is acritical part of the instrument design. ASKAP is an ar-ray of arrays and this aspect of the system architectureis particularly important and challenging, where sig-nal processing is distributed across some 4000 separateFPGA devices and 8000 separate ADCs. The stationreference and timing system (Figure 18) must ensuresynchronous sampling and processing of the data acrossthe entire array and throughout the whole DSP chain.ASKAP’s timing system is based on a general architec-ture used at CSIRO’s observatories (Hoyle & Mirtschin,2015). It distributes a 16 MHz reference clock to phasereference the signals from the antennas and a serial-encoded timestamp to enable absolute timing of syn-chronous events across the array.The 16 MHz reference clock is provided by a GPS(global positioning system)-disciplined rubidium timestandard. The rubidium clock provides the necessaryshort-term stability while the GPS corrections ensureabsolute long-term stability of the reference. Also presenton site is a hydrogen maser that can serve as an externalreference input to the rubidium sources. This is not usedfor typical ASKAP operation, but is used for VLBI (verylong baseline interferometry) experiments (e.g., Kadleret al., 2016).The rubidium time standard provides a one pulse persecond (1PPS) signal as well as the 16 MHz referenceclock to the ASKAP timing reference control computer(ATRCC), which generates a serial-encoded timestampknown as binary atomic time (BAT). The BAT serialdata stream is phase-locked to the 16 MHz reference clockand encodes a 64-bit integer representing the count ofmicroseconds since the epoch MJD 0.0, which is 00:00 UTon 17 November 1858.The BAT serial stream and the 16 MHz referenceclock are distributed together in a star topology fromthe ATRCC by the optical time and reference distributor,Figure 18, to the rest of the ASKAP DSP hardware sub-systems (digital receivers, beamformers and correlators).The transmission is by single-mode, length-matched op-tical fibres using optical splitters. The BAT signal istransported using Manchester encoding to ensure no DCbias for compatibility with optical transmission.As shown in Figure 18, the system has dual-parallel0
Hotan et al.
Figure 18.
Station reference: ASKAP’s master clock generation and event timing system. The ATRCC (ASKAP Timing ReferenceControl Computer) block is an industrial PC with a custom PCIe card that generates a serial encoded timestamp referred to as BinaryAtomic Time (BAT) that provides a precise, absolute time reference with a resolution of 1 µs. redundancy, and is remotely configurable via the OpticalTime and Reference Distributor. Either of the redundanttime standards can be selected to provide the primary16 MHz and BAT references for distribution to ASKAPDSP endpoints. The final stage in the time and referencedistribution occurs at the digital receiver, beamformeror correlator subsystems where the optically distributed16 MHz and BAT signals are used to generate all thetiming signals needed for the subsystem.In each beamformer and correlator subsystem thetiming signals are received by the Timing ReferenceDistributor (TRD). The digital receiver subsystems usethe Low-noise Reference Distributor (LRD). The archi-tecture of the TRD and LRD are shown in Figure 19.In the LRD and TRD the 16 MHz clock is multipliedup to 64 MHz and the 64 MHz and BAT signal are dis-tributed by CAT-6 cables to all modules in the subsys-tem. These signals are used to generate synchronousprocessing events in all Redback and Dragonfly modules.Each TRD can supply signals to eight Redback modules.The LRD has an additional low noise distributor forthe 64 MHz clock, which uses length-matched coaxialcable to distribute a low phase-noise copy of the 64 MHzsignal to the 12 digital receiver Dragonfly modules. Thisgenerates a high quality ADC sample clock with lowjitter.
ASKAP’s digital system is very modular and was de-signed to exceed the nominal bandwidth of 300 MHz fora full hardware deployment. For reasons outlined here,the existing system does not have a full complement ofhardware and does not exercise all possible configura-tion flexibility, though more may be unlocked in futureupgrades.
The digital receiver outputs 384 channels of 1 MHz band-width each. Processing all of these would require 8 beam-formers per antenna, but only the first 7 were installed.Wiring for the final beamformer is in place, but theconstruction budget did not allow the additional boardsto be manufactured and they were not needed to meetthe nominal bandwidth specification. Similarly, only 6of the full 8 sets of correlator redbacks were installed.Since the correlator works in blocks of 48 MHz, meetingthe nominal 300 MHz specification would have requiredinstallation and configuration of a partial block. Whilepossible, this would have increased the complexity ofcontrol system logic and configuration management, foronly a few percent improvement in continuum sensitivity.Using all available test hardware and reducing the stockof spare parts, it may be possible to install a 7th correla-tor block. However, the data ingest and image processingsystem was only designed to work with up to 300 MHzof bandwidth, so expanding to 336 MHz will be left for
SKAP System Description Figure 19.
End-point of the reference/timing distribution system. The low-noise reference distributor (LRD) generates and distributesa 64 MHz reference clock for the digital receiver sample clocks plus 12 × (64 MHz + BAT) signals for synchronous event timing. Thetiming reference distributor (TRD) is the same hardware as the LRD without the separate 12-way 64 MHz reference distributor. TheTRD is used for synchronous timing of events in the beamformer and correlator subsystems. a future upgrade once operational experience shows howmuch spare capacity exists in the post-correlator stages. The flexibility of ASKAP’s digital processing systemallows each beamformer FPGA to operate in a differ-ent frequency resolution mode with arbitrary channelsfrom the digital receiver. However, the complex channelrouting logic required to implement a generalised config-uration manager is prohibitive, and the control systemis currently restricted to a single frequency resolutionmode with contiguous bandwidth across all hardware.A summary of the available modes and the resultingbandwidth available for imaging is shown in Table 5.Regardless of frequency resolution, the 6-block correlatoralways outputs 15552 spectral channels.To reduce the amount of storage space required forinitial capture of data intended for continuum science,the ingest pipeline software can perform online averagingdown to 1 MHz frequency resolution. This reduces thestorage footprint by a factor of 54 and can be used tobalance resource requirements when scheduling observa-tions. This averaging mode is only intended for use withthe full bandwidth, not zoom modes.At the time of writing a 10 s integration interval isused in the correlator accumulation stage, reducing thedata volume by a factor of two, for a total of 1 . / sleaving the observatory. If pilot surveys show that thisaccumulation interval does not introduce any significantsmearing, it will be used for large-scale surveys as well.We have tested cycle times down to roughly 7 secondssuccessfully with the existing system, so there is someroom for improvement. Attempts to record with the nominal 5-second cadence lead to data loss that willrequire more investigation to understand. An upgradeto the ingest computing hardware at Pawsey scheduledfor November 2020 should also improve performance. At the time of design, ASKAP presented an extremecomputational challenge. Short-term storage of the cor-relator output was deemed impossible (and still presentschallenges, though advances in technology have madetemporary storage possible) and efforts were made tominimise resource-intensive aspects of downstream pro-cessing. One such limitation is that the outer six anten-nas were not expected to be used in spectral line mode,since the size of the resulting cube would be too large.Advances in visualisation and analysis software make itdesirable to consider creating high-resolution sub-cubes,at least for specific regions of interest.
The use of digital beamforming allows ASKAP to alterthe illumination of the reflector antennas (and thereforethe primary beam shape and direction) by changingnumerical coefficients. This level of flexibility is of greatbenefit, but also significantly increases the complexity ofthe system. Unlike a mechanical feed, ASKAP’s primarybeam shape and receiver noise can change due to driftsin the complex gains of the elements that combine toform a beam.Radio astronomy imaging algorithms typically assumethat the primary beam is fixed, symmetric, and identi-cal from one antenna to the next. For ASKAP, extra2
Hotan et al. calibration steps are needed to ensure that these assump-tions are met. The on-dish calibration system (ODC),described in Section 5.6, is used to monitor the complexgain of each PAF element. Although this does not pro-vide a direct measure of the beam shape, it can be usedto compensate for individual-element gain changes sothat beams are stable over time.ASKAP uses measurements of an astronomical refer-ence source to determine the numerical weights used bythe digital beamformers. A strong source is placed in thedesired beam centre for each of 36 offset positions andthe array covariance matrix ˆ R n + s is recorded. Theseobservations are compared with a noise field ˆ R n thatdoes not contain any strong sources. We compute the maximum sensitivity (maxSNR) beam-former weights following the procedure described byHotan et al. (2014) and McConnell et al. (2016), usingthe Sun as our reference source. The estimated arrayresponse ˆ a of the PAF to a far-field signal incident fromthe direction of the desired beam is given by ˆ a = u where u is the dominant eigenvector of ( ˆ R n + s − ˆ R n )(Jeffs et al., 2008). The maxSNR weights (Lo et al., 1966;Applebaum, 1976) are then computed as w = ˆ R − n ˆ a . (6)The noise-plus-signal covariance estimate ˆ R n + s ismeasured via (4) with the antenna pointed so that thecalibration source (Sun) is at the same position withinthe field of view as the desired beam pointing. The noise covariance estimate ˆ R n is measured with the antennapointing at an ‘empty’ piece of sky, typically 15° southof the calibration source.To ensure there is only one dominant eigenvec-tor, we use only X-polarisation ports to make X-polarisation beams and only Y-polarisation ports tomake Y-polarisation beams. One eigenvalue equationis solved for the sub-array of X-polarisation ports andthen a second eigenvalue equation is solved for the near-independent sub-array of Y-polarisation ports. Thismeans that the polarisation states and angles of theX and Y beams will be tied to the inherently linear andorthogonal polarisations of the X and Y elements of thechequerboard array. Commissioning measurements haveshown that this technique results in beams with low andstable polarisation leakages (Sault, 2014, 2015; Denget al., 2017) and that ASKAP already has excellentpolarisation imaging capability (Anderson et al., 2018).Weights are calculated independently for each 1 MHzchannel of each beam, which can lead to a randomisedphase response. As described in McConnell et al. (2016),we adjust the phase of the weights at each 1 MHz channelto ensure a smooth phase response with frequency foreach beam. Digital beamforming with 1 MHz resolution introducesquasi-periodic bandpass errors with a 1 MHz period.To work around this, for spectral-line observations, wesometimes reduce the beamformer resolution by fixingthe beamformer weights to be identical over sub-bandsof N × N is an odd integer. This increasesthe magnitude of the bandpass errors, but ensures theirsmoothness over the larger N × Some of these changes were required to improve therobustness and reliability of the beamforming process foroperational deployment. However, one great advantageof digital beamforming is that it allows control overaspects of the primary beam that would normally befixed by construction. One such example is the relativephase of the two orthogonal polarisations.We use the ODC system and the technique of Chip-pendale & Anderson (2019) to adjust the beamformerweights so that the XY phase of each dual-polarisationbeam pair is near zero by design. Calibrating XY phaseup-front in the beamforming allows polarisation calibra-tion and imaging to take place in the standard ASKAPsoftware pipeline, so that polarisation studies can bemade commensally with all ASKAP observations.
ASKAP’s beamformer (see Section 7.2) can simulta-neously form 36 dual-polarisation beams compared toBETA’s nine. However, as the number of processedbeams increases towards the full 36, the number of portsthat can be weighted into each beam reduces towards 60according to Table 4. This is due to hardware resourcelimitations in the beamformer.If all ports were functional and had the same gain andphase response, we could just select the weights withthe largest amplitude when selecting the finite numberof weights M that can be processed by the beamformer. SKAP System Description
In the case of a malfunctioning port, with near zero gaineven after attenuator adjustment, the covariance matri-ces are poorly conditioned and can further complicatethe solution for maxSNR beamformer weights. To avoidthis problem, we use a diagonal loading process to avoidsingularities in the inversion of the full covariance matrix.This adds a small amount of white noise to the noisecovariance matrix before its inversion and applicationin (6) to calculate the weights w = (cid:16) ˆ R n + α I (cid:17) − ˆ a . (7)The diagonal loading places an upper limit on the weightthat will be ascribed to any malfunctioning port withnear-zero gain.We use diagonal loading in a two-step process torobustly exclude malfunctioning ports from the beam-former weights. First, we select M ports with the largestsignal contribution to the beam defined by (7). The sig-nal contribution of each port is estimated with the aidof the ODC (see Section 9.5). Second, we recalculate thebasic maxSNR weights using (6) on a reduced M × M covariance matrix of the selected ports.The first step of the process tries to select the portsthat are important to the beam, using diagonal loadingto exclude malfunctioning ports. The second step appliesthe basic maxSNR algorithm, with no diagonal loadingand therefore no bias, to the selected ports.There is a sizeable literature on the appropriate se-lection of α in the diagonal loading process. We use anempirical technique guaranteed only to make improve-ments upon the maxSNR solution. We set α = β median(diag( ˆ R n )) (8)and iterate the above solution over diagonal loadingvalues of β = [0 , . , . , . , . β = 0, is equivalent tothe maxSNR solution with no diagonal loading, so theiterative process is guaranteed to achieve or improveupon the sensitivity of the basic maxSNR solution.Comparison of SEFDs achieved by this two-step pro-cess to those achieved by the raw maxSNR algorithmshow that it improves sensitivity over the raw maxSNRalgorithm in almost all scenarios. The improvement is most noticeable when there are low-gain ports with sig-nificant energy falling on them due to the focal planefield of the desired PAF beam. We use the technique of Chippendale et al. (2016a) todecouple the PAF element gains from the beamformerweights and estimate the focal plane field s matched bybeamformer weights ws = d ∗ cal ◦ w . (9)Here ◦ is the Hadamard (element-wise) product and d ∗ cal is the complex conjugate of the PAF’s responseto a plane wave. The plane-wave response is estimatedby correlating each PAF element with a copy of thebroadband calibration noise that is radiated into thePAF d cal = h x paf x ∗ rtn ih| x rtn |i . (10)Here x paf is a vector of the PAF element voltages sam-pled at the digital receiver and x rtn is the sampled copyof the calibration noise that has been sent out to theantenna and reflected back to the digital receiver via aFaraday mirror as shown in Figure 10. We commonlyrefer to d cal as the ODC response. In fact d cal are theconjugate field match beamformer weights that wouldform an aperture array beam matched to the ODCsource.Ideally, the calibration source would present a planewave to the PAF, illuminating all ports equally and inphase. In this case, s is a good estimate of the focalplane field excited when the telescope observes a pointsource in the centre of the beam created with weights w . The ports with the largest component amplitudes in s are the ports with most significant focal plane fieldexcitation by the desired signal. They are the ports withthe largest signal contribution to the beam. We infrequently observe the Sun to make new beam-former weight solutions; typically after several monthshave passed or when significant changes have been madeto the hardware. Between these solution intervals we usethe ODC system to correct existing beamformer weightsfor changes in the complex gains of the PAF elements.This is effected by the Hadamard product of the originalweights by the Hadamard (element-wise) division (cid:11) ofODC responses at the beamforming (subscript 0) andupdate (subscript 1) epochs w = d ∗ cal,0 (cid:11) d ∗ cal,1 ◦ w . (11)The phase component of this correction is applied in-dependently for each 1 MHz channel, but the median4 Hotan et al. value over the 336 MHz beamformer bandwidth is usedin amplitude. This forces the correction to be frequencyindependent in amplitude despite the ODC source hav-ing some undesired variation in spectral shape with dishantenna motion. We will revise the need for this smooth-ing in amplitude when the stability of the noise sourceis improved. The problem appears to be related to me-chanical stability of the ODC vertex antenna and its RFconnection to the noise source.Since the ODC system is constantly injecting a low-level broadband reference signal into the feed (Chippen-dale et al., 2018), the array covariance measurementsused for beamforming also capture a measure of the com-plex response of each element to the ODC signal. Whenwe wish to update the beamformer weights, we recorda smaller data product from part of the beamformerknown as the calibration correlator (see Sections 5.6and 7.2). Updated weights are calculated via (11) anduploaded to the beamformer.This process requires only a small amount of data andsince the beamformer weights are loaded via a double-buffered mechanism in the beamformer firmware, it ispossible to update the weights without interrupting anobservation in progress. The minimum update timescaleis set by the time taken to compute, upload and switchto a new set of weights (a few seconds). However, theweights are currently left fixed during scheduling blocks.We perform an update before each observing session andleave the weights fixed until it is necessary to restartthe digital receivers, typically when changing samplingbands.Currently, the ODC correction in (11) stops beamsfrom being completely destroyed by unpredictable delaychanges that occur with each full reset of the ADCs. Inthe future we hope to use the ODC to stabilise PAFbeam (voltage) patterns to better than 1 % at the halfpower points (Hayman et al., 2010).
Basic maxSNR weights can be biased at frequencies withstrong RFI. If there is RFI stronger than the beamformercalibration source in the measurement of ˆ R n + s , theresulting beam will try to point in the direction of theRFI instead of the calibration source. These weights willhave a low gain in the desired beam direction at thefrequencies with strong RFI. The astronomical signal inthe intended beam direction cannot be recovered whenusing these weights, even with careful flagging of theRFI-impacted data in the time domain.We avoid this problem by interpolating valid beam-former weights for ASKAP at interference-affected chan-nels following the technique in Chippendale & Hellbourg(2017). This uses iterative fitting of low-order polynomi-als to the weight amplitude spectrum of each individualport, taking advantage of the inherent smoothness of maxSNR weight amplitudes with frequency. Optimisation for sensitivity tends to create strongerprimary beam side lobes than a mechanical feed andallows the beam shape to take on coma distortion offaxis. Both of these can introduce undesirable effects inthe imaging process.Holography observations show that the total inten-sity beam shape is consistent with a circular Gaussianmodel to within roughly 10 %, according to measuredeccentricities (McConnell, 2016) for a beam spacing of0 . To assist with beam optimisation research, we have de-veloped a raster grid holography method that provides ameasure of the beam shapes over a large fraction of thefield of view (Hotan, 2016). Conducting a full measure-ment requires roughly six hours of observing time. Thiscan be used to measure the stability and reliability of thebeamforming process and assess the impact of changesto the constraints used when computing beamformerweights.One of the next planned improvements to the imag-ing software will allow primary beam correction in themosaicking stage using beam shapes measured via holog-raphy, instead of the circular Gaussian model currentlyapplied.
Digital beamforming provides the freedom to place pri-mary beams anywhere within the field of view of theantenna. However, in practice it is important to max-imise the packing efficiency of the beams and developstandard arrangements that minimise sensitivity fluc-tuation across the field of view and also tessellate wellwhen covering larger areas with multiple pointings.We call the arrangement of beams within the field ofview the footprint. Two footprints are in common useon ASKAP: square_6x6 that follows a square grid and
SKAP System Description Figure 20.
The two commonly-used footprints: square_6x6 (left)and closepack36 (right). In this case both have beam spacings(pitch) of 1 .
0° and a position angle of zero. The scales are indegrees, and celestial north (west) is to the top (right) of bothdiagrams. The + symbol represents the optical axis of the reflectorand the lines extending from this position ending in filled circlesrepresent the pointing shift required to optimally interleave thefootprint by using other observations to fill in the least sensitiveparts. The square arrangement has larger intrinsic sensitivityvariation but requires only one interleaving position, while theclosepack arrangement is more uniform initially but requires twointerleaving positions for improvement. closepack36 that offsets alternating rows to providemore uniform sensitivity as shown in Figure 20.Note that neither of these footprints has a beam lo-cated on the optical axis of the antenna, which wouldhave been desirable for calibration purposes. Havingan extra beam would have allowed the use of a hexag-onal arrangement containing 37 beams, including oneon the optical axis. One of the future enhancementswe are considering would allow the exchange of band-width for additional beams. As part of this work, wemay investigate adding one more beamformer engine tothe standard full-bandwidth mode if resources on theFPGAs will allow it.
Since ASKAP will spend most of its time conducting all-sky surveys, it is useful to define a standard tiling schemeto cover the celestial sphere with the footprints describedabove. In consultation with ASKAP’s survey scienceteams , we have developed a tiling scheme in whichmost of the sky is covered by bands of footprints stackedend-to-end at constant declination (see Figure 21). Thisbecomes less efficient towards the celestial poles, so at 72degrees above or below the equator we switch to polarcaps consisting of a square grid that is truncated atthe overlap with the outermost declination band. Thisapproach allows a fixed beam footprint to be used for theentire survey. Methods such as HEALPix (Górski et al.,2005) were considered, but deemed unsuitable. HEALPixwas designed to efficiently partition a spherical surfaceby allowing the shape of each partition to vary. Matching Figure 21.
Two views of the survey tiling scheme: equatorial(left) and polar (right). The boundaries of the two polar zones atΛ − and Λ + are indicated in the left-hand panel. The whole sphereis shown here; the tiling can be defined for any coordinate system(Equatorial, Galactic, etc) and then transformed into the naturaloperating frame. the different partition shapes would greatly increase theoperational overheads associated with beamforming.The number of tiles required to cover the sky dependson the beam spacing within each tile, which can be ad-justed to trade between survey speed and uniformity ormitigate widefield polarisation leakage. This optimisa-tion is also a function of the observing frequency.The first ASKAP all-sky survey, known as the RapidASKAP Continuum Survey (RACS), used a square_6x6 footprint with beam spacing of 1 .
05° at a centre fre-quency of 888 MHz. RACS consisted of 904 tiles ob-served for 15 min each, covering the entire sky south of+40° declination. A future publication will describe theforthcoming data release and source catalogue.
Beamformer weights are stored in a custom HDF5 (TheHDF Group, 2020) file format. Currently, each schedul-ing block specifies which weights file to use. We have de-veloped, and are now commissioning, a database servicethat tracks weights files and can automatically retrievethe best available weights for a requested footprint andfrequency setup. It is important to ensure that detailsof the beamformer weights are included in the metadataassociated with an observation, to provide a completerecord of the system configuration.
10 RAW DATA INGEST
Visibility data leaves the ASKAP correlator hardwareas UDP packets which are transmitted over a standardEthernet network. An EPICS IOC running at the obser-vatory gathers data from each correlator block and trans-mits it to the Pawsey supercomputing centre. Pawseyhosts a dedicated ingest cluster for ASKAP that is iso-lated from general purpose activity to ensure predictable6
Hotan et al. load.Each UDP packet sent from the observatory containsone fine channel (packed in a non-contiguous way), onebeam and a range of baselines (the whole baseline spaceis split into four groups). A specially-designed ingestpipeline software application receives the data using mul-tiple threads, tags it with metadata from the telescopecontrol system and reorders it into the form requiredfor further processing. Data sent from each correlatorhardware module are staggered in time to spread thenetwork transmission load out over the full accumulationcycle. The ingest pipeline can also perform basic taskssuch as channel averaging if required.Data are written in parallel to a series of CASA
MeasurementSet (version 2, Kemball & Wieringa, 2000)files on a Lustre file system that is common to all ingestnodes. We need to split each of the PAF beams into itsown measurement set to avoid writing to a single file atan excessive rate. For full spectral resolution mode, wemust further divide each beam into six frequency chan-nel groups. ASKAP’s image processing pipeline buildsa complete image by combining data from all of theseindependent measurement sets.Although the performance specifications of the filesystem are adequate for capturing ASKAP data at fullrate, we have run into practical problems where otherprocesses operating on the same Lustre hardware canimpact performance.
11 SCIENCE DATA PROCESSING
The plan (Cornwell et al., 2016) for ASKAP sciencedata processing (SDP) was to have a near-real-timeautonomous pipeline. This would handle the calibration,imaging, cataloguing and archiving of the data, in away that would keep up with the observing rate. Theplan hinged on the ability to forward-predict calibrationparameters for the array using a sky model and applythe solution on-the-fly.Through commissioning and early science, we wereunable to bootstrap this method of calibration and in-stead reverted to the use of dedicated observations ofa standard calibrator source in each beam. Calibrationcan therefore be applied offline and improved iterativelywith additional flagging, which has been essential tomeeting data quality requirements. Online calibrationis now considered a possible future improvement, oncea suitable sky model has been obtained from the firstall-sky surveys calibrated using traditional methods.Using dedicated calibration observations increases op-erational overheads. Due to the large number of beams,the current calibration procedure uses about 2.5 hoursof observing time and must be done whenever the beam-former weights are changed or updated, which is typically http://lustre.org once every day or two. This is feasible for survey ob-servations, but adds scheduling complexity and reducesoverall efficiency by about 10 percent. Although the timecommitment for per-beam calibration using a referencesource is significant, the storage requirements can bemitigated by keeping only the bandpass solution or dis-carding visibility data from beams when they are notpointing at the calibrator.The batch-processing pipeline that has been devel-oped provides the functionality required to process pilotsurvey data to a science-ready state. This section de-scribes the high-level functionality of the pipelines, withmore detail provided elsewhere (Whiting, 2020).The pipeline and the software it runs are collectivelyknown as ASKAPsoft. The calibration and imaging soft-ware of ASKAPsoft has been custom-written to addressthe particular imaging requirements of ASKAP, espe-cially imaging over a wide field-of-view and in near-real-time. The processing pipeline currently runs on
Galaxy , theCray XC30 at the Pawsey Supercomputing Centre. Itcreates a workflow from a series of connected computejobs, running under the Slurm workload manager, withappropriate dependencies to ensure completion in thecorrect order.The processing is mostly performed on individualbeams independently, before combining images in a lin-ear mosaic. The top-level workflow for a single beamcan be expressed as follows:1. Bandpass calibration is done using a separate obser-vation of PKS B1934 − The core ASKAPsoft imaging software is available asthe
Yandasoft package https://bitbucket.csiro.au/projects/ASKAPSDP/repos/yandasoft . Yanda is the Wajarri word for image . SKAP System Description
Most ASKAP observations are calibrated in two steps.First, the flux scale and frequency-dependent complexgains (the “bandpass”) are determined from a dedi-cated observation of PKS B1934 − −
638 and observes for several min-utes per beam. Although we currently assume a simplepoint source model for the calibrator source, Heywoodet al. (2020) have shown that slight improvements couldbe made using a model of the full field that includes con-fusing sources within the primary beam. The calibrationparameters derived from this observation are applied tothe raw science data prior to any other processing.Since the bandpass gains are determined from a dif-ferent observation, some refinement to them is oftennecessary due to time-variable factors such as atmo-spheric and ionospheric conditions that can impact thecalibration solution at a low level. This is currently donethrough a self-calibration process, using the continuum- averaged dataset. A continuum image is made, thencalibration is performed using a shallow sky model de-rived from that image. The complex gains are solved forwithin short ( ∼ Wide-field imaging is enabled via the W -projection algo-rithm (Cornwell et al., 2005), allowing handling of image-plane effects due to non-coplanar baselines. This givesaccurate imaging over the field of view of a single beam.Multiple beams could be combined via AW -projection(Bhatnagar et al., 2008, 2013), taking into account theaperture illumination, although this approach is cur-rently not used due to its large memory requirements.As described above, the aperture correction is insteadapplied when mosaicking the individual beams together.The near-real-time imaging requirement has driven anumber of design decisions within the code. The weight-ing of visibility data prior to imaging is traditionallydone with an initial pass over the data to determine thedistribution of visibilities as a function of baseline lengthand hence the weighting function, and then a subsequentpass for the imaging. To prevent repeated iteration overthe large ASKAP data products, the ASKAPsoft codeincorporates the preconditioning approach (Rau, 2010),where the re-weighting is performed after the construc-tion of the dirty image and point spread function (PSF).ASKAPsoft imaging makes use of robust weighting, simi-lar to that described in Briggs (1995), allowing a tradeoffbetween the suppression of sidelobes and the overall sen-sitivity.Continuum imaging is performed using Taylor-termmulti-frequency synthesis (Rau, 2010; Rau & Cornwell,2011). This creates an image of the total intensity ata reference frequency, along with images of the spec-tral variation across the band, from which can be deter-mined the spectral index (and curvature, if enough termsare used) of sources. For spectral imaging, each chan-nel is imaged independently, using either the averaged(continuum) data to create “continuum cubes”, or thefull-resolution data to create spectral cubes. Continuumcubes can be made in all Stokes parameters, allowingpolarisation imaging.The imaging software has distributed processing de-signed in to the code at a low level. This allows theprocessing to be distributed over many compute nodes.The continuum imaging has the gridding distributed per8 Hotan et al. coarse channel and Taylor term, with the deconvolutioncurrently performed on the head node after combina-tion of all channels. The spectral imaging is performedindependently on each channel, and so is extremely par-allel, allowing distribution over as many nodes as canbe accommodated.It should be noted that this extremely parallel caseassumes a coarse form of Doppler correction done perbeam and with only whole-channel offsets. This may bea limiting factor for some science applications.Images are currently stored in FITS format (Penceet al., 2010) for broad compatibility and because thescience data archive was designed around this format.ASKAP data cubes can be very large (tens of terabytes)and it is possible that future research and developmentmay provide a more efficient data storage format foruse in parallel processing environments with remotevisualisation services. If a new standard is established, wewill investigate upgrading ASKAP’s software platformsto incorporate it.
As part of the pipeline processing, we make cataloguesof the sources in the images, using the Selavy source-finder (Whiting & Humphreys, 2012). Selavy uses thesource-finding algorithms of Duchamp (Whiting, 2012),a source-finder developed for three-dimensional datasets.It can produce catalogues of spectral-line sources fromthe spectral cubes, largely following the Duchampmethodology, as well as catalogues of continuum sources.The continuum source-finding produces two differentcatalogues. The first is the catalogue of islands in the im-age. These are groupings of image pixels that are abovethe detection threshold. Each island is parameterisedby its location and shape, determined solely from thedetected pixels. To each island, we fit some number oftwo-dimensional Gaussians — these are the components ,and the second catalogue reports their parameters. Eachcomponent will have a location, flux and shape, as wellas information on the spectral index. This is obtained byfitting the Gaussian in question to the Taylor-1 image,which gives the product of the Taylor-0 flux and thespectral index.Source finding is an active topic of research withinthe broader community and some of the ASKAP surveyscience teams will produce their own catalogues usingdifferent methods and tools. These are referred to as“value-added” data products. For example, the SoFiAspectral-line source-finding software developed by Serraet al. (2015a) is being used on neutral hydrogen (HI)emission-line pilot survey data, and the FLASHfinderline-finding software developed by Allison et al. (2012) isbeing used to search for HI and OH absorption lines onsightlines to bright radio continuum sources in ASKAPspectral-line cubes.
12 DATA ARCHIVING AND ACCESS
Science data products produced by the ASKAPsoftpipeline are archived and made available to the astron-omy community by the CSIRO ASKAP Science DataArchive (CASDA ). Data deposited into CASDA aremade available to the survey science team (SST) mem-bers initially, and then made publicly available afterquality control and validation.CASDA is described in detail in Chapman et al. (2017)and Huynh et al. (2020). In brief, CASDA is implementedacross two data centres, Pawsey and the CSIRO datacentre in Canberra. The CSIRO data centre runs theCSIRO data access portal, an enterprise-wide systemthat archives and provides access to data across manyareas of CSIRO research, including ASKAP/CASDA.The ASKAP data are deposited on to tape and diskstorage at Pawsey, with storage managed by the next-generation archive system (NGAS) (Wu et al., 2013).Functions such as metadata search and authenticationare implemented at the Canberra data centre, whilefunctions that need to be close to the data, such as datadeposit and data access, are implemented at Pawsey.At full operations the ingest data rate of ASKAP isexpected to be approximately 200 TB / day, or more than70 PB / year, for 100 % duty cycle. Given this extremelyhigh data rate, the full resolution uncalibrated visibilitiesare not stored. The data products archived and servedby CASDA are:• calibrated visibilities (continuum resolution);• images and cubes, including intermediate imagingproducts such as clean models;• catalogues; and• metadata from ASKAPsoft and quality assessment.In addition, derived data products such as momentmaps and spectra of detected sources will also bestored in CASDA, once they are produced by ASKAP-soft (planned in future development). Given 15 TB / dayto 20 TB / day of expected science data products fromASKAPsoft we expect to have more than 5 PB / year ofdata flowing into CASDA during full operations. CASDAhas a current allocation of 10 PB of long term tape stor-age (+10 PB for redundancy) at Pawsey.CASDA data search and access is possible throughboth the data access portal web user interface andvirtual observatory (VO) services. CASDA implements awide suite of VO services to maximise the usability andinteroperability of ASKAP data products. For example,the table access protocol (TAP) can be used to searchfor specific observations via an application such as TOP-CAT (Taylor, 2005). Uploaded catalogues can be filteredfor desired radio sources using the astronomical dataquery language (ADQL). Python scripts can be used https://research.csiro.au/casda https://data.csiro.au SKAP System Description . ACASDA module has also recently been added to theastropy astroquery package.
13 PERFORMANCE MEASUREMENTS
In this section we provide a broad overview of mea-sured key performance parameters. Detailed analysisof ASKAP pilot surveys will yield further performanceinformation that will be summarised in future papers.
The sensitivity of each ASKAP antenna depends onthe physical area A of the reflector aperture and theeffective system temperature T sys /η achieved by thereceiver as installed on the reflector. Here T sys is thesystem equivalent noise temperature and η is the antennaefficiency.Both T sys and η vary with beamformer weights fora PAF receiver, so we report results for beamformerweights that maximise sensitivity (as used in operation;see Section 9). Figure 22 gives T sys /η across the ASKAPband measured with observations of PKS B1934 − T sys /η when summarisingASKAP performance.The midband T sys of an ASKAP PAF beam, excludingspillover contribution from the ground, is 50 K (Chip-pendale et al., 2016a). This measurement was made withthe PAF on the ground pointing at the zenith, but usingthe same MaxSNR beamformer weights as are appliedwhen the PAF is installed on the reflector.Astronomers characterise sensitivity of each antennaplus receiver system in terms of the system equivalentflux density (SEFD) defined bySEFD = 2 kT sys ηA . (12)This is the flux an unpolarised point source must haveto yield a signal-to-noise ratio of one at the outputof a given beam. The factor of two appears in (12)because a single-polarisation beam can only collect halfof the total incident flux from an unpolarised source(Wrobel & Walker, 1999). ASKAP’s nominal T sys /η of75 K corresponds to an SEFD of approximately 1800 Jyfor a single ASKAP antenna.The sensitivity of the telescope can also be defined bythe effective area divided by the system temperature S (Ω) = A e T sys . (13) https://github.com/csiro-rds/casda-samples https://astroquery.readthedocs.io Figure 22.
Effective system temperature T sys /η across theASKAP band. The median value over all antennas is plotted(red) for a beam close to the antennas’ boresight. The black curveis a polynomial fitted to the spectrum. Sensitivity estimates in theshaded frequency ranges are difficult because of persistent RFI.The ASKAP operating bands are also shown. This is generally a function of direction Ω within thefield of view and antenna efficiency η achieved togetherwith the PAF receiver. The sensitivity of the full ASKAParray evaluates to 54 m / K for a beam close to theboresight.
ASKAP’s field of view is set primarily by the size ofthe array of receptors at the focal plane and the sizeand focal ratio of the reflector. It is also modified by theaperture illumination, which is in turn controlled by thebeamformer weights.A measurement of the field of view with maxSNRweights was made using the ASKAP-12 array and re-ported in McConnell (2017b). More recently, that mea-surement was compared with the observed variation ofimage noise over the field of view. Figure 23 shows thiscomparison in one dimension as an east-west profile ofimage sensitivity across the median of a sample contain-ing 44 36-beam fields. The observations were made withthe square_6x6 footprint using a beam spacing of 1 . field of view at 850 MHz.The field of view reduces to 15 deg at 1700 MHz due tothe current limit of 36 dual-polarisation beams. The fieldof view at the high end of the band may be increased to30 deg in the future by adding or reconfiguring digital0 Hotan et al.
Figure 23.
The sensitivity profile over the field of view. The blackline traces the observed sensitivity (see text), and the red lineshows an analytic approximation of the sensitivity estimated fromSEFD observations. These data were obtained from observationsusing the square_6x6 footprint with a beam pitch of 1 .
05° at acentre frequency of 888 MHz. hardware to process more than 36 beams.
As seen in Figure 23 the shape of ASKAP’s field of viewis very different from that of a dish with a single feed,which is approximately Gaussian. The traditional surveyspeed figure of merit (1) should not be calculated usingthe half power field of view. When the sensitivity variessignificantly over the field of view, (1) should be viewedas a point estimate that is integrated over the field ofview to yield the survey speed (Bunton & Hay, 2010)SS = Z S (Ω) d Ω . (14)Equivalently, (1) may be used with the maximum sensi-tivity S max and a survey speed weighted field of viewΩ FoV = 1 S Z S (Ω) d Ω . (15)All ASKAP survey speed and field of view numbers arecalculated by (14) and (15) in this paper.Using this definition the survey speed of ASKAP iscurrently 91 400 m deg K − at 800 MHz with a smoothreduction to 44 200 m deg K − at 1700 MHz. It maybe possible to double survey speed at the high end ofthe band in future by processing more than 36 dual-polarisation beams. However, with current hardware theextra beams would come at the expense of bandwidth.As a point of comparison, evaluating (14) for a Gaus-sian beam yields a survey speed of 0 . G S , whereΩ G is the half power field of view of the Gaussian beam.Using this measure, the survey speed of ASKAP at1 . T sys /η of40 K (Hobbs et al., 2020).
14 RADIO FREQUENCYINTERFERENCE
ASKAP’s location on a protected radio quiet site makesit possible to observe in parts of the spectrum that are nolonger usable at other radio telescopes. In particular, theband from 700 MHz to 1080 MHz is almost entirely freeof persistent RFI. Occasionally, atmospheric conditionscan lead to ducting of distant mobile communicationssignals, but this can be predicted to some extent basedon meteorological data (Indermuehle et al., 2018b) andimpacts relatively few observations.Satellites and aircraft have been the most commonsources of RFI in observations to date (Indermuehleet al., 2016). Aircraft automatic dependent surveillancebroadcast (ADS-B) transponders transmit at 1090 MHzand are visible when flights pass over the site. This signalis strong, but has a relatively low duty cycle so futurereal-time interference mitigation methods may be ableto remove it.Satellite navigation systems are visible between ap-proximately 1150 MHz to 1300 MHz. Occupancy of in-terference approaches 100 % at several places within thisrange and typical flagging algorithms leave very littleusable data. The band from 1500 MHz to 1620 MHz alsocontains significant satellite interference, with up to 70 %occupancy.In future, adaptive beamforming could be used as aform of RFI mitigation on ASKAP, by placing nullsat the locations of transmitting satellites (Black et al.,2015; Hellbourg et al., 2014, 2016). This would involvechanging the beams during an observation, which wouldneed main-beam constraints and corrections for patternrumble bias (Jeffs et al., 2008) to avoid invalidating arraycalibration solutions.
15 TELESCOPE OPERATIONS
In a departure from other Australia Telescope NationalFacility (ATNF) telescopes, 75 % of observing time onASKAP has been pre-allocated to several internationalsurvey science teams with five-year observing plans. Theremaining 25 % of available time will be allocated toguest science projects, subject to peer review throughthe ATNF time allocation committee (TAC). ASKAPwill be operated by CSIRO staff and astronomers willnot interact directly with the telescope. Survey plans andguest science proposals will be converted into schedulingblocks that are observed, calibrated and imaged as partof operations. Science-ready images will be made avail-able on CASDA. This means that calibration, imaging,archiving and initial quality control all fall within thedomain of the observatory.
SKAP System Description Table 6
ASKAP science operations timeline
Dec 2016 Early science beginsJuly 2018 Early science endsFeb 2019 Full array operationalJuly 2019 Pilot Surveys Phase I observing beginsMay 2020 Pilot Surveys Phase I observing endsQ1 2021 Pilot Surveys Phase II planned startQ3 2021 Pilot Surveys Phase II planned endQ4 2021 Full Surveys planned startMulti-year surveys are commonly considered legacyprojects at other observatories and are usually done afterthe telescope has been operational for many years. SinceASKAP is conducting large-scale survey projects as oneof its first activities, we devised a sequence of steps totest the telescope’s readiness. See Table 6 for a timelineshowing significant milestones on the road to full surveyoperations. Descriptions of the various phases can befound below.One of the most important lessons from the early sci-ence program was the need to maintain close engagementwith the science community. Taking on the responsibil-ity of providing science-ready data products means thatthe observatory must know exactly what is requiredby its users. We developed a commissioning team thatincluded representatives from the observatory and thesurvey science teams working together.Due to the high data rate, we do not plan to keep allvisibilities long term, although we will archive visibilitiesthat have been calibrated and averaged to 1 MHz fre-quency resolution. We have also found it useful to keepraw calibration data, which accumulates at a more mod-est rate than the science observations. Outputs from theimaging pipeline are tested for quality and reprocessedif necessary, with raw data being deleted to make wayfor the next project once the image products have beenarchived.
The ASKAP early science program started with 12 an-tennas equipped with Mk II PAFs in 2016. We plannedto run two survey projects on behalf of all the survey sci-ence teams, but found that this was not feasible so earlyin the life of the instrument. Instead we devoted timeto smaller test observations for each science team, usingthe data to test ASKAP’s image processing software andidentify problems with the visibility data. These testobservations were separated by commissioning periodsof a few weeks duration, where antennas were added tothe array and features necessary for full operations wereimplemented.During this time, various existing software tools wereused to cross-check and supplement ASKAP’s custom processing software as it was developed. Eventually wewere able to complete a modest continuum survey knownas the cosmology survey, which is available on CASDA.Many publications arose from the early science pro-gram, as a result of identifying projects that the telescopecould excel at with only a third of its collecting area.These include a continuum survey and detection of coldgas outflows from the small magellanic cloud (McClure-Griffiths et al., 2018; Joseph et al., 2019), studies ofneutral hydrogen in nearby galaxy groups (Reynoldset al., 2019; Lee-Waddell et al., 2019; Elagali et al., 2019;Kleiner et al., 2019; For et al., 2019), studies of absorp-tion lines (Glowacki et al., 2019; Allison et al., 2020),continuum observations of the galaxy and mass assem-bly (GAMA) G23 field (Leahy et al., 2019), searches fortransient and variable sources (Bhandari et al., 2018),and a few targeted studies of individual radio galaxies(e.g. Seymour et al., 2020), including the polarisationcharacteristics of Centaurus A (Anderson et al., 2018).ASKAP was also the first telescope to localise a non-repeating fast radio burst to its host galaxy (Bannisteret al., 2019b).
Integration of all 36 antennas with Mk II PAFs anddigital backends was completed in February 2019. Soonafter, we commenced a series of test observations de-signed to verify all the modes required for survey opera-tions with the full array. This involved relatively simpleimaging of representative fields. On 15 July 2019, webegan a pilot survey program to demonstrate readinessfor extended science operations, and to highlight placesof improvement for sustainable full operations. Pilotsurveys involved observing small parts (100 hours perproject) of the larger survey plans developed by thesurvey science teams. The pilot survey concept arose asa natural extension of the early science program. Reach-ing peak operational efficiency is a process that takestime and these pilot surveys provide useful science datawhile improvements and updates to the telescope’s coresystems are ongoing.There are several goals associated with pilot surveys:• Provide representative data to each SST• Test SST survey strategies• Push operational limits to find pressure points• Develop operational plans for full surveys• Assess and improve processing pipelinesObservations for phase I of these pilot surveys werecompleted on 14 May 2020, although processing is ex-pected to take several additional months. Upon com-pletion of pilot surveys phase I observations we entereda 6-month consolidation period, in which the priorityfor telescope access shifted back to development, testingand maintenance. This provided time to consider lessons2
Hotan et al. learned and deploy various software and firmware up-dates, while processing the data backlog. The primarygoal of pilot surveys phase I was to demonstrate pro-duction of science-ready data products. Phase II of pi-lot surveys will focus on optimising commensality andoverall data processing efficiency. We are not currentlyplanning a third phase of pilot surveys, but this may beconsidered if significant issues arise from phase II.Full survey operations are expected to begin by theend of 2021 and ramp up in efficiency by 10 percent eachyear, reaching a limit of 70 percent efficiency in 2024.
The ASKAP control system implements a flexible ob-servation management system built around the conceptof observing procedures and scheduling blocks. Proce-dures are implemented as Python scripts using a libraryof functions that provide access to the hardware. Thesimplest (and most commonly used) procedure simplytracks a source position specified in equatorial coordi-nates. More complicated procedures have been developedfor bandpass calibration and holography observationswhere on-the-fly calculation of offset tracking positionsis required.A scheduling block is specified as a list of parametersand an associated procedure. The list of parameterscurrently includes source coordinates and system config-uration options. Eventually, we plan to expand this listto include data processing parameters, which will allowautomated execution of the image processing pipelineupon completion of a scheduling block. All-sky surveysrequire roughly 1000 individual fields, but we have foundin practice that scheduling each one individually (ratherthan specifying multiple fields in a single block) is themost effective approach, since it provides a level of faulttolerance and allows for more predictable timing.
Over the course of the pilot survey period, variousscheduling requirements have been incorporated to meetthe needs of different survey science cases. Developing aworking model for efficient and effective scheduling is anongoing process that depends on the level of system sta-bility, robustness and automation. For now, scheduling iscarried out in one of two modes: (i) transit-centric, or (ii)target of opportunity. The first approach applies largelyto long tracks, where the requirement is to schedule along observation while the target is above the horizonand pair it automatically with a bandpass calibrationobservation. The second approach is more suitable forshort tracks such as the 15 min observations used for theRapid ASKAP Continuum Survey. It selects an appro-priate target from a specified list based on a series ofconstraints (e.g. above the horizon, solar distance, lunar distance, hour angle, etc.). The scheduling algorithmsare being refined and evolved as we gain experience withthe requirements of survey projects and ensure the sys-tem is able to handle these in a way that is as automatedand efficient as possible.Our current intention is to aim for autonomousscheduling, in that the scheduling tool will detect systemstatus and environment and have access to the pool ofsurvey observations as well as survey constraints andhistorical observational data. It would then use this in-formation to decide on the next observation to scheduleand ensure the telescope is in the right state to completethis observation. The scheduling tool would also pro-vide a forecast of the coming observations if the systemstatus remained the same, but it would maintain theflexibility to adapt as necessary based on the changingsystem and environment. We intend to use the nextperiod of pilot survey observations to prototype thisapproach and determine its feasibility for full surveys,ultimately maximising automation as much as possiblewhich will help us to maximise efficiency gains. Recentdevelopments by the ASKAP team to enable automatedarray start-ups based on the input parset have been akey milestone towards this vision for scheduling, andencourage us that our goal of autonomous schedulingmay be realised for ASKAP.
ASKAP presents an incredible big-data challenge, mark-ing a noticeable shift in the ways astronomical datacan be managed and interacted with. It is quite diffi-cult to gauge data quality in meaningful ways whendata sizes are tens of terabytes, including complexmulti-dimensional axes across time, frequency, anten-nas, beams and polarisations. While working within theconstraints of no longer being able to visualise all data,effort has been invested to ensure that certain aspects ofthe data can be visualised and used for diagnostic pur-poses, particularly to ensure that the system is workingas expected in a timely way.For the raw data, a series of plots have been designedand implemented to allow insight into the recorded dataregardless of their size. Though currently the amount oftime for processing does scale with data size (improvingthis is the subject of ongoing investigation), we are ableto obtain these raw data diagnostic plots automaticallyin close to real time for short observations, which hasbeen important for determining system status. Theseplots are extensions of standard astronomical visualisa-tions of the visibilities, but necessarily more complexdue to the amount of data needing to be represented.The exact form of ASKAP diagnostics is under devel-opment as part of the continuing transition betweencommissioning and operations. The ultimate goal is toassess data quality in an automated way, making use of
SKAP System Description Figure 24.
Example ASKAP raw data diagnostic plots for Scheduling Block IDs 13830 and 14997, demonstrating visualisations ofcorrelation amplitude vs. frequency ( top-left ), correlation amplitude vs. time ( top-right ), and time-frequency waterfall of autocorrelationfor all beams and antennas ( bottom-left ) for SBID 13830. For comparison, we show the same diagnostic plot for SBID 14997. Thesediagnostics give us a direct insight into the recorded data prior to processing, and are used to inform overall data quality assessment. Inthe top two plots, each line represents a baseline colour-coded by baseline length (where red is shortest and purple is longest). In theautocorrelation waterfall plots, orange and red sections indicate beams on particular antennas that may be affected by faulty PAFdominos (in this case, the same bad YY elements are seen on ak12/ak13 for both observations), while the four vertical green stripesacross entire beams trace the less-sensitive beams at the corners of the field of view. Dark red stripes indicate a missing antenna, andany antennas with visually-outlying amplitudes are likely to have scaling issues. The vertical spikes seen in amplitude vs. channel andthe baseline-average waterfall plot are RFI. anomaly detection algorithms and other machine learn-ing techniques where feasible, and this is the subjectof further investigation. Currently, diagnostic plots arevisually inspected where necessary and allow humanpattern recognition to spot anomalous conditions.Some examples of diagnostic plots for a continuumdataset (SBID 13830 and 14997) are shown in Figure 24,demonstrating visualisation of (correlation) amplitudevs. time, amplitude vs. frequency channel, and time- frequency autocorrelation waterfall visualisations thatcombine the two. In particular, the last two plots are anattempt to visualise the contents of the entire datasetat once, showing a time-frequency waterfall plot of auto-correlation amplitude for each antenna and beam withinthe array. This is extremely effective at highlightingparticular hardware issues, but less so for issues thatdepend on cross-correlation baselines. Although in thisvisualisation each sub-plot is too small to be used for4
Hotan et al. quantitative analysis, broad trends or hardware faultsstand out clearly against a background of relatively uni-form behaviour. Conversely, the data used to generatethese visualisations is extremely rich in diagnostic infor-mation, and will likely form a useful input dataset foreither classification or machine-learning algorithms.Extensive quality control is also done post-imaging.Diagnostics are being developed by the science teamsand will be incorporated into the standard observatory-driven processing pipelines over time. CASDA includesa validation stage that must occur prior to data release.Currently this requires human inspection of diagnosticinformation and selection of a three-tier quality grade. Asdiagnostic algorithms improve, we will seek to automatethis process as much as possible.
16 FUTURE ENHANCEMENTS
One of the first upgrades to ASKAP’s systems couldbe a coherent transient detection module that improvesthe sensitivity of fast radio burst searches. This wouldprovide access to an alternative search data stream fromthe correlator rather than the beamformers, increasingthe sensitivity to fast transients by a factor of six andproviding near-instantaneous arcsecond localisation.Another possible improvement is the addition of atied-array module to create phased array beams fromall 36 antennas. This would allow the full array to beused for very long baseline interferometry, pulsar timingexperiments, and the search for extraterrestrial intelli-gence.Satellite interference breaks up ASKAP bands 2 and3 into clean sections smaller than the telescope’s instan-taneous bandwidth, so it could be beneficial to split theobserving band itself into two or more disconnected spec-tral windows. Although this is not currently possible, weare planning to develop such a split-band mode throughsoftware and control system upgrades. This would onlybe possible within one of the band-limiting filters (seeTable 3). Split-band mode would also allow simultaneousobservations of widely-separated spectral lines and mayalso be beneficial for determining spectral indices androtation measures.Another possible improvement would be allowing theexchange of bandwidth for additional PAF beams. Thiswould be particularly beneficial at the high-frequencyend of ASKAP’s frequency range since, above 850 MHz,36 dual-polarisation beams cannot fully sample the30 deg field of view of the PAF (Bunton & Hay, 2010;McConnell, 2017b). At 1700 MHz, the field of view iscurrently limited to 15 deg . Processing more beams maytherefore double the current field of view at 1700 MHz.Additional beams would also reduce the need for inter-leaved observations, but further analysis of the effectsof correlated noise between closely spaced beams maybe needed to understand the exact benefit (Serra et al., 2015b; McConnell, 2017a,b).An additional 48 MHz correlator block could be in-stalled to bring the instantaneous bandwidth to 336 MHz.To achieve the full 384 MHz bandwidth a full complementof Redback modules for the beamformer and correlatorare needed. The additional data rate would be difficultto support on the current supercomputing platform andwould diminish the amount of spare hardware modules,so any such upgrade needs careful consideration. Alter-natively, upgrading the correlator with new hardwarewould release sufficient hardware to expand the beam-former (since the same physical boards are currentlyused in both subsystems). An option for a new correla-tor is the Gemini (Kooistra et al., 2017) board (currentlyin development) which is a good match to the currentRedback modules. An upgraded correlator would allowincreased frequency resolution across the band to 9 . . f /D similar to Parkes, which has been tested successfullywith an ASKAP PAF (Chippendale et al., 2016b; Chip-pendale & Hellbourg, 2017; Deng et al., 2017; Reynoldset al., 2017). With the additional collecting area, theinstantaneous sensitivity of ASKAP increases by 56 %.Survey speed is also increased, but more beams may berequired to compensate for the reduced primary beamsize.Sensitivity could also be improved significantly byupdating the room-temperature LNA design with newtransistors (Shaw & Hay, 2015; Weinreb & Shi, 2019) orby scaling up the manufacturability and affordability ofcryogenic PAF technology like that under developmentfor the Parkes 64 m telescope (Dunning et al., 2016,2019), as mentioned earlier in Section 1.1.
17 CONCLUSION
ASKAP is one of the first radio telescopes to employPAF technology, giving the telescope a wide field ofview, rapid survey speed and excellent polarisation char-acteristics. This has been demonstrated in several pilot
SKAP System Description
18 ACKNOWLEDGEMENTS
The Australian SKA Pathfinder is part of the AustraliaTelescope National Facility which is managed by CSIRO.Operation of ASKAP is funded by the Australian Govern-ment with support from the National Collaborative ResearchInfrastructure Strategy. ASKAP uses the resources of thePawsey Supercomputing Centre. Establishment of ASKAP, the Murchison Radio-astronomy Observatory and the PawseySupercomputing Centre are initiatives of the Australian Gov-ernment, with support from the Government of WesternAustralia and the Science and Industry Endowment Fund.We acknowledge the Wajarri Yamatji as the traditional own-ers of the Observatory site.
REFERENCES
ACMA 2014, Radiocommunications assignment andlicensing instruction MS 32, Coordination of Appa-ratus Licensed Services Within the Mid West RadioQuiet Zone Western Australia,
Abeywickrema S., Allen G., Ardern K., Schinckel A.,Leitch A., Wilson C., Beresford R., 2013, in 2013 Asia-Pacific Symposium on Electromagnetic Compatibility(APEMC). pp 1–4, doi:10.1109/APEMC.2013.7360649Allison J. R., Sadler E. M., Whiting M. T., 2012, PASA,29, 221Allison J. R., et al., 2015, MNRAS, 453, 1249Allison J. R., et al., 2017, MNRAS, 465, 4450Allison J. R., et al., 2020, MNRAS, 494, 3627Anderson C., et al., 2018, Galaxies, 6, 127Applebaum S. P., 1976, IEEE Trans. Antennas Propag.,24, 585Bannister K., Zackay B., Qiu H., James C., ShannonR., 2019a, FREDDA: A fast, real-time engine forde-dispersing amplitudes, astrophysics source codelibrary, record ascl:1906.003
Bannister K. W., et al., 2019b, Science, 365, 565Beresford R., Bunton J., 2013, in 2013 Asia-Pacific Sym-posium on Electromagnetic Compatibility (APEMC).pp 1–4, doi:10.1109/APEMC.2013.7360638Beresford R., Li L., 2013, in 2013 Asia-Pacific Sympo-sium on Electromagnetic Compatibility (APEMC).pp 1–4, doi:10.1109/APEMC.2013.7360637Beresford R., Cheng W., Roberts P., 2017a, in2017 XXXIInd General Assembly and Scien-tific Symposium of the International Unionof Radio Science (URSI GASS). pp 1–4,doi:10.23919/URSIGASS.2017.8105422Beresford R., et al., 2017b, in 2017 XXXIInd GeneralAssembly and Scientific Symposium of the Interna-tional Union of Radio Science (URSI GASS). pp 1–4,doi:10.23919/URSIGASS.2017.8105423Beresford R., Ferris D., Cheng W., Hampson G., BuntonJ., Chippendale A., Kanapathippillai J., 2017c, in 2017International Topical Meeting on Microwave Photon-ics (MWP). pp 1–4, doi:10.1109/MWP.2017.8168641Bhandari S., et al., 2018, MNRAS, 478, 1784Bhatnagar S., Cornwell T. J., Golap K., Uson J. M.,2008, A&A, 487, 4196
Hotan et al.
Bhatnagar S., Rau U., Golap K., 2013, ApJ, 770, 91Black R. A., Jeffs B. D., Warnick K. F., Hellbourg G.,Chippendale A., 2015, in 2015 IEEE Signal Processingand Signal Processing Education Workshop (SP/SPE).pp 261–266, doi:10.1109/DSP-SPE.2015.7369563Bowman J. D., Rogers A. E. E., Monsalve R. A.,Mozdzen T. J., Mahesh N., 2018, Nature, 555, 67Briggs D. S., 1995, PhD thesis, New Mexico Institute ofMining and Technology,
Brown A. J., et al., 2014, in 2014 Interna-tional Conference on Electromagnetics in Ad-vanced Applications (ICEAA). pp 268–271,doi:10.1109/ICEAA.2014.6903860Bunton J. D., 2003a, SKA Memo 40, Figure of meritfor SKA survey speed.
Bunton J. D., 2003b, ALMA Memo 447, Multi-resolutionFX correlator. http://library.nrao.edu/public/memos/alma/main/memo447.pdf
Bunton J. D., Hay S. G., 2010, in 2010 InternationalConference on Electromagnetics in Advanced Applica-tions. pp 728–730, doi:10.1109/ICEAA.2010.5651120CSIRO 2008, Public specification, ASKAP antenna spec-ification and operating parameters. CSIRO,
Chapman J. M., Dempsey J., Miller D., Heywood I.,Pritchard J., Sangster E., Whiting M., Dart M., 2017,in Lorente N. P. F., Shortridge K., Wayth R., eds,Astronomical Society of the Pacific Conference Se-ries Vol. 512, Astronomical Data Analysis Softwareand Systems XXV. pp 73, http://aspbooks.org/custom/publications/paper/512--0073.html
Chippendale A. P., Anderson C., 2019, ACES Memo 19,On-dish calibration of XY phase for ASKAP phasedarray feeds. CSIRO,
Chippendale A. P., Hellbourg G., 2017, in 2017International Conference on Electromagnetics inAdvanced Applications (ICEAA). pp 948–951,doi:10.1109/ICEAA.2017.8065413Chippendale A., Wormnes K., 2013, in 2013 Asia-Pacific Symposium on Electromagnetic Compatibility(APEMC). pp 1–4, doi:10.1109/APEMC.2013.7360640Chippendale A. P., Colegate T. M., O’Sullivan J. D.,2007, SKA Memo 92, SKAcost: a tool for SKAcost and performance estimation.
Chippendale A., O’Sullivan J., Reynolds J., Gough R.,Hayman D., Hay S., 2010, in 2010 IEEE InternationalSymposium on Phased Array Systems and Technology.pp 648–652, doi:10.1109/ARRAY.2010.5613298Chippendale A. P., Hayman D. B., Hay S. G., 2014,PASA, 31, e019 Chippendale A. P., et al., 2015, in 2015 InternationalSymposium on Antennas and Propagation (ISAP).pp 1–4 ( arXiv:1509.05489 )Chippendale A. P., et al., 2016a, in 2016 10th EuropeanConference on Antennas and Propagation (EuCAP).pp 1–5, doi:10.1109/EuCAP.2016.7481741Chippendale A. P., Beresford R. J., Deng X., LeachM., Reynolds J. E., Kramer M., Tzioumis T., 2016b,in 2016 International Conference on Electromagnet-ics in Advanced Applications (ICEAA). pp 909–912,doi:10.1109/ICEAA.2016.7731550Chippendale A. P., Button C., Lourenço L., 2018,ACES Memo 18, Measuring ASKAP’s on-dish cal-ibration signal level and its impact on beam sensitiv-ity. CSIRO,
Cornwell T. J., Golap K., Bhatnagar S., 2005, inShopbell P., Britton M., Ebert R., eds, Astronom-ical Society of the Pacific Conference Series Vol.347, Astronomical Data Analysis Software and Sys-tems XIV. pp 86, http://aspbooks.org/custom/publications/paper/347--0086.html
Cornwell T., Humphreys B., Lenc E., Voronkov M.,Whiting M., Mitchell D., Ord S., Collins D., 2016,Technical report, ASKAP science processing. CSIRO,
DeBoer D. R., et al., 2009, Proc. IEEE, 97, 1507Deller A. T., et al., 2011, PASP, 123, 275Deng X., et al., 2017, PASA, 34, e026Dunning A., Bowen M. A., Hayman D. B., Kanapathip-pillai J., Kanoniuk H., Shaw R. D., Severs S., 2016, in2016 46th European Microwave Conference (EuMC).pp 1568–1571, doi:10.1109/EuMC.2016.7824657Dunning A., et al., 2019, in PAF Workshop 2019.Elagali A., et al., 2019, MNRAS, 487, 2797Elmer M., Jeffs B. D., Warnick K. F., 2014, IEEE Trans.Antennas Propag., 62, 6067Feng Q.-Q., Li Z.-C., Li G.-Y., 2010, in Proc. SPIE. p.782022, doi:10.1117/12.866765For B. Q., et al., 2019, MNRAS, 489, 5723Frost O. L., 1972, Proc. IEEE, 60, 926Glowacki M., et al., 2019, MNRAS, 489, 4926Górski K. M., Hivon E., Banday A. J., Wand elt B. D.,Hansen F. K., Reinecke M., Bartelmann M., 2005,ApJ, 622, 759Gupta N., Johnston S., Feain I., Cornwell T.,2008, Memo 1, The initial array configurationfor ASKAP. CSIRO,
Guzman J. C., Humphreys B., 2010, in Proc. SPIE. p.77401J, doi:10.1117/12.856962Guzman J., et al., 2019, ASKAPsoft: ASKAP sciencedata processor software, astrophysics source code li-
SKAP System Description ascl:1912.003
Hampson G., et al., 2012. pp 807–809,doi:10.1109/ICEAA.2012.6328742Hampson G. A., Brown A., Bunton J. D.,Neuhold S., Chekkala R., Bateman T., TuthillJ., 2014, in Proc. XXXIth URSI GeneralAssembly and Scientific Symposium. pp 1–4,doi:10.1109/URSIGASS.2014.6930062Harvey-Smith L., et al., 2016, MNRAS, 460, 2180Hay S., 2010, International Journal of Microwave andOptical Technology, 5, 375Hay S. G., O’Sullivan J. D., 2008, Radio Science, 43Hay S. G., O’Sullivan J. D., Kot J. S., GranetC., Grancea A., Forsyth A. R., Hayman D. H.,2007, in The Second European Conference on An-tennas and Propagation, EuCAP 2007. pp 1–5,doi:10.1049/ic.2007.0899Hay S. G., O’Sullivan J. D., Mittra R., 2011, IEEE Trans.Antennas Propag., 59, 1828Hayman D. B., Chippendale A., Qiao R., BuntonJ. D., Beresford R. J., Roberts P., Axtens P.,2010, in 2010 International Conference on Electro-magnetics in Advanced Applications. pp 418–421,doi:10.1109/ICEAA.2010.5653177Hellbourg G., Chippendale A. P., Kesteven M. J., JeffsB. D., 2014, in 2014 IEEE Global Conference on Signaland Information Processing (GlobalSIP). pp 1286–1290, doi:10.1109/GlobalSIP.2014.7032330Hellbourg G., Bannister K., Hotan A., 2016, in 2016Radio Frequency Interference (RFI). IEEE ConferenceSeries. pp 37–42, doi:10.1109/RFINT.2016.7833528Heywood I., et al., 2016, MNRAS, 457, 4160Heywood I., et al., 2020, MNRAS, 494, 5018Hobbs G., et al., 2016, MNRAS, 456, 3948Hobbs G., et al., 2020, PASA, 37, e012Hotan A. W., 2016, ACES Memo 11, Holographic Mea-surement of ASKAP Primary Beams. CSIRO,
Hotan A. W., et al., 2014, PASA, 31, e041Hoyle S. A., Mirtschin P. L., 2015, in Proc. 15thInt. Conf. on Accelerator and Large ExperimentalPhysics Control Systems (ICALEPCS’15). pp 660–663,doi:doi:10.18429/JACoW-ICALEPCS2015-WEM301Huynh M., Dempsey J., Whiting M. T., Ophel M., 2020,in Ballester P., Ibsen J., Solar M., Shortridge K.,eds, Astronomical Society of the Pacific ConferenceSeries Vol. 522, Astronomical Data Analysis Softwareand Systems XXVII. pp 263, http://aspbooks.org/custom/publications/paper/522--0263.html
Indermuehle B., Harvey-Smith L., Wilson C., ChowK., 2016, in 2016 Radio Frequency Interference(RFI) workshop. IEEE Conference Series. pp 43–48,doi:10.1109/RFINT.2016.7833529Indermuehle B. T., Harvey-Smith L., Marquarding M., Reynolds J., 2018a, in Proc. SPIE. p. 107041W,doi:10.1117/12.2311926Indermuehle B. T., Harvey-Smith L., MarquardingM., Reynolds J., 2018b, in Proc. SPIE. p. 107042S,doi:10.1117/12.2311917Jeffs B. D., Warnick K. F., Landon J., Waldron J.,Jones D., Fisher J. R., Norrod R. D., 2008, IEEE J.Sel. Topics Signal Process., 2, 635Johnston S., Grey A., 2006, SKA Memo 72, Surveys withthe xNTD and CLAR.
Johnston S., et al., 2007, PASA, 24, 174Joseph T. D., et al., 2019, MNRAS, 490, 1202Kadler M., et al., 2016, Nature Physics, 12, 807Kemball A., Wieringa M., 2000, AIPS++ Memo 229,
MeasurementSet definition version 2.0 . NRAO, https://casa.nrao.edu/Memos/229.html
Kleiner D., et al., 2019, MNRAS, 488, 5352Kooistra E., Hampson G. A., Gunst A. W., Bunton J. D.,Schoonderbeek G. W., Brown A., 2017, in 2017 XXXI-Ind General Assembly and Scientific Symposium of theInternational Union of Radio Science (URSI GASS).pp 1–4, doi:10.23919/URSIGASS.2017.8104976Leahy D. A., et al., 2019, PASA, 36, e024Lee-Waddell K., et al., 2019, MNRAS, 487, 5248Lo Y. T., Lee S. W., Lee Q. H., 1966, Proc. IEEE, 54,1033McClure-Griffiths N. M., et al., 2018, Nature Astronomy,2, 901McConnell D., 2016, ACES Memo 10, Field structureat the focus of paraboloidal reflectors and compari-son with ASKAP beams. CSIRO,
McConnell D., 2017a, ACES Memo 14, EstimatingASAKP beam to beam correlation. CSIRO,
McConnell D., 2017b, ACES Memo 15, Observing withASKAP: Optimisation for survey speed. CSIRO,
McConnell D., et al., 2016, PASA, 33, e042Moss V. A., et al., 2017, MNRAS, 471, 2952Pence W. D., Chiappetti L., Page C. G., Shaw R. A.,Stobie E., 2010, A&A, 524, A42Rau U., 2010, PhD thesis, New Mexico Institute ofMining and Technology,
Rau U., Cornwell T. J., 2011, A&A, 532, A71Reynolds J. E., 1994, Technical Report 39.3/040, Arevised flux scale for the AT compact array. Aus-tralia Telescope National Facility,
Reynolds T. N., Staveley-Smith L., Rhee J., WestmeierT., Chippendale A. P., Deng X., Ekers R. D., Kramer8
Hotan et al.
M., 2017, PASA, 34, e051Reynolds T. N., et al., 2019, MNRAS, 482, 3591Rice S. O., 1982, Proc. IEEE, 70, 692Sarkissian J. M., Reynolds J. E., Hobbs G., Harvey-Smith L., 2017, PASA, 34, e027Sault B., 2014, ACES Memo 2, Initial characterisationof BETA polarimetric response. CSIRO,
Sault B., 2015, ACES Memo 7, Widefield polarimet-ric considerations for ASKAP. CSIRO,
Schinckel A. E., et al., 2011, in Asia-Pacific MicrowaveConference 2011. pp 1178–1181Serra P., et al., 2015a, MNRAS, 448, 1922Serra P., et al., 2015b, MNRAS, 452, 2680Seymour N., et al., 2020, PASA, 37, e013Shaw R. D., Hay S. G., 2015, in 2015 9th EuropeanConference on Antennas and Propagation (EuCAP).pp 1–4, https://ieeexplore.ieee.org/document/7228365
Shaw R. D., Hay S. G., Ranga Y., 2012, in2012 International Conference on Electromag-netics in Advanced Applications. pp 438–441,doi:10.1109/ICEAA.2012.6328666Taylor M. B., 2005, in Shopbell P., Britton M., Ebert R.,eds, Astronomical Society of the Pacific ConferenceSeries Vol. 347, Astronomical Data Analysis Softwareand Systems XIV. p. 29The HDF Group 1997-2020, Hierarchical data format,version 5,
Tingay S. J., et al., 2013, PASA, 30, e007Tuthill J., Hampson G., Bunton J., Brown A., NeuholdS., Bateman T., de Souza L., Joseph J., 2012,in 2012 International Conference on Electromag-netics in Advanced Applications. pp 1067–1070,doi:10.1109/ICEAA.2012.6328788Tuthill J., Hampson G., Bunton J. D., Harris F. J.,Brown A., Ferris R., Bateman T., 2015, in 2015 IEEESignal Processing and Signal Processing EducationWorkshop (SP/SPE). pp 255–260, doi:10.1109/DSP-SPE.2015.7369562Van Veen B. D., Buckley K. M., 1988, IEEE ASSPMagazine, 5, 4Wayth R. B., et al., 2018, PASA, 35Weinreb S., Shi J., 2019, in PAF Workshop 2019. https://events.mpifr-bonn.mpg.de/indico/event/108/session/17/contribution/27
Whiting M. T., 2012, MNRAS, 421, 3242Whiting M. T., 2020, in Ballester P., Ibsen J.,Solar M., Shortridge K., eds, Astronomical So-ciety of the Pacific Conference Series Vol. 522,Astronomical Data Analysis Software and Sys-tems XXVII. p. 469, http://aspbooks.org/custom/publications/paper/522-0469.html
Whiting M., Humphreys B., 2012, PASA, 29, 371Wilson C., Storey M., Tzioumis T., 2013, in 2013 Asia-Pacific Symposium on Electromagnetic Compatibility(APEMC). pp 1–4, doi:10.1109/APEMC.2013.7360651Wilson C., Chow K., Harvey-Smith L., Inder-muehle B., Sokolowski M., Wayth R., 2016. Proc.2016 International Conference on Electromagneticsin Advanced Applications (ICEAA). pp 922–923,doi:10.1109/ICEAA.2016.7731554Wrobel J. M., Walker R. C., 1999, in Taylor G. B., CarilliC. L., Perley R. A., eds, Astronomical Society of thePacific Conference Series Vol. 180, Synthesis Imagingin Radio Astronomy II. p. 171, https://ui.adsabs.harvard.edu/abs/1999ASPC..180..171Whttps://ui.adsabs.harvard.edu/abs/1999ASPC..180..171W